Photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM

Abstract

The invention relates to a photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM. The method comprises the following steps: preprocessing a historical meteorological data set; the preprocessed historical meteorological data set is divided into a plurality of clusters through a clustering algorithm, the clusters represent different types of meteorological data sets, and the clustering center in each cluster is typical meteorological characteristic data corresponding to the cluster; based on each type of meteorological data set, respectively determining optimal hyper-parameters of a plurality of Bi-LSTM models by adopting a multi-strategy improved Red-beak-Blue-Tapie optimization algorithm, and obtaining a photovoltaic power prediction sub-model of each type of meteorological data; and selecting a corresponding photovoltaic power prediction sub-model according to the similarity between the real-time meteorological data and the typical meteorological characteristic data, inputting the real-time meteorological data into the corresponding photovoltaic power prediction sub-model, and outputting a photovoltaic power prediction result. The method can quickly and accurately predict the output power of the photovoltaic power generation system, and can provide data support for a power system.

Description

Technical Field

[0001] The present application relates to the technical field of photovoltaic power short-term prediction, and in particular to a photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM. Background Art

[0002] At present, with the rapid development of deep learning, it has become a trend in photovoltaic power prediction to improve the accuracy and reliability of photovoltaic power prediction by adopting advanced mathematical modeling and machine learning techniques.

[0003] Many existing machine learning algorithms have a slow speed in training prediction models and low prediction accuracy. In addition, the determination of prediction algorithm hyperparameters is extremely cumbersome. Therefore, a technical solution is needed to efficiently determine the prediction algorithm hyperparameters, speed up the prediction model training speed, and improve the prediction accuracy. Summary of the invention

[0004] Based on this, it is necessary to provide a photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM, which includes:

[0005] S1: Acquire a historical meteorological data set of a forecast area, and preprocess the historical meteorological data set;

[0006] S2: dividing the preprocessed historical meteorological data set into multiple clusters by a clustering algorithm, each cluster represents a meteorological data set of a different category, and the cluster center in each cluster is the typical meteorological characteristic data corresponding to the cluster;

[0007] S3: Based on each type of meteorological data set, the multi-strategy improved red-billed blue magpie optimization algorithm is used to determine the optimal hyperparameters of multiple Bi-LSTM models, and the photovoltaic power prediction sub-model for each type of meteorological data is obtained;

[0008] S4: Acquire real-time meteorological data of the prediction area, select a corresponding photovoltaic power prediction sub-model according to the similarity between the real-time meteorological data and the typical meteorological characteristic data, input the real-time meteorological data into the corresponding photovoltaic power prediction sub-model, and output the photovoltaic power prediction result.

[0009] Beneficial effects: This method can quickly and accurately predict the output power of photovoltaic power generation systems, provide data support for power systems, help dispatching departments formulate power generation, peak load regulation and dispatching plans, effectively improve the accuracy and robustness of photovoltaic power prediction, and provide new possibilities for intelligent management and operation optimization of photovoltaic power generation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0011] Figure 1 This is a flow chart of a photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM in an embodiment of the present application.

[0012] Figure 2 This is a comparison chart of the photovoltaic power predicted by each model on a sunny day and the actual power in the embodiments of the present application.

[0013] Figure 3 This is a comparison chart of the photovoltaic power predicted by each model on a cloudy day and the actual power in the embodiments of the present application.

[0014] Figure 4 This is a comparison chart of the photovoltaic power predicted by each model on a rainy day and the actual power in the embodiments of the present application. DETAILED DESCRIPTION

[0015] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.

[0016] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0017] like Figure 1 As shown, this embodiment provides a photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM, the method comprising:

[0018] S1: Obtain the historical meteorological dataset of the forecast area and preprocess the historical meteorological dataset.

[0019] In this embodiment, the historical meteorological data set includes meteorological data with a sampling interval of 15 minutes, a time span of 5 years, and an absolute value of the Pearson correlation coefficient with the direct normal radiation of the sun greater than 0.2.

[0020] In this embodiment, the meteorological data in the historical meteorological data set includes: temperature, azimuth, cloud opacity, DHI, DNI, relative humidity, wind direction, wind speed, and zenith angle.

[0021] Furthermore, the calculation formula of the Pearson correlation coefficient is:

[0022]

[0023] Where P represents the Pearson correlation coefficient, P∈[-1,1], P=0 means the two variables are uncorrelated, P<0 means the two variables are negatively correlated, P>0 means the two variables are positively correlated, the larger the absolute value of P, the stronger the correlation between the two variables; n represents the number of samples; x i represents the meteorological data of the i-th sample; y i represents the direct normal solar radiation of the i-th sample; Represents the average value of each sample meteorological data; Represents the average value of direct normal solar radiation for each sample.

[0024] Furthermore, the preprocessing of the historical meteorological data set includes:

[0025] Step 1: Eliminate the meteorological data of the period without photovoltaic power in the historical meteorological data set;

[0026] Filling the missing data in the historical meteorological data set with the mean value of the meteorological data;

[0027] Using the triple standard deviation principle to remove outliers in the historical meteorological data set;

[0028] Step 2: Normalize the meteorological data in the historical meteorological data set after elimination and filling. The normalization calculation formula is:

[0029]

[0030] Among them, X scale represents the normalized meteorological data; X represents the actual value of the meteorological data; X min Indicates the minimum value of meteorological data; X max Indicates the maximum value of the meteorological data.

[0031] S2: Divide the preprocessed historical meteorological data set into multiple clusters through a clustering algorithm, each cluster represents a meteorological data set of a different category, and the cluster center in each cluster is the typical meteorological characteristic data corresponding to the cluster.

[0032] In this embodiment, the clustering algorithm is a fuzzy C-means clustering algorithm (FCM algorithm). Fuzzy C-means clustering has three key parameters: a fixed number of clusters, a centroid for each cluster, and each data point belongs to the cluster closest to the centroid. Fuzzy C-means clustering obtains the cluster center by minimizing the objective function.

[0033] The objective function is essentially the sum of the Euclidean distances from each point to each cluster (the sum of squared errors). Through iterative operations, the error value of the objective function is gradually reduced until the objective function converges and the clustering result is obtained.

[0034] The objective function is expressed as:

[0035]

[0036] Among them, J m represents the objective function; n represents the number of samples; C represents the number of cluster centers; Represents the meteorological data x in the mth cluster i For cluster center c i The membership degree of x i represents the meteorological data of the i-th sample; c j represents the jth cluster center.

[0037] The process of fuzzy C-means clustering algorithm is:

[0038] Step 1: Select the number of clusters C, select the appropriate mth cluster, and initialize the matrix U0 determined by the membership function (initialized with a random value between [0, 1]);

[0039] Step 2: Calculate the cluster center using the following formula:

[0040]

[0041] Step 3: Calculate the new membership matrix. The calculation formula is:

[0042]

[0043] Among them, u ij represents the new membership matrix; c e represents the e-th cluster center;

[0044] Step 4: Repeat steps 3-4, iterate the membership matrix, and minimize the objective function:

[0045] Compareij and u i(j+1) , if the change of the two is less than a certain threshold, step 5 is the termination condition, then the algorithm stops, otherwise go to step 3;

[0046] Step 5: In this embodiment, the original data set is divided into three categories: sunny day, rainy day, and cloudy day, so as to predict the output power in the three weather conditions respectively, thereby improving the prediction accuracy.

[0047] The clustering effect is evaluated according to the silhouette coefficient, and the best clustering algorithm is selected to cluster the meteorological data set. The calculation of the silhouette coefficient s(i) includes:

[0048]

[0049] Among them, a(i) is the average distance between sample i and other samples in the same cluster, which is called cohesion; b(i) is the average distance between sample i and all samples in other clusters, which is called separation. The average value of the silhouette coefficients of all samples is the final silhouette coefficient, which ranges from [-1,1]. The closer the value is to 1, the smaller the distance within the cluster, the larger the distance between clusters, and the better the clustering effect.

[0050] S3: Based on each type of meteorological data set, a multi-strategy improved red-billed blue magpie optimization algorithm is used to determine the optimal hyperparameters of multiple Bi-LSTM models, and a photovoltaic power prediction sub-model for each type of meteorological data is obtained.

[0051] Specifically, the step includes:

[0052] S3.1: Divide any type of meteorological data set into a training set and a validation set in proportion;

[0053] S3.2: Take the meteorological data in the training set as input and the photovoltaic power as output;

[0054] S3.3: Based on the input and output, the optimal hyperparameters of the Bi-LSTM model are determined by using a multi-strategy improved red-billed blue magpie optimization algorithm, and the photovoltaic power prediction sub-model corresponding to the meteorological data set is obtained;

[0055] S3.4: Validate the photovoltaic power prediction sub-model corresponding to the meteorological data set based on the validation set.

[0056] In this embodiment, the multi-strategy improved red-billed blue magpie optimization algorithm includes:

[0057] Step 1: Use the hyperparameters of the Bi-LSTM model as the population and initialize the population using the good point set;

[0058] Initializing the population using a good point set includes:

[0059] It is difficult to ensure the uniformity, diversity and stability of the initial population by obtaining individuals in the population through random initialization. The good point set can generate a series of uniform and stable initial points. Using the good point set to initialize the population can effectively improve the diversity and uniformity of the population and improve the algorithm search performance. Therefore, in this embodiment, the good point set is used to initialize the population to clarify the dimension d (i.e., the number of elements in each solution vector) of the optimal hyperparameter and the size N (i.e., the number of individuals in the population).

[0060] Generate a good point set: According to the definition of a good point set, for each dimension d, a set of real numbers or real number sequences needs to be selected (where s is the number of real numbers selected according to specific needs, usually s can be equal to d or other suitable values), forming a good point set. These real numbers or real number sequences should be closely related to the dimension d to ensure that the generated point set is evenly distributed in the d-dimensional space. For each index h (1≤h≤d), the good point set is expressed as:

[0061]

[0062] Among them, P d (h) represents the good point set; d represents the dimension; They respectively represent the first real number, the second real number, and the sth real number selected in the d-dimensional space, which are related to the dimension d and are used to generate the good point set; h represents the index variable.

[0063] Generate an initial population from a good point set: Since the good point set itself may not be sufficient to directly generate the points required for the entire population, one or more points can be selected from the good point set as the starting point, and then random perturbations are added to each point to generate other points in the population. The random perturbations are determined based on the upper and lower bounds of the decision variables of each point.

[0064] The initial population generated by the good point set is more evenly distributed, which improves the diversity of the population, enhances the algorithm's search ability and optimization effect, and the good point set error is smaller, which greatly avoids the algorithm from falling into local optimality.

[0065] Solving the problem based on the initial solution, the quality of the initial solution will directly affect the algorithm's performance. Therefore, the reverse solution is considered simultaneously in the process of initializing the population of the good point set, and individuals with better fitness are retained as the solution of the algorithm, expanding the population search space, improving the quality of the initial solution, increasing the possibility of algorithm optimization and reducing the blindness of the solution process. While using the good point set to initialize the population, the reverse learning mechanism strategy is used to calculate the reverse solution of the red-billed blue magpie individual. The calculation formula is:

[0066] OP i =w·(ub j +lb j )+a i ;

[0067] Among them, OP i Represents individual a i The reverse solution of i represents the i-th red-billed blue magpie individual; w represents the third random number; ub j represents the upper bound of the j-th decision variable; lb j represents the lower bound of the j-th decision variable;

[0068] Step 2: In the process of searching for food, red-billed blue magpies usually act in small groups (2 to 5) or flocks (more than 10) to improve search efficiency. This adaptability and flexibility enables red-billed blue magpies to adopt a variety of hunting strategies according to environmental conditions and available resources to ensure adequate food supply;

[0069] In the population, the red-billed blue magpie searches for food in small groups or groups; the mathematical expression for searching for food in small groups is:

[0070]

[0071] Among them, X i (t+1) represents the position of the i-th individual in the small group at the t+1 iteration; X i (t) represents the position of the i-th individual in the small group at the tth iteration; p represents the number of individuals in the randomly selected small group; X k (t) represents the position of the kth individual in the randomly selected small group at the tth iteration; X rs (t) represents the search position randomly selected by the small group at the tth iteration; Rand1 represents the first random number;

[0072] The mathematical representation of searching for food in a flock is:

[0073]

[0074] Among them, X i (t+1)' represents the position of the i-th individual in the population at the t+1 iteration; X i (t)′ represents the position of the i-th individual in the population at the tth iteration; q represents the number of individuals in the randomly selected population; X k (t)′ represents the position of the kth individual in the population randomly selected at the tth iteration; X rs (t)′ represents the search position randomly selected by the group at the tth iteration; Rand2 represents the second random number;

[0075] Step 3: In small group actions, the main target is usually small prey or plants; when small food is found, the mathematical model for attacking prey is expressed as:

[0076]

[0077] Among them, X food (t) represents the location of the food at the tth iteration; Randn1 represents the first random number used to generate a standard normal distribution (mean value is 0, standard deviation is 1); CF represents the dynamic change coefficient, which is used to adjust the speed and intensity of the individual moving towards the food; T represents the total number of iterations;

[0078] In group action, the main target is usually large prey; when large food is found, the mathematical model of attacking prey is expressed as:

[0079]

[0080] Among them, Randn2 represents the second random number used to generate a standard normal distribution (mean 0, standard deviation 1); This predatory hunting behavior emphasizes the variety of strategies and skills possessed by the red-billed blue magpie, making it a versatile predator that is good at successfully obtaining food in a variety of situations.

[0081] Step 4: After obtaining food, store the remaining food for future consumption to ensure a stable food supply when food is scarce. The process of storing food retains the information of the solution, making it easier for individuals to find the global optimal value and obtain the best hyperparameters. The mathematical representation of storing food is:

[0082]

[0083] in, represents the fitness of the i-th individual before the position is updated, Represents the fitness of the i-th individual after the position is updated.

[0084] In this embodiment, inspired by the red-billed blue magpie attacking prey, the Levy flight strategy is introduced in the attack prey phase of the red-billed blue magpie optimization algorithm to enhance convergence. Assuming that they can use the Levy flight strategy to capture prey, the mathematical model is expressed as:

[0085]

[0086] Where T is the current iteration, and are the current positions of the i-th red-billed blue magpie and the random r-th red-billed blue magpie, is the new position of the i-th red-billed blue magpie, is the best position in the optimized population of red-billed blue magpie, r3 and r4 are the fourth and fifth random numbers between (0,1), respectively, C1=2r4(1-T / T max) is the random jump intensity that measures the flight intensity of the velociraptor; L F represents the Lévy flight function; u and v are the first normal distribution random number and the second normal distribution random number respectively, β is the default constant, which is equal to 1.5; σ and τ(·) represent the distribution scale parameter and gamma function related to the Lévy flight distribution respectively.

[0087] The multi-strategy improved red-billed blue magpie optimization algorithm is based on the foraging behavior of the red-billed blue magpie, and especially emphasizes the cooperation and competition among individuals. In the process of finding food, individuals will constantly adjust their strategies to avoid falling into local optimality and improve global search capabilities.

[0088] S4: Acquire real-time meteorological data of the prediction area, select a corresponding photovoltaic power prediction sub-model according to the similarity between the real-time meteorological data and the typical meteorological characteristic data, input the real-time meteorological data into the corresponding photovoltaic power prediction sub-model, and output the photovoltaic power prediction result.

