CN117469105A - Blade icing identification method and system based on condition variation self-encoder - Google Patents
Blade icing identification method and system based on condition variation self-encoder Download PDFInfo
- Publication number
- CN117469105A CN117469105A CN202311092110.0A CN202311092110A CN117469105A CN 117469105 A CN117469105 A CN 117469105A CN 202311092110 A CN202311092110 A CN 202311092110A CN 117469105 A CN117469105 A CN 117469105A
- Authority
- CN
- China
- Prior art keywords
- data
- condition
- wind turbine
- working condition
- blade icing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 87
- 230000002159 abnormal effect Effects 0.000 claims abstract description 100
- 238000012549 training Methods 0.000 claims abstract description 48
- 230000005856 abnormality Effects 0.000 claims abstract description 34
- 239000013074 reference sample Substances 0.000 claims abstract description 31
- 238000012216 screening Methods 0.000 claims abstract description 27
- 238000012360 testing method Methods 0.000 claims abstract description 25
- 239000000203 mixture Substances 0.000 claims abstract description 21
- 238000003745 diagnosis Methods 0.000 claims abstract description 10
- 238000005259 measurement Methods 0.000 claims abstract description 10
- 238000009826 distribution Methods 0.000 claims description 50
- 238000001514 detection method Methods 0.000 claims description 36
- 239000000523 sample Substances 0.000 claims description 33
- 230000008859 change Effects 0.000 claims description 27
- 238000004422 calculation algorithm Methods 0.000 claims description 14
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000003860 storage Methods 0.000 claims description 7
- 238000000638 solvent extraction Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 230000001133 acceleration Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 17
- 230000008569 process Effects 0.000 description 15
- 230000008901 benefit Effects 0.000 description 5
- 238000013461 design Methods 0.000 description 5
- 238000007476 Maximum Likelihood Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000002349 favourable effect Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 208000024891 symptom Diseases 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000012614 Monte-Carlo sampling Methods 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011058 failure modes and effects analysis Methods 0.000 description 1
- 230000003862 health status Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000010977 unit operation Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
- F03D—WIND MOTORS
- F03D80/00—Details, components or accessories not provided for in groups F03D1/00 - F03D17/00
- F03D80/40—Ice detection; De-icing means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/72—Wind turbines with rotation axis in wind direction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Combustion & Propulsion (AREA)
- Sustainable Development (AREA)
- Sustainable Energy (AREA)
- Chemical & Material Sciences (AREA)
- Mechanical Engineering (AREA)
- Wind Motors (AREA)
Abstract
The application provides a blade icing identification method and system based on a condition variation self-encoder, wherein the method comprises the following steps: acquiring historical data related to weather conditions, tower vibration conditions, output power characteristics and wind meter measurement results of a region where the wind turbine generator is located; removing abnormal data and unsteady state data from the historical data, and dividing working conditions; clustering the data under each working condition through a Gaussian mixture model, and screening a reference sample from the clustering clusters; training the condition variation self-encoder through a training set to obtain a reference model; and inputting test data of the target wind turbine to be detected into a reference model, positioning abnormal parameters by calculating reconstruction probability, and inputting the abnormal parameters into an abnormal feature knowledge semantic network corresponding to the blade icing mode for diagnosis and reasoning. The method can accurately identify whether the blade has icing abnormality or not, and the identification result has higher accuracy.
Description
Technical Field
The application relates to the technical field of wind power generation, in particular to a blade icing identification method and system based on a condition variation self-encoder.
Background
Along with the improvement of the popularity of the wind turbine, people pay more attention to the health state of the wind turbine so as to ensure the safe and stable production of the wind turbine. Under certain meteorological conditions, the wind turbine blade may be frozen, for example, the blade is frozen in rime and rime. The fan blade icing recognition not only brings the safety production problem, but also can influence the power output of the unit, so that when blade icing abnormality occurs, the abnormality needs to be recognized timely and accurately so as to carry out relevant operation and maintenance.
In the related art, when whether a unit blade is frozen or not is identified, abnormality identification is generally performed by two modes of numerical simulation and numerical simulation. However, the recognition mode in the related art has few considered factors, and in the non-ideal state in the actual application, the recognition result may deviate, so that the accuracy of blade icing recognition is low.
Disclosure of Invention
The present application aims to solve, at least to some extent, one of the technical problems in the related art.
Therefore, a first object of the present application is to provide a method for identifying blade icing based on a condition variable self-encoder, which uses a variable self-encoder model to reconstruct variable working condition data, establishes an anomaly detection model with multi-feature parameter fusion, can accurately locate abnormal parameters caused by blade icing under variable working condition, and performs reasoning on the located abnormal parameters through a related knowledge semantic network to obtain an identification result. Therefore, the accuracy and the reliability of identifying the icing abnormality of the unit blade are improved.
A second object of the present application is to propose a blade icing recognition system based on a condition-variable self-encoder;
a third object of the present application is to propose a non-transitory computer readable storage medium.
To achieve the above object, a first aspect of the present application provides a method for identifying blade icing based on a condition-variable self-encoder, the method comprising the steps of:
aiming at the influence of blade icing on the running of the wind turbine, historical data related to the weather condition of the region where the wind turbine is located, the vibration condition of the tower drum, the output power characteristic and the measurement result of the wind meter are obtained;
eliminating five types of abnormal data and unsteady state data determined according to a preset standard from the historical data, and dividing the historical data according to the working condition characteristic set and an equidistant working condition dividing mode;
based on the similarity of data samples, clustering the data under each working condition through a Gaussian Mixture Model (GMM), screening out a reference sample from one or more kinds of clustering clusters under each working condition, and partitioning a training set from the reference sample;
analyzing the running state of the wind turbine under the variable working condition, constructing a reference model of the wind turbine under the variable working condition based on a condition variation self-encoder, and training the condition variation self-encoder through the training set to obtain the reference model for detecting the blade icing abnormal parameters;
Inputting test data of a target wind turbine to be detected into the reference model, calculating reconstruction probability serving as a characteristic index of blade icing recognition abnormality detection through the reference model, positioning abnormal parameters based on the reconstruction probability, and inputting the abnormal parameters into a pre-constructed abnormal characteristic knowledge semantic network corresponding to a blade icing mode for diagnosis and reasoning so as to identify whether the blade icing abnormality occurs in the target wind turbine.
Optionally, in one embodiment of the present application, the acquired historical data includes: reactive power average value, generating capacity average value, wind speed average value measured by a mechanical wind meter, wind speed average value measured by an ultrasonic wind meter, wind speed average value in a plurality of time periods, hub rotation speed average value, generator rotation speed average value, torque feedback average value, power grid voltage average value, power grid three-phase current average value, power grid line voltage average value, power grid outlet line current average value, tower vibration acceleration average value and active power output quantity of the converter.
Optionally, in an embodiment of the present application, the removing unsteady state data from the historical data includes: the output power of the wind turbine generator is represented by an expression comprising the change rate of power, and the difference value of the output power at two adjacent moments is calculated; estimating the change rate of power by means of the average value of the sample statistics in the time window, and calculating a confidence interval in which the true value of the change rate of power is positioned by adopting an interval estimation method; and under the condition that the confidence interval does not comprise zero, judging that the wind turbine generator is in an unsteady state working condition in the time window, and eliminating corresponding unsteady state data.
Optionally, in an embodiment of the present application, the operating condition feature set includes a plurality of operating condition feature parameters including, but not limited to: wind speed, wind direction, moment of torsion, rotational speed and ambient temperature, based on operating mode feature set, through equidistant operating mode division's mode to the historical data carries out the operating mode and divides, includes: determining the maximum value and the minimum value of each operation condition characteristic parameter in the self variation range, and obtaining the corresponding operation condition dividing interval of each operation condition characteristic parameter; based on the maximum value, the minimum value and the working condition dividing interval corresponding to each operation working condition characteristic parameter, respectively carrying out equal interval working condition dividing on the historical data, and determining each divided working condition in an intersection mode; and removing invalid working conditions from all the divided working conditions, and judging whether the number of the remaining valid working conditions is larger than a preset minimum threshold value.
