Photovoltaic power generation power prediction method based on HGS-VMD-ENN ultra-short-term chaos prediction model

Abstract

The invention relates to a photovoltaic power generation power prediction method based on an HGS-VMD-ENN ultra-short-term chaos prediction model, and the method comprises the steps: designing a fitness function which considers weighted permutation entropy and decomposition loss, taking the value minimization of the fitness function as an optimization target, employing HGS-VMD to adaptively determine a VMD optimal decomposition parameter combination, and carrying out the optimization of the VMD optimal decomposition parameter combination. The multi-scale decomposition module is used for performing multi-scale decomposition on photovoltaic power generation power to obtain photovoltaic power generation power decomposition components; according to the photovoltaic power generation power decomposition component, a phase space matrix of the photovoltaic power generation power decomposition component is constructed by adopting a PSR phase space reconstruction method, and an HGS-VMD-ENN ultra-short-term chaos prediction model is constructed in combination with a brain emotion neural network; and according to the established HGS-VMD-ENN ultra-short-term chaos prediction model, photovoltaic power ultra-short-term prediction is carried out. Compared with the prior art, the method can solve the problems of large randomness and poor prediction precision caused by traditional VMD empirical parameter selection, and improves the stability and precision of photovoltaic power generation power prediction.

Description

Technical Field

[0001] The invention relates to the technical field of photovoltaic power generation prediction, and in particular to a photovoltaic power generation power prediction method based on a HGS-VMD-ENN ultra-short-term chaos prediction model. Background Art

[0002] In recent years, with the acceleration of energy transformation, the installed capacity of solar power generation is expected to exceed 800 million kilowatts in 2030. At present, the frequent occurrence of power mutation events and the intermittent and volatile nature of photovoltaic power generation will inevitably bring severe challenges to the safe and economic operation of the power system after the grid is connected to a high proportion of photovoltaic power. Therefore, improving the ultra-short-term prediction accuracy of photovoltaic power generation is of great significance to the safe and stable operation of the power grid, optimized dispatching management and the balance of power supply and demand.

[0003] In existing research, scholars at home and abroad use signal decomposition technology to reduce the volatility of photovoltaic power time series to improve prediction accuracy. This method has been widely used in photovoltaic power prediction. Among many signal decomposition technologies, VMD (variational mode decomposition) is a completely non-recursive signal processing method. It decomposes complex signals into multiple relatively stable decomposition components by iteratively searching for the optimal solution of variational mode. Its decomposition effect is better than other decomposition methods such as empirical mode decomposition and wavelet decomposition. The combination of VMD key parameters (K, α) is crucial to the VMD decomposition effect, and the VMD decomposition components and residual components directly affect the final prediction results of the model. However, in practical applications, the traditional experience of setting VMD parameters is highly subjective. When the subsequence frequencies are similar, it is difficult to effectively determine the most appropriate decomposition parameters, resulting in poor prediction accuracy. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a photovoltaic power generation prediction method based on the HGS (hunger games search)-VMD-ENN (emotional neural network) ultra-short-term chaotic prediction model, which can solve the problems of large randomness and poor prediction accuracy caused by the selection of traditional VMD empirical parameters.

[0005] The purpose of the present invention can be achieved by the following technical scheme: A photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model comprises the following steps:

[0006] S1. Design a fitness function that takes weighted permutation entropy and decomposition loss into consideration, and take the minimum value of the fitness function as the optimization goal. Adopt HGS-VMD to adaptively determine the best VMD decomposition parameter combination, which is used to perform multi-scale decomposition of photovoltaic power generation to obtain photovoltaic power generation power decomposition components.

[0007] S2. According to the decomposition components of photovoltaic power generation, the phase space matrix is ​​constructed by using the PSR (phase space reconstruction) method, and the HGS-VMD-ENN ultra-short-term chaos prediction model is constructed in combination with the brain emotion neural network;

[0008] S3. Based on the established HGS-VMD-ENN ultra-short-term chaos prediction model, ultra-short-term prediction of photovoltaic power is performed.

[0009] Furthermore, the step S1 specifically adopts the entropy weight method to fuse the weighted permutation entropy and the decomposition loss as a new fitness function of the HGS-VMD multi-scale decomposition, which can reduce the influence of the decomposition loss on the prediction result while ensuring the regularity of the decomposition components.

[0010] Furthermore, the fitness function in step S1 is specifically:

[0011]

[0012] Among them, Y ij is the indicator matrix composed of weighted permutation entropy and decomposition loss, P ij is the proportion of the jth indicator under the indicator matrix Y.

[0013] Furthermore, the process of using HGS-VMD to adaptively determine the optimal decomposition parameter combination of VMD in step S1 includes:

[0014] S11, using HGS algorithm to solve VMD constrained optimization problem, taking the minimum value of fitness function as optimization goal, determining the best decomposition parameters of VMD through HGS-VMD adaptive parameter optimization, and calculating the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2 of each individual parameter combination according to the fitness value after VMD decomposition;

[0015] S12. Update the position of individual parameter combinations according to the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2. The algorithm terminates after the maximum number of iterations. The current output is the VMD optimal decomposition parameter combination (K, α).

[0016] Furthermore, the formulas for calculating the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2 of each individual parameter combination in step S11 are respectively:

[0017] VC=sech(F(i)-DF)

[0018]

[0019] W1=(1-exp(-|hungry(i)-SH|))×r×2

[0020]

[0021] Among them, F(i) is the fitness value of the individual parameter combination, DF is the optimal fitness value, sech(x)=2 / (e x +e -x ) is a hyperbolic function, b = 100*(1+r), r is a random number in [0,1], SH is the sum of the hunger levels of all individual parameter combinations, and N is the total number of individuals.

[0022] Furthermore, the formula for updating the position of the individual parameter combination in step S12 is:

[0023]

[0024] Among them, TP is the optimal parameter position in the iteration process, vb=2*shrink*r-shrink, and shrink is the convergence factor.

[0025] Furthermore, the step S2 comprises the following steps:

[0026] S21, reconstructing the phase space of the one-dimensional photovoltaic power generation time series g(t), and obtaining the phase space matrix according to the embedding dimension m and the time delay τ;

[0027] S22. Divide the sample sets for model training and testing, use the embedding dimension m as the number of ENN model input layers, use the phase space matrix G as the ENN model input matrix, train the model training sample set through ENN, and then use the test sample set to perform model simulation verification to construct the HGS-VMD-ENN ultra-short-term chaos prediction model.

[0028] Furthermore, the expression of the phase space matrix in step S21 is:

[0029]

[0030] Wherein, M is the number of delay vectors, M=N-mτ+τ.

[0031] Furthermore, the process of performing ultra-short-term photovoltaic power prediction in step S3 includes: stimulating the input signal G n Entering the thalamus, n represents the number of columns of the input reconstruction matrix, 1≤n≤m, the amygdala receives the input signal SC from the sensory cortex n , and receives the extended signal G from the thalamus m+1 , combined with the outputs of the amygdala and orbitofrontal cortex, the final output value is obtained, which is the photovoltaic power prediction result.

[0032] Furthermore, the extended signal G from the thalamus m+1 Specifically:

[0033] G m+1 =max(G n )

[0034] The final output value is:

[0035] E=E a -E o

[0036]

[0037] Among them, E a and E o It is the output of the amygdala and orbitofrontal cortex link.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] The present invention uses the Hunger Game search algorithm HGS for variational mode decomposition VMD core parameter optimization, and designs an HGS-VMD fitness function that considers weighted permutation entropy and decomposition loss to decompose the photovoltaic signal into a series of relatively stable decomposition components; then the phase space reconstruction PSR is used to map each decomposition component to a high-dimensional phase space to construct the optimal number of input nodes of the brain emotion neural network ENN model, and the prediction results of each decomposition component are superimposed to obtain the final prediction value. Thus, an HGS-VMD-ENN ultra-short-term chaos prediction model is established, which is used for ultra-short-term prediction of photovoltaic power, and can solve the problems of large randomness and poor prediction accuracy caused by the selection of traditional VMD empirical parameters, and improve the prediction accuracy.

[0040] The present invention adopts the entropy weight method to fuse weighted permutation entropy and decomposition loss, and uses this as a new fitness function of HGS-VMD multi-scale decomposition, and uses the HGS algorithm to solve the VMD constraint optimization problem, takes the minimum value of the new fitness function as the optimization goal, and determines the VMD optimal decomposition parameter through HGS-VMD adaptive parameter optimization, which is used to perform multi-scale decomposition of photovoltaic power generation. Thus, the photovoltaic power generation is decomposed by the data preprocessing method of HGS-VMD multi-scale decomposition, which can improve the regularity of VMD decomposition components while reducing the influence of residual components on prediction results and reducing volatility during prediction.

[0041] The present invention utilizes the HGS-VMD method to decompose photovoltaic power generation into different frequency components, and respectively models them for prediction, reconstructs the phase space of each decomposed component, extracts a regular motion trajectory from the intricate time series, and constructs the optimal number of input nodes of the ENN. The advantages of the ENN, such as strong nonlinear mapping ability and less prediction time, are effectively combined, and the photovoltaic power generation can be predicted stably, quickly and reliably. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0043] Figure 2 It is a schematic diagram of the application process of the embodiment;

[0044] Figure 3 Schematic diagram of the topological structure of the HGS-VMD-ENN ultra-short-term chaos prediction model in the present invention;

[0045] Figure 4 This is a diagram showing the decomposition results of the HGS-VMD data preprocessing method in the embodiment;

[0046] Figure 5a and 5b Graph showing photovoltaic power generation prediction results under different weather conditions in the embodiment. DETAILED DESCRIPTION

[0047] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Example

[0049] like Figure 1 As shown, a photovoltaic power generation power prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model includes the following steps:

[0050] S1. Design a fitness function that takes weighted permutation entropy and decomposition loss into consideration, and take the minimum value of the fitness function as the optimization goal. Adopt HGS-VMD to adaptively determine the best VMD decomposition parameter combination, which is used to perform multi-scale decomposition of photovoltaic power generation to obtain photovoltaic power generation power decomposition components.

[0051] S2. According to the decomposition components of photovoltaic power generation, the phase space matrix is ​​constructed by using the PSR phase space reconstruction method, and the HGS-VMD-ENN ultra-short-term chaos prediction model is constructed in combination with the brain emotion neural network;

[0052] S3. Based on the established HGS-VMD-ENN ultra-short-term chaos prediction model, ultra-short-term prediction of photovoltaic power is performed.

[0053] This embodiment applies the above solution, such as Figure 2 As shown, the main contents are:

[0054] 1) Design a fitness function that takes weighted permutation entropy and decomposition loss into consideration, and take the minimum value of this fitness function as the optimization goal. Adopt HGS-VMD to adaptively determine the optimal decomposition parameter combination (K, α) of VMD. Based on the obtained optimal decomposition parameter combination, perform multi-scale decomposition of photovoltaic power generation. The specific steps are as follows:

[0055] 1.1) The entropy weight method is used to fuse weighted permutation entropy and decomposition loss as a new fitness function for HGS-VMD multi-scale decomposition. This ensures the regularity of the decomposition components while reducing the impact of the decomposition loss on the prediction results. The corresponding fitness function is:

[0056]

[0057] Where Y ij is the indicator matrix composed of weighted permutation entropy and decomposition loss; P ij is the proportion of the jth indicator under the indicator matrix Y.

[0058] 1.2) The HGS algorithm is used to solve the VMD constrained optimization problem, with the minimum Fitness value as the optimization goal, and the optimal VMD decomposition parameters are determined through HGS-VMD adaptive parameter optimization. According to the fitness value after VMD decomposition, the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2 of each individual parameter combination are calculated:

[0059] VC=sech(F(i)-DF) (2)

[0060]

[0061] W1=(1-exp(-|hungry(i)-SH|))×r×2 (4)

[0062]

[0063] Where: F(i) is the fitness value of the individual parameter combination; DF is the optimal fitness value; sech(x) = 2 / (e x +e -x ) is a hyperbolic function; b = 100*(1+r), r is a random number in [0,1]; SH is the sum of the hunger levels of all individual parameter combinations; N is the total number of individuals.

[0064] 1.3) Update the position of individual parameter combinations according to the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2:

[0065]

[0066] Where: TP is the optimal parameter position in the iteration process; vb = 2*shrink*r-shrink; where shrink is the convergence factor.

[0067] 1.4) The algorithm terminates after the maximum number of iterations. At this time, the output is the optimal decomposition parameter of VMD, and the photovoltaic power generation power is decomposed at multiple scales based on the optimal decomposition parameter combination.

[0068] 2) According to the photovoltaic power decomposition components obtained in step 1), the phase space matrix is ​​constructed using the PSR method, and the HGS-VMD-ENN ultra-short-term chaos prediction model is constructed in combination with the brain emotion neural network. The specific steps are as follows:

[0069] 2.1) Reconstruct the phase space of the one-dimensional photovoltaic power time series g(t). According to the embedding dimension m and time delay τ, the phase space matrix expression is obtained as follows:

[0070]

[0071] Wherein, M is the number of delay vectors, M=N-mτ+τ.

[0072] 2.2) Divide the sample sets for model training and testing, use the embedding dimension m as the number of ENN model input layers, use the phase space matrix G as the ENN model input matrix, and after training the model training sample set through ENN, use the test sample set to perform model simulation verification.

[0073] 3) Based on the HGS-VMD-ENN ultra-short-term chaos prediction model established in step 2) (such as Figure 3 The trained model is used to realize the ultra-short-term prediction of photovoltaic power. The specific steps are as follows:

[0074] 3.1) Stimulus input signal Gn (n represents the number of columns of the input reconstruction matrix, 1≤n≤m) enters the thalamus, and the amygdala receives the input signal SC from the sensory cortex n , and receives the extended signal G from the thalamus m+1 . Among them G m+1 The calculation method is:

[0075] G m+1 =max(G n ) (8)

[0076] 3.2) The final output value is calculated according to the following formula:

[0077] E=E a -E o (9)

[0078] Where E a and E o are the outputs of the amygdala and orbitofrontal cortex links. They are defined as follows:

[0079]

[0080] To verify the effectiveness of this solution, this embodiment compares the photovoltaic power prediction value output by the HGS-VMD-ENN ultra-short-term chaos prediction model with the actual value. Based on the historical power generation data of a photovoltaic power generation system, this embodiment selects the power generation data of a photovoltaic power station for example simulation analysis, selects the data from 7:00 to 19:00 every day, and selects 145 sampling points every day.

[0081] First, according to step 1), a fitness function that considers weighted permutation entropy and decomposition loss is designed, and the minimum value is taken as the optimization goal. HGS-VMD is used to adaptively determine the optimal decomposition parameter combination (K, α) of VMD. Based on the obtained optimal decomposition parameter combination, the photovoltaic power generation power is decomposed at multiple scales. The decomposition results are shown in Figure 4 shown.

[0082] According to step 2) and step 3), an ultra-short-term chaotic prediction method for photovoltaic power generation based on HGS-VMD-ENN is constructed, and the key parameters of the example are set as follows:

[0083] The number of nodes in the network input layer is set to the phase space reconstruction embedding dimension m, the output neuron is 1, and the number of hidden layer nodes is 13. In the design of the HGS-VMD algorithm, the population size is 50, the maximum number of iterations is 50, the number of optimization parameters is 2, the number of decomposition layers K is [2,20], the penalty factor α is [100,3000], and the mutation control variable VC2 is 0.05. Then the photovoltaic power generation is decomposed. Method 1 takes the minimum value of the designed fitness function as the optimization goal to optimize the parameter combination; Method 2 takes the minimum weighted permutation entropy value as the optimization objective function to optimize the parameter combination. The results are shown in Table 1.

[0084] Table 1

[0085]

[0086] It can be seen from Table 1 that the decomposition loss of HGS-VMD multi-scale decomposition using the fitness function proposed in this scheme is much smaller than the decomposition loss of method 2. Figure 4 It can be seen that the HGS-VMD multi-scale decomposition results proposed in this scheme can describe the fluctuation characteristics at different time scales.

[0087] The historical data of the first 21 days are selected as the training set, and the data of the last 3 days are selected as the prediction set for single-step (5 minutes in advance) cyclic rolling prediction. Based on the cyclic prediction idea of ​​adding the latest measured data and removing the most distant measured data, the photovoltaic output power time series is continuously updated, and the model input-output mapping relationship is learned using the training set and the prediction set to realize the prediction of photovoltaic output power. The prediction results are as follows: Figure 5a , 5b shown.

[0088] In order to objectively evaluate the prediction effect of the proposed method, the mean absolute percentage error E MAPE , Root mean square error E RMSE and the correlation coefficient E R 2 Evaluate the prediction accuracy.

[0089] Table 2

[0090]

[0091] pass Figure 5a , 5bIt can be seen from the prediction results that the prediction method proposed in this scheme does not have obvious fluctuation errors during the period of sudden change of photovoltaic power generation, and the prediction results are closer to the true value. It can be seen from Table 2 that compared with the PSR-ENN prediction method, the prediction method proposed in this scheme has higher prediction accuracy after HGS-VMD multi-scale decomposition. In weather with severe power fluctuations, the prediction accuracy is significantly improved. It can be seen that through the above example analysis, the ultra-short-term chaotic prediction method of photovoltaic power generation based on HGS-VMD-ENN proposed in this scheme has achieved good prediction results, with high prediction accuracy and stability.

[0092] In summary, this scheme uses the Hunger Game search algorithm HGS for variational mode decomposition VMD core parameter optimization, and designs a HGS-VMD fitness function that considers weighted permutation entropy and decomposition loss to decompose the photovoltaic signal into a series of relatively stable decomposition components; uses phase space reconstruction PSR to map each decomposition component to a high-dimensional phase space to construct the optimal number of input nodes of the brain emotion neural network ENN model, and superimposes the prediction results of each decomposition component to obtain the final prediction value. The prediction performance of this scheme is more stable and more accurate than the traditional prediction method, which effectively reduces the prediction volatility of photovoltaic power generation and improves the prediction performance.

Claims

1. A photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model, characterized in that: The following steps are involved: S1. Design a fitness function that takes weighted permutation entropy and decomposition loss into consideration, and take the minimum value of the fitness function as the optimization goal. Adopt HGS-VMD to adaptively determine the best VMD decomposition parameter combination, which is used to perform multi-scale decomposition of photovoltaic power generation to obtain photovoltaic power generation power decomposition components. S2. According to the decomposition components of photovoltaic power generation, the PSR method is used to construct its phase space matrix, and the HGS-VMD-ENN ultra-short-term chaos prediction model is constructed in combination with the brain emotion neural network; S3. Based on the established HGS-VMD-ENN ultra-short-term chaos prediction model, ultra-short-term prediction of photovoltaic power is performed.

2. A photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 1, characterized in that: The step S1 specifically adopts the entropy weight method to fuse the weighted permutation entropy and the decomposition loss as a new fitness function of the HGS-VMD multi-scale decomposition, which can reduce the impact of the decomposition loss on the prediction results while ensuring the regularity of the decomposition components.

3. The photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 2 is characterized in that: The fitness function in step S1 is specifically: Among them, Y ij is the indicator matrix composed of weighted permutation entropy and decomposition loss, P ij is the proportion of the jth indicator under the indicator matrix Y.

4. The photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 1 is characterized in that: The process of using HGS-VMD to adaptively determine the optimal VMD decomposition parameter combination in step S1 includes: S11, using HGS algorithm to solve VMD constrained optimization problem, taking the minimum value of fitness function as optimization goal, determining the best decomposition parameters of VMD through HGS-VMD adaptive parameter optimization, and calculating the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2 of each individual parameter combination according to the fitness value after VMD decomposition; S12. Update the position of individual parameter combinations according to the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2. The algorithm terminates after the maximum number of iterations. The current output is the VMD optimal decomposition parameter combination (K, α).

5. The photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 4 is characterized in that: The formulas for calculating the variation control value VC, hunger degree hungry(i), hunger weights W1 and W2 of each individual parameter combination in step S11 are: VC=sech(F(i)-DF) W1=(1-exp(-|hungry(i)-SH|))×r×2 Among them, F(i) is the fitness value of the individual parameter combination, DF is the optimal fitness value, sech(x)=2 / (e x +e -x ) is a hyperbolic function, b = 100*(1+r), r is a random number in [0,1], SH is the sum of the hunger levels of all individual parameter combinations, and N is the total number of individuals.

6. A photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 5, characterized in that: The formula for updating the position of the individual parameter combination in step S12 is: Among them, TP is the optimal parameter position in the iteration process, vb=2*shrink*r-shrink, and shrink is the convergence factor.

7. The photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 1 is characterized in that: The step S2 comprises the following steps: S21, reconstructing the phase space of the one-dimensional photovoltaic power generation time series g(t), and obtaining the phase space matrix according to the embedding dimension m and the time delay τ; S22. Divide the sample sets for model training and testing, use the embedding dimension m as the number of ENN model input layers, use the phase space matrix G as the ENN model input matrix, train the model training sample set through ENN, and then use the test sample set to perform model simulation verification to construct the HGS-VMD-ENN ultra-short-term chaos prediction model.

8. The photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 7 is characterized in that: The expression of the phase space matrix in step S21 is: Wherein, M is the number of delay vectors, M=N-mτ+τ.

9. The photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 8 is characterized in that: The process of performing ultra-short-term photovoltaic power prediction in step S3 includes: stimulating the input signal G n Entering the thalamus, n represents the number of columns of the input reconstruction matrix, 1≤n≤m, the amygdala receives the input signal SC from the sensory cortex n , and receives the extended signal G from the thalamus m+1 , combined with the outputs of the amygdala and orbitofrontal cortex, the final output value is obtained, which is the photovoltaic power prediction result.

10. A photovoltaic power generation prediction method based on the HGS-VMD-ENN ultra-short-term chaos prediction model according to claim 9, characterized in that: The extended signal G from the thalamus m+1 Specifically: G m+1 =max(G n ) The final output value is: E=E a -AND o Among them, E a and E o It is the output of the amygdala and orbitofrontal cortex link.

Images