CN114282646B - Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement - Google Patents
Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement Download PDFInfo
- Publication number
- CN114282646B CN114282646B CN202111432905.2A CN202111432905A CN114282646B CN 114282646 B CN114282646 B CN 114282646B CN 202111432905 A CN202111432905 A CN 202111432905A CN 114282646 B CN114282646 B CN 114282646B
- Authority
- CN
- China
- Prior art keywords
- feature extraction
- bilstm
- optical power
- whale
- data
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 65
- 230000003287 optical effect Effects 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 title claims abstract description 18
- 230000006872 improvement Effects 0.000 title claims description 8
- 241000283153 Cetacea Species 0.000 claims abstract description 48
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 31
- 238000005457 optimization Methods 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 16
- 238000010606 normalization Methods 0.000 claims abstract description 14
- 238000005311 autocorrelation function Methods 0.000 claims abstract description 12
- 238000013507 mapping Methods 0.000 claims abstract description 12
- 239000013598 vector Substances 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 230000009191 jumping Effects 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 4
- 230000008602 contraction Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000013136 deep learning model Methods 0.000 description 4
- 238000010248 power generation Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000013179 statistical model Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- 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
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a system for predicting optical power based on two-stage feature extraction and improved BiLSTM, wherein the method comprises the following steps: s1, performing shallow feature extraction on optical power data by using a partial autocorrelation function, and performing normalization processing; s2, constructing a CNN network, and sending the processed data into the CNN for deep feature extraction; s3, constructing a BiLSTM model; s4, introducing Lorenz mapping to improve the initial population of the whale algorithm, and optimizing parameters of BiLSTM by adopting an improved whale optimization algorithm; and S5, sending the data extracted by the CNN depth features to an improved BiLSTM for prediction. According to the invention, through the two-stage feature extraction of shallow feature extraction and depth feature extraction, the correlation between features can be further mined, noise and unstable components of the optical power data can be filtered out, and the data obtained by processing the noise and unstable components are sent into an improved BiLSTM model for optical power prediction, so that the prediction precision can be effectively improved.
Description
Technical Field
The invention belongs to the technical field of optical power prediction, and particularly relates to an optical power prediction method and an optical power prediction system based on two-stage feature extraction and BiLSTM improvement.
Background
With the decrease of non-renewable energy reserves, the development of clean renewable energy has become a research hotspot in recent years. The application of optical power energy as one of renewable energy sources to the field of power generation has been widely focused and studied. However, the grid is hindered from utilizing the optical power energy for large-scale photovoltaic power generation due to the disadvantage that the optical power energy has instability. The optical power energy is accurately predicted, the power grid can be better guided to perform power generation, scheduling and other works, and factors with great threat to the safe operation of the power grid are prevented. Therefore, accurate and reliable optical power predictions are necessary.
At present, there are four types of optical power prediction models, namely a statistical model, a traditional machine learning model, a deep learning model and a hybrid model. Statistical models require data to be stationary, whereas traditional machine learning models have insufficient modeling processing power for multidimensional data input, which results in limited accuracy in their predictions of light power. The deep learning model can extract key features with the greatest influence on the target vector from mass data, and improves photovoltaic power prediction accuracy. The hybrid model can comprehensively consider the advantages of the deep learning model and the intelligent evolution algorithm, and has better performance in the field of optical power prediction.
Disclosure of Invention
The invention aims to: aiming at the defects of the prior art, the invention provides a method and a system for predicting optical power based on two-stage feature extraction and improved BiLSTM, through the two-stage feature extraction of shallow feature extraction and depth feature extraction, the correlation between features can be further excavated, noise and unstable components of optical power data can be filtered out, and the data obtained by processing the correlation can be sent into an improved BiLSTM model for optical power prediction, so that the prediction precision can be effectively improved.
The technical scheme is as follows: the invention provides a light power prediction method based on two-stage feature extraction and improved BiLSTM, which specifically comprises the following steps:
(1) Collecting optical power data, performing shallow feature extraction on the optical power data by using a partial autocorrelation function, and then performing normalization processing;
(2) Constructing a CNN network, sending output data of shallow feature extraction into the CNN network for deep feature extraction, and dividing the output data into a training set and a testing set;
(3) Constructing a BiLSTM model;
(4) Improving a whale optimization algorithm, introducing Lorenz mapping to generate a whale initial population, and sending training data into the improved whale optimization algorithm to optimize the learning rate and the hidden layer node number of the BiLSTM model constructed in the step (3);
(5) The optical power is predicted using a BiLSTM model with optimized parameters.