Optionally, in an embodiment of the present application, the clustering, by the gaussian mixture model GMM, the data under each of the partitioned working conditions includes: estimating parameters of the Gaussian mixture model GMM through a maximum expected value algorithm; determining the number of sub-models of the Gaussian mixture model through a red pool information criterion AIC so as to cluster the state types of the wind turbine generator; the screening the reference sample from one or more kinds of cluster under each working condition comprises the following steps: and comparing the data samples under different clusters, and selecting a group of data with highest average running efficiency as a reference sample.
Optionally, in an embodiment of the present application, the calculating, by the reference model, a reconstruction probability as a feature index of blade icing recognition anomaly detection includes: obtaining a test sample, obtaining a first parameter and a second parameter in Gaussian distribution of each hidden variable through an encoder of the conditional variation self-encoder, and sampling a preset number of sample points for each hidden variable; calculating a third parameter and a fourth parameter in a reconstruction variable likelihood distribution corresponding to each hidden variable through a decoder of the conditional variation self-encoder; based on the third parameter and the fourth parameter, an average value of log likelihood of the test sample under a hidden variable condition is counted.
Optionally, in an embodiment of the present application, constructing the abnormal feature knowledge semantic network corresponding to the blade icing mode includes: combining a plurality of analysis methods to obtain knowledge of fault identification of blade icing of the wind turbine generator; and converting the blade icing fault knowledge into a structured language, and modeling the structured language through a preset open source tool based on the concept of the ontology so as to generate the abnormal feature knowledge semantic network.
To achieve the above object, a second aspect of the present application further proposes a blade icing identification system based on a condition-variable self-encoder, comprising the following modules:
the acquisition module is used for acquiring historical data related to weather conditions, tower vibration conditions, output power characteristics and wind meter measurement results of the region where the wind turbine generator is located aiming at the influence of blade icing on the operation of the wind turbine generator;
the dividing module is used for eliminating five types of abnormal data and unsteady state data determined according to a preset standard from the historical data, and dividing the historical data according to the working condition characteristic set and an equidistant working condition dividing mode;
the screening module is used for clustering the data under each working condition through the Gaussian mixture model GMM based on the similarity of the data samples, screening a reference sample from one or more types of clustering clusters under each working condition, and dividing a training set from the reference sample;
the training module is used for analyzing the running state of the wind turbine generator under the variable working condition, constructing a reference model of the wind turbine generator under the variable working condition based on the condition variation self-encoder, and training the condition variation self-encoder through the training set to obtain the reference model for detecting the blade icing abnormal parameters;
The identification module is used for inputting the test data of the target wind turbine to be detected into the reference model, calculating the reconstruction probability through the reference model to serve as a characteristic index for detecting abnormal blade icing recognition, positioning abnormal parameters based on the reconstruction probability, inputting the abnormal parameters into a pre-constructed abnormal characteristic knowledge semantic network corresponding to the blade icing mode to carry out diagnosis and reasoning so as to identify whether the target wind turbine has abnormal blade icing.
In order to achieve the above embodiments, an embodiment of a third aspect of the present application further proposes a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the blade icing identification method based on the condition-variable self-encoder in the above embodiments.
The technical scheme provided by the embodiment of the application at least brings the following beneficial effects: the method and the device have the advantages that steady state judgment, working condition division and reference sample screening are firstly carried out, invalid abnormal data, unsteady state data and a small amount of tiny abnormal data points mixed in historical data of the wind turbine generator are removed, the cleaned data are used for training a subsequent abnormal detection model, and the accuracy of the abnormal data detected by the trained abnormal detection model can be improved. And then reconstructing variable working condition data by adopting a variable self-encoder model, establishing an anomaly detection model with multi-characteristic parameter fusion, and positioning blade icing anomaly parameters of the wind turbine generator under the variable working condition. Therefore, the method can establish an accurate reference model under the conditions that the working conditions are changed and deviate from the design working conditions, is favorable for timely and accurately detecting abnormal parameters under the variable working conditions, and improves the accuracy of identifying blade icing anomalies of the wind turbine generator under the variable working conditions. And finally, carrying out reasoning on the positioned abnormal parameters through a related knowledge semantic network to acquire a recognition result, judging whether the fan is abnormal due to blade icing or not currently, and determining specific abnormal parameters. Therefore, the method improves the accuracy and reliability of the identification of the icing abnormality of the unit blade, is convenient for eliminating the icing abnormality of the blade in time and ensures the normal operation of the unit.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow chart of a blade icing identification method based on a condition-variable self-encoder according to an embodiment of the present application;
fig. 2 is a flowchart of a method for removing unsteady state data according to an embodiment of the present application;
FIG. 3 is a flowchart of a method for dividing the working conditions of historical data according to an embodiment of the present application;
fig. 4 is a flowchart of a method for screening a reference sample of a wind turbine generator set according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a CVAE model according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of a blade icing identification system based on a condition-variable self-encoder according to an embodiment of the present application.
Detailed Description
Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative and intended to explain the present invention and should not be construed as limiting the invention.
The blade icing identification method and system based on the condition variation self-encoder provided by the embodiment of the invention are described in detail below with reference to the accompanying drawings.
Fig. 1 is a flowchart of a blade icing identification method based on a condition-variable self-encoder according to an embodiment of the present application, as shown in fig. 1, the method includes the following steps:
step S101, aiming at the influence of blade icing on the running of the wind turbine, historical data related to weather conditions of the region where the wind turbine is located, vibration conditions of a tower drum, output power characteristics and wind meter measurement results are obtained.
It should be noted that, the operation state of the wind turbine (may be simply referred to as a fan in the present application) mainly shows three characteristics: pneumatic, mechanical and power characteristics, and when the fan blade is frozen, the three characteristics are affected correspondingly. Therefore, various influences of blade icing on the running of the wind turbine are considered, and relevant fan running history data are collected so as to conduct subsequent abnormal identification.
It can be understood that mass monitoring data can be accumulated in the running process of the wind turbine, and complex mapping rules between the normal running state and the same parameters of the equipment are often included in the monitoring data. With the development of deep learning model technology, the deep learning model is utilized to mine data information, and a solution is provided for solving the modeling of the multivariable coupling relation. Based on the method, the condition variation self-encoder is trained through the collected historical data, blade icing abnormality identification is carried out through a training completion model, and a specific process is described later.
Specifically, the icing can change the surface structure of the blade, so that the aerodynamic performance of the fan is directly reduced, the wind energy utilization rate is reduced, and the power generation of the wind turbine generator is influenced. Meanwhile, the blades are frozen to be irregular, unbalance among the blades of each fan can be caused, and vibration of the tower barrel of the fan tower is aggravated. In addition, when the fan in cold weather is frozen, mechanical anemometers in the unit also increase damping of mechanical equipment due to the fact that the fan is frozen, and accuracy of wind measurement is reduced.
Therefore, when historical data are collected by considering the influence of blade icing on the running of the wind turbine, the historical data are mainly analyzed and obtained from the aspects of subjective weather conditions, tower vibration conditions, wind energy output power conditions, wind meter conditions and the like. Therefore, the condition variation self-encoder is trained through collected historical data related to blade icing, and the trained model can detect blade icing.
In one embodiment of the present application, the collected historical data includes: reactive power average value, generating capacity average value, wind speed average value measured by a mechanical wind meter, wind speed average value measured by an ultrasonic wind meter, wind speed average value in a plurality of time periods, hub rotation speed average value, generator rotation speed average value, torque feedback average value, power grid voltage average value, power grid three-phase current average value, power grid line voltage average value, power grid outlet line current average value, tower vibration acceleration average value and active power output quantity of the converter.
In this embodiment, two different types of anemometers are set to collect wind speed conditions in an environment where a fan is located, and a mean value of data recorded by each anemometer in a certain period or a mean value of historical wind speed data overall is calculated. The wind speed average value in the plurality of time periods may be a wind speed average value in a relatively close time period after the current blade icing identification procedure is started, so as to determine a current wind speed condition, for example, the wind speed average value in the plurality of time periods may include a wind speed average value in 30s and a wind speed average value in 5 min.
In the specific implementation, the historical operation data of the wind turbine can be obtained by reading data detected by each detection device preset on the wind turbine, calling detection data stored in a centralized control system of a wind power station where the wind turbine is located and other modes. And then, screening parameters related to the conversion characteristics, the vibration characteristics and the output power characteristics of the blade from the acquired various data.
