CN111143934B - Structural deformation prediction method based on time convolution network - Google Patents
Structural deformation prediction method based on time convolution network Download PDFInfo
- Publication number
- CN111143934B CN111143934B CN201911370873.0A CN201911370873A CN111143934B CN 111143934 B CN111143934 B CN 111143934B CN 201911370873 A CN201911370873 A CN 201911370873A CN 111143934 B CN111143934 B CN 111143934B
- Authority
- CN
- China
- Prior art keywords
- time
- convolution
- convolution network
- monitoring data
- structural deformation
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012544 monitoring process Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 8
- 238000010606 normalization Methods 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 6
- 230000001364 causal effect Effects 0.000 claims description 6
- 238000002790 cross-validation Methods 0.000 claims description 5
- 230000010339 dilation Effects 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000007787 long-term memory Effects 0.000 abstract description 2
- 238000005065 mining Methods 0.000 abstract description 2
- 238000012795 verification Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 8
- 238000012545 processing Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a structural deformation prediction method based on a time convolution network, which comprises the following steps: 1) Measuring actual deformation monitoring data of the structure, and performing preprocessing such as interpolation, normalization and the like; 2) Extracting characteristics of the deformation monitoring data of the structure processed in the step 1), and fully mining internal characteristics of the time sequence through a time convolution network; 3) Connecting the time features extracted in the step 2) to a full connection layer to obtain prediction output; 4) And obtaining the optimal combination parameters through analysis and cross verification of the model super parameters, and taking the prediction result of the optimal parameters as a final prediction result. According to the method, the historical information of the monitoring data can be effectively utilized, long-term memory is obtained, the time characteristics of the data are fully mined, training parameters can be reduced through one-dimensional expansion convolution operation, and the result shows that the method has high prediction accuracy and can provide scientific basis for analyzing the structural deformation trend.
Description
Technical Field
The invention belongs to the field of structural deformation monitoring, and relates to a structural deformation prediction method based on a time convolution network.
Background
With the rapid development of society, the demands of large buildings such as subways, bridges, tunnels and the like are also increasing, and the safety of construction and operation of the large buildings is particularly important. At present, deformation monitoring is mainly carried out on key parts of a structure by a modern and automatic technology, a reasonable monitoring prediction model is established by analyzing a monitoring data rule, and the deformation development trend of the structure can be mastered timely and accurately, so that the method is an important mode and means for evaluating the structural performance and monitoring the safety of the structure. Because deformation monitoring data generally has characteristics such as instability and nonlinearity, the existing structural deformation prediction model is lack of mining internal characteristics of a time sequence, and can not effectively utilize historical data to make effective judgment on future deformation trend. These problems all lead to larger prediction errors, poor model applicability and serious influence on the judgment of the structural state. Therefore, an accurate and effective structure deformation prediction model is established, measures are taken in advance to prevent disasters through analysis of prediction results, and scientific prediction has important significance for reducing life and property loss and guaranteeing structure safety.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a structural deformation prediction method based on a time convolution network.
In order to achieve the above object, the method for predicting structural deformation based on a time convolution network according to the present invention comprises the following steps:
1) Measuring actual deformation monitoring data of the structure and preprocessing the actual deformation monitoring data;
2) Extracting time characteristics of the deformation monitoring data of the structure subjected to the pretreatment in the step 1) through a time convolution network;
3) Obtaining a structural deformation prediction result through the full connection layer for the time characteristics of the structural deformation monitoring data obtained in the step 2);
4) And 3) carrying out model hyper-parameter analysis and experimental cross-validation on the structural deformation prediction result obtained in the step 3) to determine the optimal model parameters, and taking the prediction result corresponding to the optimal model parameters as the final prediction result of the structural deformation.
In step 1), when the actual deformation monitoring data of the structure has data loss, error data or the number of data samples is smaller than the preset number, interpolation and normalization processing are performed on the actual deformation monitoring data of the structure.
In step 2), the input S of each residual block in the time convolution network (i,1) Outputs S from the previous residual block (i-1,L) The expansion factor d of each residual block increases exponentially, d=2 i I is the number of convolution layers, the result S of residual connection of each layer t (i,j) The method comprises the following steps:
wherein S is (i,j) Representing the output of the time convolution network in the ith residual block and the jth convolution layer,the output result of the dilation convolution for each convolution layer is represented, and ReLu represents the activation function.
Features are extracted through a time convolution network, a causal expansion convolution is utilized to increase the receptive field while preventing future information leakage, and then historical information of monitoring data is utilized to extract time features of the data.
When the residual input and the residual output have different dimensions, a convolution of 1x1 is added to ensure that the residual input and the residual output have the same dimensions.