Further, the partial autocorrelation function in the step (1) is:
s t ＝φ i1 s t-1 +φ i2 s t-2 +…φ ij s t-k ...+φ ii s t-i +u t (1)
wherein s is t Is a time series, s t-k (k=1, 2, …, i) represents a time series with a hysteresis order of k order, phi ij Represents the j (j=1, 2, …, i) th regression coefficient in the i-th order autoregressive equation, u t Is a residual sequence.
Further, the step (2) includes the steps of:
(21) Sending the light power data subjected to shallow layer feature extraction and normalization treatment to a convolution layer, and carrying out feature extraction by the convolution layer;
(22) Nonlinear mapping is carried out on the output of the convolution layer by using a sigmoid function as an activation function;
(23) Taking the output result of the step (22) as the input of a pooling layer, and performing data dimension reduction on the input data at the pooling layer;
(24) The convolution layer and the pooling layer are stacked, and the number of stacked layers is two;
(25) The Flatten layer integrates the extracted features and outputs the integrated features to the improved BiLSTM model for prediction.
Further, the implementation process of the step (3) is as follows:
(31) Establishing a forgetting door unit f t The formula is as follows:
f t ＝σ*(w f *[h t-1 ,x t ]+b f ) (3)
wherein w is f And b f Respectively the weight and bias input by the forgetting gate, sigma is a sigmoid activation function, h t-1 For inputting information at the last moment, x t Inputting the current time;
i t ＝σ(w i *[h t-1 ,x t ]+b i ) (5)
wherein w is c And w i Respectively as intermediate variablesAnd input i t Weight parameters of (2); b c And b i Intermediate variables +.>And input i t Is offset from (a); tanh is an activation function of tanh;
(32) The output gate unit is built, and the formula is as follows:
o t ＝σ*(w o *[h t-1 ,x t ]+b o ) (7)
h t ＝o t *tanh(C t ) (8)
wherein w is o And b o The weight and bias of the output gate input are respectively C t To control gate output, h t Calculating the output value of the unit for the LSTM;
the total output value of the BiLSTM computing unit at the time t is the sum of the output values of the forward LSTM unit and the backward LSTM unit:
wherein,,is a vector concatenation operation.
Further, the implementation process of the step (4) is as follows:
(41) Initializing parameter population scale, iteration times and upper and lower boundaries of whale positions of a whale optimization algorithm, and introducing Lorenz mapping to generate an initial population:
wherein x, y and z respectively represent variables related to convection intensity, horizontal temperature difference and vertical temperature difference; the parameters ζ, η,representing parameters relating to the pram majority, rayleigh number and container size, respectively;
(42) Calculating the fitness value of all individuals in the population to obtain the current optimal whale individual position vector;
(43) Introducing parameter p and convergence factor A, p is [0,1]Is a random number of (a) and the convergence factor a=a (2*r 1 -1)；r 1 Is [0,1]A is gradually reduced from 2 to 0 by the random number in between; at each iteration, the values of p and the convergence factor A are calculated, when p.ltoreq.0.5 and |A|<1, jumping to step (44); when p is>When 0.5 and |A| is not less than 1, skipping to the step (45); when p is>0.5, jumping to the step (46);
(44) Performing contraction bounding iteration update on individual whale position vectors;
x(t+1)＝x rand -A*|C*x rand -x t | (13)
wherein x is t The individual position at the t-th iteration, the wobble factor C is [0,2]A random number, x rand Is a random individual in a whale population;
(45) Performing random search predation iterative update on individual whale position vectors; performing spiral predation iterative updating on individual whale position vectors:
x(t+1)＝x best -A*|C*x best -x t | (14)
wherein x is best Representing the optimal individuals in the current population;
(46) Performing spiral predation iterative updating on individual whale position vectors:
x(t+1)＝d best *exp(bl)*cos(2πl)+x best (15)
wherein d best ＝|x best -x t The representation represents that the individual x is distant from the optimal individual x before the location update best B is a constant, l is the interval [ -1,1]A random number on the table;
(47) The iteration times are added with 1, whether the maximum iteration times of the algorithm are reached or not is judged, if the maximum iteration times are reached, the step (48) is skipped, and otherwise, the step (42) is entered;
(48) The improved whale optimization algorithm optimizes the learning rate and hidden layer node number of the BiLSTM.
Based on the same inventive concept, the present invention also provides an optical power prediction system based on two-stage feature extraction and improved BiLSTM, comprising:
shallow feature extraction unit: extracting partial autocorrelation function features of the original optical power time sequence;
depth feature extraction unit: the optical power data used for carrying out normalization processing after shallow layer feature extraction is subjected to deep feature extraction by using a CNN network unit;
improving whale algorithm to optimize BiLSTM unit: the improved whale optimization algorithm is used for optimizing the learning rate and the hidden layer node number of the BiLSTM;
prediction unit: and sending the output data of the depth feature extraction unit into BiLSTM containing the optimized parameters for prediction, and then outputting the optical power prediction result.
The beneficial effects are that: compared with the prior art, the invention has the beneficial effects that: 1. according to the invention, through two-stage feature extraction of shallow feature extraction and depth feature extraction, the correlation between features is further excavated, and noise and unstable components of optical power data are filtered out; the improved BiLSTM model is used for predicting the data extracted by the depth features, the advantages of the deep learning model and the intelligent evolution algorithm are comprehensively considered, and the light power prediction accuracy is effectively improved; 2. according to the invention, lorenz mapping is introduced on the initialization of an original whale optimization algorithm, so that the diversity of population is enriched, the convergence rate of the algorithm is accelerated, and the improved whale optimization algorithm is used for optimizing the parameters of the BiLSTM, so that the performance of the BiLSTM model can be effectively improved.
Drawings
FIG. 1 is a flow chart of a method for optical power prediction based on two-stage feature extraction and BiLSTM improvement;
FIG. 2 is a schematic diagram of a two-stage feature extraction and improved BiLSTM based optical power prediction system;
FIG. 3 is a graph comparing actual values with predicted values using the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings.
The invention provides an optical power prediction method based on two-stage feature extraction and improved BiLSTM, which comprises the steps of firstly carrying out shallow feature extraction on optical power data by using a partial autocorrelation function, then carrying out normalization processing, secondly carrying out deep feature extraction on the optical power data subjected to shallow feature processing and normalization processing by using a CNN network, and then sending the data subjected to deep feature extraction into an improved BiLSTM model for prediction. And meanwhile, generating an initial population by using Lorenz mapping for the whale optimization algorithm, and optimizing parameters of the BiLSTM model by using the improved whale optimization algorithm. Through the two-stage feature extraction of shallow feature extraction and depth feature extraction, the correlation between features can be further excavated, noise and unstable components of the optical power data can be filtered out, the processed data are sent into an improved BiLSTM model for optical power prediction, and prediction accuracy can be effectively improved. As shown in fig. 1, the method specifically comprises the following steps:
step 1: collecting optical power data, and performing shallow feature extraction and normalization processing on the optical power data by using a partial autocorrelation function.
And calculating the correlation between the optical power time sequence and the hysteresis value of the optical power time sequence by using a partial autocorrelation function, and extracting the characteristics. The partial autocorrelation function expression is:
s t ＝φ i1 s t-1 +φ i2 s t-2 +…+φ ii s t-i +u t (1)
wherein s is t Is a time series, s t-k (k=1, 2, …, i) represents a time series with a hysteresis order of k order, phi ij Represents the j (j=1, 2, …, i) th regression coefficient in the i-th order autoregressive equation, u t Is a residual sequence.
Carrying out normalization processing on the light power data extracted by the shallow layer characteristics:
f * ＝f i -f min /f max -f min i＝1,2,…,n (2)
wherein f i Indicating the optical power value at the i-th time, f min Representing the minimum value, f, in the optical power data max Representing the maximum value in the optical power data.
Step 2: and constructing a CNN network, and sending the data subjected to shallow feature processing and normalization processing into the CNN for deep feature extraction.
(2.1) transmitting the light power data subjected to shallow layer feature extraction and normalization treatment to a convolution layer, and performing feature extraction by the convolution layer;
(2.2) non-linearly mapping the convolutional layer output using a sigmoid function as an activation function;
(2.3) taking the output result of (2.2) as the input of a pooling layer, and performing data dimension reduction on the input data at the pooling layer;
(2.4) stacking the convolution layer and the pooling layer, wherein the number of the stacking layers is two;
(2.5) the Flatten layer integrates the extracted features and outputs them to the modified BiLSTM.
Step 3: the BiLSTM model is constructed, and the specific construction steps are as follows:
(3.1) establishing a forgetting door Unit f t The formula is as follows:
f t ＝σ*(w f *[h t-1 ,x t ]+b f ) (3)
wherein w is f And b f Respectively the weight and bias input by the forgetting gate, sigma is a sigmoid activation function, h t-1 For inputting information at the last moment, x t Is input for the current time.
i t ＝σ(w i *[h t-1 ,x t ]+b i ) (5)
Wherein w is c And w i Respectively as intermediate variablesAnd input i t Weight parameters of (2); b c And b i Intermediate variables +.>And input i t Is offset from (a); tanh is the activation function of tanh.
(3.2) establishing an output gate unit, the formula is as follows:
o t ＝σ*(w o *[h t-1 ,x t ]+b o ) (7)
h t ＝o t *tanh(C t ) (8)
wherein w is o And b o The weight and bias of the output gate input are respectively C t To control gate output, h t The output value of the unit is calculated for the present LSTM.
The total output value of the BiLSTM computing unit at the time t is the sum of the output values of the forward LSTM unit and the backward LSTM unit, and the specific expressions are (9) - (11):
wherein,,is a vector concatenation operation.
Step 4: the whale optimization algorithm was modified and then the parameters of the BiLSTM were optimized using the modified whale optimization algorithm. The specific improvement steps are as follows:
(4.1) initializing the parameter population scale, the iteration times and the upper and lower boundaries of the whale position of the whale optimization algorithm.
Aiming at the defect that population diversity of a whale optimization algorithm is not abundant, lorenz mapping is introduced to generate an initial population, and a Lorenz mapping formula is as follows:
wherein x, y and z respectively represent variables related to convection intensity, horizontal temperature difference and vertical temperature difference; the parameters ζ, η,representing parameters relating to the pram majority, rayleigh number and container size, respectively.
And (4.2) calculating the fitness value of all individuals in the population to obtain the current optimal whale individual position vector.
(4.3) introducing a parameter p and a convergence factor A, p being [0,1]Is a random number of (a) and the convergence factor a=a (2*r 1 -1)；，r 1 Is [0,1]The random number in between, a, gradually decreases from 2 to 0. At each iteration, the values of p and the convergence factor A are calculated, when p.ltoreq.0.5 and |A|<1, jumping to the step (4.4); when p is>When 0.5 and |A| is not less than 1, skipping to the step (4.5); when p is>And (4.6) skipping to the step (0.5).
(4.4) performing contraction bounding iteration update on individual whale position vectors:
x(t+1)＝x rand -A*|C*x rand -x t | (13)
wherein x is t The individual position at the t-th iteration, the wobble factor C is [0,2]A random number, x rand Is a random individual in the whale population.
(4.5) performing random search predation iterative updating on individual whale position vectors; performing spiral predation iterative updating on individual whale position vectors:
x(t+1)＝x best -A*|C*x best -x t | (14)
wherein x is best Representing the optimal individual in the current population.
(4.6) performing spiral predation iterative updating on individual whale position vectors:
x(t+1)＝d best *exp(bl)*cos(2πl)+x best (15)
wherein d best ＝|x best -x t The representation represents that the individual x is distant from the optimal individual x before the location update best B is a constant, l is the interval [ -1,1]Random numbers on the same.
And (4.7) adding 1 to the iteration number, judging whether the maximum iteration number of the algorithm is reached, if so, jumping to the step (4.8), otherwise, entering the step (4.2).
(4.8) optimizing the learning rate and the hidden layer node number of the BiLSTM by improving a whale optimizing algorithm.
Step 5: and (3) sending the integrated output data of the flat layer into the improved BiLSTM model for prediction, and outputting an optical power prediction result.
Four indices, MAE (mean absolute error), RMSE (root mean square error), R (correlation coefficient), MAPE (mean absolute percent error), were used to evaluate the performance of the proposed model, and the MAE, RMSE, R and MAPE calculation formulas were as follows:
wherein N is the total number of samples, o i Represents the (th) true value, p, at the (th) moment i Representing the p-th predicted value at the i-th instant,mean value representing true value, +.>Representing the average of the predicted values.
As shown in fig. 2, the present invention further provides a light power prediction system based on two-stage feature extraction and improved BiLSTM, comprising: the shallow layer feature extraction unit is used for extracting partial autocorrelation function features of the original optical power time sequence; the depth feature extraction unit is used for carrying out depth feature extraction on the light power data subjected to shallow feature extraction and normalization processing by using the CNN network unit; improving whale algorithm to optimize BiLSTM unit: the improved whale optimization algorithm is used for optimizing the learning rate and the hidden layer node number of the BiLSTM; prediction unit: and sending the output data of the depth feature extraction unit into BiLSTM containing the optimized parameters for prediction, and then outputting the optical power prediction result.
The invention samples the light power data once every 5min from eight points in the morning to five points in the afternoon in 2021, 5 months, 1 day to 5 months, 30 days, and 3600 data points in total in a Shanghai city creep area, and experimental simulation is carried out to verify the effect of the model provided by the invention. FIG. 3 is a schematic diagram of two-stage feature extraction and improvement of the true and predicted values of the BiLSTM model, wherein the training and test sets are according to 7: 3.
TABLE 1 prediction Performance index of optical power for different models
Table 1 shows the performance indexes obtained by two-stage feature extraction and sending the two-stage feature extraction to different models, and from Table 1, it can be seen that the two-stage feature extraction and improved BiLSTM hybrid model provided by the invention has excellent performance on optical power prediction.
Claims (4)
1. A two-stage feature extraction and improved BiLSTM based optical power prediction method comprising the steps of:
(1) Collecting optical power data, performing shallow feature extraction on the optical power data by using a partial autocorrelation function, and then performing normalization processing;
(2) Constructing a CNN network, sending output data of shallow feature extraction into the CNN network for deep feature extraction, and dividing the output data into a training set and a testing set;
(3) The BiLSTM model is constructed, and the implementation process is as follows:
establishing a forgetting door unit f t The formula is as follows:
f t ＝σ*(w f *[h t-1 ,x t ]+b f ) (3)
wherein w is f And b f Respectively the weight and bias input by the forgetting gate, sigma is a sigmoid activation function, h t-1 For inputting information at the last moment, x t Inputting the current time;
i t ＝σ(w i *[h t-1 ,x t ]+b i ) (5)
wherein w is c And w i Respectively as intermediate variablesAnd input i t Weight parameters of (2); b c And b i Intermediate variables +.>And input i t Is offset from (a); tanh is an activation function of tanh;
the output gate unit is built, and the formula is as follows:
o t ＝σ*(w o *[h t-1 ,x t ]+b o ) (7)
h t ＝o t *tanh(C t ) (8)
wherein w is o And b o The weight and bias of the output gate input are respectively C t To control gate output, h t Calculating the output value of the unit for the LSTM;
the total output value of the BiLSTM computing unit at the time t is the sum of the output values of the forward LSTM unit and the backward LSTM unit:
h t ＝h t (1) ⊕h t (2) (11)
wherein ∈is vector concatenation operation;
(4) Improving a whale optimization algorithm, introducing Lorenz mapping to generate a whale initial population, and sending training data into the improved whale optimization algorithm to optimize the learning rate and the hidden layer node number of the BiLSTM model constructed in the step (3); the specific implementation process is as follows:
(41) Initializing parameter population scale, iteration times and upper and lower boundaries of whale positions of a whale optimization algorithm, and introducing Lorenz mapping to generate an initial population:
wherein x, y and z respectively represent variables related to convection intensity, horizontal temperature difference and vertical temperature difference; the parameters ζ, η,representing parameters relating to the pram majority, rayleigh number and container size, respectively;
(42) Calculating the fitness value of all individuals in the population to obtain the current optimal whale individual position vector;
(43) Introducing parameter p and convergence factor A, p is [0,1]Is a random number of (a) and the convergence factor a=a (2*r 1 -1)；r 1 Is [0,1]A is gradually reduced from 2 to 0 by the random number in between; at each iteration, the values of p and the convergence factor A are calculated, when p.ltoreq.0.5 and |A|<1, jumping to step (44); when p is>When 0.5 and |A| is not less than 1, skipping to the step (45); when p is>0.5, jumping to the step (46);
(44) Performing contraction bounding iteration update on individual whale position vectors;
x(t+1)＝x rand -A*|C*x rand -x t | (13)
wherein x is t The individual position at the t-th iteration, the wobble factor C is [0,2]A random number, x rand Is a random individual in a whale population;
(45) Performing random search predation iterative update on individual whale position vectors; performing spiral predation iterative updating on individual whale position vectors:
x(t+1)＝x best -A*|C*x best -x t | (14)
wherein x is best Representing the optimal individuals in the current population;
(46) Performing spiral predation iterative updating on individual whale position vectors:
x(t+1)＝d best *exp(bl)*cos(2πl)+x best (15)
wherein d best ＝|x best -x t The representation represents that the individual x is distant from the optimal individual x before the location update best B is a constant, l is the interval [ -1,1]A random number on the table;
(47) The iteration times are added with 1, whether the maximum iteration times of the algorithm are reached or not is judged, if the maximum iteration times are reached, the step (48) is skipped, and otherwise, the step (42) is entered;
(48) Optimizing the learning rate and the hidden layer node number of the BiLSTM by using an improved whale optimization algorithm;
(5) The optical power is predicted using a BiLSTM model with optimized parameters.
2. The two-stage feature extraction and improvement BiLSTM based optical power prediction method of claim 1 wherein said partial autocorrelation function of step (1) is:
s t ＝φ i1 s t-1 +φ i2 s t-2 +…φ ij s t-k ...+φ ii s t-i +u t (1)
wherein s is t Is a time series, s t-k (k=1, 2, …, i) represents a time series with a hysteresis order of k order, phi ij Represents the j (j=1, 2, …, i) th regression coefficient in the i-th order autoregressive equation, u t Is a residual sequence.
3. The two-stage feature extraction and improvement BiLSTM based optical power prediction method of claim 1, wherein said step (2) comprises the steps of:
(21) Sending the light power data subjected to shallow layer feature extraction and normalization treatment to a convolution layer, and carrying out feature extraction by the convolution layer;
(22) Nonlinear mapping is carried out on the output of the convolution layer by using a sigmoid function as an activation function;
(23) Taking the output result of the step (22) as the input of a pooling layer, and performing data dimension reduction on the input data at the pooling layer;
(24) The convolution layer and the pooling layer are stacked, and the number of stacked layers is two;
(25) The Flatten layer integrates the extracted features and outputs the integrated features to the improved BiLSTM model for prediction.
4. A bi-stage feature extraction and improved BiLSTM based optical power prediction system employing the method of any of claims 1-3, comprising:
shallow feature extraction unit: extracting partial autocorrelation function features of the original optical power time sequence;
depth feature extraction unit: the optical power data used for carrying out normalization processing after shallow layer feature extraction is subjected to deep feature extraction by using a CNN network unit;
improving whale algorithm to optimize BiLSTM unit: the improved whale optimization algorithm is used for optimizing the learning rate and the hidden layer node number of the BiLSTM;
prediction unit: and sending the output data of the depth feature extraction unit into BiLSTM containing the optimized parameters for prediction, and then outputting the optical power prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111432905.2A CN114282646B (en) | 2021-11-29 | 2021-11-29 | Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111432905.2A CN114282646B (en) | 2021-11-29 | 2021-11-29 | Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114282646A CN114282646A (en) | 2022-04-05 |
| CN114282646B true CN114282646B (en) | 2023-08-25 |
Family
ID=80870193
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111432905.2A Active CN114282646B (en) | 2021-11-29 | 2021-11-29 | Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114282646B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114912673A (en) * | 2022-04-29 | 2022-08-16 | 河海大学 | Water level prediction method based on whale optimization algorithm and long-term and short-term memory network |
| CN115660032B (en) * | 2022-08-31 | 2024-05-10 | 华能江苏综合能源服务有限公司 | BI-LSTM neural network integration attention mechanism-based building roof photovoltaic power prediction method |
| CN115630726B (en) * | 2022-09-01 | 2024-01-30 | 华能江苏综合能源服务有限公司 | Roof photovoltaic power prediction method based on VMD-BILSTM neural network fused attention mechanism |
| CN116565863B (en) * | 2023-07-10 | 2023-09-26 | 南京师范大学 | Short-term photovoltaic output prediction method based on space-time correlation |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3454260A1 (en) * | 2017-09-11 | 2019-03-13 | Tata Consultancy Services Limited | Bilstm-siamese network based classifier for identifying target class of queries and providing responses thereof |
| CN110110930A (en) * | 2019-05-08 | 2019-08-09 | 西南交通大学 | A kind of Recognition with Recurrent Neural Network Short-Term Load Forecasting Method improving whale algorithm |
| CN112667816A (en) * | 2020-12-31 | 2021-04-16 | 华中师范大学 | Deep learning-based aspect level emotion analysis method and system |
| CN112817653A (en) * | 2021-01-22 | 2021-05-18 | 西安交通大学 | Cloud-side-based federated learning calculation unloading computing system and method |
| CN112966073A (en) * | 2021-04-07 | 2021-06-15 | 华南理工大学 | Short text matching method based on semantics and shallow features |
| CN113239181A (en) * | 2021-05-14 | 2021-08-10 | 廖伟智 | Scientific and technological literature citation recommendation method based on deep learning |
| CN113326969A (en) * | 2021-04-29 | 2021-08-31 | 淮阴工学院 | Short-term wind speed prediction method and system based on improved whale algorithm for optimizing ELM |
| CN113468803A (en) * | 2021-06-09 | 2021-10-01 | 淮阴工学院 | Improved WOA-GRU-based flood flow prediction method and system |
| CN113536982A (en) * | 2021-06-28 | 2021-10-22 | 长江大学 | Intelligent prediction method, system, equipment and medium for residual life of main reducer |
| CN113627071A (en) * | 2021-06-26 | 2021-11-09 | 西安科技大学 | Coal-fired boiler NO based on whale algorithm optimization long-time memory networkXSoft measurement method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10902738B2 (en) * | 2017-08-03 | 2021-01-26 | Microsoft Technology Licensing, Llc | Neural models for key phrase detection and question generation |
| US11562251B2 (en) * | 2019-05-16 | 2023-01-24 | Salesforce.Com, Inc. | Learning world graphs to accelerate hierarchical reinforcement learning |
-
2021
- 2021-11-29 CN CN202111432905.2A patent/CN114282646B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3454260A1 (en) * | 2017-09-11 | 2019-03-13 | Tata Consultancy Services Limited | Bilstm-siamese network based classifier for identifying target class of queries and providing responses thereof |
| CN110110930A (en) * | 2019-05-08 | 2019-08-09 | 西南交通大学 | A kind of Recognition with Recurrent Neural Network Short-Term Load Forecasting Method improving whale algorithm |
| CN112667816A (en) * | 2020-12-31 | 2021-04-16 | 华中师范大学 | Deep learning-based aspect level emotion analysis method and system |
| CN112817653A (en) * | 2021-01-22 | 2021-05-18 | 西安交通大学 | Cloud-side-based federated learning calculation unloading computing system and method |
| CN112966073A (en) * | 2021-04-07 | 2021-06-15 | 华南理工大学 | Short text matching method based on semantics and shallow features |
| CN113326969A (en) * | 2021-04-29 | 2021-08-31 | 淮阴工学院 | Short-term wind speed prediction method and system based on improved whale algorithm for optimizing ELM |