Step S102, eliminating five types of abnormal data and unsteady state data determined according to a preset standard from the historical data, and dividing the historical data according to the working condition characteristic set in an equally-spaced working condition dividing mode.
The five types of abnormal data determined according to the preset standard refer to a large number of abnormal points which do not accord with the normal output characteristics of the wind turbine generator set in the measured operation data of the wind turbine generator set due to the reasons of wind abandoning electricity limiting, communication equipment failure, extreme weather, blade pollution, failure of a wind speed sensor and the like in the operation process of the wind turbine generator set. The unsteady state data refer to data generated when the wind turbine generator is in an unsteady state working condition caused by unexpected factors.
Specifically, when abnormal data which does not accord with the output characteristics of the wind turbine generator are removed from the acquired historical operation data. As a possible implementation manner, firstly, determining abnormal data to be removed, specifically referring to industry standards for determining that a wind generating set is in a normal operation condition, and generating operation data in the following five types of scenes: the data in the case where the external conditions other than the wind speed exceed the operating range of the wind turbine, the data during which the failure occurs to cause the wind turbine to be inoperable, the data in the case where the wind turbine is manually shut down or in a test or maintenance operation state, the data during which the test equipment fails or performance is degraded, for example, blade icing and pollution, and the data in the case where the wind direction exceeds a prescribed measurement sector are taken as abnormal data that do not conform to the output characteristics of the wind turbine. And then eliminating all operation data generated in the time period corresponding to the five conditions from the historical operation data.
Therefore, the abnormal data are removed before the abnormal detection of the running state is carried out by the training detection model, the data used for training the subsequent abnormal detection model are generated when the wind turbine generator is in the healthy state and are all data which are collected under the normal running condition of the wind turbine generator and are not damaged, and the accuracy of the abnormal detection result is improved.
Further, unsteady state data are removed from the historical operation data. Specifically, in the actual power generation production process of the wind turbine generator, due to the influence of weather factors such as start and stop of a fan, shielding of obstacles, rapid change of wind speed and the like, a scene in which unstable working conditions and steady working conditions occur alternately may exist when the wind turbine generator is operated, so that a large amount of unstable data is mixed in historical operation data, and stronger consistency between input and output parameters of a system cannot be ensured under the unstable working conditions, so that the unstable data need to be removed before an abnormal operation state detection model is established.
In an embodiment of the present application, in order to more clearly describe a specific implementation process of removing non-steady state data in the present application, an exemplary description is provided below of a method for removing non-steady state data in the embodiment of the present application. Fig. 2 is a flowchart of a method for eliminating unsteady state data according to an embodiment of the present application, as shown in fig. 2, the method includes the following steps:
Step S201, the output power of the wind turbine generator is represented by an expression comprising the rate of change of power, and the difference value of the output power at two adjacent moments is calculated.
It should be noted that, because the output power of the wind turbine generator has the characteristics of small fluctuation in the steady state process and large fluctuation in the unsteady state process, the application selects the output power as the characteristic index of steady state discrimination. When the judgment is specifically performed, the output power of the wind turbine generator is expressed in the form of a formula, and the output power is expressed as an expression containing the change rate of the power as shown in the following formula:
wherein p is t For the measured value of power at time t, μ is the initial true value of power, m is the rate of change of power, ε is the random error of power, and ε obeys the normal distribution.
Further, the difference between the output powers at two adjacent moments is calculated.
Specifically, the difference between the steady-state operation and the unsteady-state operation of the wind turbine generator is whether the load response rate is equal to 0, and the load response rate can be represented by the change rate m of power. In this embodiment, in order to facilitate the estimation of the m value, the difference Δp between the powers of two adjacent moments is calculated, that is, the above formula is subtracted at two adjacent moments to obtain the difference, as shown in the following formula:
Δp＝p t -p t-1 ＝m+(ε t -ε t-1 )
Step S202, estimating the change rate of the power by the mean value of the sample statistic in the time window, and calculating the confidence interval where the true value of the change rate of the power is located by adopting an interval estimation method.
In particular, due to epsilon t ～N(0,σ 2 ) The statistic Δp is expected to be m, i.e., Δp to N (m, 2σ) 2 ). According to the property of the time sequence, m can be estimated by adopting the mean value of the sample statistics in the time window, namely, the m can be estimated by the following formula:
where h is the number of samples in the sliding time window.
Further, in order to ensure the reliability of the estimation, the present embodiment determines, by an interval estimation method, a confidence interval in which the true value of the rate of change of the power (i.e., m) is located.
For example, the present embodiment adopts the interval estimation mode to perform the estimation, namely, the following formula (4) shows:
wherein,and->Is two statistics at a given level of salience α. The confidence level that (1-alpha) exists considers that the true value of m is located in the confidence interval +.>And (3) inner part.
And step S203, under the condition that the confidence interval does not comprise zero, judging that the wind turbine generator is in an unsteady state working condition in a time window, and eliminating corresponding unsteady state data.
Continuing to refer to the interval estimation formula in the step S202, judging whether the wind turbine generator is in an unsteady state working condition in the time windows t-1 to t according to whether the confidence interval does not include zero, namely if the confidence interval If the data does not include 0, judging that the wind turbine generator is in an unsteady state working condition in the time period, wherein the operation data in the time period are unsteady state data, and eliminating the unsteady state data in the time period.
Therefore, by the method for eliminating the unsteady state data, unsteady state data generated under the unsteady state working condition which does not meet strong consistency are eliminated, the complexity of the subsequent modeling and training process according to the historical operation data is reduced, the time and energy consumed by accurate modeling are reduced, and the convenience of the abnormality detection method is improved.
Furthermore, after the data screening processing, the screened historical data is subjected to working condition division in an equally-spaced working condition division mode based on a preset working condition feature set.
The working condition characteristic set comprises a plurality of operating condition characteristic parameters, wherein the operating condition characteristic parameters are operating parameters which directly or indirectly affect the state characteristic parameters of the wind turbine generator, and can be used for distinguishing different working conditions, namely, the operating condition characteristic parameters can be approximately regarded as boundary conditions of working condition division.
In one embodiment of the present application, the constructed working condition feature set may include five operating condition feature parameters selected by comparing the influence degree of each parameter in the historical data on the state feature of the unit, including wind speed, wind direction, torque, rotation speed and ambient temperature. In order to more clearly illustrate a specific implementation process of the working condition division in the present application, an exemplary description is provided below with a working condition division method in an embodiment of the present application. Fig. 3 is a flowchart of a method for dividing working conditions of historical data according to an embodiment of the present application, as shown in fig. 3, the method includes the following steps:
Step S301, determining the maximum value and the minimum value of each operation condition characteristic parameter in the self-variation range, and obtaining the condition dividing interval corresponding to each operation condition characteristic parameter.
Specifically, traversing the screened historical operation data, comparing and determining the maximum value and the minimum value of each operation condition characteristic parameter in the self-variation range, and then acquiring the preset condition dividing interval of each operation condition characteristic parameter. The working condition dividing interval of the characteristic parameters can be set according to the characteristics of the current wind turbine generator, the actual transformation range and data distribution characteristics of each parameter and the accuracy requirement of anomaly detection, for example, when the accuracy requirement is higher, the working condition dividing interval is finer, and the working condition dividing interval of the characteristic parameters of each operation working condition can be set to be shorter.
The working condition dividing intervals corresponding to the characteristic parameters of each operation working condition can be the same or different.
Step S302, based on the maximum value, the minimum value and the working condition dividing interval corresponding to each operation working condition characteristic parameter, equally-spaced working condition dividing is carried out on the historical data respectively, and the divided working conditions are determined in an intersection mode.
Specifically, after determining the interval of each characteristic parameter division, the following formula is used for equally-spaced working condition division:
C i ＝(S i ，S i +ΔS)∩(D i ，D i +ΔD)∩(M i ，M i +ΔM)∩(N i ，N i +ΔN)∩(T i ，T i +ΔT)
s.t.S i ＜＜S min ，S i +ΔS＞＞S max ，D i ＜＜D min ，D i +ΔD＞＞D max ，M i ＜＜M min ，M i +ΔM＞＞M max ，
N i ＜＜N min ，N i +ΔN＞＞N max ，T i ＜＜T min ，T i +ΔT＞＞T max
wherein S is wind speed, D is wind direction, M is torque, N is rotational speed, and T is ambient temperature. From the formula and the description of the characteristic parameters, T min And T max Respectively represent the minimum and maximum values of the ambient temperature, P min And P max The minimum value and the maximum value of the load are respectively represented, the delta T and the delta P respectively represent the environment temperature and the working condition dividing interval of the load, and other characteristic parameters can be similar, and are not repeated here.