In step 3), time features S extracted by the time convolution network t (n,L) Obtaining the structure deformation prediction result through the full connection layerThe method comprises the following steps:
wherein w and b are parameters obtained by training, S t (n,L) G (·) is a linear activation function for the output of the time characteristic;
in the step 4), after the super parameter analysis, the optimal parameters of the model are as follows: the convolution kernel size is 8, the number of convolution kernels is 16, the expansion coefficients are {1,2,4,8,16,32}, the training times are 500, the optimization function is adam, and the loss function is MSE.
The invention has the following beneficial effects:
when the structural deformation prediction method based on the time convolution network is specifically operated, the time feature extraction is carried out on the deformation monitoring data of the actual structure after the pretreatment in the step 1) through the time convolution network, then the extracted time feature is fully connected with a layer to obtain a structural deformation prediction result, then the model super-parameter analysis and the experimental cross-validation are carried out on the structural deformation prediction result to determine the optimal model parameters, and the prediction result corresponding to the optimal model parameters is used as the final prediction result of the structural deformation, so that the historical data of the monitoring data are effectively utilized, the internal features of the data are fully excavated, the leakage of information can be effectively prevented by utilizing causal expansion convolution, meanwhile, the receptive field of the observation data is enlarged, the accurate prediction of the structural deformation is realized, and a scientific basis is made for the judgment of the structural state. Experimental results show that the predicted value and the true value of the invention are very close, the invention can provide scientific basis for the structural deformation trend, and has important significance for reducing life and property loss and guaranteeing structural safety.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a graph of the pretreatment results in the present invention;
FIG. 3 is a graph of the results of a time convolution network prediction in accordance with the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawing figures:
referring to fig. 1, the method for predicting structural deformation based on a time convolution network according to the present invention comprises the following steps:
1) Measuring actual deformation monitoring data of the structure and preprocessing the actual deformation monitoring data;
2) Extracting time characteristics of the deformation monitoring data of the structure subjected to the pretreatment in the step 1) through a time convolution network;
3) Obtaining a structural deformation prediction result through the full connection layer for the time characteristics of the structural deformation monitoring data obtained in the step 2);
4) And 3) carrying out model hyper-parameter analysis and experimental cross-validation on the structural deformation prediction result obtained in the step 3) to determine the optimal model parameters, and taking the structural deformation prediction result corresponding to the optimal model parameters as the final predicted structural deformation result.
In step 1), when the actual deformation monitoring data of the structure has data loss, error data or the number of data samples is smaller than the preset number, interpolation and normalization processing are performed on the actual deformation monitoring data of the structure.
In step 2), the input S of each residual block in the time convolution network (i,1) Outputs S from the previous residual block (i-1,L) The expansion factor d of each residual block increases exponentially, d=2 i I is the number of convolution layers, the result S of residual connection of each layer t (i,j) The method comprises the following steps:
wherein S is (i,j) Representing the output of the time convolution network in the ith residual block and the jth convolution layer,the output result of the dilation convolution for each convolution layer is represented, and ReLu represents the activation function.
Features are extracted through a time convolution network, a causal expansion convolution is utilized to increase the receptive field while preventing future information leakage, and then historical information of monitoring data is utilized to extract time features of the data.
When the residual input and the residual output have different dimensions, a convolution of 1x1 is added to ensure that the residual input and the residual output have the same dimensions.
In step 3), time rollsTime feature S of product network extraction t (n,L) Obtaining the structure deformation prediction result through the full connection layerThe method comprises the following steps:
wherein w and b are parameters obtained by training, S t (n,L) G (·) is a linear activation function for the output of the time characteristic;
in the step 4), after the super parameter analysis, the optimal parameters of the model are as follows: the convolution kernel size is 8, the number of convolution kernels is 16, the expansion coefficients are {1,2,4,8,16,32}, the training times are 500, the optimization function is adam, and the loss function is MSE.
According to the invention, the interpolation processing of the small sample data can effectively prevent the influence of data loss and error data on the prediction result, the normalization can accelerate the speed of gradient descent to solve the optimal solution, and the accuracy is improved to a certain extent. Because the structural deformation data has correlation, the causal convolution is realized by the time convolution network through padding, and the causal convolution ensures that future information cannot be used for the prediction of the previous time steps, thereby effectively preventing information leakage. The training parameters can be reduced through one-dimensional expansion convolution operation, the receptive field can be effectively increased through expansion convolution, the sequence length observed by a network is increased under the condition that the calculated amount is basically unchanged, long-term memory is obtained, the internal features of a time sequence are deeply mined, meanwhile, the accuracy of a model can be improved through residual connection, and finally, the prediction result of the optimal parameters is used as the final prediction result through analysis of super parameters.