| CN113239181A (en) * | 2021-05-14 | 2021-08-10 | 廖伟智 | Scientific and technological literature citation recommendation method based on deep learning |
| CN113468803A (en) * | 2021-06-09 | 2021-10-01 | 淮阴工学院 | Improved WOA-GRU-based flood flow prediction method and system |
| CN113627071A (en) * | 2021-06-26 | 2021-11-09 | 西安科技大学 | Coal-fired boiler NO based on whale algorithm optimization long-time memory networkXSoft measurement method |
| CN113536982A (en) * | 2021-06-28 | 2021-10-22 | 长江大学 | Intelligent prediction method, system, equipment and medium for residual life of main reducer |
Non-Patent Citations (1)
| Title |
|---|
| Multi-scale clustering analysis based modeling of photovoltaic power characteristics and its application in prediction;GUAN Lin等;《Automation of Electric Power Systems》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114282646A (en) | 2022-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114282646B (en) | Optical power prediction method and system based on two-stage feature extraction and BiLSTM improvement | |
| CN110175386B (en) | Method for predicting temperature of electrical equipment of transformer substation | |
| US20230196076A1 (en) | Method for optimally selecting flood-control operation scheme based on temporal convolutional network | |
| CN111401755B (en) | Multi-new-energy output scene generation method, device and system based on Markov chain | |
| CN108596327B (en) | Seismic velocity spectrum artificial intelligence picking method based on deep learning | |
| CN110084424A (en) | A kind of Methods of electric load forecasting based on LSTM and LGBM | |
| CN109978283B (en) | Photovoltaic power generation power prediction method based on branch evolution neural network | |
| CN109146162B (en) | A kind of probability wind speed forecasting method based on integrated Recognition with Recurrent Neural Network | |
| CN113449464B (en) | Wind power prediction method based on improved deep extreme learning machine | |
| Liu et al. | A fault diagnosis intelligent algorithm based on improved BP neural network | |
| CN113688869B (en) | Photovoltaic data missing reconstruction method based on generation countermeasure network | |
| CN113468817B (en) | Ultra-short-term wind power prediction method based on IGOA (insulated gate bipolar transistor) optimized ELM (ELM) | |
| CN114897129A (en) | Photovoltaic power station short-term power prediction method based on similar daily clustering and Kmeans-GRA-LSTM | |
| CN112330487A (en) | Photovoltaic power generation short-term power prediction method | |
| CN113361803A (en) | Ultra-short-term photovoltaic power prediction method based on generation countermeasure network | |
| CN114330100A (en) | Short-term photovoltaic power probability interval prediction method | |
| CN115561005A (en) | Chemical process fault diagnosis method based on EEMD decomposition and lightweight neural network | |
| CN113255366A (en) | Aspect-level text emotion analysis method based on heterogeneous graph neural network | |
| CN117154690A (en) | Photovoltaic power generation power prediction method and system based on neural network | |
| CN116415177A (en) | Classifier parameter identification method based on extreme learning machine | |
| CN116646929A (en) | PSO-CNN-BILSTM-based short-term wind power prediction method | |
| CN111192158A (en) | Transformer substation daily load curve similarity matching method based on deep learning | |
| CN118134284A (en) | Deep learning wind power prediction method based on multi-stage attention mechanism | |
| CN114386666A (en) | Wind power plant short-term wind speed prediction method based on space-time correlation | |
| CN117374941A (en) | Photovoltaic power generation power prediction method based on neural network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240119 Address after: 518000 1104, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Patentee after: Shenzhen Hongyue Enterprise Management Consulting Co.,Ltd. Address before: 223003 Jiangsu Huaian economic and Technological Development Zone, 1 East Road. Patentee before: HUAIYIN INSTITUTE OF TECHNOLOGY |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240403 Address after: 334000, No. 42 Fanshang, Dongwang Village, Qingshui Township, Guangxin District, Shangrao City, Jiangxi Province Patentee after: Zhang Yimin Country or region after: China Address before: 518000 1104, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Patentee before: Shenzhen Hongyue Enterprise Management Consulting Co.,Ltd. Country or region before: China |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240509 Address after: 334000, Room 05, 11th Floor, Building 2, No. 9 Tianyou Avenue, High speed Rail Economic Experimental Zone, Shangrao City, Jiangxi Province Patentee after: Jiangxi Bobang New Energy Technology Co.,Ltd. Country or region after: China Address before: 334000, No. 42 Fanshang, Dongwang Village, Qingshui Township, Guangxin District, Shangrao City, Jiangxi Province Patentee before: Zhang Yimin Country or region before: China |
|
| TR01 | Transfer of patent right |