As can be seen from the working condition dividing formula, the ith working condition C is divided by the application i The method is characterized in that the method is determined by the intersection of the dividing conditions of the characteristic parameters of each operation condition, and the corresponding operation condition is divided according to the intersection of the characteristic parameters of each operation condition in the respective dividing interval.
And step S303, removing invalid working conditions from all the divided working conditions, and judging whether the number of the remaining valid working conditions is larger than a preset minimum threshold value.
It should be noted that, because the working condition division based on the equal interval often occurs a part of invalid working conditions with no samples or fewer samples, in this embodiment, after the working condition division is performed at equal intervals, part of invalid working conditions are also removed.
Specifically, the number of samples in each working condition after division is detected, if the number of samples is lower than a preset threshold value, the working condition is judged to be an invalid working condition, the invalid working condition is removed from all the working conditions after division until the number of remaining valid working conditions is greater than a preset proportion of the total number of theoretical working conditions, for example, when the number of remaining valid working conditions is greater than 30% of the total number of theoretical working conditions, the collected data samples are judged to be meaningful relative to the current working condition division result, and thus the working condition division is completed.
In one embodiment of the present application, if it is determined that the number of remaining effective working conditions is less than the preset minimum threshold, the process returns to step S101, and the collection and screening of the historical data are performed again. The method can acquire more abundant historical data by more various data acquisition modes and prolonging the time period aimed at by data acquisition, and adjust the mode of eliminating the data.
Step S103, based on the similarity of the data samples, clustering the data under each working condition through a Gaussian Mixture Model (GMM), screening out a reference sample from one or more kinds of clustering clusters under each working condition, and partitioning a training set from the reference sample.
Wherein the gaussian mixture model (Gaussian Mixture Model, GMM for short) is a linear combination of a plurality of gaussian distribution functions, and things can be precisely quantized by gaussian probability density functions.
Specifically, a reference sample is screened from historical operation data after working conditions are divided through a Gaussian mixture model. The reference sample is a data sample of the wind turbine generator in a normal state or a healthy running state. However, due to the influence of external conditions, boundaries of the normal running state are changed in different working condition scenes, and it is difficult to uniformly define the scale of the normal running state. Therefore, the working condition division is carried out, the characteristics of things can be accurately quantized by means of the Gaussian mixture model GMM, the data samples under the single working condition are clustered according to the similarity by the GMM, and data in a certain class or a plurality of classes of clustering clusters are selected as reference samples according to a pre-defined screening standard.
The GMM combines the advantages of a parameter estimation method and a non-parameter estimation method, is not limited to a specific probability density function form, and can approximate any continuous distribution with any precision under the condition that the number of submodels is enough, so that the method and the device determine a plurality of different distributions contained under a historical operation data set through the GMM, namely, generate different clustering clusters.
In particular, in one embodiment of the present application, in order to more clearly illustrate a specific implementation process of screening a reference sample by using GMM in the present application, an exemplary method for screening a reference sample is described below. Fig. 4 is a flowchart of a method for screening a reference sample of a wind turbine generator set according to an embodiment of the present application, where for each working condition that is defined above, the method may be used to screen the reference sample. As shown in fig. 4, the method comprises the steps of:
In step S401, parameters of the gaussian mixture model GMM are estimated by a maximum expectation algorithm.
Specifically, the maximum Expectation-Maximization algorithm (EM) algorithm is an iterative algorithm, the EM algorithm can be used to estimate parameters of a probability model containing hidden variables, and the maximum likelihood parameter value is obtained through iteration, so that the objective function of the EM algorithm adopted in the application is as follows:
in this embodiment, when the GMM parameter is estimated by the EM algorithm, the coarse value of the GMM parameter is estimated first, and then the likelihood function is maximized using the obtained coarse value.
For example, in the embodiment of the present application, the likelihood function of the GMM is calculated first, and if N parameters in the GMM model need to be estimated, the maximum likelihood functions of the N parameters are solved first, then the GMM parameters are estimated by the EM algorithm, that is, the maximum likelihood functions of the N parameters are maximized, specifically, the initial values of the N parameters may be designated first, and iteration is performed by the EM algorithm until the N parameters or the log likelihood functions converge, so as to obtain estimated parameter values.
And step S402, determining the number of submodels of the Gaussian mixture model through a red pool information criterion AIC so as to cluster the state category of the target wind turbine.
The physical meaning corresponding to the number of the GMM submodels is the number of the state categories of the wind turbine generator. When the number of the GMM submodels is 1, the GMM submodels are equivalent to Gaussian distribution, and the change of the operation level of the wind turbine generator set in the application, faults in the operation process and slow aging of the wind turbine generator set under normal conditions are considered, wherein a historical data sample possibly comprises a plurality of operation modes, and the probability distribution is expressed as superposition combination of a plurality of Gaussian distributions.
Specifically, in order to determine the number of GMM submodels, AIC evaluation criteria are selected in the embodiment of the present application. The red pool information criterion (Akaike's Information Criterion, AIC for short) is an evaluation method based on the concept of entropy and used for weighing complexity of a model and fine fitting data, and is defined as the following formula:
AIC＝2K-2ln(L)
where K is the number of sub-models and ln (L) represents the log-likelihood function of the model.
It should be noted that as the number of submodels increases, the complexity of the model increases, and the AIC value decreases and increases. Therefore, the embodiment of the application selects the number of sub-models corresponding to the minimum AIC as the optimal number of sub-models of the GMM. Therefore, the embodiment of the application clusters the data samples under the single working condition according to the similarity by estimating the parameters of the GMM and determining the number of the submodels by combining with the AIC criterion. And further, the data samples under each working condition after division can be clustered in sequence through the GMM.
Step S403, screening out a reference sample from one or more kinds of clusters under each working condition according to a preset screening standard.
In the embodiment of the present application, the reference is defined as a parameter value corresponding to the highest unit operation efficiency that can be actually achieved under the current operation boundary condition, so that the preset screening standard in the present application may be to select the data with the highest operation efficiency as the reference sample. When the reference samples are specifically screened, the data samples under different clustering clusters determined by the GMM can be compared according to each working condition, a group of data with highest average operation efficiency is selected as the reference samples, probability density distribution of the data can be further estimated, and a reference interval is determined. And then the standard samples under each working condition after division are sequentially screened out in the same way.
Further, summarizing the reference samples under each working condition obtained in the manner in the above embodiment, obtaining reference sample data, and dividing the obtained reference samples into a training set and a test set according to a certain proportion, for example, dividing the summarized reference samples into the training set and the test set according to a proportion of 8:2, where the training set is used for subsequent training of the condition variable self-encoder, and the test set is used for testing and evaluating the trained condition variable self-encoder model.
And step S104, analyzing the running state of the wind turbine under the variable working condition, constructing a reference model of the wind turbine under the variable working condition based on the condition variable self-encoder, and training the condition variable self-encoder through a training set to obtain the reference model for detecting the blade icing abnormal parameters.
The variable working condition refers to the condition that the wind turbine generator is in a working condition change or in an operation scene deviating from a design working condition. According to the method, the running state of the wind turbine generator under the variable working condition is analyzed, and the mode of building the reference model under the variable working condition is determined.
Specifically, as most of the running state characteristic indexes of the wind turbine generator are related to boundary conditions and have coupling relations with each other, the wind turbine generator can monitor the change of running parameters, and the change of the running parameters is not only related to the health condition of the wind turbine generator, but also influenced by external working conditions. Therefore, in one embodiment of the application, when analyzing the variable working condition operation state of the wind turbine generator, the relation between each characteristic index and boundary variable of the wind turbine generator under the variable working condition is constructed by analyzing the historical operation data. The characteristic indexes can comprise primary parameters, secondary performance indexes, external working condition variables and unit state variables which can be directly monitored. The primary parameters can be directly obtained through sensor measuring points arranged at each position of the wind turbine generator, and the method comprises the following steps: temperature, wind speed, vibration, etc. The secondary performance index is needed to be obtained through parameter soft measurement, is more sensitive to the change of the state of the unit, and can rapidly locate the position where the abnormality occurs. External operating condition variables, including: environmental parameters and rotational speed of the unit. The unit state variables reflect the health status and original design capabilities of the unit itself.
Furthermore, the running state model of the wind turbine generator under the variable working condition is expressed as a logp in the form of conditional probability distribution θ (X, y|c) indicating that the values of the primary parameter X and the secondary performance index Y are predicted given the operating variable C. After determining the characteristic index to be monitored, using the parameters of the screened data sample identification probability model to select representative maximum likelihood estimation solution, wherein the method specifically comprises the following steps:
wherein N represents the number of training samples, and x is used in the application for simplicity of expression without causing ambiguity (i) Denoted as x. Because the operation parameters and performance indexes are also influenced by the unit state variables Z except the working condition variables, the probability model not only contains the observation variables, but also contains the hidden variables, and the logp is directly solved θ (x, y|c) is difficult and is not suitable for learning conditional probability distributions directly from data. Therefore, according to the Bayesian theorem, the probability distribution of the observable variable is realized by the conditional probability distribution of the hidden variable, namely, the logp θ (x, y|c) can be expressed in a form as shown in the following formula:
logp θ (x,y|c)＝logp θ (x,y|z,c)+logp θ (z|c)-logp θ (z|x,y,c)
in combination with the formula, when the characteristic indexes of the units are coupled with each other, complex nonlinear relation exists between the variables, and the posterior probability logp of the hidden variable z θ (zI x, y, c) is not a simple probability distribution. Especially when the conditional probability logp in the formula θ When the parameters in (z|x, y, c) are calculated by neural network, the EM algorithm is no longer applicable. Therefore, the method adopts the improved condition variable self-encoder to establish the reference model under the variable working condition of the unit.
Therefore, the method and the device determine that the self-encoder based on the condition variation is more suitable for constructing the reference model under the variable working condition by analyzing the running state of the wind turbine under the variable working condition. For more clear explanation of the principle that the improved condition variable self-encoder is adopted to establish a reference model under the variable working condition of the unit, the variable self-encoder is described below:
a Variational self-Encoder (VAE) is a deep learning generation model, integrates the advantage of deep learning on the basis of a probability model, and can autonomously learn probability distribution obeyed by data and generate similar data. The network architecture of the VAE is divided into two parts, encoder and decoder.
Wherein the encoder, also called inference network, functions by p θ And (z|x) mapping the original data x to a low-dimensional hidden space, and encoding the original data x into hidden variables z to realize feature extraction and dimension reduction. The decoder, also called the generation network, functions to distribute p from hidden variables θ (z) sample z, through p θ (x|z) reconstruct the original data x.
Decoder p θ (x|z) integration of a multi-layer neural network, which can improve the expressive power on complex nonlinear relationships, but the posterior probability p of the hidden variable z θ (z|x) is thus more complex and cannot obtain a resolved form of the distribution. The VAE thus constructs an optimizable q using variational reasoning φ (z|x) is used to approximately represent the true posterior probability p in the encoder θ (z|x). The observation variable x log likelihood can be written as shown in the following formula:
logp θ (x)＝D KL (q φ (z|x)||p θ (z|x))+L ELBO (θ,φ；x)
wherein D is KL Representing KL divergence for measuring similarity of two probability distributions in the same event spaceThe value is constantly non-negative. The problem of maximum log likelihood estimation of the observation variable x is therefore equivalent to maximizing the variation lower bound L ELBO (θ, φ; x) as shown in the following formula.
logp θ (x)≥L ELBO (θ,φ；x)
The condition that the equation equal sign holds is that the variation approximation posterior probability is equal to the true posterior probability. L (L) ELBO (θ, φ; x) can continue to expand as shown in the following equation.
The formula contains two items: the first term is the reconstruction error of the observed variable x by approximating the posterior distribution q φ (z|x) sampling the hidden variable z and calculating the log likelihood logp θ (x|z) for describing the difference between the generated sample and the real sample; the second term is the approximate posterior distribution q of the hidden variable φ (z|x) a priori distribution p θ KL dispersion between (z) constrains the posterior distribution to be close to the a priori distribution, which can be seen as a hidden variable z regularization coefficient. The VAE adopts SGVB training algorithm to optimally solve the parameters phi and theta, and simultaneously meets the following two conditions: firstly, training data and reconstruction data are made to be as equal as possible; second, let z be a posterior distribution q φ (z|x) approximates the a priori distribution p θ (z)。
VAE generally assumes a posterior distribution q of z φ (z|x) obeys a Gaussian distribution, a priori distribution p θ (z) obeying a standard normal distribution, likelihood p θ (x|z) a multivariate gaussian distribution or bernoulli distribution is selected based on the characteristics of the variables. The parameter estimation of the model directly using Monte Carlo sampling generates a large variance, and in order to reduce the variance, the hidden variable is represented by a method of heavy parameter transformation, which is represented by a micro equation and a random variable, as shown in the following formula.
z＝g(x，y，c，ε) with ε～p(ε)
The probability of first pass satisfies p θ (z)＝N(0,I)，q φ (z|x)＝N(z；μ,σ 2 I) Under the condition, the expression of the lower bound of the variationThe equivalent is the following formula:
in the embodiment of the application, because the running state model of the wind turbine generator is the conditional probability distribution lovp of the characteristic indexes X and Y under the condition of the working condition variable C θ (x, y|c) monitoring. Therefore, the method introduces a condition variable to improve the original network structure based on the structure of the standard VAE, and establishes a condition variable self-Encoder (CVAE) model for predicting the reference data under the variable working condition.
In particular, when constructing the reference model based on the condition variation self-encoder, as a possible implementation manner, the embodiment of the application constructs a network structure of the CVAE shown in fig. 5. As shown in fig. 5, the model includes an encoding module 1, a decoding module 2, and an acquisition and transmission module 3 for the operating condition (C). Wherein the objective of CVAE is to solve the log-likelihood of the condition log-gp θ (x, y|c) value of the parameter at maximum. Under the variable working condition, firstly, the lovp is firstly used for θ (x) Extended to conditional log likelihood logp θ (x, y|c), written as the following formula:
logp θ (x,y|c)＝D KL (q φ (z|x,y,c)||p θ (z|x,y,c))+L ELBO (θ,φ；x,y,c)
then, according to the basic principle of VAE, the variation lower boundary L of the conditional log likelihood is calculated ELBO (θ, φ; x, y, c) expands to the following expression:
in this embodiment, assuming that the prior distribution of hidden variables obeys an isotropic multivariate gaussian, the prior probability does not involve parameters, as shown in the following formula:
p(z)＝N(0,I)
conditional likelihood p of observed variable θ (x, y|z, c) obeys a multivariate gaussian distribution as shown in the following equation:
wherein the parameters are derived from the output of the multi-layer neural network in the encoder, σ θ (z, c) is assumed to be constant, μ θ (z,c)＝W θ [z,c]+b θ ，σ θ (z,c)＝const。
Due to the true posterior probability p θ The form of (z|x, y, c) is a multi-element Gaussian model with diagonal variance, and in order to simplify the estimation of the posterior probability, the encoder distribution q is represented by using a multi-element Gaussian distribution with diagonal variance structure φ (z|x, y, c) as shown in the following formula.
q φ (z|x,y,c)＝N(μ φ (x,y,c),σ φ (x,y,c) 2 I)
The parameterized variance approximates the distribution as shown in the following equation:
z＝μ φ (x，y，x)+σ φ (x，y，c)⊙ε with ε～N(0，I)
the parameters can be calculated by the neural network as well, and the calculation mode is shown in the following formula:
therefore, the reference model of the target wind turbine generator set based on the condition variation self-encoder under the variable working condition is built, then the CVAE model is trained through sample data in the training set obtained in the step S103, and the reference model for detecting blade icing of the wind turbine generator set under the variable working condition can be obtained after the training is completed.
The specific process of training the CVAE model may refer to the process of training the deep learning network model in the related art, and the implementation principle is similar, which is not limited herein.
