Wind power prediction method and system based on improved dual-objective grey wolf optimization algorithm

Abstract

The invention discloses a wind power prediction method and system based on an improved dual-objective grey wolf optimization algorithm, and relates to the technical field of wind power prediction, and the method comprises the steps: calculating a division threshold value of different wind speed fluctuation processes of a wind power plant in each season; constructing a fluctuation process division model based on the division threshold; dividing the historical wind speed data based on a fluctuation process division model to obtain wind speed data after wind speed fluctuation division; acquiring wind power historical power data corresponding to the wind speed data after the wind speed fluctuation division; a wind power prediction model based on CNN-BiLSTM is constructed; training the wind power prediction model through an improved dual-target grey wolf optimization algorithm; and performing wind power prediction through the trained wind power prediction model. The wind power prediction model provided by the invention has the advantages of higher prediction precision, stronger anti-interference capability, high efficiency and the like, and short-term wind power prediction with high prediction precision and high reliability can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to a wind power prediction method and system based on an improved dual-objective grey wolf optimization algorithm. Background Art

[0002] As a clean and renewable energy source, wind power generation is becoming increasingly important in the energy structure, and its large-scale access and effective utilization have become urgent issues. However, the randomness, unpredictability and seasonal distribution of wind power generation have brought great difficulties to the stable operation of the power grid. High-precision wind power forecasting is needed to assist the operation and optimization of the system. There are still many problems and deficiencies in current research. Summary of the invention

[0003] The purpose of the present invention is to provide a wind power prediction method and system based on an improved dual-objective grey wolf optimization algorithm, so as to achieve short-term wind power prediction with high prediction accuracy and high reliability, and provide technical support and data support for the stable operation of the power grid.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A wind power prediction method based on an improved dual-objective grey wolf optimization algorithm, comprising:

[0006] Calculate the division thresholds for different wind speed fluctuation processes in each season;

[0007] Constructing a fluctuation process partition model based on the partition threshold;

[0008] Dividing the historical wind speed data based on the fluctuation process division model to obtain wind speed data after wind speed fluctuation division;

[0009] Obtain historical wind power data corresponding to the wind speed data after wind speed fluctuation division;

[0010] Constructing a wind power prediction model based on CNN-BiLSTM; the input of the wind power prediction model includes wind speed data after wind speed fluctuation division and wind power historical power data, and the output is wind power prediction power;

[0011] The wind power prediction model is trained by improving the dual-objective grey wolf optimization algorithm;

[0012] Wind power prediction is performed using the trained wind power prediction model.

[0013] To achieve the above object, the present invention also provides the following solution:

[0014] A wind power prediction system based on an improved dual-objective grey wolf optimization algorithm, comprising:

[0015] A division threshold calculation module is used to calculate the division thresholds of different wind speed fluctuation processes in each season of the wind farm;

[0016] A fluctuation process partition model construction module, used to construct a fluctuation process partition model based on the partition threshold;

[0017] A wind speed data division module, used to divide the historical wind speed data based on the fluctuation process division model to obtain wind speed data after wind speed fluctuation division;

[0018] A wind power historical power data acquisition module is used to acquire wind power historical power data corresponding to wind speed data after wind speed fluctuation division;

[0019] A wind power prediction model building module is used to build a wind power prediction model based on CNN-BiLSTM; the input of the wind power prediction model includes wind speed data after wind speed fluctuation division and wind power historical power data, and the output is wind power prediction power;

[0020] A model training module, used for training the wind power prediction model by improving the dual-objective grey wolf optimization algorithm;

[0021] The wind power prediction module is used to predict wind power using a trained wind power prediction model.

[0022] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0023] The present invention considers the factors affecting wind power under different seasonal division thresholds, and constructs a wind power prediction model based on CNN-BiLSTM with strong feature extraction capability, reduced overfitting, and captured sequence dependencies, and uses the multi-objective gray wolf optimization algorithm improved by elite reverse learning to adjust the hyperparameters of the wind power prediction model. The wind power prediction model provided by the present invention has the advantages of higher prediction accuracy, stronger anti-interference ability, high efficiency, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0025] Figure 1 A flow chart of a wind power prediction method based on an improved dual-objective grey wolf optimization algorithm provided by the present invention;

[0026] Figure 2 is another flow chart of a wind power prediction method based on an improved dual-objective grey wolf optimization algorithm;

[0027] Figure 3 Schematic diagram of the forecast results output by the wind power forecast model for the last 10 days of spring before the wind speed threshold is divided;

[0028] Figure 4 Schematic diagram of the forecast results output by the wind power forecast model for the last 10 days of summer before the wind speed threshold is divided;

[0029] Figure 5 Schematic diagram of the forecast results output by the wind power forecast model for the last 10 days of autumn before the wind speed threshold is divided;

[0030] Figure 6 Schematic diagram of the forecast results output by the wind power forecast model for the last 10 days of winter before the wind speed threshold is divided;

[0031] Figure 7 This is a schematic diagram of the forecast results output by the wind power forecast model for the last 10 days of spring after the wind speed threshold is divided;

[0032] Figure 8 This is a schematic diagram of the forecast results output by the wind power forecast model for the last 10 days of summer after the wind speed threshold is divided;

[0033] Fig. 9 This is a schematic diagram of the forecast results output by the wind power forecast model for the last 10 days of autumn after the wind speed threshold is divided;

[0034] Fig.10 This is a schematic diagram of the prediction results output by the wind power prediction model for the last 10 days of winter after the wind speed threshold is divided;

[0035] Fig.11 This is a schematic diagram of the forecast results for the last 10 days of a year without dividing into seasons before the wind speed threshold is divided;

[0036] Fig.12 This is a schematic diagram of the forecast results for the last 10 days of the year without dividing the seasons after the wind speed threshold is divided. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] The purpose of the present invention is to provide a wind power prediction method and system based on an improved dual-objective grey wolf optimization algorithm, which is used to predict short-term wind power and provide technical and data support for the stable operation of the power grid.

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Embodiment 1

[0041] like Figure 1-Figure 2 As shown, the wind power prediction method based on the improved dual-objective grey wolf optimization algorithm provided in this embodiment includes the following steps:

[0042] S1: Calculate the division thresholds for different wind speed fluctuation processes in each wind farm in each season.

[0043] Obtain historical data of the wind farm, including wind speed data in spring, summer, autumn and winter with a time scale of 15 minutes. Then extract the peak wind speed of all fluctuations in the wind farm from the wind speed data of each season and record it as but:

[0044]

[0045] In the formula, are the peak wind speeds in a wind farm in a certain season; α∈{1,2,3,4} represent spring, summer, autumn and winter respectively; m is the total number of times the peak wind speed occurs in a wind farm in a certain season.

[0046] The kernel density estimation method and cumulative distribution function are applied to compensate for the peak characteristics of different wind speed fluctuation processes, and the zero output threshold, medium fluctuation threshold and large fluctuation threshold of wind speed fluctuation processes in different seasons are selected.

[0047] The selection of threshold is divided into three steps:

[0048] Step 1: Statistical distribution of peak values. Extract the peak values ​​of all fluctuation processes of each wind farm. Statistical peak distribution.

[0049] Step 2: Calculate the probability density function through the kernel density estimation method, and then use the cumulative distribution function to calculate the wind speed corresponding to the values ​​of 0.25, 0.5 and 0.75 to determine the zero output threshold of the wind speed fluctuation process in different seasons Medium fluctuation threshold and large fluctuation threshold

[0050] Probability density function of kernel density estimation method The calculation formula is:

[0051]

[0052] In the formula, is the βth sample point; is the point at which its probability is evaluated; h is the bandwidth.

[0053] Cumulative distribution function The calculation formula is:

[0054]

[0055] Where χ(t) is the probability density function.

[0056] Step 3: If the zero output threshold calculated by the kernel density estimation method and the cumulative distribution function in the second step is too large, the threshold is corrected and set as the wind turbine cut-in wind speed.

[0057] S2: Constructing a fluctuation process partition model based on the partition threshold.

[0058] The division of a wind speed fluctuation process is that its local minimum value starts from a threshold value lower than the zero output threshold, the medium fluctuation threshold value and the large fluctuation threshold value, passes through one or more wind speed peaks, and ends when the final wind speed local minimum value is lower than the wind speed threshold value. The three wind speed division thresholds determined in step S1 construct the fluctuation process division model:

[0059]

[0060] In the formula, Z a,0 , Z a,1 , Z a,2 , Z a,3 are the wind speed data column vectors for the wind speed zero output process, small fluctuation process, medium fluctuation process and large fluctuation process respectively; v max It is the maximum value of the wind speed sequence in each wind speed process.

[0061] S3: dividing the historical wind speed data based on the fluctuation process division model to obtain wind speed data after wind speed fluctuation division.

[0062] The wind speed model data set at time t constructed by the wind speed data obtained in step S1 is recorded as V′ t , is the column vector of wind speed data, then V′ t It is expressed as:

[0063] V′ t =[v′ t-8640 ,v′ t-8639 ,...,v′ t-1 ] T

[0064] In the formula, v′t-8640 ,v′ t-8639 ,…,v′ t-1 It is the historical wind speed data of 15 minutes in a quarter before time t.

[0065] V′ t Before dividing the wind speed sequence into fluctuation processes, it is necessary to remove abnormal wind speed values. First, the abnormal wind speed data identification formula is:

[0066]

[0067] In the formula, t i ,t i-1 Wind speed data at the moment.

[0068] Then, its adjacent normal value is used for replacement update, and the update formula is:

[0069]

[0070] Finally, in the formula, t i The wind speed data is updated at all times. is abnormal wind speed data and The updated Choose normal data that is farther away; t i-1 ,t i ,t i+1 Wind speed data before the moment update; V ab is the abnormal wind speed data set.

[0071] The updated wind speed model dataset is V t =[v t-8640 ,v t-8639 ,...,v t-1 ] T , divide the data set V by constructing the wave process partition model t The wind speed data set of large fluctuation process, medium fluctuation process, small fluctuation process and zero output process after a certain season is obtained, which is recorded as V t z , is the column vector of wind speed data, then:

[0072]

[0073] In the formula, is the column vector of different wind speed fluctuation processes; φ is the total number of wind speed fluctuation processes; n∈{0,1,2,3} represents the wind speed zero output process, small fluctuation process, medium fluctuation process and large fluctuation process respectively.

[0074] S4: Obtain historical wind power data corresponding to the wind speed data after wind speed fluctuation division.

[0075] By V t z The wind power corresponding to each wind speed fluctuation process is recorded as but:

[0076]

[0077] In the formula, is the wind power column vector of different wind speed fluctuation processes.

[0078] S5: Construct a wind power prediction model based on CNN-BiLSTM; the input of the wind power prediction model includes wind speed data after wind speed fluctuation division and wind power historical power data, and the output is wind power prediction power.

[0079] The wind power corresponding to the wind speed data set after a certain season division is obtained, wherein a part of the divided wind speed data and the historical wind power data corresponding to each wind speed fluctuation process are selected to be divided into the model input data of the training set; the remaining part of the historical wind power data corresponding to each wind speed fluctuation process is divided into the model output comparison data of the training set.

[0080] When building a wind power prediction model based on CNN-BiLSTM, the wind power prediction data is the time point data with a time scale of 15 minutes, where the prediction model output at time t is recorded as Y t ,but:

[0081] Y t =[P t ] T

[0082] Where P t is the predicted wind power at time t.

[0083] The prediction model input at time t is recorded as X t , is the wind speed data and wind power historical data after the wind speed fluctuations of each 15-minute period in a certain season before time t, then the matrix X t Expressed as

[0084] The working process of the wind power prediction model based on CNN-BiLSTM is as follows:

[0085] First, the wind speed data and wind power historical data after wind speed fluctuation division are used as the feature input of the CNN model, then:

[0086]

[0087]

[0088] In the formula, is the output of the convolutional layer; * is the inner product; W γ is the weight matrix of the convolution kernel; η is the bias term of the convolution layer; f is the activation function; is the output of the pooling layer; max represents the pooling operation.

[0089] Then, the pooled result Processed as the input of BiLSTM, in the forward LSTM:

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096] In the backward LSTM:

[0097]

[0098]

[0099]

[0100]

[0101]

[0102]

[0103] Where σ is the activation function; W κ , W c are the forget gate weight matrix and the unit-to-output gate weight matrix of the BiLSTM model respectively; W j , W o is the hidden input gate weight matrix; κ , j , o , c They are respectively the bias vector of the current hidden layer forget gate, the current input gate bias vector, the bias vector of the current hidden layer input gate, and the bias vector of the current hidden layer memory unit. Represent the forget gate output vector, input gate output vector, output gate output vector, candidate memory cell state, and memory cell state of the forward and backward LSTM layers respectively; Represent the hidden state vectors of the forward and backward LSTM respectively.

[0104] Finally, the outputs of the forward and backward LSTM layers are linearly superimposed according to certain weights, and the output of the CNN-BiLSTM model is obtained through the fully connected layer and recorded as Y t ,but:

[0105]

[0106] Where W fc is the weight matrix of the fully connected layer; s a 、s b is a constant where s a +s b =1;λ fc is the bias vector of the fully connected layer.

[0107] The input-output mapping relationship of the prediction model is:

[0108] Y t =f(X t )

[0109] Among them, f(·) is the input-output mapping function of the prediction model.

[0110] S6: Training the wind power prediction model by improving the dual-objective grey wolf optimization algorithm.

[0111] A multi-objective improved gray wolf optimization algorithm is applied to train wind power prediction models for four different seasons. The individual positions of gray wolves in the multi-objective improved gray wolf optimization algorithm reflect the various hyperparameters in the short-term wind power prediction model based on the fluctuation process division threshold in different seasons. The process of changing the improved weight values ​​of the short-term wind power prediction model based on the fluctuation process division threshold is the process of changing the values ​​of the individual positions of gray wolves. The Pareto non-inferior solution selection strategy is used to find all dimensions of the optimal position of the individual gray wolf in the optimal compromise solution of the gray wolf population to complete the training of the prediction model.