[0089] Specifically, the step includes:

[0090] Real-time weather data includes real-time weather data every 15 minutes during the forecast day for the forecast area;

[0091] The Euclidean distance is used to calculate the similarity between the real-time meteorological data and the typical meteorological characteristic data, and the calculation formula is:

[0092]

[0093] Where D(·) represents the Euclidean distance; g represents real-time meteorological data; l represents typical meteorological characteristic data; g z Indicates the zth parameter in real-time meteorological data; l z represents the zth parameter in the typical meteorological characteristic data; Z represents the number of parameters;

[0094] A photovoltaic power prediction sub-model corresponding to the typical meteorological characteristic data having the smallest Euclidean distance to the real-time meteorological data is selected; the real-time meteorological data for every 15 minutes on the forecast day is input into the corresponding photovoltaic power prediction sub-model, and the predicted photovoltaic power is output.

[0095] This embodiment aims at the shortcomings of the traditional red-billed blue magpie optimization algorithm (RBMO algorithm) and provides a multi-strategy improved red-billed blue magpie optimization algorithm (IRBMO algorithm). The difference and advantages of the algorithm are as follows:

[0096] 1. Improvement of population initialization method:

[0097] In the traditional RBMO algorithm, the population is usually initialized by random distribution, which may lead to insufficient diversity of the initial population and easily fall into the local optimal solution;

[0098] The IRBMO algorithm introduces a set of good points for population initialization, which can ensure that the population has high diversity in the entire search space, avoid excessive concentration of initial points, and thus reduce the risk of falling into a local optimal solution.

[0099] 2. Introduction of reverse learning mechanism:

[0100] The traditional RBMO algorithm only relies on the local search and global search capabilities of the current solution when updating individuals, which may show limitations in complex search spaces;

[0101] The IRBMO algorithm introduces a reverse learning mechanism, which generates a reverse solution to the current solution in the individual update phase, and selects the better solution for replacement by comparing the original solution and the reverse solution. This mechanism enhances the diversity of the algorithm, while helping to avoid local optimal solutions and improve the quality of the overall solution.

[0102] 3. Application of Levi's flight strategy:

[0103] The traditional RBMO algorithm itself has certain search capabilities, but it may be insufficient in terms of jumping and global exploration capabilities, especially in complex problems.

[0104] The IRBMO algorithm enhances the individual's jumping ability in the search space by introducing the Lévy flight strategy; Lévy flight generates ultra-distance jumps through random steps, so that the individual can jump out of the local area, expand the search range, enhance the global search ability, and effectively avoid falling into the local optimum.

[0105] 4. Algorithm advantages:

[0106] The IRBMO algorithm can more effectively find high-quality solutions in the search space by improving population initialization, introducing reverse learning and Levy flight strategy, especially performing better in complex multi-modal function optimization problems.

[0107] Compared with the traditional RBMO algorithm, the IRBMO algorithm has stronger global exploration capabilities and can avoid early convergence to suboptimal solutions;

[0108] By comprehensively using good point sets, reverse learning and Levy flight strategy, the IRBMO algorithm can improve the quality of solutions while ensuring diversity, and ultimately find a better global solution.

[0109] By improving population initialization, introducing reverse learning and Levy flight strategy, the IRBMO algorithm has significant advantages over the traditional RBMO algorithm.

[0110] In this embodiment, RMSE (root mean square error) and MAE (mean absolute error) are used as evaluation indicators, and a comparative data table of RMSE and MAE evaluation indicators of different models under different weather types is provided;

[0111] Table 1 is a comparison table of prediction performance index data;

[0112]

[0113] It can be seen from Table 1 that the FCM-IRBMO-BILSTM model proposed in this embodiment has excellent performance in both RMSE and MAE evaluation indicators under different weather types. Figure 2 , Figure 3 , Figure 4 It can be seen that the FCM-IRBMO-BILSTM model proposed in this embodiment has a high prediction accuracy of photovoltaic power under different weather types, indicating that the FCM-IRBMO-BILSTM model has higher accuracy and robustness in photovoltaic power prediction, and has certain value in the research on fully improving the operating efficiency of photovoltaic systems.

[0114] The photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM provided in this embodiment can quickly and accurately predict the output power of the photovoltaic power generation system, provide data support for the power system, help the dispatching department to formulate power generation, peak regulation and dispatching plans, effectively improve the accuracy and robustness of photovoltaic power prediction, and provide new possibilities for the intelligent management and operation optimization of photovoltaic power generation systems.

[0115] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims

1. A photovoltaic power short-term prediction method based on FCM-IRBMO-BILSTM, characterized in that: include: S1: Acquire a historical meteorological data set of a forecast area, and preprocess the historical meteorological data set; S2: dividing the preprocessed historical meteorological data set into multiple clusters by a clustering algorithm, each cluster represents a meteorological data set of a different category, and the cluster center in each cluster is the typical meteorological characteristic data corresponding to the cluster; S3: Based on each type of meteorological data set, the multi-strategy improved red-billed blue magpie optimization algorithm is used to determine the optimal hyperparameters of multiple Bi-LSTM models, and the photovoltaic power prediction sub-model for each type of meteorological data is obtained; S4: Acquire real-time meteorological data of the prediction area, select a corresponding photovoltaic power prediction sub-model according to the similarity between the real-time meteorological data and the typical meteorological characteristic data, input the real-time meteorological data into the corresponding photovoltaic power prediction sub-model, and output the photovoltaic power prediction result.

2. The photovoltaic power short-term prediction method according to claim 1, characterized in that: The historical meteorological data set includes meteorological data with a sampling interval of 15 minutes, a time span of 5 years, and an absolute value of the Pearson correlation coefficient with the direct normal radiation of the sun greater than 0.

2.

3. The photovoltaic power short-term prediction method according to claim 1, characterized in that: The calculation formula of Pearson correlation coefficient is: Where P represents the Pearson correlation coefficient, P∈[-1,1], P=0 means the two variables are uncorrelated, P<0 means the two variables are negatively correlated, and P>0 means the two variables are positively correlated; n represents the number of samples; x i represents the meteorological data of the i-th sample; y i represents the direct normal solar radiation of the i-th sample; Represents the average value of each sample meteorological data; Represents the average value of direct normal solar radiation for each sample.

4. The photovoltaic power short-term prediction method according to claim 1, characterized in that: The preprocessing of the historical meteorological data set includes: Step 1: Eliminate the meteorological data of the period without photovoltaic power in the historical meteorological data set; Filling the missing data in the historical meteorological data set with the mean value of the meteorological data; Using the triple standard deviation principle to remove outliers in the historical meteorological data set; Step 2: Normalize the meteorological data in the historical meteorological data set after elimination and filling. The normalization calculation formula is: Among them, X scale represents the normalized meteorological data; X represents the actual value of the meteorological data; X min Indicates the minimum value of meteorological data; X max Indicates the maximum value of the meteorological data.

5. The photovoltaic power short-term prediction method according to claim 1, characterized in that: The clustering algorithm is a fuzzy C-means clustering algorithm, which gradually reduces the error value of the objective function through iterative operations until the objective function converges to obtain a clustering result; The objective function is expressed as: Among them, J m represents the objective function; n represents the number of samples; C represents the number of cluster centers; Represents the meteorological data x in the mth cluster i For cluster center c i The membership degree of x i represents the meteorological data of the i-th sample; c j represents the jth cluster center.

6. The photovoltaic power short-term prediction method according to claim 1, characterized in that S3 include: S3.1: Divide any type of meteorological data set into a training set and a validation set in proportion; S3.2: Take the meteorological data in the training set as input and the photovoltaic power as output; S3.3: Based on the input and output, the optimal hyperparameters of the Bi-LSTM model are determined by using a multi-strategy improved red-billed blue magpie optimization algorithm, and the photovoltaic power prediction sub-model corresponding to the meteorological data set is obtained; S3.4: Validate the photovoltaic power prediction sub-model corresponding to the meteorological data set based on the validation set.