Step 105, inputting test data of a target wind turbine to be detected into a reference model, calculating reconstruction probability through the reference model to serve as a characteristic index of blade icing recognition abnormality detection, positioning abnormal parameters based on the reconstruction probability, and inputting the abnormal parameters into a pre-constructed abnormal characteristic knowledge semantic network corresponding to a blade icing mode to perform diagnosis and reasoning so as to identify whether the target wind turbine has blade icing abnormality.
Wherein the reconstruction probability is a reconstruction error term in the loss function of the CVAE model. It should be noted that, the reconstructed probability distribution of the variable is not the variable itself, the reconstructed data includes more information of the variable, including posterior distribution and generation likelihood of the hidden variable, for the CVAE reference model, the smaller the reconstructed probability of the test data, the larger the difference between the data and the reference working condition is, so the application selects the reconstructed probability as the characteristic index of anomaly detection.
Specifically, the reconstruction probability is calculated through the reference model which is completed through training, and is used as the characteristic index of anomaly detection. As a possible implementation manner, when calculating the reconstruction probability, a test sample is obtained first, a first parameter and a second parameter in gaussian distribution of each hidden variable are obtained through an encoder of a conditional variation self-encoder, a preset number of sample points are sampled for each hidden variable, a third parameter and a fourth parameter in reconstruction variable likelihood distribution corresponding to each hidden variable are calculated through a decoder of the conditional variation self-encoder, and then an average value of log likelihood of the test sample under the condition of the hidden variable is counted based on the third parameter and the fourth parameter.
Specifically, the test sample may be sample data for calculating a reconstruction probability, which is selected from historical operation data of the current target wind turbine to be detected. After inputting the test sample X to the CVAE reference model, obtaining a first parameter mu in the Gaussian distribution of the hidden variable through an encoder of the CVAE φ And a second parameter sigma φ And samples L points for the hidden variable Z. Then, a third parameter mu of the likelihood distribution of the reconstruction variable corresponding to each hidden variable Z is calculated by a decoder θ And a fourth parameter sigma θ Finally, calculating average value of log-generation likelihood of test sample under hidden variable condition to obtain reconstruction probability, and specific calculation methodThe formula may be represented by the following formula:
further, after a reference model under a variable working condition is generated and the reconstruction probability is calculated, real-time operation data of the target wind turbine to be detected are obtained, differences between the current data and the reference data are compared, and abnormality judgment is carried out.
As a possible implementation manner, the abnormal operation data is positioned according to the reconstruction probability and the deviation degree between the actual value and the reconstruction value of the real-time operation data, and the method comprises the following steps of estimating the reconstruction probability distribution of the KDE statistical training sample through the kernel density, and taking the lower limit of the corresponding confidence interval when the confidence coefficient is equal to a preset value as the threshold value of abnormality detection. Then, judging whether the reconstruction probability of the real-time operation data is abnormal according to the threshold value of the abnormality detection, and calculating the deviation degree of each parameter in the real-time operation data by the following formula under the condition that the reconstruction probability is abnormal:
Wherein v is k Representing the normalization coefficient, x k Representing the actual value of the parameter k,representing the reconstructed value of parameter k. And finally, comparing the deviation degree of each parameter, and determining the parameter generating the abnormality.
Specifically, a kernel density estimation (kernel density estimation, abbreviated as KDE) is adopted to count a reconstruction probability distribution for training samples, and a lower limit of a corresponding confidence interval when the confidence is equal to 95% is used as a threshold value for anomaly detection, wherein a specific calculation formula is as follows:
the training sample may be sample data from which a threshold value for calculating abnormal detection is extracted from the acquired real-time operation data. In the wind turbine generator, parameters such as wind speed, rotating speed and the like are mutually coupled, disturbance of local parameters can be transmitted to other parameters in the system, and tiny deviation of early anomalies of certain parameters can be pulled to a normal level under the control and regulation effects of the system, so that positioning of parameter anomalies is affected.
Therefore, the abnormal change of the unit local system unit can be found in advance by adopting the abnormal detection based on the reconstruction probability. On the basis, a difference vector between the reconstruction output and the original input parameters is calculated, the relative deviation of the reconstruction output and the original input parameters is used for further positioning the abnormality of the parameters, and the calculation formula of the parameter deviation degree is shown as follows:
Wherein v is k The normalized coefficient is represented and is obtained by reconstructing absolute deviation statistics between output and input signals of the training samples. X is x k Andrepresenting the actual and reconstructed values of the parameter k, respectively.
In this embodiment, the reconstruction probability distribution of the KDE statistical training sample is estimated through the kernel density, after the threshold value of anomaly detection is obtained, the reconstruction probability of the real-time operation data is calculated, the reconstruction probability of the real-time operation data is compared with the threshold value of anomaly detection, and once the reconstruction probability of the real-time operation data is detected to be abnormal, the current abnormal state of the target wind turbine generator is determined. Further, the deviation degree of each operation parameter in the operation data, such as the fan torque, is calculated through the formula, and the abnormality of the parameters is further positioned through comparing the deviation degree of each parameter in the system. For example, the deviation of each operation parameter is compared with a preset deviation threshold, and when the deviation of a certain operation parameter is greater than the preset deviation threshold, the operation parameter is determined to be abnormal.
Furthermore, the detected abnormal parameters are input into a pre-constructed abnormal feature knowledge semantic network corresponding to the blade icing mode to carry out diagnosis and reasoning so as to identify whether the blade icing abnormality occurs in the target wind turbine generator.
In one embodiment of the present application, when the knowledge semantic network is pre-built before the anomaly identification is performed, the method may be implemented by the following steps: firstly, combining a plurality of analysis methods to obtain knowledge of fault identification of blade icing of a wind turbine generator; and then converting the blade icing fault knowledge into a structured language, and modeling the structured language through a preset open source tool based on the concept of the ontology so as to generate an abnormal feature knowledge semantic network.
Specifically, in this embodiment, a fault diagnosis knowledge base for blade icing identification is constructed based on ontology concepts, abnormal parameters are input into an abnormal feature knowledge semantic network to be inferred, after judgment logic is satisfied, it is determined that blade icing abnormality occurs in a wind turbine generator system currently, relevant staff is reminded of the blade icing abnormality, and the abnormality is timely checked.
In specific implementation, expert knowledge of blade icing of a fan can be obtained through a plurality of analysis methods such as fault mode and influence analysis FMEA and fault tree analysis FTA, and an ontology in the field of blade icing is constructed by combining the obtained knowledge, so that an ontology knowledge base of blade icing is generated. For example, the related fault knowledge of the unstructured photovoltaic generator set is expressed through a computer recognizable structured language, and then a model is built by using a source opening tool prot, so that a blade icing diagnosis knowledge base based on a body, namely an abnormal characteristic knowledge semantic network, is formed.
Further, after the abnormal parameters are input into the abnormal feature knowledge semantic network, determining fault symptoms according to the abnormal parameters, and then inquiring whether the fault symptoms belong to a blade icing abnormal mode in the abnormal feature knowledge semantic network through SPARQL inquiry sentences, if so, judging that the target wind turbine generator has blade icing abnormality currently.
Furthermore, the blade icing abnormality alarm information can be sent to the mobile terminal of the relevant staff through wireless communication connection, so that the staff can know the blade icing abnormality in time and perform relevant removal operation.
Therefore, whether the blade is frozen or not can be accurately identified, the reliability of the identification result is high, and the worker can timely remove the blade from being frozen.
In summary, according to the blade icing identification method based on the condition variation self-encoder, steady state judgment, working condition division and reference sample screening are performed first, invalid abnormal data, unsteady state data and a small amount of tiny abnormal data points mixed in historical data of a wind turbine generator are removed, the cleaned data are used for training a subsequent abnormal detection model, and accuracy of abnormal data detection of the trained abnormal detection model can be improved. And then reconstructing variable working condition data by adopting a variable self-encoder model, establishing an anomaly detection model with multi-characteristic parameter fusion, and positioning blade icing anomaly parameters of the wind turbine generator under the variable working condition. Therefore, the method can establish an accurate reference model under the conditions that the working conditions are changed and deviate from the design working conditions, is favorable for timely and accurately detecting abnormal parameters under the variable working conditions, and improves the accuracy of identifying blade icing anomalies of the wind turbine generator under the variable working conditions. And finally, carrying out reasoning on the positioned abnormal parameters through a related knowledge semantic network to acquire a recognition result, judging whether the fan is abnormal due to blade icing or not currently, and determining specific abnormal parameters. Therefore, the method improves the accuracy and reliability of the identification of the icing abnormality of the unit blade, is convenient for eliminating the icing abnormality of the blade in time, and ensures the normal operation of the unit.
In order to implement the above embodiment, the present application further provides a blade icing identifying system based on a condition-variable self-encoder, and fig. 6 is a schematic structural diagram of the blade icing identifying system based on the condition-variable self-encoder according to the embodiment of the present application, as shown in fig. 6, where the system includes an obtaining module 100, a dividing module 200, a screening module 300, a training module 400, and an identifying module 500.
The acquiring module 100 is configured to acquire, for an influence of blade icing on operation of the wind turbine, historical data related to weather conditions of an area where the wind turbine is located, vibration conditions of a tower, output power characteristics and measurement results of a wind meter.
The dividing module 200 is configured to remove five types of abnormal data and unsteady state data determined according to a preset standard from the historical data, and divide the historical data according to an equally-spaced working condition dividing manner based on a working condition feature set.
The screening module 300 is configured to cluster the data under each of the partitioned working conditions by using the gaussian mixture model GMM based on the similarity of the data samples, screen a reference sample from one or more kinds of clusters under each of the working conditions, and partition a training set from the reference sample.
The training module 400 is used for analyzing the running state of the wind turbine generator under the variable working condition, constructing a reference model of the wind turbine generator under the variable working condition based on the condition variation self-encoder, and training the condition variation self-encoder through a training set to obtain the reference model for detecting the blade icing abnormal parameters.
The identifying module 500 is configured to input test data of a target wind turbine generator to be detected to a reference model, calculate a reconstruction probability through the reference model as a feature index of blade icing recognition anomaly detection, locate an anomaly parameter based on the reconstruction probability, and input the anomaly parameter to a pre-constructed anomaly feature knowledge semantic network corresponding to a blade icing mode for diagnosis and reasoning so as to identify whether the target wind turbine generator has blade icing anomalies.
Optionally, in one embodiment of the present application, the partitioning module 200 is specifically configured to: the output power of the wind turbine generator is represented by an expression containing the change rate of power, and the difference value of the output power at two adjacent moments is calculated; estimating the change rate of the power by the mean value of the sample statistics in the time window, and calculating the confidence interval where the true value of the change rate of the power is located by adopting an interval estimation method; and under the condition that the confidence interval does not comprise zero, judging that the wind turbine generator is in an unsteady state working condition in a time window, and eliminating corresponding unsteady state data.
Optionally, in one embodiment of the present application, the operating condition feature set comprises a plurality of operating condition feature parameters including, but not limited to: wind speed, wind direction, torque, rotational speed, and ambient temperature, the dividing module 200 is specifically configured to: determining the maximum value and the minimum value of each operation condition characteristic parameter in the self variation range, and obtaining the corresponding condition dividing interval of each operation condition characteristic parameter; based on the maximum value, the minimum value and the working condition dividing interval corresponding to each operation working condition characteristic parameter, respectively dividing the historical data at equal intervals, and determining each divided working condition in an intersection mode; and removing invalid working conditions from all the divided working conditions, and judging whether the number of the remaining valid working conditions is larger than a preset minimum threshold value.
Optionally, in one embodiment of the present application, the screening module 300 is specifically configured to: estimating parameters of the Gaussian mixture model GMM through a maximum expected value algorithm; determining the number of submodels of the Gaussian mixture model through a red pool information criterion AIC so as to cluster the state types of the wind turbine generator; and comparing the data samples under different clusters, and selecting a group of data with highest average running efficiency as a reference sample.
Optionally, in one embodiment of the present application, the identification module 500 is specifically configured to: acquiring a test sample, acquiring a first parameter and a second parameter in Gaussian distribution of each hidden variable through an encoder of a conditional variation self-encoder, and sampling a preset number of sample points for each hidden variable; calculating a third parameter and a fourth parameter in a reconstruction variable likelihood distribution corresponding to each hidden variable through a decoder of a conditional variation self-encoder; based on the third parameter and the fourth parameter, an average value of log likelihood of the test sample under the condition of hidden variables is counted.
Optionally, in one embodiment of the present application, the identification module 500 is further configured to: combining a plurality of analysis methods to obtain knowledge of fault identification of blade icing of the wind turbine generator; the blade icing fault knowledge is converted into a structured language, and the structured language is modeled through a preset open source tool based on the concept of the ontology, so that an abnormal feature knowledge semantic network is generated.
It should be noted that the foregoing explanation of the embodiment of the blade icing identification method based on the condition variation self-encoder is also applicable to the system of the embodiment, and will not be repeated here
In summary, in the blade icing identification system based on the condition variation self-encoder according to the embodiment of the application, steady state discrimination, working condition division and reference sample screening are performed first, invalid abnormal data, unsteady state data and a small amount of tiny abnormal data points mixed in historical data of a wind turbine generator are removed, the cleaned data are used for training a subsequent abnormal detection model, and accuracy of abnormal data detection of the trained abnormal detection model can be improved. And then reconstructing variable working condition data by adopting a variable self-encoder model, establishing an anomaly detection model with multi-characteristic parameter fusion, and positioning blade icing anomaly parameters of the wind turbine generator under the variable working condition. Therefore, the system can establish an accurate reference model under the conditions that the working conditions are changed and deviate from the design working conditions, is favorable for timely and accurately detecting abnormal parameters under the variable working conditions, and improves the accuracy of identifying blade icing anomalies of the wind turbine generator under the variable working conditions. And finally, carrying out reasoning on the positioned abnormal parameters through a related knowledge semantic network to acquire a recognition result, judging whether the fan is abnormal due to blade icing or not currently, and determining specific abnormal parameters. Therefore, the system improves the accuracy and reliability of the identification of the icing abnormality of the unit blade, is convenient for eliminating the icing abnormality of the blade in time and ensures the normal operation of the unit.
In order to implement the above embodiments, the present application further proposes a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method for identifying blade icing based on a condition-variable self-encoder as described in any of the above embodiments.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "plurality" is at least two, such as two, three, etc., unless explicitly defined otherwise.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and additional implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present application.
Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.
It is to be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. As with the other embodiments, if implemented in hardware, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.
In addition, each functional unit in each embodiment of the present application may be integrated in one processing module, or each unit may exist alone physically, or two or more units may be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product.
The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like. Although embodiments of the present application have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the application, and that variations, modifications, alternatives, and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the application.
Claims (10)
1. The blade icing identification method based on the condition variation self-encoder is characterized by comprising the following steps of:
aiming at the influence of blade icing on the running of the wind turbine, historical data related to the weather condition of the region where the wind turbine is located, the vibration condition of the tower drum, the output power characteristic and the measurement result of the wind meter are obtained;
eliminating five types of abnormal data and unsteady state data determined according to a preset standard from the historical data, and dividing the historical data according to the working condition characteristic set and an equidistant working condition dividing mode;
based on the similarity of data samples, clustering the data under each working condition through a Gaussian Mixture Model (GMM), screening out a reference sample from one or more kinds of clustering clusters under each working condition, and partitioning a training set from the reference sample;
Analyzing the running state of the wind turbine under the variable working condition, constructing a reference model of the wind turbine under the variable working condition based on a condition variation self-encoder, and training the condition variation self-encoder through the training set to obtain the reference model for detecting the blade icing abnormal parameters;
inputting test data of a target wind turbine to be detected into the reference model, calculating reconstruction probability serving as a characteristic index of blade icing recognition abnormality detection through the reference model, positioning abnormal parameters based on the reconstruction probability, and inputting the abnormal parameters into a pre-constructed abnormal characteristic knowledge semantic network corresponding to a blade icing mode for diagnosis and reasoning so as to identify whether the blade icing abnormality occurs in the target wind turbine.
2. The blade icing identification method as claimed in claim 1, wherein said historical data comprises: reactive power average value, generating capacity average value, wind speed average value measured by a mechanical wind meter, wind speed average value measured by an ultrasonic wind meter, wind speed average value in a plurality of time periods, hub rotation speed average value, generator rotation speed average value, torque feedback average value, power grid voltage average value, power grid three-phase current average value, power grid line voltage average value, power grid outlet line current average value, tower vibration acceleration average value and active power output quantity of the converter.