The method is suitable for predicting the structural deformation, as shown in fig. 1, the structural deformation prediction model is subjected to interpolation and normalization processing according to measured data, the data effectiveness is improved, the influence on the prediction precision is reduced, the time characteristics are extracted through a time convolution network, the prediction output is obtained through full connection, and the final prediction result of the structural deformation is determined through an optimal parameter model; as shown in fig. 2, the trend of the structure after the data preprocessing is consistent with that of the original data, and the structure can be well matched even at the inflection point. As shown in fig. 3, the accuracy of the model is verified through a test set, and the predicted value is very close to the measured value, so that the method is an effective structure deformation prediction model; as shown in table 1, by analyzing the model hyper-parameters, an optimal parameter combination of the model is found by using a cross-validation method, and the corresponding prediction result is used as a final prediction result, as shown in fig. 3.
TABLE 1
Claims (6)
1. The structural deformation prediction method based on the time convolution network is characterized by comprising the following steps of:
1) Measuring actual deformation monitoring data of the structure and preprocessing the actual deformation monitoring data;
2) Extracting time characteristics of the deformation monitoring data of the structure subjected to the pretreatment in the step 1) through a time convolution network;
3) Obtaining a structural deformation prediction result through the full connection layer for the time characteristics of the structural deformation monitoring data obtained in the step 2);
4) Performing model hyper-parameter analysis and experimental cross-validation on the structural deformation prediction result obtained in the step 3) to determine an optimal model parameter, and taking the prediction result corresponding to the optimal model parameter as a final prediction result of structural deformation;
in step 2), the input S of each residual block in the time convolution network (i,1) Outputs S from the previous residual block (i-1,L) The expansion factor d of each residual block increases exponentially, d=2 i I is the number of convolution layers, each layer is leftResults of the ligation S t (i,j) The method comprises the following steps:
wherein S is (i,j) Representing the output of the time convolution network in the ith residual block and the jth convolution layer,the output result of the dilation convolution for each convolution layer is represented, and ReLu represents the activation function.
2. The method for predicting structural deformation based on time convolution network according to claim 1, wherein in step 1), when the actual deformation monitoring data of the structure has data loss, error data or the number of data samples is smaller than a preset number, interpolation and normalization are performed on the actual deformation monitoring data of the structure.
3. The method for predicting structural deformation based on time convolution network according to claim 1, wherein features are extracted through the time convolution network, a receptive field is increased while future information leakage is prevented by using causal dilation convolution, and then time features of data are extracted by using history information of monitoring data.
4. The method of claim 1, wherein when the residual input and the residual output have different dimensions, a 1x1 convolution is added to ensure that the residual input and the residual output have the same dimensions.
5. The method for predicting structural deformation based on time convolution network according to claim 1, wherein in step 3), the time feature S extracted by the time convolution network t (n,L) Obtaining the structure deformation prediction result through the full connection layerThe method comprises the following steps:
wherein w and b are parameters obtained by training, S t (n,L) G (·) is the linear activation function for the output of the time characteristic.
6. The method for predicting structural deformation based on time convolution network according to claim 1, wherein in step 4), after the super-parametric analysis, the optimal parameters of the model are: the convolution kernel size is 8, the number of convolution kernels is 16, the expansion coefficients are {1,2,4,8,16,32}, the training times are 500, the optimization function is adam, and the loss function is MSE.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911370873.0A CN111143934B (en) | 2019-12-26 | 2019-12-26 | Structural deformation prediction method based on time convolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911370873.0A CN111143934B (en) | 2019-12-26 | 2019-12-26 | Structural deformation prediction method based on time convolution network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111143934A CN111143934A (en) | 2020-05-12 |
| CN111143934B true CN111143934B (en) | 2024-04-09 |
Family
ID=70520685
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911370873.0A Active CN111143934B (en) | 2019-12-26 | 2019-12-26 | Structural deformation prediction method based on time convolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111143934B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112668775A (en) * | 2020-12-25 | 2021-04-16 | 西安翔迅科技有限责任公司 | Air quality prediction method based on time sequence convolution network algorithm |
| CN113178073A (en) * | 2021-04-25 | 2021-07-27 | 南京工业大学 | Traffic flow short-term prediction optimization application method based on time convolution network |
| CN114048835A (en) * | 2021-08-01 | 2022-02-15 | 北京工业大学 | Road base strain analysis method based on time sequence convolution network |
| CN114091729A (en) * | 2021-10-21 | 2022-02-25 | 中国电建集团华东勘测设计研究院有限公司 | Multi-step wind power prediction method based on singular spectrum decomposition and time convolution neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20180054414A (en) * | 2016-11-14 | 2018-05-24 | 삼성전자주식회사 | Method and apparatus for analysis of facial image |
| CN108509978A (en) * | 2018-02-28 | 2018-09-07 | 中南大学 | The multi-class targets detection method and model of multi-stage characteristics fusion based on CNN |
| CN109614981A (en) * | 2018-10-17 | 2019-04-12 | 东北大学 | The Power System Intelligent fault detection method and system of convolutional neural networks based on Spearman rank correlation |
| CN110032975A (en) * | 2019-04-15 | 2019-07-19 | 禁核试北京国家数据中心 | A kind of pick-up method of seismic phase |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7822697B2 (en) * | 2006-09-29 | 2010-10-26 | Globvision Inc. | Method and apparatus for infrastructure health monitoring and analysis wherein anomalies are detected by comparing measured outputs to estimated/modeled outputs by using a delay |
-
2019
- 2019-12-26 CN CN201911370873.0A patent/CN111143934B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20180054414A (en) * | 2016-11-14 | 2018-05-24 | 삼성전자주식회사 | Method and apparatus for analysis of facial image |
| CN108509978A (en) * | 2018-02-28 | 2018-09-07 | 中南大学 | The multi-class targets detection method and model of multi-stage characteristics fusion based on CNN |
| CN109614981A (en) * | 2018-10-17 | 2019-04-12 | 东北大学 | The Power System Intelligent fault detection method and system of convolutional neural networks based on Spearman rank correlation |
| CN110032975A (en) * | 2019-04-15 | 2019-07-19 | 禁核试北京国家数据中心 | A kind of pick-up method of seismic phase |
Non-Patent Citations (2)
| Title |
|---|
| 基于全卷积神经网络的建筑物屋顶自动提取;刘文涛;李世华;覃驭楚;;地球信息科学学报(11);全文 * |
| 基于时间序列与神经网络的软岩隧道变形预测模型及其应用;熊兴吾;孟茁超;;交通科学与工程(02);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111143934A (en) | 2020-05-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111143934B (en) | Structural deformation prediction method based on time convolution network | |
| CN110866314B (en) | Method for predicting residual life of rotating machinery of multilayer bidirectional gate control circulation unit network | |
| CN109657937B (en) | Product reliability evaluation and service life prediction method based on degradation data | |
| CN110610035A (en) | Rolling bearing residual life prediction method based on GRU neural network | |
| CN110569566B (en) | Method for predicting mechanical property of plate strip | |
| CN112884747A (en) | Automatic bridge crack detection system integrating cyclic residual convolution and context extractor network | |
| CN110909413A (en) | Concrete panel rock-fill dam horizontal displacement prediction accuracy method and system | |
| CN114997313B (en) | Abnormality detection method for ocean on-line monitoring data | |
| CN110990788A (en) | Bearing residual life prediction method based on ternary wiener process | |
| CN116432699A (en) | Method for predicting silicon content of blast furnace molten iron based on time sequence chart packing network | |
| CN111914490B (en) | Pump station unit state evaluation method based on depth convolution random forest self-coding | |
| CN117521512A (en) | Bearing residual service life prediction method based on multi-scale Bayesian convolution transducer model | |
| CN114493023A (en) | Diagnosis and interpolation method in abnormal water regime data based on RF-Adaboost model | |
| CN110993132B (en) | Transient monitoring method for supporting fatigue monitoring function of nuclear power plant | |
| CN115017015B (en) | Method and system for detecting abnormal behavior of program in edge computing environment | |
| CN116662800A (en) | Rolling bearing fault diagnosis method based on self-adaptive attention mechanism | |
| CN115081485B (en) | AI-based magnetic flux leakage internal detection data automatic analysis method | |
| CN117010723A (en) | Carbon emission prediction method, system, apparatus, and computer-readable storage medium | |
| CN116244596A (en) | Industrial time sequence data anomaly detection method based on TCN and attention mechanism | |
| CN116628444A (en) | Water quality early warning method based on improved meta-learning | |
| CN113266449B (en) | Method and system for predicting air leakage situation in front of aftertreatment system | |
| CN114791553A (en) | Chip failure parameter prediction method | |
| CN111274709A (en) | Bayesian strategy-based product reliability analysis method and storage medium | |
| CN110617927A (en) | Structural settlement deformation prediction method based on EMD-SVR-WNN | |
| Zhang et al. | A Fault Diagnosis and Predictive Maintenance Algorithm for Mechanical Systems Based on Deep Learning |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210621 Address after: 710064 middle section of south 2nd Ring Road, Xi'an, Shaanxi Applicant after: CHANG'AN University Applicant after: CHINA RAILWAY FIRST SURVEY AND DESIGN INSTITUTE GROUP Co.,Ltd. Address before: 710064 middle section, south two ring road, Shaanxi, Xi'an Applicant before: CHANG'AN University |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |