Evolutionary framework based fuzzy generative adversarial network wind power prediction method and device

Abstract

The present application relates to the technical field of wind power prediction, in particular to a wind power prediction method and device based on an evolutionary framework fuzzy generative adversarial network. The method comprises: constructing a wind turbine data generation model, the wind turbine data generation model being used to generate a plurality of wind turbine data; constructing a wind turbine data rationality discrimination model, the wind turbine data rationality discrimination model being used to determine whether the generated wind turbine data meets the requirements; based on the wind turbine data generation model and the wind turbine data rationality discrimination model, constructing a sample generation optimization model based on evolutionary computation; based on the sample generation optimization model based on evolutionary computation and real wind turbine data, generating a plurality of wind turbine sample data; constructing a wind power prediction initial model, training the power prediction initial model through the wind turbine sample data, and obtaining a trained wind power prediction model; and predicting wind power through the wind power prediction model. The present application can improve the effect of wind power prediction under complex conditions.

Description

Technical Field

[0001] This invention relates to the field of wind power prediction technology, and in particular to a fuzzy adversarial generative network method and apparatus for wind power prediction based on an evolutionary framework. Background Technology

[0002] Wind power, as an important sustainable energy source, has experienced rapid development over the past few decades. By the end of 2019, the global installed capacity of wind turbines (hereinafter referred to as "wind turbines") reached 651 GW, covering more than 6% of global electricity demand. However, unlike traditional electricity, wind turbines operate in complex and variable environments, posing significant challenges to wind turbine power generation forecasting and negatively impacting the stability of power systems. To address this issue, wind power forecasting technology has been developed to predict wind power generation, thereby assisting in the stable operation of power systems. Wind power forecasting is a crucial soft-sensor technology that plays a vital role in the application of wind power and power systems.

[0003] Short-term forecasting is crucial in wind power prediction. It typically utilizes historical data from the past few hours to predict wind power output for the next few hours. Short-term wind power prediction methods mainly include mechanistic, statistical, and artificial intelligence (AI) based approaches. Mechanism-based and statistical methods often require substantial additional information (e.g., high-resolution weather data, physical data of the wind turbine's environment), which is often difficult to obtain. Therefore, AI-based methods (including shallow learning and deep learning methods) are increasingly being applied to short-term wind power prediction.

[0004] In AI-based short-term wind power prediction methods, shallow learning methods are easy to deploy and use, and have been widely researched and applied. These methods typically use historical data to build regression models to predict wind power for the next few hours. Among these methods, shallow learning methods based on neural network models, support vector regression models, and extreme learning machine models have received considerable attention. Meanwhile, with the rapid development of deep learning technology, deep learning methods have been extensively studied in the field of short-term wind power prediction in recent years. Deep learning methods introduce more hidden layers into the regression model, significantly improving model performance and thus enabling more accurate wind power prediction.

[0005] However, despite the remarkable success of many methods, some problems still need to be addressed:

[0006] (1) Although many existing methods can achieve good average prediction performance, their prediction performance will deteriorate significantly under some extreme conditions (such as drastically changing environments), which poses a great threat to the stability of power systems.

[0007] (2) Due to the limited number of samples under these extreme conditions, it is not possible to train prediction models based on artificial intelligence methods well. Summary of the Invention

[0008] This invention provides a method and apparatus for wind power prediction based on an evolutionary framework and fuzzy adversarial generative network. The technical solution is as follows:

[0009] On the one hand, a fuzzy adversarial generative network wind power prediction method based on an evolutionary framework is provided. This method is implemented by electronic devices and includes:

[0010] S1. Construct a wind turbine data generation model, which is used to generate multiple wind turbine data under different conditions;

[0011] S2. Construct a wind turbine data rationality judgment model, which is used to determine whether multiple generated wind turbine data meet the requirements of wind turbine data;

[0012] S3. Based on the wind turbine data generation model and the wind turbine data rationality judgment model, construct a sample generation optimization model based on evolutionary calculation;

[0013] S4. Based on the sample generation optimization model of the evolutionary calculation and the real wind turbine data, generate multiple wind turbine sample data;

[0014] S5. Construct an initial model for wind power prediction, and train the initial model for power prediction using the wind turbine sample data to obtain a trained wind power prediction model.

[0015] S6. Predict wind power using the wind power prediction model.

[0016] Optionally, the construction of the wind turbine data generation model in S1 includes:

[0017] S11. Preprocess the acquired real wind turbine data to form a basic data sample from real wind turbine data of different wind turbines at different times.

[0018]

[0019] in, For the nth sample after preprocessing, S iLet m be the wind power vector of all wind turbines in the field at time i, m be the number of historical wind turbine data used in the prediction process, and t be the number of wind turbine data to be predicted preset in the prediction process.

[0020] S12. A deep convolutional neural network is used to construct a wind turbine data generation model. The loss functions for the generator network and the adversarial network are as follows:

[0021]

[0022] Where G is the generator network and D is the discriminator network. The data consists of real wind turbine sample data, z is a random vector input into the generator network, E represents the expectation, and p represents the probability distribution;

[0023] S13. To enable the constructed generative network to generate wind turbine data under specific conditions as required, a generative data rationality judgment model Ω is constructed:

[0024] δ (r) =Ω(G(z))

[0025] Where G(z) represents the generated wind turbine sample data, δ (r) To determine the result, if δ (r) A value of 1 indicates that the generated sample is reasonable; if δ (r) A value of 0 indicates that the generated sample is unreasonable;

[0026] S14. Based on the generated data rationality judgment model Ω, construct a supervised deep convolutional generation network based on the judgment model. The loss function of the supervised deep convolutional generation model based on the judgment model is designed as follows:

[0027]

[0028]

[0029] Where U is the number of samples in each training iteration, u represents the u-th sample, and η is the threshold for rationality judgment;

[0030] S15. Based on the generated data rationality discrimination model Ω and the loss function of the supervised depthwise convolutional generation model based on the discrimination model, the loss function of the discrimination model is constructed as follows:

[0031]

[0032] in, It consists of real wind turbine sample data from each iteration.

[0033] Optionally, the construction of the wind turbine data rationality judgment model in S2 includes:

[0034] S21. By fitting the correlation probability density function of wind power generation from any two adjacent wind turbines using a large number of real samples, the correlation data between the two wind turbine data is calculated using the Pearson correlation coefficient:

[0035]

[0036] Where, d x and d y These are data from two wind turbines, pcc(d) x ,d y ) represents the calculated correlation value, and E represents the expected value;

[0037] S22. Construct the antecedent of the membership function of the fuzzy logic system. Input the obtained correlation data. First, define the term "similarity" for the antecedent of the membership function and use the commonly used triangular or trapezoidal membership function.

[0038] I: Fix the peak point of the relevant probability density function to the vertex of the triangle. Based on the principle that the area of ​​a triangle equals the area of ​​its probability density function, the left endpoint P1 of the trigonometric function is determined:

[0039]

[0040] Where ψ(x) is the probability density function of x. It is the left region of ψ(x). It is the left region of the membership function of the triangle, and P represents the point to be optimized, which refers to the coordinate value of any point on the base of the triangle function;

[0041] II: Right endpoint of the membership function of a triangle, right point P r Represented as:

[0042]

[0043] in, It is the right region of the probability density function. It is the right region of the membership functions of the triangle;

[0044] Thus, the antecedent membership function μ of the fuzzy logic system is obtained. ant ;

[0045] S23. Define two types of fuzzy logic system rules to determine whether generated samples are reasonable:

[0046] Type I: If the power correlation of all adjacent wind turbines is similar, then the generated sample is reasonable;

[0047] Type II: If the power correlation between any adjacent wind turbines is dissimilar, the generated sample is unreasonable;

[0048] S24. Based on the rules in S23, construct the membership functions of the consequents of the triangle, thereby obtaining the membership functions μ of the consequents of the fuzzy logic system. csq ;

[0049] S25. Perform fuzzy inference based on the defined membership functions of the antecedent and consequent, and the rules:

[0050]

[0051] in, Let R be the sample data of the wind turbines that need to be judged, and R be the rules of the fuzzy logic system. It is the result value of defuzzification, representing the probability that the target sample meets the requirements of the wind turbine data;

[0052] S26, according to The value is used to determine whether the input wind turbine sample data is reasonable.

[0053] Optionally, in step S3, based on the wind turbine data generation model and the wind turbine data rationality judgment model, a sample generation optimization model based on evolutionary computation is constructed, including:

[0054] S31. Use the wind turbine data generation model as a population generation module for evolutionary computation;

[0055] S32. Use the wind turbine data rationality judgment model as the selection condition module for evolutionary calculation;

[0056] S33. Use the wind turbine data generation model as a crossover and mutation module for evolutionary computation;

[0057] S34. Based on the evolutionary calculation modules constructed above, wind turbine sample data under multiple specific conditions are generated through iterative loops:

[0058]

[0059] Among them, G re To construct the evolutionary computational generative model, This is real wind turbine sample data. This refers to wind turbine sample data under specific conditions generated through planned iterations.

[0060] Optionally, the initial model for wind power prediction in step S5, which involves training the initial model using the wind turbine sample data to obtain a trained wind power prediction model, includes:

[0061] S51. Mix the real wind turbine sample data with the generated wind turbine sample data and use it as training data for the short-term power prediction model:

[0062] S T =S O US G

[0063] Among them, S O For the original wind turbine data, S G To generate a large amount of wind turbine data under special conditions, S T Data for the final short-term wind power prediction model;

[0064] S52, Based on the training data S T Training a short-term wind power prediction model:

[0065]

[0066] Among them, f nn This is a short-term wind power prediction model based on neural networks. This is the prediction result for short-term wind power.

[0067] On the other hand, a fuzzy adversarial generative network (AGN) wind power prediction device based on an evolutionary framework is provided. This device is applied to the fuzzy adversarial generative network wind power prediction method based on an evolutionary framework. The device includes:

[0068] The first construction module is used to construct a wind turbine data generation model, which is used to generate multiple wind turbine data under different conditions;

[0069] The second construction module is used to construct a wind turbine data rationality judgment model, which is used to determine whether the generated multiple wind turbine data meet the requirements of wind turbine data.

[0070] The third construction module is used to construct a sample generation optimization model based on evolutionary calculation, based on the wind turbine data generation model and the wind turbine data rationality judgment model.

[0071] The generation module is used to generate multiple wind turbine sample data based on the sample generation optimization model derived from the evolutionary calculation and real wind turbine data.

[0072] The training module is used to construct an initial model for wind power prediction. The initial model is trained using the wind turbine sample data to obtain a trained wind power prediction model.

[0073] The prediction module is used to predict wind power using the wind power prediction model.

[0074] Optionally, the first building module is configured to:

[0075] S11. Preprocess the acquired real wind turbine data to form a basic data sample from real wind turbine data of different wind turbines at different times.

[0076]

[0077] in, For the nth sample after preprocessing, S i Let m be the wind power vector of all wind turbines in the field at time i, m be the number of historical wind turbine data used in the prediction process, and t be the number of wind turbine data to be predicted preset in the prediction process.

[0078] S12. A deep convolutional neural network is used to construct a wind turbine data generation model. The loss functions for the generator network and the adversarial network are as follows:

[0079]

[0080] Where G is the generator network and D is the discriminator network. The data consists of real wind turbine sample data, z is a random vector input into the generator network, E represents the expectation, and p represents the probability distribution;

[0081] S13. To enable the constructed generative network to generate wind turbine data under specific conditions as required, a generative data rationality judgment model Ω is constructed:

[0082] δ (r) =Ω(G(z))

[0083] Where G(z) represents the generated wind turbine sample data, δ (r) To determine the result, if δ (r) A value of 1 indicates that the generated sample is reasonable; if δ (r) A value of 0 indicates that the generated sample is unreasonable;

[0084] S14. Based on the generated data rationality judgment model Ω, construct a supervised deep convolutional generation network based on the judgment model. The loss function of the supervised deep convolutional generation model based on the judgment model is designed as follows:

[0085]

[0086]

[0087] Where U is the number of samples in each training iteration, u represents the u-th sample, and η is the threshold for rationality judgment;

[0088] S15. Based on the generated data rationality discrimination model Ω and the loss function of the supervised depthwise convolutional generation model based on the discrimination model, the loss function of the discrimination model is constructed as follows:

[0089]

[0090] in, It consists of real wind turbine sample data from each iteration.

[0091] Optionally, the second building module is used for:

[0092] S21. By fitting the correlation probability density function of wind power generation from any two adjacent wind turbines using a large number of real samples, the correlation data between the two wind turbine data is calculated using the Pearson correlation coefficient:

[0093]

[0094] Where, d x and d y These are data from two wind turbines, pcc(d) x ,d y ) represents the calculated correlation value, and E represents the expected value;

[0095] S22. Construct the antecedent of the membership function of the fuzzy logic system. Input the obtained correlation data. First, define the term "similarity" for the antecedent of the membership function and use the commonly used triangular or trapezoidal membership function.

[0096] I: Fix the peak point of the relevant probability density function to the vertex of the triangle. Based on the principle that the area of ​​a triangle equals the area of ​​its probability density function, the left endpoint P1 of the trigonometric function is determined:

[0097]

[0098] Where ψ(x) is the probability density function of x. It is the left region of ψ(x). It is the left region of the membership function of the triangle, and P represents the point to be optimized, which refers to the coordinate value of any point on the base of the triangle function;

[0099] II: Right endpoint of the membership function of a triangle, right point P r Represented as:

[0100]

[0101] in, It is the right region of the probability density function. It is the right region of the membership functions of the triangle;

[0102] Thus, the antecedent membership function μ of the fuzzy logic system is obtained. ant ;

[0103] S23. Define two types of fuzzy logic system rules to determine whether generated samples are reasonable:

[0104] Type I: If the power correlation of all adjacent wind turbines is similar, then the generated sample is reasonable;

[0105] Type II: If the power correlation between any adjacent wind turbines is dissimilar, the generated sample is unreasonable;

[0106] S24. Based on the rules in S23, construct the membership functions of the consequents of the triangle, thereby obtaining the membership functions μ of the consequents of the fuzzy logic system. csq ;

[0107] S25. Perform fuzzy inference based on the defined membership functions of the antecedent and consequent, and the rules:

[0108]

[0109] in, Let R be the sample data of the wind turbines that need to be judged, and R be the rules of the fuzzy logic system. It is the result value of defuzzification, representing the probability that the target sample meets the requirements of the wind turbine data;

[0110] S26, according to The value is used to determine whether the input wind turbine sample data is reasonable.

[0111] Optionally, the third building module is used for:

[0112] S31. Use the wind turbine data generation model as a population generation module for evolutionary computation;

[0113] S32. Use the wind turbine data rationality judgment model as the selection condition module for evolutionary calculation;

[0114] S33. Use the wind turbine data generation model as a crossover and mutation module for evolutionary computation;

[0115] S34. Based on the evolutionary calculation modules constructed above, wind turbine sample data under multiple specific conditions are generated through iterative loops:

[0116]

[0117] Among them, G re To construct the evolutionary computational generative model, This is real wind turbine sample data. This refers to wind turbine sample data under specific conditions generated through planned iterations.

[0118] Optionally, the training module is used for:

[0119] S51. Mix the real wind turbine sample data with the generated wind turbine sample data and use it as training data for the short-term power prediction model:

[0120] S T =S O US G

[0121] Among them, S O For the original wind turbine data, S G To generate a large amount of wind turbine data under special conditions, S T Data for the final short-term wind power prediction model;

[0122] S52, Based on the training data S T Training a short-term wind power prediction model:

[0123]

[0124] Among them, f nn This is a short-term wind power prediction model based on neural networks. This is the prediction result for short-term wind power.

[0125] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described fuzzy adversarial generative network wind power prediction method based on an evolutionary framework.

[0126] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement the above-described fuzzy adversarial generative network wind power prediction method based on an evolutionary framework.

[0127] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0128] In this embodiment of the invention, the method can predict short-term wind power. Based on an evolutionary computation framework, it utilizes a fuzzy inference system and a generative adversarial network to generate wind turbine data under complex conditions, thereby using sufficient data to train a short-term wind power prediction model. First, a wind turbine data rationality judgment model is constructed based on a fuzzy logic system to address the correlation between wind turbines. Then, a supervised deep convolutional generative adversarial network is proposed to generate a large amount of wind turbine data. Finally, a wind turbine complex condition data generation framework based on an evolutionary framework is constructed, which can generate a large amount of wind turbine data under complex conditions. The generated wind turbine data under complex conditions is used as training samples to train the short-term wind power prediction model, improving the accuracy of wind power prediction under complex conditions. Attached Figure Description

[0129] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0130] Figure 1 This is a flowchart of a fuzzy adversarial generative network wind power prediction method based on an evolutionary framework, provided by an embodiment of the present invention.

[0131] Figure 2 This is a flowchart of a fuzzy adversarial generative network wind power prediction method based on an evolutionary framework, provided by an embodiment of the present invention.

[0132] Figure 3 This is a block diagram of a fuzzy adversarial generative network wind power prediction device based on an evolutionary framework provided in an embodiment of the present invention;

[0133] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0134] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0135] This invention provides a fuzzy adversarial generative network (GFCN) wind power prediction method based on an evolutionary framework. This method can be implemented using an electronic device, such as a terminal or a server. In this embodiment, a wind farm with 24 wind turbines is used, and data from these turbines is used for short-term wind power prediction. The sampling interval for the wind turbine data is 30 minutes, and historical wind power data from the past four hours is used to predict wind power data for the next two hours. Figure 1The flowchart of the fuzzy adversarial generative network wind power prediction method based on an evolutionary framework is shown below. Figure 2 The flowchart shown illustrates that the processing flow of this method may include the following steps:

[0136] S1. Construct a wind turbine data generation model. The wind turbine data generation model is used to generate multiple wind turbine data under different conditions.

[0137] Optionally, the construction of the wind turbine data generation model in S1 includes:

[0138] S11. Preprocess the acquired real wind turbine data to form a basic data sample from real wind turbine data of different wind turbines at different times.

[0139]

[0140] in, For the nth sample after preprocessing, S i Let be the wind power vector of all wind turbines in the entire field at time i, m be the number of historical wind turbine data used in the prediction process, and t be the preset number of wind turbine data to be predicted in the prediction process. It should be noted that when using the wind power prediction model, m historical wind turbine data are used to predict t future wind turbine data. Therefore, the samples used to train the initial wind power prediction model also need to follow this rule, with each sample including m historical wind turbine data and t predicted ground truth values. Therefore, in the above expression, It includes m+t data points. In a specific implementation, one feasible setting is as follows: S i The dimension is 24, the value of m is 8, and the value of t is 4.

[0141] S12. A deep convolutional neural network is used to construct a wind turbine data generation model. The loss functions for the generator network and the adversarial network are as follows:

[0142]

[0143] Where G is the generator network and D is the discriminator network. The data consists of real wind turbine sample data, z is a random vector input into the generator network, E represents the expectation, and p represents the probability distribution;

[0144] S13. To enable the constructed generative network to generate wind turbine data under specific conditions as required, a generative data rationality judgment model Ω is constructed:

[0145] δ (r) =Ω(G(z))

[0146] Where G(z) represents the generated wind turbine sample data, δ (r) To determine the result, if δ(r) A value of 1 indicates that the generated sample is reasonable; if δ (r) A value of 0 indicates that the generated sample is unreasonable;

[0147] S14. Based on the data rationality judgment model Ω, construct a supervised deep convolutional generation network based on the judgment model. The loss function of the supervised deep convolutional generation model based on the judgment model is designed as follows:

[0148]

[0149]

[0150] Where U is the number of samples in each training iteration, u represents the u-th sample, and η is the threshold for rationality judgment;

[0151] S15. Based on the loss functions of the generated data rationality discrimination model Ω and the supervised depthwise convolutional generation model based on the discrimination model, the loss function of the discrimination model is constructed as follows:

[0152]

[0153] in, It consists of real wind turbine sample data from each iteration.

[0154] S2. Construct a wind turbine data rationality judgment model. The wind turbine data rationality judgment model is used to determine whether the generated multiple wind turbine data meet the requirements of wind turbine data.

[0155] Optionally, the wind turbine data rationality judgment model in S2 includes:

[0156] S21. By fitting the correlation probability density function of wind power generation from any two adjacent wind turbines using a large number of real samples, the correlation data between the two wind turbine data is calculated using the Pearson correlation coefficient:

[0157]

[0158] Where, d x and d y These are data from two wind turbines, pcc(d) x ,d y ) represents the calculated correlation value, and E represents the expected value;

[0159] S22. Construct the antecedent of the membership function of the fuzzy logic system. Input the obtained correlation data. First, define the term "similarity" for the antecedent of the membership function and use the commonly used triangular or trapezoidal membership function.

[0160] I: Fix the peak point of the relevant probability density function to the vertex of the triangle. Based on the principle that the area of ​​a triangle equals the area of ​​its probability density function, the left endpoint P1 of the trigonometric function is determined:

[0161]

[0162] Where ψ(x) is the probability density function of x. It is the left region of ψ(x). It is the left region of the membership function of the triangle, and P represents the point to be optimized, which refers to the coordinate value of any point on the base of the triangle function;

[0163] II: Right endpoint of the membership function of a triangle, right point P r Represented as:

[0164]

[0165] in, It is the right region of the probability density function. It is the right region of the membership functions of the triangle;

[0166] Thus, the antecedent membership function μ of the fuzzy logic system is obtained. ant ;

[0167] S23. Define two types of fuzzy logic system rules to determine whether generated samples are reasonable:

[0168] Type I: If the power correlation of all adjacent wind turbines is similar, then the generated sample is reasonable;

[0169] Type II: If the power correlation between any adjacent wind turbines is dissimilar, the generated sample is unreasonable;

[0170] S24. Based on the rules in S23, construct the membership functions of the consequents of the triangle, thereby obtaining the membership functions μ of the consequents of the fuzzy logic system. csq ;

[0171] S25. Perform fuzzy inference based on the defined membership functions of the antecedent and consequent, and the rules:

[0172]

[0173] in, Let R be the sample data of the wind turbines that need to be judged, and R be the rules of the fuzzy logic system. It is the result value of defuzzification, representing the probability that the target sample meets the requirements of the wind turbine data;

[0174] S26, according to The value is used to determine whether the input wind turbine sample data is reasonable.

[0175] In one feasible implementation, a reasonableness threshold can be preset. When determining whether the wind turbine sample data is reasonable, the wind turbine sample data can be... The value is compared with a reasonableness threshold. If the wind turbine sample data... If the value is greater than or equal to the reasonableness threshold, the wind turbine sample data is judged to be reasonable; if the wind turbine sample data... If the value is less than the reasonableness threshold, the wind turbine sample data is judged to be unreasonable.

[0176] S3. Based on the wind turbine data generation model and the wind turbine data rationality judgment model, construct a sample generation optimization model based on evolutionary computation.

[0177] Optionally, the sample generation optimization model based on evolutionary computation, constructed in S3 based on the wind turbine data generation model and the wind turbine data rationality judgment model, may include the following steps 31-34:

[0178] S31. Use the wind turbine data generation model as a population generation module for evolutionary computation;

[0179] S32. Use the wind turbine data rationality judgment model as the selection condition module for evolutionary calculation;

[0180] S33. Use the wind turbine data generation model as a crossover and mutation module for evolutionary computation;

[0181] S34. Based on the evolutionary calculation modules constructed above, wind turbine sample data under multiple specific conditions are generated through iterative loops:

[0182]

[0183] Among them, G re To construct the evolutionary computational generative model, This is real wind turbine sample data. This refers to wind turbine sample data under specific conditions generated through planned iterations. This ensures a sufficient number of samples can be generated. A short-term power prediction model that meets the required prediction accuracy is trained. In a specific implementation, a feasible setup is as follows: initial samples... The sample size was chosen to be 5000, and samples were generated. The sample size was also chosen to be 5000.

[0184] S4. Based on the evolutionary computation-based sample generation optimization model and real wind turbine data, generate multiple wind turbine sample data.

[0185] In one feasible implementation, real wind turbine data is input into an evolutionary computational sample generation optimization model, which generates wind turbine sample data under different conditions.

[0186] S5. Construct an initial model for wind power prediction. Train the initial model using wind turbine sample data to obtain a trained wind power prediction model.

[0187] Optionally, in S5, the initial model for wind power prediction is constructed by training the initial model using wind turbine sample data to obtain a trained wind power prediction model, including:

[0188] S51. Mix the real wind turbine sample data with the generated wind turbine sample data and use it as training data for the short-term power prediction model:

[0189] S T =S O US G

[0190] Among them, S O For the original wind turbine data, S G To generate a large amount of wind turbine data under special conditions, S T Data for the final short-term wind power prediction model;

[0191] S52, based on training data S T Training a short-term wind power prediction model:

[0192]

[0193] Among them, f nn This is a short-term wind power prediction model based on neural networks. This is a prediction of short-term wind power. In this implementation, It consists of four values: power predictions for the next 30 minutes, 60 minutes, 90 minutes, and 120 minutes. It uses a trained f... nn It can then perform wind power prediction under complex conditions.

[0194] S6. Predict wind power using a wind power prediction model.

[0195] In this embodiment of the invention, the method can predict short-term wind power. Based on an evolutionary computation framework, it utilizes a fuzzy inference system and a generative adversarial network to generate wind turbine data under complex conditions, thereby using sufficient data to train a short-term wind power prediction model. First, a wind turbine data rationality judgment model is constructed based on a fuzzy logic system to address the correlation between wind turbines. Then, a supervised deep convolutional generative adversarial network is proposed to generate a large amount of wind turbine data. Finally, a wind turbine complex condition data generation framework based on an evolutionary framework is constructed, which can generate a large amount of wind turbine data under complex conditions. The generated wind turbine data under complex conditions is used as training samples to train the short-term wind power prediction model, improving the accuracy of wind power prediction under complex conditions.

[0196] Figure 3 This is a block diagram of a fuzzy adversarial generative network wind power prediction device based on an evolutionary framework, according to an exemplary embodiment. (Refer to...) Figure 3 The device 300 includes:

[0197] The first construction module 310 is used to construct a wind turbine data generation model, which is used to generate multiple wind turbine data under different conditions;

[0198] The second construction module 320 is used to construct a wind turbine data rationality judgment model, which is used to determine whether the generated multiple wind turbine data meet the requirements of wind turbine data.

[0199] The third construction module 330 is used to construct a sample generation optimization model based on evolutionary calculation based on the wind turbine data generation model and the wind turbine data rationality judgment model.

[0200] The generation module 340 is used to generate multiple wind turbine sample data based on the sample generation optimization model of the evolutionary calculation and real wind turbine data.

[0201] The training module 350 is used to construct an initial model for wind power prediction. The initial model for power prediction is trained using the wind turbine sample data to obtain a trained wind power prediction model.

[0202] The prediction module 360 ​​is used to predict wind power through the wind power prediction model.

[0203] Optionally, the first building module 310 is configured to:

[0204] S11. Preprocess the acquired real wind turbine data to form a basic data sample from real wind turbine data of different wind turbines at different times.

[0205]

[0206] in, For the nth sample after preprocessing, S i Let m be the wind power vector of all wind turbines in the field at time i, m be the number of historical wind turbine data used in the prediction process, and t be the number of wind turbine data to be predicted preset in the prediction process.

[0207] S12. A deep convolutional neural network is used to construct a wind turbine data generation model. The loss functions for the generator network and the adversarial network are as follows:

[0208]

[0209] Where G is the generator network and D is the discriminator network. The data consists of real wind turbine sample data, z is a random vector input into the generator network, E represents the expectation, and p represents the probability distribution;

[0210] S13. To enable the constructed generative network to generate wind turbine data under specific conditions as required, a generative data rationality judgment model Ω is constructed:

[0211] δ (r) =Ω(G(z))

[0212] Where G(z) represents the generated wind turbine sample data, δ (r) To determine the result, if δ (r) A value of 1 indicates that the generated sample is reasonable; if δ (r) A value of 0 indicates that the generated sample is unreasonable;

[0213] S14. Based on the generated data rationality judgment model Ω, construct a supervised deep convolutional generation network based on the judgment model. The loss function of the supervised deep convolutional generation model based on the judgment model is designed as follows:

[0214]

[0215]

[0216] Where U is the number of samples in each training iteration, u represents the u-th sample, and η is the threshold for rationality judgment;

[0217] S15. Based on the generated data rationality discrimination model Ω and the loss function of the supervised depthwise convolutional generation model based on the discrimination model, the loss function of the discrimination model is constructed as follows:

[0218]

[0219] in, It consists of real wind turbine sample data from each iteration.

[0220] Optionally, the second building module 320 is used for:

[0221] S21. By fitting the correlation probability density function of wind power generation from any two adjacent wind turbines using a large number of real samples, the correlation data between the two wind turbine data is calculated using the Pearson correlation coefficient:

[0222]

[0223] Where, d x and d y These are data from two wind turbines, pcc(d) x ,d y ) represents the calculated correlation value, and E represents the expected value;

[0224] S22. Construct the antecedent of the membership function of the fuzzy logic system. Input the obtained correlation data. First, define the term "similarity" for the antecedent of the membership function and use the commonly used triangular or trapezoidal membership function.

[0225] I: Fix the peak point of the relevant probability density function to the vertex of the triangle. Based on the principle that the area of ​​a triangle equals the area of ​​its probability density function, the left endpoint P1 of the trigonometric function is determined:

[0226]

[0227]

[0228] Where ψ(x) is the probability density function of x. It is the left region of ψ(x). It is the left region of the membership function of the triangle, and P represents the point to be optimized, which refers to the coordinate value of any point on the base of the triangle function;

[0229] II: Right endpoint of the membership function of a triangle, right point P r Represented as:

[0230]

[0231] in, It is the right region of the probability density function. It is the right region of the membership functions of the triangle;

[0232] Thus, the antecedent membership function μ of the fuzzy logic system is obtained. ant ;

[0233] S23. Define two types of fuzzy logic system rules to determine whether generated samples are reasonable:

[0234] Type I: If the power correlation of all adjacent wind turbines is similar, then the generated sample is reasonable;

[0235] Type II: If the power correlation between any adjacent wind turbines is dissimilar, the generated sample is unreasonable;

[0236] S24. Based on the rules in S23, construct the membership functions of the consequents of the triangle, thereby obtaining the membership functions μ of the consequents of the fuzzy logic system. csq ;

[0237] S25. Perform fuzzy inference based on the defined membership functions of the antecedent and consequent, and the rules:

[0238]

[0239] in, Let R be the sample data of the wind turbines that need to be judged, and R be the rules of the fuzzy logic system. It is the result value of defuzzification, representing the probability that the target sample meets the requirements of the wind turbine data;

[0240] S26, according to The value is used to determine whether the input wind turbine sample data is reasonable.

[0241] Optionally, the third building module 330 is used for:

[0242] S31. Use the wind turbine data generation model as a population generation module for evolutionary computation;

[0243] S32. Use the wind turbine data rationality judgment model as the selection condition module for evolutionary calculation;

[0244] S33. Use the wind turbine data generation model as a crossover and mutation module for evolutionary computation;

[0245] S34. Based on the evolutionary calculation modules constructed above, wind turbine sample data under multiple specific conditions are generated through iterative loops:

[0246]

[0247] Among them, G re To construct the evolutionary computational generative model, This is real wind turbine sample data. This refers to wind turbine sample data under specific conditions generated through planned iterations.

[0248] Optionally, the training module 350 is used for:

[0249] S51. Mix the real wind turbine sample data with the generated wind turbine sample data and use it as training data for the short-term power prediction model:

[0250] S T =S O US G

[0251] Among them, S O For the original wind turbine data, S G To generate a large amount of wind turbine data under special conditions, S T Data for the final short-term wind power prediction model;

[0252] S52, Based on the training data S T Training a short-term wind power prediction model:

[0253]

[0254] Among them, f nn This is a short-term wind power prediction model based on neural networks. This is the prediction result for short-term wind power.

[0255] In this embodiment of the invention, the method can predict short-term wind power. Based on an evolutionary computation framework, it utilizes a fuzzy inference system and a generative adversarial network to generate wind turbine data under complex conditions, thereby using sufficient data to train a short-term wind power prediction model. First, a wind turbine data rationality judgment model is constructed based on a fuzzy logic system to address the correlation between wind turbines. Then, a supervised deep convolutional generative adversarial network is proposed to generate a large amount of wind turbine data. Finally, a wind turbine complex condition data generation framework based on an evolutionary framework is constructed, which can generate a large amount of wind turbine data under complex conditions. The generated wind turbine data under complex conditions is used as training samples to train the short-term wind power prediction model, improving the accuracy of wind power prediction under complex conditions.

[0256] Figure 4 This is a schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present invention. The electronic device 400 may vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 401 and one or more memories 402. The memory 402 stores at least one instruction, which is loaded and executed by the processor 401 to implement the steps of the above-described fuzzy adversarial generative network wind power prediction method based on an evolutionary framework.

[0257] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the aforementioned fuzzy adversarial generative network wind power prediction method based on an evolutionary framework. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0258] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0259] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A wind power prediction method based on an evolutionary framework using fuzzy adversarial generative networks, characterized in that, The method includes: S1. Construct a wind turbine data generation model, which is used to generate multiple wind turbine data under different conditions; S2. Construct a wind turbine data rationality judgment model, which is used to determine whether multiple generated wind turbine data meet the requirements of wind turbine data; S3. Based on the wind turbine data generation model and the wind turbine data rationality judgment model, construct a sample generation optimization model based on evolutionary calculation; S4. Based on the sample generation optimization model of the evolutionary calculation and the real wind turbine data, generate multiple wind turbine sample data; S5. Construct an initial model for wind power prediction, and train the initial model for power prediction using the wind turbine sample data to obtain a trained wind power prediction model. S6. Predict wind power using the wind power prediction model; Wherein, S1 includes: S11. Preprocess the acquired real wind turbine data to form a basic data sample from real wind turbine data of different wind turbines at different times. S12. Use deep convolutional neural networks to build a wind turbine data generation model; S13. In order to enable the constructed generator network to generate wind turbine data under specific conditions as required, a data generation rationality judgment model Ω is constructed. S14. Based on the data rationality judgment model Ω, construct a supervised deep convolutional generative network based on the judgment model; S15. Based on the generated data rationality discrimination model Ω and the loss function of the supervised deep convolutional generation model based on the discrimination model, construct the loss function of the discrimination model; Wherein, S2 includes: S21. Fit the correlation probability density function of wind power generation generated by any two adjacent wind turbines using a large number of real samples, and use the Pearson correlation coefficient to calculate the correlation data between the two wind turbine data. S22. Construct the antecedent of the membership function of the fuzzy logic system. Input the obtained correlation data. First, define the term "similarity" for the antecedent of the membership function and use the commonly used triangular or trapezoidal membership function. S23. Define two types of fuzzy logic system rules to determine whether generated samples are reasonable: S24. Based on the rules in S23, construct the membership functions of the consequents of the triangle, thereby obtaining the membership functions μ of the consequents of the fuzzy logic system. csq ; S25. Perform fuzzy inference based on the defined membership functions of the antecedent and consequent, and the rules: S26. Based on the defuzzification result The value is used to determine whether the input fan sample data is reasonable; Wherein, S3 includes: S31. Use the wind turbine data generation model as a population generation module for evolutionary computation; S32. Use the wind turbine data rationality judgment model as the selection condition module for evolutionary calculation; S33. Use the wind turbine data generation model as a crossover and mutation module for evolutionary computation; S34. Based on the various modules of evolutionary computation constructed above, wind turbine sample data under multiple specific conditions are generated through iterative loops.

2. The method according to claim 1, characterized in that, The formula for preprocessing the acquired real wind turbine data in S11 to form a basic data sample from real wind turbine data of different wind turbines at different times is expressed as follows; in, For the nth sample after preprocessing, S i Let m be the wind power vector of all wind turbines in the field at time i, m be the number of historical wind turbine data used in the prediction process, and t be the number of wind turbine data to be predicted preset in the prediction process. S12 uses a deep convolutional neural network to construct a wind turbine data generation model. The loss functions for the generator network and the adversarial network are as follows: Where G is the generator network and D is the discriminator network. The data consists of real wind turbine sample data, z is a random vector input into the generator network, E represents the expectation, and p represents the probability distribution; The data rationality judgment model Ω constructed in S13 is as follows: d (r) =Ω(G(z)) Where G(z) represents the generated wind turbine sample data, δ (r) To determine the result, if δ (r) A value of 1 indicates that the generated sample is reasonable; if δ (r) A value of 0 indicates that the generated sample is unreasonable; Based on the data rationality judgment model Ω generated in S14, a supervised deep convolutional generation network based on the judgment model is constructed. The loss function of the supervised deep convolutional generation model based on the judgment model is designed as follows: Where U is the number of samples in each training iteration, u represents the u-th sample, and η is the threshold for rationality judgment; The loss function of the discrimination model constructed in S15 based on the loss function of the generated data rationality discrimination model Ω and the loss function of the supervised depthwise convolutional generation model based on the discrimination model is as follows: in, It consists of real wind turbine sample data from each iteration.

3. The method according to claim 1, characterized in that, S21 involves fitting the correlation probability density function of wind power generation from any two adjacent wind turbines using a large number of real samples, and calculating the correlation data between the two wind turbine data using the Pearson correlation coefficient, as shown in the following formula: Where, d x and d y These are data from two wind turbines, pcc(d) x ,d y ) represents the calculated correlation value, and E represents the expected value; The preceding part of the membership function of the fuzzy logic system in S22 is constructed by inputting the obtained correlation data. First, the terminology of the membership function preceding part is defined as similarity. Commonly used triangular or trapezoidal membership functions are adopted and defined as follows. I: Fix the peak point of the relevant probability density function to the vertex of the triangle. Based on the principle that the area of ​​a triangle equals the area of ​​its probability density function, the left endpoint P1 of the trigonometric function is determined: Where ψ(x) is the probability density function of x. It is the left region of ψ(x). It is the left region of the membership function of the triangle, and P represents the point to be optimized, which refers to the coordinate value of any point on the base of the triangle function; II: Right endpoint of the membership function of a triangle, right point P r Represented as: in, It is the right region of the probability density function. It is the right region of the membership functions of the triangle; Thus, the antecedent membership function μ of the fuzzy logic system is obtained. ant ; S23 defines two types of fuzzy logic system rules to determine whether the generated samples are reasonable. The rules are as follows: Type I: If the power correlation of all adjacent wind turbines is similar, then the generated sample is reasonable; Type II: If the power correlation between any adjacent wind turbines is dissimilar, the generated sample is unreasonable; S25 performs fuzzy inference based on the defined membership functions of the antecedent and consequent, and the rules, as shown in the following formula: in, Let R be the sample data of the wind turbines that need to be judged, and R be the rules of the fuzzy logic system. It is the result value of defuzzification, representing the probability that the target sample meets the requirements of the wind turbine data.

4. The method according to claim 2 or 3, characterized in that, S34, based on the various modules of the evolutionary calculation constructed above, generates wind turbine sample data under multiple specific conditions through iterative loops, as shown in the following formula: Among them, G re To construct the evolutionary computational generative model, This is real wind turbine sample data. This refers to wind turbine sample data under specific conditions generated through planned iterations.

5. The method according to claim 1, characterized in that, The initial model for wind power prediction in step S5 involves training the initial model using the wind turbine sample data to obtain a trained wind power prediction model, including: S51. Mix the real wind turbine sample data with the generated wind turbine sample data and use it as training data for the short-term power prediction model: S T =S O ∪S G Among them, S O For the original wind turbine data, S G To generate a large amount of wind turbine data under special conditions, S T Data for the final short-term wind power prediction model; S52, Based on the training data S T Training a short-term wind power prediction model: Among them, f nn This is a short-term wind power prediction model based on neural networks. This is the prediction result for short-term wind power.

6. A wind power prediction device based on an evolutionary framework and a fuzzy adversarial generative network, wherein the fuzzy adversarial generative network wind power prediction device based on an evolutionary framework is used to implement the wind power prediction method based on an evolutionary framework and a fuzzy adversarial generative network as described in any one of claims 1-5, characterized in that, The device includes: The first construction module is used to construct a wind turbine data generation model, which is used to generate multiple wind turbine data under different conditions; The second construction module is used to construct a wind turbine data rationality judgment model, which is used to determine whether the generated multiple wind turbine data meet the requirements of wind turbine data. The third construction module is used to construct a sample generation optimization model based on evolutionary calculation, based on the wind turbine data generation model and the wind turbine data rationality judgment model. The generation module is used to generate multiple wind turbine sample data based on the sample generation optimization model derived from the evolutionary calculation and real wind turbine data. The training module is used to construct an initial model for wind power prediction. The initial model is trained using the wind turbine sample data to obtain a trained wind power prediction model. The prediction module is used to predict wind power using the wind power prediction model.

7. The apparatus according to claim 6, characterized in that, The first building module is used for: S11. Preprocess the acquired real wind turbine data to form a basic data sample from real wind turbine data of different wind turbines at different times. in, For the nth sample after preprocessing, S i Let m be the wind power vector of all wind turbines in the field at time i, m be the number of historical wind turbine data used in the prediction process, and t be the number of wind turbine data to be predicted preset in the prediction process. S12. A deep convolutional neural network is used to construct a wind turbine data generation model. The loss functions for the generator network and the adversarial network are as follows: Where G is the generator network and D is the discriminator network. The data consists of real wind turbine sample data, z is a random vector input into the generator network, E represents the expectation, and p represents the probability distribution; S13. To enable the constructed generative network to generate wind turbine data under specific conditions as required, a generative data rationality judgment model Ω is constructed: d (r) =Ω(G(z)) Where G(z) represents the generated wind turbine sample data, δ (r) To determine the result, if δ (r) A value of 1 indicates that the generated sample is reasonable; if δ (r) A value of 0 indicates that the generated sample is unreasonable; S14. Based on the generated data rationality judgment model Ω, construct a supervised deep convolutional generation network based on the judgment model. The loss function of the supervised deep convolutional generation model based on the judgment model is designed as follows: Where U is the number of samples in each training iteration, u represents the u-th sample, and η is the threshold for rationality judgment; S15. Based on the generated data rationality discrimination model Ω and the loss function of the supervised depthwise convolutional generation model based on the discrimination model, the loss function of the discrimination model is constructed as follows: in, It consists of real wind turbine sample data from each iteration.

8. The apparatus according to claim 6, characterized in that, The second building module is used for: S21. By fitting the correlation probability density function of wind power generation from any two adjacent wind turbines using a large number of real samples, the correlation data between the two wind turbine data is calculated using the Pearson correlation coefficient: Where, d x and d y These are data from two wind turbines, pcc(d) x ,d y ) represents the calculated correlation value, and E represents the expected value; S22. Construct the antecedent of the membership function of the fuzzy logic system. Input the obtained correlation data. First, define the term "similarity" for the antecedent of the membership function and use the commonly used triangular or trapezoidal membership function. I: Fix the peak point of the relevant probability density function to the vertex of the triangle. Based on the principle that the area of ​​a triangle equals the area of ​​its probability density function, the left endpoint P1 of the trigonometric function is determined: Where ψ(x) is the probability density function of x. It is the left region of ψ(x). It is the left region of the membership function of the triangle, and P represents the point to be optimized, which refers to the coordinate value of any point on the base of the triangle function; II: Right endpoint of the membership function of a triangle, right point P r Represented as: in, It is the right region of the probability density function. It is the right region of the membership functions of the triangle; Thus, the antecedent membership function μ of the fuzzy logic system is obtained. ant ; S23. Define two types of fuzzy logic system rules to determine whether generated samples are reasonable: Type I: If the power correlation of all adjacent wind turbines is similar, then the generated sample is reasonable; Type II: If the power correlation between any adjacent wind turbines is dissimilar, the generated sample is unreasonable; S24. Based on the rules in S23, construct the membership functions of the consequents of the triangle, thereby obtaining the membership functions μ of the consequents of the fuzzy logic system. csq ; S25. Perform fuzzy inference based on the defined membership functions of the antecedent and consequent, and the rules: in, Let R be the sample data of the wind turbines that need to be judged, and R be the rules of the fuzzy logic system. It is the result value of defuzzification, representing the probability that the target sample meets the requirements of the wind turbine data; S26, according to The value is used to determine whether the input wind turbine sample data is reasonable.

9. The apparatus according to claim 7 or 8, characterized in that, The third building module is used for: S31. Use the wind turbine data generation model as a population generation module for evolutionary computation; S32. Use the wind turbine data rationality judgment model as the selection condition module for evolutionary calculation; S33. Use the wind turbine data generation model as a crossover and mutation module for evolutionary computation; S34. Based on the evolutionary calculation modules constructed above, wind turbine sample data under multiple specific conditions are generated through iterative loops: Among them, G re To construct the evolutionary computational generative model, This is real wind turbine sample data. This refers to wind turbine sample data under specific conditions generated through planned iterations.

10. The apparatus according to claim 6, characterized in that, The training module is used for: S51. Mix the real wind turbine sample data with the generated wind turbine sample data and use it as training data for the short-term power prediction model: S T =S O ∪S G Among them, S O For the original wind turbine data, S G To generate a large amount of wind turbine data under special conditions, S T Data for the final short-term wind power prediction model; S52, Based on the training data S T Training a short-term wind power prediction model: Among them, f nn This is a short-term wind power prediction model based on neural networks. This is the prediction result for short-term wind power.

Images