3. The blade icing identification method as recited in claim 1, wherein said removing non-stationary data from said historical data comprises:
the output power of the wind turbine generator is represented by an expression comprising the change rate of power, and the difference value of the output power at two adjacent moments is calculated;
estimating the change rate of power by means of the average value of the sample statistics in the time window, and calculating a confidence interval in which the true value of the change rate of power is positioned by adopting an interval estimation method;
and under the condition that the confidence interval does not comprise zero, judging that the wind turbine generator is in an unsteady state working condition in the time window, and eliminating corresponding unsteady state data.
4. The blade icing identification method of claim 1 wherein said set of operating condition characteristics comprises a plurality of operating condition characteristic parameters including, but not limited to: wind speed, wind direction, moment of torsion, rotational speed and ambient temperature, based on operating mode feature set, through equidistant operating mode division's mode to the historical data carries out the operating mode and divides, includes:
determining the maximum value and the minimum value of each operation condition characteristic parameter in the self variation range, and obtaining the corresponding operation condition dividing interval of each operation condition characteristic parameter;
Based on the maximum value, the minimum value and the working condition dividing interval corresponding to each operation working condition characteristic parameter, respectively carrying out equal interval working condition dividing on the historical data, and determining each divided working condition in an intersection mode;
and removing invalid working conditions from all the divided working conditions, and judging whether the number of the remaining valid working conditions is larger than a preset minimum threshold value.
5. The blade icing recognition method according to claim 1, wherein the clustering of the data for each of the divided conditions by the gaussian mixture model GMM includes:
estimating parameters of the Gaussian mixture model GMM through a maximum expected value algorithm;
determining the number of sub-models of the Gaussian mixture model through a red pool information criterion AIC so as to cluster the state types of the wind turbine generator;
the screening the reference sample from one or more kinds of cluster under each working condition comprises the following steps:
and comparing the data samples under different clusters, and selecting a group of data with highest average running efficiency as a reference sample.
6. The blade icing recognition method according to claim 1, wherein the calculating of the reconstruction probability by the reference model as a characteristic index of blade icing recognition abnormality detection comprises:
Obtaining a test sample, obtaining a first parameter and a second parameter in Gaussian distribution of each hidden variable through an encoder of the conditional variation self-encoder, and sampling a preset number of sample points for each hidden variable;
calculating a third parameter and a fourth parameter in a reconstruction variable likelihood distribution corresponding to each hidden variable through a decoder of the conditional variation self-encoder;
based on the third parameter and the fourth parameter, an average value of log likelihood of the test sample under a hidden variable condition is counted.
7. The blade icing identification method as claimed in claim 1, wherein constructing the abnormal feature knowledge semantic network corresponding to the blade icing mode comprises:
combining a plurality of analysis methods to obtain knowledge of fault identification of blade icing of the wind turbine generator;
and converting the blade icing fault knowledge into a structured language, and modeling the structured language through a preset open source tool based on the concept of the ontology so as to generate the abnormal feature knowledge semantic network.
8. A blade icing identification system based on a condition-variable self-encoder, comprising:
the acquisition module is used for acquiring historical data related to weather conditions, tower vibration conditions, output power characteristics and wind meter measurement results of the region where the wind turbine generator is located aiming at the influence of blade icing on the operation of the wind turbine generator;
The dividing module is used for eliminating five types of abnormal data and unsteady state data determined according to a preset standard from the historical data, and dividing the historical data according to the working condition characteristic set and an equidistant working condition dividing mode;
the screening module is used for clustering the data under each working condition through the Gaussian mixture model GMM based on the similarity of the data samples, screening a reference sample from one or more types of clustering clusters under each working condition, and dividing a training set from the reference sample;
the training module is used for analyzing the running state of the wind turbine generator under the variable working condition, constructing a reference model of the wind turbine generator under the variable working condition based on the condition variation self-encoder, and training the condition variation self-encoder through the training set to obtain the reference model for detecting the blade icing abnormal parameters;
the identification module is used for inputting the test data of the target wind turbine to be detected into the reference model, calculating the reconstruction probability through the reference model to serve as a characteristic index for detecting abnormal blade icing recognition, positioning abnormal parameters based on the reconstruction probability, inputting the abnormal parameters into a pre-constructed abnormal characteristic knowledge semantic network corresponding to the blade icing mode to carry out diagnosis and reasoning so as to identify whether the target wind turbine has abnormal blade icing.
9. The system according to claim 8, wherein the partitioning module is specifically configured to:
the output power of the wind turbine generator is represented by an expression comprising the change rate of power, and the difference value of the output power at two adjacent moments is calculated;
estimating the change rate of power by means of the average value of the sample statistics in the time window, and calculating a confidence interval in which the true value of the change rate of power is positioned by adopting an interval estimation method;
and under the condition that the confidence interval does not comprise zero, judging that the wind turbine generator is in an unsteady state working condition in the time window, and eliminating corresponding unsteady state data.
10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements a method of identifying blade icing based on a conditional variation self-encoder as claimed in any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311092110.0A CN117469105A (en) | 2023-08-28 | 2023-08-28 | Blade icing identification method and system based on condition variation self-encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311092110.0A CN117469105A (en) | 2023-08-28 | 2023-08-28 | Blade icing identification method and system based on condition variation self-encoder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117469105A true CN117469105A (en) | 2024-01-30 |
Family
ID=89630065
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311092110.0A Pending CN117469105A (en) | 2023-08-28 | 2023-08-28 | Blade icing identification method and system based on condition variation self-encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117469105A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118764366A (en) * | 2024-09-09 | 2024-10-11 | 中国移动紫金(江苏)创新研究院有限公司 | Gateway fault detection method, device, equipment, storage medium and product |
-
2023
- 2023-08-28 CN CN202311092110.0A patent/CN117469105A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118764366A (en) * | 2024-09-09 | 2024-10-11 | 中国移动紫金(江苏)创新研究院有限公司 | Gateway fault detection method, device, equipment, storage medium and product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114623050A (en) | Method, system and medium for detecting abnormal operation data of wind turbine generator | |
| CN110414155B (en) | Fan component temperature abnormity detection and alarm method with single measuring point | |
| Meng et al. | A hybrid method for prognostics of lithium-ion batteries capacity considering regeneration phenomena | |
| CN110362045B (en) | Marine doubly-fed wind turbine generator fault discrimination method considering marine meteorological factors | |
| CN111597651A (en) | Rolling bearing performance degradation evaluation method based on HWPSO-SVDD model | |
| CN112733283A (en) | Wind turbine generator component fault prediction method | |
| CN114912640A (en) | Method and system for detecting abnormal mode of generator set based on deep learning | |
| CN111103137A (en) | Wind turbine gearbox fault diagnosis method based on deep neural network | |
| CN109492790A (en) | Wind turbines health control method based on neural network and data mining | |
| CN111784537B (en) | Power distribution network state parameter monitoring method and device and electronic equipment | |
| CN116484751B (en) | Fatigue life assessment method and device for wind turbine generator components | |
| Bai et al. | A two-phase-based deep neural network for simultaneous health monitoring and prediction of rolling bearings | |
| CN111273125A (en) | RST-CNN-based power cable channel fault diagnosis method | |
| CN110969185A (en) | Equipment abnormal state detection method based on data reconstruction | |
| CN117674420A (en) | Power station operation state monitoring system and method based on multivariate time sequence prediction | |
| CN117471346A (en) | Method and system for determining remaining life and health status of retired battery module | |
| CN117469105A (en) | Blade icing identification method and system based on condition variation self-encoder | |
| CN115877205A (en) | Intelligent fault diagnosis system and method for servo motor | |
| CN117706369A (en) | Fuel cell life prediction method based on dynamic health index | |
| Liu et al. | Real-time comprehensive health status assessment of hydropower units based on multi-source heterogeneous uncertainty information | |
| CN114742363A (en) | Energy efficiency state evaluation method, system and medium for wind turbine generator | |
| Du et al. | A SOM based Anomaly detection method for wind turbines health management through SCADA data | |
| CN112580254B (en) | Multi-dimensional analysis and early warning method for converter transformer based on multi-level decision fusion | |
| CN114117887A (en) | Real-time assessment method, system and medium for online power generation performance of wind turbine generator | |
| CN114320770A (en) | Effective wind speed measuring method and device for wind turbine generator |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |