Photovoltaic power ultra-short-term prediction method based on improved emotion neural network
By introducing the BO-SVM and ISO algorithms into the emotional neural network and improving the short reflection paths and weights, the problem of low ultra-short-term prediction accuracy of photovoltaic power in the existing technology is solved, and higher prediction accuracy and stability are achieved.
Patent Information
- Application Number
- CN202411421701.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-12
- Publication Date
- 2025-01-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the ultra-short-term prediction of photovoltaic power, the short reflection path has poor nonlinear fitting ability to input data, resulting in low prediction accuracy.
The Bayesian Optimization-Support Vector Machine (BO-SVM) and Improved Snake Optimization (ISO) algorithms are used to improve the short reflection paths and weights of the emotional neural network, and the improved ENN model is obtained through iterative training for photovoltaic power prediction.
Effectively reduce the impact of non -linear fluctuations of photovoltaic power data on predictive accuracy, making photovoltaic prediction output more stable and more accurate.
Smart Images
Figure CN119358738A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of photovoltaic power prediction, and in particular to an ultra-short-term photovoltaic power prediction method based on an improved emotional neural network. Background Art
[0002] With the proposal of the dual carbon goals, photovoltaic power generation has attracted attention from all walks of life due to its pollution-free and renewable characteristics. However, due to the intermittent and random characteristics of photovoltaic power generation, large-scale grid connection will affect the power quality of the power system. Therefore, accurate prediction of photovoltaic power is of great significance to the operation and dispatch of the power grid.
[0003] Compared with machine learning, the emerging emotion learning model in recent years has been widely used by domestic and foreign scholars in the fields of structural control, nonlinear system control, etc. due to its better performance in response speed and feedback regulation, and it also has good response ability to different time series. ENN (emotion neural network) evolved from the emotion learning model inherits the good performance of the former in model regulation. ENN does not need to determine the number of hidden layer neurons in the modeling process, and the modeling difficulty is lower and the training time is shorter. It has a stronger nonlinear response speed than traditional artificial neural networks when the minute-level time series mutates, so it is more conducive to the prediction of ultra-short-term photovoltaic power. However, due to its own structural characteristics, ENN also has certain defects in the data processing process. Its short reflection path has poor nonlinear fitting ability for input data, resulting in low prediction accuracy. At present, due to the mismatch between its own fitting accuracy and response speed, the photovoltaic power prediction accuracy of ENN needs to be further improved. 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 ultra-short-term prediction method based on an improved emotional neural network. Based on Bayesian optimization-support vector machine (BO-SVM) and improved snake optimization (ISO), ENN is improved and used for photovoltaic power prediction, which can effectively reduce the impact of nonlinear fluctuations of photovoltaic power data on prediction accuracy, making the photovoltaic prediction output more stable and more accurate.
[0005] The purpose of the present invention can be achieved by the following technical solution: A photovoltaic power ultra-short-term prediction method based on an improved emotional neural network comprises the following steps:
[0006] S1. Collect the time series data of photovoltaic power generation sample, normalize the photovoltaic power generation time series, use the maximum Lyapunov exponent method to analyze the chaotic characteristics of the photovoltaic power time series, obtain the phase space parameters through the improved correlation coefficient (CC) method, and reconstruct the phase space;
[0007] S2, inputting the phase space matrix column vector into the ENN model, improving the ENN short reflection path based on BO-SVM, updating the ENN weights based on the ISO algorithm, and obtaining the improved ENN model through iterative training;
[0008] S3, input the data to be predicted into the improved ENN model according to the chaotic column vector, obtain the prediction result by rolling single-step prediction mode, denormalize the prediction result and output it to obtain the photovoltaic power prediction data.
[0009] Furthermore, the calculation formula for obtaining the phase space parameters by the improved correlation coefficient method in step S1 is:
[0010]
[0011] Where N is the number of data points in the photovoltaic power time series, M is the number of delay vectors, R is the defined spatial distance, H(x) is the step function, and p i 、p j Reconstruct two phase points in phase space for the PV power time series.
[0012] Furthermore, in step S2, the improvement of the ENN short reflection path based on BO-SVM specifically adopts a historical data three-dimensional phase point classification plane selection method based on SVM, extracts the nonlinear characteristics of the historical data based on the distance from the three-dimensional phase point to the classification plane, and uses the BO algorithm to optimize the SVM solution accuracy.
[0013] Furthermore, the process of improving the ENN short reflection path based on BO-SVM in step S2 includes:
[0014] In the linearly inseparable SVM, in order to express the linear inseparability of high-dimensional space with the linear separability of low-dimensional space, the radial basis function (RBF) kernel function is introduced, where g is the kernel parameter, and the expression is:
[0015]
[0016] The objective problem and constraints of SVM are:
[0017]
[0018] The convex quadratic optimization of the target problem of SVM is used to find the global optimal solution, and the variables of the decision plane are obtained as follows:
[0019]
[0020] The decision plane required by SVM is:
[0021] f(x)=sgn(w T ·x+b)
[0022] In SVM, the selection of penalty coefficient C and kernel parameter g directly affects the misclassification tolerance of the data to be classified and the number of support vectors, thus determining the solution accuracy and computational complexity of the SVM algorithm. For C and g, the BO algorithm is used for optimization. The optimization process includes four operations: modeling, sampling, updating, and selection. In a given black box function to be optimized, the BO algorithm selects historical data sampling points through the acquisition function (AF), establishes the posterior distribution of the black box function with the prior function (PF) and continuously updates it to make it close to the true distribution of the unknown function.
[0023] In the BO-SVM improved ENN, the neurons AC of the amygdala m+1 Responsible for receiving improved three-dimensional phase points. These improved phase points are defined according to their spatial position and the distance from the decision plane:
[0024]
[0025] Among them, β l is the three-dimensional phase point q of the input vector l Three-dimensional phase point of improved vector The weight of; int(x) is the floor rounding function; d l is the three-dimensional phase point q of each input in SVM l The distance to the decision plane, when the distance is larger, the weight of the three-dimensional phase point in the improved signal is smaller.
[0026] Furthermore, the updating of ENN weights based on the ISO algorithm in step S2 is specifically an improvement of the Snake Optimization (SO) algorithm, and is used for iterative optimization of ENN weights.
[0027] Furthermore, the process of improving the snake optimization algorithm in step S2 includes:
[0028] Introducing the chaotic cubic operator in population initialization to ensure the non-ergodicity of the population within a certain range, expand the individual optimization search range, and accelerate the convergence speed of the algorithm during operation;
[0029] In the foraging stage, an adversarial learning mechanism is introduced to calculate the reverse solution of lens imaging by using the lens imaging method to expand the range of optional solutions.
[0030] The Cauchy-Gaussian mutation strategy is used to process the individuals with the best fitness.
[0031] Furthermore, the chaos cubic operator is specifically:
[0032] y(t+1)=4y(t) 3 -3y(t)
[0033]
[0034] Among them, X ub is the position of the individual when searching for the upper limit, X lb is the individual position when searching for the lower limit, and t is the current iteration number.
[0035] Furthermore, the inverse solution calculation formula of the lens imaging is:
[0036]
[0037] Among them, Q is the current total amount of food in the environment, is the current optimal position of the individual snake.
[0038] Furthermore, the expression of the Cauchy-Gaussian mutation strategy is:
[0039]
[0040] λ2=1-λ1
[0041]
[0042] Where T is the maximum number of iterations of the ISO algorithm, λ1 and λ2 are inertia factors, and σ is the denominator constant to ensure that the denominator of the σ function is not zero.
[0043] Furthermore, after the predicted data in step S3 is input into the improved ENN model according to the chaotic column vector, the stimulus input signal T I Entering the thalamus, I represents the number of columns of the input reconstruction matrix, 1≤I≤m, the amygdala receives the input signal A from the sensory cortex I , and receives signals from the improved short reflex pathway T from the thalamus m+1 ;
[0044] The final output value E of the improved ENN model is:
[0045]
[0046] Among them, E ais the sum of the inputs from other pathways and the improved short reflex pathway, E0 is the sum of the negative feedback from the orbitofrontal cortex module to the amygdala module, T i is the data vector input to the improved ENN, ω l and δ l is the weight.
[0047] Compared with the prior art, the present invention has the following advantages:
[0048] The present invention first analyzes the chaotic characteristics of the photovoltaic power time series, obtains the phase space parameters by improving the correlation coefficient method, and reconstructs the phase space to complete the preprocessing of the sample data; then the phase space matrix column vector is input into the ENN model, the ENN short reflection path is improved based on BO-SVM, the ENN weight is updated based on the ISO algorithm, and the improved ENN model is obtained through iterative training to complete the model training process; finally, the data to be predicted is input into the improved ENN model according to the chaotic column vector, the rolling single-step prediction mode is used to obtain the prediction result, the prediction result is denormalized and output to obtain the photovoltaic power prediction data, and the rolling prediction process is realized. Therefore, on the one hand, a historical data classification method based on BO-SVM is adopted, and the nonlinear characteristics are extracted by the distance from the historical data to the SVM classification plane, so as to improve the nonlinear fitting ability of the ENN short reflection path, and secondly, the snake optimization algorithm is improved and used for the weight optimization of the ENN to ensure that the short reflection path reasonably expresses the nonlinear characteristics of the historical data, thereby effectively reducing the influence of the nonlinear fluctuation of the photovoltaic power data on the prediction accuracy, so that the photovoltaic prediction output is more stable and more accurate.
[0049] The present invention proposes an SVM hyperparameter combination selection method based on the BO algorithm. On the basis of the BO algorithm, the influence of the number of support vectors on the SVM hyperplane solution is considered, which is conducive to the extraction of nonlinear features. In addition, on the premise of obtaining an accurate solution of the hyperplane, a nonlinear feature extraction method based on the distance from the three-dimensional phase point of historical data to the hyperplane is adopted. Compared with the traditional ENN, more historical data are more fairly mobilized to show nonlinear characteristics, thereby improving the nonlinear fitting ability of the ENN short reflection path to the input data.
[0050] The present invention also proposes an ISO algorithm and uses it for iterative optimization of ENN weights. By improving the search capability of the SO algorithm, the optimization performance of the ENN weights is improved, ensuring that the nonlinear characteristics of the historical time series extracted by the ENN short reflection path can be expressed in the prediction results with appropriate weights. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 It is a schematic diagram of the method flow of the present invention;
[0052] Figure 2 It is a schematic diagram of the topological structure of the improved ENN model in the present invention;
[0053] Figure 3 This is a comparison chart of photovoltaic power prediction results of ENN, SVM-ENN and BO-SVM-ENN in cloudy weather in the embodiment;
[0054] Figure 4 This is a comparison chart of the optimization performance of the ISO algorithm in the embodiment;
[0055] Figure 5a to Figure 5c It is a comparison chart of photovoltaic output power prediction results under different weather conditions in the embodiment. DETAILED DESCRIPTION
[0056] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0057] Example
[0058] like Figure 1 As shown, a photovoltaic power ultra-short-term prediction method based on an improved emotional neural network includes the following steps:
[0059] S1. Collect the photovoltaic power generation sample time series data, normalize the photovoltaic power generation time series data, use the maximum Lyapunov exponent method to analyze the chaotic characteristics of the photovoltaic power time series, obtain the phase space parameters through the improved correlation coefficient method, and reconstruct the phase space;
[0060] S2, inputting the phase space matrix column vector into the ENN model, improving the ENN short reflection path based on BO-SVM, updating the ENN weights based on the ISO algorithm, and obtaining the improved ENN model through iterative training;
[0061] S3, input the data to be predicted into the improved ENN model according to the chaotic column vector, obtain the prediction result by rolling single-step prediction mode, denormalize the prediction result and output it to obtain the photovoltaic power prediction data.
[0062] This embodiment applies the above solution, and the main contents are:
[0063] 1) For the photovoltaic power generation time series after data normalization, the maximum Lyapunov exponent method is used to analyze the chaotic characteristics of the photovoltaic power time series, and the phase space parameters are obtained by the improved correlation coefficient method, and the phase space is reconstructed;
[0064] 2) In order to improve the nonlinear fitting ability of ENN short reflection path, a three-dimensional phase point classification plane selection method based on SVM is adopted to extract the nonlinear characteristics of historical data based on the distance from the three-dimensional phase point to the classification plane, and the BO algorithm is used to optimize the SVM solution accuracy;
[0065] 3) To ensure that the ENN prediction results can take into account both the generalization of the output of the improved short reflex pathway and the characteristics of the output of other pathways, the Snake Optimization (SO) algorithm is improved and used for the iterative optimization of the ENN weights;
[0066] 4) According to step 1), step 2) and step 3), the ENN model based on BO-SVM and ISO improvement is obtained, and the measured data is used to realize the ultra-short-term prediction of photovoltaic power.
[0067] Specifically, step 1) reconstructs the chaotic phase space by using the improved CC method, and the corresponding solution is:
[0068]
[0069] Where N is the number of data points in the photovoltaic power time series; M is the number of delay vectors; R is the defined spatial distance; H(x) is the step function; p i 、p j Reconstruct two phase points in phase space for the PV power time series.
[0070] Step 2) includes the following process:
[0071] 2.1) In linearly inseparable SVM, in order to express the linear inseparability of high-dimensional space with the linear separability of low-dimensional space, the radial basis function (RBF) kernel function is introduced, where g is the kernel parameter, and the expression is shown in formula (2).
[0072]
[0073] Then the objective problem and constraints of SVM are shown in formula (3):
[0074]
[0075] The convex quadratic optimization of formula (3) is used to find the global optimal solution, and the variables of the decision plane are shown in formula (4):
[0076]
[0077] The decision plane required by SVM is shown in formula (5):
[0078] f(x)=sgn(w T ·x+b) (5)
[0079] 2.2) In SVM, the selection of penalty coefficient C and kernel parameter g directly affects the misclassification tolerance of the data to be classified and the number of support vectors, thereby determining the solution accuracy and computational complexity of the SVM algorithm. For C and g, the BO algorithm is used for optimization. Its working process mainly includes four operations: modeling, sampling, updating, and selection. Under a given black box function to be optimized, the BO algorithm selects historical data sampling points through the acquisition function (AF), establishes the posterior distribution of the black box function with the prior function (PF) and continuously updates it to make it close to the true distribution of the unknown function.
[0080] 2.3) In the BO-SVM improved ENN, the neurons AC of the amygdala m+1 Responsible for receiving these improved three-dimensional phase points. These improved phase points are defined according to the distance between their spatial position and the decision plane, as shown in formula (6). In this way, the characteristics of historical data can be expressed more accurately, so that the model can better utilize the information of input data in the prediction process, thereby improving the accuracy of prediction.
[0081]
[0082] Where: β l is the three-dimensional phase point q of the input vector l Three-dimensional phase point of improved vector The weight of; int(x) is the floor rounding function; d l is the three-dimensional phase point q of each input in SVM l The distance to the decision plane, when the distance is larger, the weight of the three-dimensional phase point in the improved signal is smaller.
[0083] Step 3) In order to improve the search ability of the SO algorithm and avoid falling into the local optimum, the following three improvements are designed:
[0084] 3.1) Population initialization introduces the chaotic cubic operator
[0085] The selection of initialization parameters and modes directly determines the convergence speed and search range during the operation of the algorithm, thus affecting the efficiency and performance of the algorithm. Therefore, this scheme introduces the chaotic cubic operator to improve the initialization process of the population individuals in the SO algorithm, which ensures the non-ergodicity of the population within a certain range, expands the optimization search range of individuals, and accelerates the convergence speed of the algorithm during operation.
[0086] y(t+1)=4y(t) 3 -3y(t) (7)
[0087]
[0088] Where, X ub is the position of the individual when searching for the upper limit; X lb is the individual position when searching for the lower limit; t is the current iteration number.
[0089] 3.2) Introducing adversarial learning mechanism in the foraging stage
[0090] The core idea of this method is to use lens imaging to calculate the inverse solution of lens imaging to expand the range of optional solutions. The proposed method can better explore the search space and increase the quality of the initial population individuals.
[0091]
[0092] Where: Q is the current total amount of food in the environment; X b t est is the current optimal position of the individual snake.
[0093] 3.3) The optimal individual introduces the Cauchy-Gauss mutation strategy
[0094] In order to improve the algorithm's ability to maintain population diversity, resist stagnation, and avoid falling into local optimality, this scheme adopts the Cauchy-Gaussian mutation strategy to process individuals with optimal fitness. The Cauchy-Gaussian mutation expression is shown in equations (10) and (11), where the inertia factor λ2=1-λ1.
[0095]
[0096] Where: T is the maximum number of iterations of the ISO algorithm; σ is the denominator constant, which ensures that the denominator of the σ function is not 0.
[0097] The improved ENN model is trained as Figure 2 As shown, in step 4), the stimulus input signal T I (I represents the number of columns of the input reconstruction matrix, 1≤I≤m) enters the thalamus, and the amygdala receives the input signal A from the sensory cortex I , and receives signals from the improved short reflex pathway T from the thalamus m+1 The final output value of the improved ENN model is calculated according to formula (12):
[0098]
[0099] In the formula, E a is the sum of the inputs from other pathways and the improved short reflex pathway, E0 is the sum of the negative feedback from the orbitofrontal cortex module to the amygdala module, T i is the data vector input to the improved ENN, ω l and δ l is the weight.
[0100] To verify the effectiveness of this solution, this embodiment compares the ultra-short-term photovoltaic power prediction model based on the improved emotional neural network proposed in this solution with the actual value. Based on the historical power generation data of a photovoltaic power generation system, this embodiment uses the measured power generation data of a centralized photovoltaic power station for 30 days in April 2022, with a sampling interval of 5 minutes. 07:00-19:00 every day is used as the sampling period. According to the daily cloud cover information provided, it is divided into sunny, cloudy and cloudy weather according to the national meteorological standards, and a total of 4320 sampled power data are used for analysis in the embodiment.
[0101] The parameters to be set for the BO algorithm include the number of initialization points n, the maximum number of iterations I, and the selection of prior functions and acquisition functions. The above parameter settings are shown in Table 1, and the improvement comparison of BO-SVM is shown in Figure 3 shown.
[0102] Table 1
[0103]
[0104] The ISO algorithm requires the determination of a series of parameters, including the number of male snakes N m 、Number of female snakes f , search upper bound u b , search lower bound l b , data dimension dim, maximum number of iterations T, food quantity constant c1, exploration constant c2, foraging constant c3 and constant ε, etc. The above parameter settings are shown in Table 2, and the running results are shown in Figure 4 shown.
[0105] Table 2
[0106]
[0107] The historical data of the first 22 days are selected as the training set, and the historical data of the last 8 days are selected as the training 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 to 5c As shown ( Figure 5a Comparison of photovoltaic power prediction results under sunny weather. Figure 5b Comparison of photovoltaic power prediction results under cloudy weather conditions. Figure 5c (Comparison of photovoltaic power prediction results under cloudy weather).
[0108] In order to objectively evaluate the prediction effect of the proposed method, the mean absolute percentage error M is used. MAPE and the root mean square error RRMSE The prediction accuracy is evaluated and the results are shown in Table 3.
[0109] Table 3
[0110]
[0111]
[0112] It can be seen from the data in Table 3 that when the actual value of photovoltaic power changes suddenly, the prediction results of SO-ENN produce a certain degree of prediction deviation. Compared with SO-ENN, SVM-SO-ENN has a better prediction accuracy in sunny, cloudy and overcast weather conditions. MAPE and R RMSE The values decreased by 0.62%, 1.88%, 3.42% and 8.43, 23.17, and 56.41 respectively.
[0113] Comparing the prediction errors of SVM-SO-ENN and SVM-ISO-ENN, it can be seen that the ISO algorithm improves the expression of short reflection paths in the prediction results by improving the optimization performance of the SO algorithm. MAPE and R RMSE The values are 0.46%, 1.29%, 1.48% and 7.31, 16.97, 21.64 lower than those of the SVM-SO-ENN model.
[0114] Compared with SVM-ISO-ENN, the proposed BO-SVM-ISO-ENN model has an average M MAPE and R RMSE The model proposed by the present invention reduces M by 0.87% and 16.76% respectively. MAPE and R RMSE They decreased by 1.40% and 20.15 respectively.
Claims
1. A photovoltaic power ultra-short-term prediction method based on an improved emotional neural network, characterized in that: The following steps are involved: S1. Collect the photovoltaic power generation sample time series data, normalize the photovoltaic power generation time series data, use the maximum Lyapunov exponent method to analyze the chaotic characteristics of the photovoltaic power time series, obtain the phase space parameters through the improved correlation coefficient method, and reconstruct the phase space; S2, inputting the phase space matrix column vector into the ENN model, improving the ENN short reflection path based on BO-SVM, updating the ENN weights based on the ISO algorithm, and obtaining the improved ENN model through iterative training; S3, input the data to be predicted into the improved ENN model according to the chaotic column vector, obtain the prediction result by rolling single-step prediction mode, denormalize the prediction result and output it to obtain the photovoltaic power prediction data.
2. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 1 is characterized in that: The calculation formula for obtaining the phase space parameters by the improved correlation coefficient method in step S1 is: Where N is the number of data points in the photovoltaic power time series, M is the number of delay vectors, R is the defined spatial distance, H(x) is the step function, and p i 、p j Reconstruct two phase points in phase space for the PV power time series.
3. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 1 is characterized in that: In step S2, the improvement of the ENN short reflection path based on BO-SVM specifically adopts a historical data three-dimensional phase point classification plane selection method based on SVM, extracts the nonlinear characteristics of the historical data based on the distance from the three-dimensional phase point to the classification plane, and uses the BO algorithm to optimize the SVM solution accuracy.
4. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 3 is characterized in that: The process of improving the ENN short reflection path based on BO-SVM in step S2 includes: In the linearly inseparable SVM, in order to express the linear inseparability of high-dimensional space with the linear separability of low-dimensional space, the radial basis kernel function is introduced, where g is the kernel parameter, and the expression is: The objective problem and constraints of SVM are: The convex quadratic optimization of the target problem of SVM is used to find the global optimal solution, and the variables of the decision plane are obtained as follows: The decision plane required by SVM is: f(x)=sgn(w T ·x+b) In SVM, the selection of penalty coefficient C and kernel parameter g directly affects the misclassification tolerance of the data to be classified and the number of support vectors, thus determining the solution accuracy and computational complexity of the SVM algorithm. For C and g, the BO algorithm is used for optimization. The optimization process includes four operations: modeling, sampling, updating, and selection. In a given black box function to be optimized, the BO algorithm selects historical data sampling points through the acquisition function AF, establishes the posterior distribution of the black box function with the prior function PF, and continuously updates it to make it close to the true distribution of the unknown function. In the BO-SVM improved ENN, the neurons AC of the amygdala m+1 Responsible for receiving improved 3D phase points, which are defined according to their spatial position and the distance from the decision plane: Among them, β l is the three-dimensional phase point q of the input vector l Three-dimensional phase point of improved vector The weight of; int(x) is the floor rounding function; d l is the three-dimensional phase point q of each input in SVM l The distance to the decision plane, when the distance is larger, the weight of the three-dimensional phase point in the improved signal is smaller.
5. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 1 is characterized in that: The updating of the ENN weights based on the ISO algorithm in step S2 is specifically an improvement on the snake optimization SO algorithm, and is used for iterative optimization of the ENN weights.
6. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 5 is characterized in that: The process of improving the snake optimization algorithm in step S2 includes: Introducing the chaotic cubic operator in population initialization to ensure the non-ergodicity of the population within a certain range, expand the individual optimization search range, and accelerate the convergence speed of the algorithm during operation; In the foraging stage, an adversarial learning mechanism is introduced to calculate the reverse solution of lens imaging by using the lens imaging method to expand the range of optional solutions. The Cauchy-Gaussian mutation strategy is used to process the individuals with the best fitness.
7. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 6 is characterized in that: The chaos cubic operator is specifically: y(t+1)=4y(t) 3 -3y(t) Among them, X ub is the position of the individual when searching for the upper limit, X lb is the individual position when searching for the lower limit, and t is the current iteration number.
8. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 7 is characterized in that: The reverse solution calculation formula of the lens imaging is: Among them, Q is the current total amount of food in the environment, is the current optimal position of the individual snake.
9. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 8 is characterized in that: The expression of the Cauchy-Gauss mutation strategy is: Where T is the maximum number of iterations of the ISO algorithm, λ1 and λ2 are inertia factors, and σ is the denominator constant to ensure that the denominator of the σ function is not zero.
10. The photovoltaic power ultra-short-term prediction method based on improved emotional neural network according to claim 9, characterized in that: After the predicted data in step S3 is input into the improved ENN model according to the chaotic column vector, the stimulus input signal T I Entering the thalamus, I represents the number of columns of the input reconstruction matrix, 1≤I≤m, the amygdala receives the input signal A from the sensory cortex I , and receives signals from the improved short reflex pathway T from the thalamus m+1 ; The final output value E of the improved ENN model is: Among them, E a is the sum of the inputs from other pathways and the improved short reflex pathway, E0 is the sum of the negative feedback from the orbitofrontal cortex module to the amygdala module, T i is the data vector input to the improved ENN, ω l and δ l is the weight.