7. The photovoltaic power short-term prediction method according to claim 1, characterized in that: The multi-strategy improved red-billed blue magpie optimization algorithm includes: Step 1: Use the hyperparameters of the Bi-LSTM model as the population and initialize the population using the good point set; Step 2: In the population, the red-billed blue magpies search for food in small groups or groups. A small group is 2 to 5 red-billed blue magpies, and a group is more than 10 red-billed blue magpies. The mathematical expression of searching for food in small groups is: Among them, X i (t+1) represents the position of the i-th individual in the small group at the t+1 iteration; X i (t) represents the position of the i-th individual in the small group at the tth iteration; p represents the number of individuals in the randomly selected small group; X k (t) represents the position of the kth individual in the randomly selected small group at the tth iteration; X rs (t) represents the search position randomly selected by the small group at the tth iteration; Rand1 represents the first random number; The mathematical representation of searching for food in a flock is: Among them, X i (t+1)' represents the position of the i-th individual in the population at the t+1 iteration; X i (t)′ represents the position of the i-th individual in the population at the tth iteration; q represents the number of individuals in the randomly selected population; X k (t)′ represents the position of the kth individual in the population randomly selected at the tth iteration; X rs (t)′ represents the search position randomly selected by the group at the tth iteration; Rand2 represents the second random number; Step 3: When small food is found, the mathematical model of attacking prey is expressed as: Among them, X food (t) represents the location of the food at the tth iteration; Randn1 represents the first random number used to generate the standard normal distribution; CF represents the dynamic change coefficient, which is used to adjust the speed and intensity of the individual moving towards the food; T represents the total number of iterations; When a large food is found, the mathematical model for attacking the prey is expressed as: Among them, Randn2 represents the second random number used to generate the standard normal distribution; Step 4: After obtaining the food, store the remaining food except for consumption to obtain the optimal hyperparameters; mathematically expressed as: in, represents the fitness of the i-th individual before the position is updated, Represents the fitness of the i-th individual after the position is updated.

8. The photovoltaic power short-term prediction method according to claim 7, characterized in that: Initializing the population using a good point set includes: Determine the dimensions of the optimal hyperparameters and the size of the population; In each dimensional space, select real numbers or real number sequences to form a good point set, which is expressed as: Among them, P d (h) represents the good point set; d represents the dimension; They represent the first, second, and sth real numbers selected in the d-dimensional space respectively; h represents the index variable; One or more points are selected from the best point set as starting points, and random perturbations are added to each point to generate other points in the population. The random perturbations are determined based on the upper and lower bounds of the decision variables of each point.

9. The photovoltaic power short-term prediction method according to claim 7, characterized in that: Also includes: While using the good point set to initialize the population, the reverse learning mechanism strategy is used to calculate the reverse solution of the red-billed blue magpie individual. The calculation formula is: OP i =w·(ub j +lb j )+a i ; Among them, OP i Represents individual a i The reverse solution of i represents the i-th red-billed blue magpie individual; w represents the third random number; ub j represents the upper bound of the j-th decision variable; lb j represents the lower bound of the j-th decision variable.

10. The photovoltaic power short-term prediction method according to claim 1, characterized in that S4 include: Real-time weather data includes real-time weather data every 15 minutes during the forecast day for the forecast area; The similarity between the real-time meteorological data and the typical meteorological characteristic data is calculated using Euclidean distance; Selecting a photovoltaic power prediction sub-model corresponding to the typical meteorological characteristic data having the smallest Euclidean distance to the real-time meteorological data; The real-time meteorological data of every 15 minutes during the forecast day is input into the corresponding PV power forecast sub-model, and the predicted PV power is output.

Images