Power distribution network photovoltaic generation power prediction method based on WOA-VMD and double-current network

Abstract

The invention provides a power distribution network photovoltaic power generation power prediction method based on WOA-VMD and a double-current network, and relates to the technical field of photovoltaic power generation, and the method comprises the steps: obtaining historical photovoltaic power generation power data and meteorological factors; performing data cleaning on the historical photovoltaic power generation power data; optimizing the VMD by using the WOA to obtain the optimal number of solid-state mode function components and decomposition penalty factors, and decomposing the preprocessed power generation power data by using the VMD to obtain the corresponding number of solid-state mode function components according to the optimal number of the solid-state mode function components and the decomposition penalty factors; taking the solid-state mode function components and the meteorological factors as input, and utilizing a pre-constructed BiGRU model and a pre-constructed GCN model to carry out prediction to obtain predicted values corresponding to the solid-state mode function components; and superposing the predicted values of the solid-state mode function components to obtain a final photovoltaic power generation power predicted value. According to the scheme, the problem that the photovoltaic power generation power prediction accuracy is low due to the fact that the original line load power is unstable and data relevance exists can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and in particular to a method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network. Background Art

[0002] Photovoltaic power generation is a flexible and sustainable distributed power source that plays an important role in the global clean energy demand and has significant economic and environmental advantages. At present, as a major form of solar energy utilization, the power generation process of photovoltaic power generation is affected by meteorological factors such as radiation intensity, wind speed, wind direction, temperature, and humidity, making it highly random, volatile, and uncertain. These characteristics have brought great challenges to the safe and stable operation of active distribution networks. Accurately predicting photovoltaic power is a key way to ensure the safe and stable operation of active distribution networks, reasonably arrange energy scheduling, and improve photovoltaic power consumption.

[0003] According to the different time scales of photovoltaic power prediction, photovoltaic power prediction can be divided into three prediction time scales: long-term prediction, medium-term prediction and short-term prediction. Since short-term prediction includes general types such as economic dispatch, load control and real-time dispatch, the present invention focuses on short-term prediction.

[0004] In recent years, with the development of artificial intelligence, it has been widely used in the field of photovoltaic power generation prediction. However, the current prediction of photovoltaic power generation based on machine learning relies heavily on feature engineering, which increases the size and time complexity of the model, and the performance will decrease when the data size and prediction steps increase. Moreover, in practice, the original line load power is unstable, and there is a correlation between the various collected data, which leads to the low accuracy of the existing photovoltaic power generation prediction. Summary of the invention

[0005] In view of this, in order to address the above shortcomings, it is necessary to propose a photovoltaic power prediction method for distribution network based on WOA-VMD and dual-flow network to solve the problem of low photovoltaic power prediction accuracy caused by the instability of original line load power and the correlation between data.

[0006] The present invention provides a method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network, comprising:

[0007] Obtain historical photovoltaic power data and meteorological factors passing through the distribution network;

[0008] Performing data cleaning on the historical photovoltaic power generation data to obtain pre-processed power generation data;

[0009] The variational mode decomposition algorithm VMD is optimized using the whale optimization algorithm WOA to obtain the optimal number of solid-state mode function components and decomposition penalty factor;

[0010] According to the optimal number of solid-state mode function components and the decomposition penalty factor, a variational mode decomposition algorithm VMD is applied to decompose the preprocessed power generation data to obtain solid-state mode function components whose number is the optimal number of solid-state mode function components;

[0011] Taking the solid state mode function components and the meteorological factors as input, using the pre-built BiGRU model and GCN model to perform prediction, and obtaining the prediction value corresponding to each solid state mode function component;

[0012] The predicted values ​​of the solid-state mode function components are superimposed to obtain a final predicted value of photovoltaic power generation of the distribution network.

[0013] Preferably, the data cleaning of the historical photovoltaic power generation data includes:

[0014] For missing data, repeated data and jump data, the average value of several generated powers near the data is used to replace it.

[0015] Preferably, the optimization of the variational mode decomposition algorithm VMD by using the whale optimization algorithm WOA includes:

[0016] Parameter initialization: Initialize the position, number of iterations and number of whale individual populations of the whale optimization algorithm WOA; the position of the initialized whale individual population vector contains the initial number of solid mode function components and the decomposition penalty factor, and the position of the i-th individual is expressed as follows:

[0017] X i = r·(u b -l b )+l b

[0018] In the formula, r∈[0,1] is a random number, X i The value range is [l b ,u b ],l b is the minimum value of the boundary, u b is the maximum value of the boundary;

[0019] Contraction and encirclement iteration: When the probability of whale predation mechanism p < 0.5 and |A| ≤ 1, the iterative update is expressed as follows:

[0020]

[0021] In the formula, p is a random number in the range of [0,1], t is the current iteration number, and t max is the maximum number of iterations, a is the penalty factor for whale optimization, is the solution position vector currently generated, is the optimal solution position vector, and is the coefficient vector, represents the distance vector from the whale to the prey;

[0022] When the probability of whale predation mechanism p < 0.5 and |A| > 1, the iterative update is expressed as follows:

[0023]

[0024] In the formula, is the randomly selected whale position vector;

[0025] When the probability of whale predation mechanism p≥0.5, the spiral contraction method is adopted, and the iterative update is expressed as follows:

[0026]

[0027] In the formula, To simulate the distance between the whale group and the prey, l is a random number (-1,1), b is the spiral constant, and e is a natural number;

[0028] Iteration termination: the shrinking and encircling iteration steps are executed cyclically until the maximum number of iterations is reached so that the whale optimization penalty factor is reduced to 0, thereby obtaining the optimal number of solid-state mode function components and the optimal decomposition penalty factor.

[0029] Preferably, the applying a variational mode decomposition algorithm VMD to decompose the preprocessed power generation data according to the optimal number of solid-state mode function components and the decomposition penalty factor comprises:

[0030] For the pre-processed power generation data, assuming that each mode has a finite bandwidth with a center frequency, construct a decomposition model for finding a minimum sum of the fixed bandwidths of each mode;

[0031] The optimal penalty factor and Lagrange multiplication operator obtained after optimization are introduced to solve the decomposition model; wherein, in the solution process, the multiplier alternating direction method is used to iterate the parameter update;

[0032] After each iterative solution, determine whether the accuracy requirement is met; if so, stop the iteration and obtain the solid mode function component; otherwise, return to execute the parameter update iteration process.

[0033] Preferably, the decomposition model is expressed as follows:

[0034]

[0035] Where K is the number of optimal solid-state mode function components obtained after optimization, u k ,ω k are the kth mode component and center frequency after decomposition, δ(t) is the Dirac function, * is the convolution operation, The pre-processed power generation data.

[0036] Preferably, the introducing the optimal penalty factor and Lagrange multiplication operator obtained after optimization to solve the decomposition model includes:

[0037] By introducing the optimal decomposition penalty factor and Lagrange multiplication operator obtained after optimization, the constrained problem is transformed into an unconstrained problem, and the following augmented Lagrangian expression is obtained:

[0038]

[0039] In the formula, α is the optimal decomposition penalty factor, and λ is the Lagrange multiplication operator;

[0040] By the following calculation formula, k ,ω k ,λ is iteratively updated:

[0041]

[0042] In the formula, They are is the Fourier transform of , n is the number of iterations, and γ is the noise tolerance required to meet the fidelity requirements of signal decomposition.

[0043] Preferably, the determining whether the accuracy requirement is met includes:

[0044] Determine whether the following calculation formula is satisfied. If so, the accuracy requirement is met, otherwise it is not satisfied:

[0045]

[0046] In the formula, ε is the preset accuracy.

[0047] Preferably, the method of using the solid state mode function component and the meteorological factor as input and using a pre-built BiGRU model and a GCN model to perform prediction includes:

[0048] Each solid state mode function component and the meteorological factor are used as inputs of the BiGRU model to obtain a time series information matrix;

[0049] The adjacency matrix and the feature matrix are used as inputs of the GCN model to obtain a spatial information matrix; wherein the adjacency matrix is ​​a connection matrix based on the relationship between meteorological factors and each solid state mode function component, and the feature matrix is ​​a matrix formed by regarding the values ​​of meteorological factors and each solid state mode function component as node attribute features;

[0050] The time series information matrix and the spatial information matrix are used as inputs, and the predicted values ​​of each solid state mode function component are obtained through the output of the fully connected layer.

[0051] Preferably, the method for determining the adjacency matrix includes:

[0052] The correlation coefficients between each solid state model function component and meteorological factors are calculated using the following formula:

[0053]

[0054] Among them, r i,j The correlation coefficient between the i-th solid state model function component and the j-th meteorological factor, IMF i Characterizes the i-th solid state mode function component, Y j Characterizes the jth meteorological factor, Cov(IMF i ,Y j ) is used to characterize the IMF i With Y j The covariance of Ver[IMF i ] and Ver[Y j ] respectively IMF i With Y j The variance of

[0055] The adjacency matrix is ​​obtained according to the correlation coefficients between each solid state mode function component and the meteorological factor.

[0056] Preferably, the step of using the adjacency matrix and the feature matrix as inputs of the GCN model to obtain the spatial information matrix includes obtaining the spatial information matrix using the following calculation formula:

[0057]

[0058] Wherein, A represents the adjacency matrix, The adjacency matrix after characterization, I N Represents the N-order identity matrix, H (l) represents the network output of layer l, and H (0) represents the feature matrix, σ represents the Sigmoid activation function, W (l) represents the parameters to be trained, for The degree matrix of .

[0059] It can be seen from the above technical scheme that in the WOA-WMD and dual-flow network distribution network photovoltaic power prediction method provided by the present invention, firstly, the historical photovoltaic power data and meteorological factors passing through the distribution network are obtained, and then the historical photovoltaic power data are cleaned, and the variational mode decomposition algorithm is further optimized by using the whale optimization algorithm WOA to obtain the optimal number of solid-state mode function components and decomposition penalty factors, and then the pre-processed power generation data can be decomposed using the optimal number of solid-state mode function components and decomposition penalty factors; further, the decomposed solid-state mode function components and meteorological factors are used as input, and the BiGRU model and the GCN model are used to predict the predicted values ​​corresponding to each solid-state mode function component, and the final distribution network photovoltaic power prediction value can be obtained by superimposing the predicted values ​​of each solid-state mode function component. It can be seen that this scheme uses the whale optimization algorithm WOA and the variational mode decomposition algorithm VMD to optimize and decompose the data during data processing, which effectively improves the instability of the load power in the original line and fully considers the impact of meteorological factors on the power generation. By using the GCN-BiGRU dual-flow network to predict photovoltaic power generation, the accuracy of the distribution network power generation prediction is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart of a method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network is provided in an embodiment of the present invention.

[0061] Figure 2 A prediction schematic diagram based on a BiGRU model and a GCN model is provided for an embodiment of the present invention.

[0062] Figure 3 This is a performance comparison chart of the method provided by the present invention and some similar methods when performing predictions. DETAILED DESCRIPTION

[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0064] like Figure 1 As shown, an embodiment of the present invention provides a method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network, which may include the following steps:

[0065] Step 101: Acquire historical photovoltaic power generation data and meteorological factors passing through the distribution network;

[0066] Step 102: cleaning the historical photovoltaic power generation data to obtain pre-processed power generation data;

[0067] Step 103: Optimizing the variational mode decomposition algorithm VMD using the whale optimization algorithm WOA to obtain the optimal number of solid mode function components and decomposition penalty factor;

[0068] Step 104: according to the optimal number of solid-state mode function components and the decomposition penalty factor, the preprocessed power generation data is decomposed using the variational mode decomposition algorithm VMD to obtain solid-state mode function components whose number is the optimal number of solid-state mode function components;

[0069] Step 105: using the solid state model function components and meteorological factors as input, using the pre-built BiGRU model and GCN model to perform predictions, and obtaining the prediction values ​​corresponding to each solid state model function component;

[0070] Step 106: superimpose the predicted values ​​of the functional components of each solid state mode to obtain the final predicted value of photovoltaic power generation of the distribution network.

[0071] In this embodiment, the whale optimization algorithm WOA and the variational mode decomposition algorithm VMD are used to optimize and decompose the data during data processing, which effectively improves the instability of the load power in the original line and fully considers the impact of meteorological factors on the power generation power. By using the GCN-BiGRU dual-flow network to predict photovoltaic power generation, the accuracy of the distribution network power generation prediction is improved.

[0072] For step 101, historical photovoltaic power generation data and meteorological factors passing through the distribution network are obtained;

[0073] In this embodiment, the historical photovoltaic power generation data sequence X(t) passing through the distribution network is obtained, and the historical data of the meteorological factor NWP data in the photovoltaic power station is obtained, such as light, humidity, temperature, etc.

[0074] In step 102, historical photovoltaic power generation data is cleaned to obtain pre-processed power generation data;

[0075] Since the acquired historical photovoltaic power generation data may be missing, jump, or repeated, it is considered to clean up the historical photovoltaic power generation data. Specifically, the missing data, repeated data, and jump data can be replaced with the average value of several power generation data near the data, thereby obtaining the preprocessed power generation data X~(t).

[0076] For step 103, the variational mode decomposition algorithm VMD is optimized using the whale optimization algorithm WOA to obtain the optimal number of solid-state mode function components and decomposition penalty factor;

[0077] In this embodiment, WOA is considered to optimize the data of photovoltaic power decomposed by VMD to obtain the best K solid-state mode function components IMF and decomposition penalty factor α, that is, to obtain the best parameters K and α.

[0078] Specifically, the parameters are initialized first. For example, the position, number of iterations and number of whale individual populations of the whale optimization algorithm WOA are initialized; the position of the initialized whale individual population vector contains the initial number of solid mode function components and the decomposition penalty factor, that is, the initial position of the whale individual population vector is given [initial IMF component number K0, initial decomposition penalty factor α0]. The i-th individual position can be expressed as follows:

[0079] X i = r·(u b -l b )+l b

[0080] In the formula, r∈[0,1] is a random number, X i The value range is [l b ,u b ],l b is the minimum value of the boundary, u b is the maximum value of the boundary;

[0081] Then, the iterative update is performed through the shrinking and surrounding mechanism. Specifically, when the probability of the whale predation mechanism p < 0.5 and |A| ≤ 1, the shrinking and surrounding iterative update can be expressed as follows:

[0082]

[0083] In the formula, p is a random number in the range of [0,1], t is the current iteration number, and t max is the maximum number of iterations, a is the whale optimization penalty factor, which decreases linearly from 2 to 0 during the iteration process. is the solution position vector currently generated, is the optimal solution position vector, and is the coefficient vector, represents the distance vector from the whale to the prey;

[0084] When the probability of whale predation mechanism p < 0.5 and |A| > 1, the iterative update is expressed as follows:

[0085]

[0086] In the formula, is the randomly selected whale position vector;

[0087] When the probability of whale predation mechanism p≥0.5, the spiral contraction method is adopted, and the iterative update is expressed as follows:

[0088]

[0089] In the formula, To simulate the distance between the whale group and the prey, l is a random number (-1,1), b is the spiral constant, and e is a natural number;

[0090] Finally, it is determined whether the iteration termination condition is met. Specifically, the shrinking and encircling iteration steps are executed cyclically until the maximum number of iterations is reached so that the whale optimization penalty factor a is reduced to 0, and the optimal number of solid mode function components K and the optimal decomposition penalty factor α are obtained.

[0091] In step 104, according to the optimal number of solid-state mode function components and the decomposition penalty factor, the preprocessed power generation data is decomposed using a variational mode decomposition algorithm VMD to obtain solid-state mode function components having the optimal number of solid-state mode function components;

[0092] In this step, after obtaining the optimal number of solid-state mode function components K and the optimal decomposition penalty factor α through step 103, consider using the optimal parameters to apply VMD to decompose the preprocessed power generation data to obtain K IMF components representing the characteristics of photovoltaic power generation data.

[0093] Specifically, step 104 can be implemented in the following manner:

[0094] For the pre-processed power generation data, assuming that each mode has a finite bandwidth with a center frequency, a decomposition model is constructed to find the minimum sum of the fixed bandwidths of each mode;

[0095] The optimal penalty factor and Lagrange multiplication operator obtained after optimization are introduced to solve the decomposition model; in the solution process, the multiplier alternating direction method is used to iterate the parameter update;

[0096] After each iterative solution, determine whether the accuracy requirement is met; if so, stop the iteration and obtain the solid mode function component; otherwise, return to execute the parameter update iteration process.

[0097] In this embodiment, first, for the pre-processed power generation data Assuming that each mode has a finite bandwidth with a center frequency, we now look for a mode that minimizes the sum of the estimated bandwidths of each mode. The specific model can be expressed as follows:

[0098]

[0099] Where K is the number of optimal solid-state mode function components obtained after optimization, u k ,ω k are the kth mode component and center frequency after decomposition, δ(t) is the Dirac function, * is the convolution operation, To pre-process the power generation data.

[0100] Secondly, to solve the above model, consider introducing the optimal decomposition penalty factor α obtained after WOA optimizes VMD to reduce the impact of Gaussian noise. At the same time, the Lagrangian multiplication operator λ is introduced to transform the constrained problem into an unconstrained problem, and the following augmented Lagrangian expression is obtained:

[0101]

[0102] In the formula, α is the optimal decomposition penalty factor, and λ is the Lagrange multiplication operator;

[0103] The parameter u is further updated iteratively by the multiplier alternating direction method k ,ω k ,λ, the specific update process is as follows:

[0104]

[0105] In the formula, They are u i (t),λ(t), is the Fourier transform of , n is the number of iterations, and γ is the noise tolerance required to meet the fidelity requirements of signal decomposition.

[0106] Finally, it is determined whether the accuracy requirement is met. Specifically, for a given iteration accuracy ε>0, if If the accuracy requirement is met, the iteration is stopped and the solid state mode function component is obtained. Otherwise, the iteration accuracy requirement is not met and the execution returns to u k ,ω k , the iterative process of updating the λ parameter.

[0107] For step 105, the solid state mode function components and meteorological factors are used as inputs, and the pre-built BiGRU model and GCN model are used for prediction to obtain the prediction values ​​corresponding to each solid state mode function component;

[0108] In this step, we consider constructing a GCN model and a BiGRU model respectively, and then use the two models to predict the solid mode function component IMF to obtain the predicted value. Figure 2 As shown in the schematic diagram, this step outputs the time series information matrix through BiGRU, outputs the spatial information matrix through GCN, and then converts it into a vector through the fully connected layer. Finally, the number of output variables is set to output the corresponding number of predicted values. Specifically, step 105 can be implemented as follows:

[0109] Each solid state mode function component and meteorological factor are used as the input of the BiGRU model to obtain the time series information matrix;

[0110] The adjacency matrix and the feature matrix are used as the input of the GCN model to obtain the spatial information matrix; the adjacency matrix is ​​a connection matrix based on the relationship between meteorological factors and each solid state mode function component, and the feature matrix is ​​a matrix that regards the values ​​of meteorological factors and each solid state mode function component as node attribute features;

[0111] The time series information matrix and the spatial information matrix are taken as input, and the predicted values ​​of each solid state mode function component are obtained through the output of the fully connected layer.

[0112] In this embodiment, the BiGRU model is considered to be used to extract the time series information of photovoltaic power generation. Specifically, when the BiGRU model is used to obtain the time series information matrix, the BiGRU model corresponding to time t can be expressed as follows:

[0113] Reset gate: r t =σ(W r ·[h t-1 ,x t ])

[0114] Update gate: z t =σ(W z ·[h t-1 ,x t ])

[0115] Hidden state:

[0116] Output Gate:

[0117] In the formula, r t is the reset gate, z t is the update gate, x t is the current input, which is a sequence of solid state model function components IMF and meteorological factors. is the current input and the previous hidden layer state summary, h t is the hidden layer output after updating the memory, W ris the weight matrix of the reset gate, W z is the weight matrix of the update gate, W h is the weight matrix of the hidden layer, σ is the Sigmoid function, and tanh is the hyperbolic tangent activation function.

[0118] Furthermore, we consider using the GCN model to extract the spatial correlation characteristics of meteorological data and IMF components. Specifically, when determining the adjacency matrix, it can be obtained as follows:

[0119] First, the Pearson correlation coefficient is used to calculate the correlation coefficient r for each solid model function component IMF and meteorological factor. The specific calculation formula is as follows:

[0120]

[0121] Among them, r i,j The correlation coefficient between the i-th solid state model function component and the j-th meteorological factor, IMF i Characterize the i-th solid mode function component, i = 1, 2, ..., K, Y j represents the jth meteorological factor, j = 1, 2, ..., M, M is the total number of meteorological factors, Cov(IMF i ,Y j ) is used to characterize the IMF i With Y j The covariance of Ver[IMF i ] and Ver[Y j ] respectively IMF i With Y j The variance of

[0122] Then, according to the correlation coefficient between each solid state model function component and the meteorological factor, the adjacency matrix representing the spatial correlation between the meteorological factor and the solid state model function component IMF is constructed.

[0123] The construction operation formula for the graph in the GCN model can be expressed as:

[0124] G=(V,E)

[0125] V=(NWP,IMF1,IMF2,…,IMF K )

[0126] Where V represents the meteorological factors and the IMF component sequences, and E represents the relationship set between each meteorological factor and the IMF component node;

[0127] The connection matrix of meteorological factors and IMF components in the photovoltaic power station is represented by an N×N dimensional matrix A, which is called the adjacency matrix; the values ​​of meteorological factors and IMF components in the photovoltaic power station are regarded as node attribute features, and the N×F dimensional feature matrix X is constructed, where F represents the time series length of meteorological factors and IMF components. In this way, the spatial information matrix can be obtained using the following calculation formula:

[0128]

[0129] Among them, A represents the adjacency matrix, The adjacency matrix after characterization, I N Represents the N-order identity matrix, H (l) represents the network output of layer l, and H (0) represents the feature matrix X, σ represents the Sigmoid activation function, W (l) represents the parameters to be trained, for The degree matrix of .

[0130] Furthermore, after obtaining the time series information matrix and the spatial information matrix, they are used as inputs and the predicted values ​​of each solid state mode function component are obtained through the output of the fully connected layer.

[0131] For example, assuming that the number of IMF components is 5, then:

[0132] The first input: the input received by the BiGRU layer: light, humidity..., IMF1, and the time series information matrix is ​​obtained; GCN input: adjacency matrix A, the information contained in the feature matrix X includes light, humidity..., IMF1, and the spatial information matrix is ​​obtained.

[0133] Second input: Input received by the BiGRU layer: light, humidity, ..., IMF2, and the time series information matrix is ​​obtained; GCN input: adjacency matrix A, the information contained in the feature matrix X includes light, humidity, ..., IMF2, and the spatial information matrix is ​​obtained.

[0134] And so on.

[0135] The training is completed after 5 times, and then the spatial information matrix and time series information matrix of each component are input into the fully connected layer to obtain the predicted values ​​of each IMF component.

[0136] For step 106, the predicted values ​​of the functional components of each solid state mode are superimposed to obtain the final predicted value of photovoltaic power generation of the distribution network.

[0137] In this step, by superimposing various IMF components, the final predicted value of the photovoltaic power generation power of the distribution network can be obtained.

[0138] The effect of this scheme is explained below with reference to specific experiments.

[0139] The 2019-2020 photovoltaic power station data of a local distribution network is used, with a resolution of 15 minutes, and meteorological factors include temperature, humidity, rainfall, etc. The model is trained and constructed based on this solution, with a training set of 7000, a validation set of 2000, and a test set of 1000. As shown in Table 1 and Figure 3 The following is a comparison of the error indicators of this scheme and similar algorithms:

[0140] Table 1

[0141]

[0142]

[0143] In the table, WOA-VMD-DSN is the solution provided by the present invention, RMSE, MAE and R 2 They are the error evaluation indicators root mean square error, mean absolute error and R 2 Coefficient of determination.

[0144] From Table 1 and Figure 3 It can be seen that the error index of the method proposed in the present invention is significantly better than that of the models in other control groups, which shows that the present scheme has good performance and is better than similar algorithms.

[0145] The present specification also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed in a computer, the computer is caused to execute a method in any one of the embodiments in the specification.

[0146] The present specification also provides a computing device, including a memory and a processor, wherein executable codes are stored in the memory, and when the processor executes the executable codes, a method in any embodiment of the present specification is implemented.

[0147] Since the device embodiments provided by the present invention are based on the same inventive concept as the method embodiments of this specification, the specific contents can be found in the description of the method embodiments of this specification and will not be repeated here.

[0148] The modules or units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. The above disclosure is only the preferred embodiment of the present invention, and of course it cannot be used to limit the scope of the rights of the present invention. Those skilled in the art can understand that all or part of the processes of the above embodiment are implemented, and the equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A photovoltaic power generation prediction method for distribution network based on WOA-VMD and dual-flow network, characterized in that: include: Obtain historical photovoltaic power data and meteorological factors passing through the distribution network; Performing data cleaning on the historical photovoltaic power generation data to obtain pre-processed power generation data; The variational mode decomposition algorithm VMD is optimized using the whale optimization algorithm WOA to obtain the optimal number of solid-state mode function components and decomposition penalty factor; According to the optimal number of solid-state mode function components and the decomposition penalty factor, a variational mode decomposition algorithm VMD is applied to decompose the preprocessed power generation data to obtain solid-state mode function components whose number is the optimal number of solid-state mode function components; Taking the solid state mode function components and the meteorological factors as input, using the pre-built BiGRU model and GCN model to perform prediction, and obtaining the prediction value corresponding to each solid state mode function component; The predicted values ​​of the solid-state mode function components are superimposed to obtain a final predicted value of photovoltaic power generation of the distribution network.

2. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network according to claim 1 is characterized in that: The data cleaning of the historical photovoltaic power generation data includes: For missing data, repeated data and jump data, the average value of several generated powers near the data is used to replace it.

3. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and dual-flow network according to claim 1, characterized in that: The method of optimizing the variational mode decomposition algorithm VMD by using the whale optimization algorithm WOA includes: Parameter initialization: Initialize the position, number of iterations and number of whale individual populations of the whale optimization algorithm WOA; the position of the initialized whale individual population vector contains the initial number of solid mode function components and the decomposition penalty factor, and the position of the i-th individual is expressed as follows: X i =r·(u b -l b )+l b In the formula, r∈[0,1] is a random number, X i The value range is [l b ,u b ],l b is the minimum value of the boundary, u b is the maximum value of the boundary; Contraction and encirclement iteration: When the probability of whale predation mechanism p < 0.5 and |A| ≤ 1, the iterative update is expressed as follows: In the formula, are all random numbers, and the value range is [0,1], t is the current iteration number, t max is the maximum number of iterations, a is the penalty factor for whale optimization, is the solution position vector currently generated, is the optimal solution position vector, and is the coefficient vector, represents the distance vector from the whale to the prey; When the whale predation mechanism probability p < 0.5 and |A| > 1, the iterative update is expressed as follows: In the formula, is the randomly selected whale position vector; When the probability of whale predation mechanism p≥0.5, the spiral contraction method is adopted, and the iterative update is expressed as follows: In the formula, To simulate the distance between the whale group and the prey, l is a random number (-1,1), b is the spiral constant, and e is a natural number; Iteration termination: the shrinking and encircling iteration steps are executed cyclically until the maximum number of iterations is reached so that the whale optimization penalty factor is reduced to 0, thereby obtaining the optimal number of solid-state mode function components and the optimal decomposition penalty factor.

4. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and dual-flow network according to claim 1, characterized in that: The step of applying a variational mode decomposition algorithm VMD to decompose the preprocessed power generation data according to the optimal number of solid-state mode function components and the decomposition penalty factor comprises: For the pre-processed power generation data, assuming that each mode has a finite bandwidth with a center frequency, construct a decomposition model for finding a minimum sum of the fixed bandwidths of each mode; The optimal penalty factor and Lagrange multiplication operator obtained after optimization are introduced to solve the decomposition model; wherein, in the solution process, the multiplier alternating direction method is used to iterate the parameter update; After each iterative solution, determine whether the accuracy requirement is met; if so, stop the iteration and obtain the solid mode function component; otherwise, return to execute the parameter update iteration process.

5. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network according to claim 4 is characterized in that: The decomposition model is expressed as follows: Where K is the number of optimal solid-state mode function components obtained after optimization, u k ,ω k are the kth mode component and center frequency after decomposition, δ(t) is the Dirac function, * is the convolution operation, The pre-processed power generation data.

6. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network according to claim 5, characterized in that: The optimal penalty factor and Lagrange multiplication operator obtained after the optimization are introduced to solve the decomposition model, including: By introducing the optimal decomposition penalty factor and Lagrangian multiplication operator obtained after optimization, the constrained problem is transformed into an unconstrained problem, and the following augmented Lagrangian expression is obtained: In the formula, α is the optimal decomposition penalty factor, and λ is the Lagrange multiplication operator; By the following calculation formula, k ,ω k ,λ is iteratively updated: In the formula, They are is the Fourier transform of , n is the number of iterations, and γ is the noise tolerance required to meet the fidelity requirements of signal decomposition.

7. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network according to claim 6, characterized in that: The determining whether the accuracy requirement is met includes: Determine whether the following calculation formula is satisfied. If so, the accuracy requirement is met, otherwise it is not satisfied: In the formula, ε is the preset accuracy.

8. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and dual-flow network according to claim 1, characterized in that: The method uses the solid state mode function component and the meteorological factor as input and uses the pre-built BiGRU model and GCN model to make predictions, including: Each solid state mode function component and the meteorological factor are used as inputs of the BiGRU model to obtain a time series information matrix; The adjacency matrix and the feature matrix are used as inputs of the GCN model to obtain a spatial information matrix; wherein the adjacency matrix is ​​a connection matrix based on the relationship between meteorological factors and each solid state mode function component, and the feature matrix is ​​a matrix formed by regarding the values ​​of meteorological factors and each solid state mode function component as node attribute features; The time series information matrix and the spatial information matrix are used as inputs, and the predicted values ​​of each solid state mode function component are obtained through the output of the fully connected layer.

9. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network according to claim 8, characterized in that: The method for determining the adjacency matrix comprises: The correlation coefficients between each solid state model function component and meteorological factors are calculated using the following formula: Among them, r i,j The correlation coefficient between the i-th solid state model function component and the j-th meteorological factor, IMF i Characterizes the i-th solid state mode function component, Y j Characterizes the jth meteorological factor, Cov(IMF i ,Y j ) is used to characterize the IMF i With Y j The covariance of Ver[IMF i ] and Ver[Y j ] respectively IMF i With Y j The variance of The adjacency matrix is ​​obtained according to the correlation coefficients between each solid state mode function component and the meteorological factor.

10. The method for predicting photovoltaic power generation in a distribution network based on WOA-VMD and a dual-flow network according to claim 9, characterized in that: The adjacency matrix and the feature matrix are used as inputs of the GCN model to obtain the spatial information matrix, including using the following calculation formula to obtain the spatial information matrix: Wherein, A represents the adjacency matrix, A~ represents the processed adjacency matrix, and I N Represents the N-order identity matrix, H (l) represents the network output of layer l, and H (0) represents the feature matrix, σ represents the Sigmoid activation function, W (l) represents the parameters to be trained, and D~ is the degree matrix of A~.

Images