[0112] The initialization stage of the multi-objective gray wolf optimization algorithm was improved through the elite reverse learning method. In the gray wolf optimization initialization stage, two gray wolf populations, namely the gray wolf initialization population and the dynamic gray wolf reverse population, were compared and the optimal a solutions were selected as the initial solutions. This improved the model prediction accuracy, expanded the search area of ​​the gray wolf population and improved the search efficiency of the gray wolf optimization algorithm.

[0113] Gray wolf population initialization is divided into three steps:

[0114] Step 1: Randomly initialize a gray wolf population with i gray wolf individuals, each with d-dimensional position information.

[0115] Step 2: Improve the Grey Wolf Optimization Algorithm through elite reverse learning and introduce dynamic reverse points:

[0116] Among them, the current position of the gray wolf individual is recorded as The dynamic reverse point introduced by the elite reverse learning method is:

[0117]

[0118]

[0119] In the formula, ν1, ν2 are random numbers between (0, 0.5), which conform to the uniform distribution; are the minimum and maximum values ​​of the j-th dimension position information of the μ-th gray wolf individual; is the j-dimensional position information of the μth gray wolf individual, μ∈[1,a], j∈[1,d].

[0120] The reverse gray wolf population with a gray wolf individuals and d-dimensional position information is obtained through the dynamic reverse point of the elite reverse learning method.

[0121] Step 3: Compare the fitness values ​​of the initial gray wolf population and the position information of each gray wolf in its dynamic reverse gray wolf population, and select a gray wolf individuals with better fitness from high to low to form the final initialization population. The gray wolf population W improved by the elite reverse learning method is expressed as:

[0122]

[0123] In the formula, d represents the maximum position dimension of each individual gray wolf, and refers to the hyperparameters of CNN-BiLSTM to be optimized, namely, the number of iterations of the CNN-BiLSTM algorithm, the learning rate, the number of convolutional layer filter outputs, and the number of hidden nodes in the bidirectional LSTM layer.

[0124] The fitness calculation formula is:

[0125]

[0126] In the formula, is the fitness of the μth gray wolf; The number of power samples for wind power prediction; is the actual value of power at time t.

[0127] The process of using the improved Grey Wolf optimization algorithm to train the wind power prediction model is as follows:

[0128] After determining the maximum number of iterations and the stopping condition, the wolf group optimized by the elite reverse learning method is used for iterative optimization. If the model prediction accuracy standard is met, the training is completed. On the contrary, if the accuracy does not meet the preset standard, the process will continue to repeat until the condition is met or the number of iterations of the gray wolf optimization algorithm exceeds the maximum number specified, at which time the training will also end.

[0129] Step 1: Initialize the gray wolf population W and set the maximum number of iterations iter max .

[0130] Step 2: If it is less than the maximum number of iterations, determine the three alpha wolves W by calculating the fitness values ​​of the top three elite reverse learning methods fi , W se , W th , and then update the position of the candidate gray wolf.

[0131] Step 3: Use the following formula to let the gray wolves hunt prey. When |B|>1, the gray wolves try to disperse in various areas and search for prey to conduct a global search. When |B|<1, the gray wolves will concentrate on hunting prey, which means that the wolf pack will shrink as much as possible and fall into a local optimal solution.

[0132]

[0133]

[0134] In the formula, τ represents the number of iterations; A and B are coefficients; represents the position information of the prey in the jth dimension at the τth iteration; It represents the position information of the μth gray wolf in the jth dimension at the τth iteration. The calculation formulas of A and B are as follows:

[0135] A=2·ψ2

[0136] B=2ζ·ψ1-ζ

[0137] Where ψ1 and ψ2 are random values ​​in (0,1); ζ is a convergence factor that decreases linearly from 2 to 0 with the number of iterations.

[0138] Step 4: Use the following formula to perform gray wolf hunting. The gray wolf has the ability to identify the location of potential prey (optimal solution), and the search process mainly relies on W fi , W se , W th Gray wolves conduct supervised hunting.

[0139]

[0140]

[0141]

[0142]

[0143]

[0144]

[0145]

[0146] In the formula, Represents W fi , W se , W th The j-th dimension position information of the three wolves at the τ-th iteration; Represents W fi , W se , W th The updated position of the candidate wolf after supervision; A1, A2, A3 are random numbers affected by ψ2; B1, B2, B3 are random numbers affected by ψ1 and ζ; is the final position of the candidate wolf after the position is updated.

[0147] Step 5: Use the Pareto non-inferior solution selection strategy and update the current gray wolf position. In order to meet the accuracy and robustness of the prediction model, it is necessary to use the fuzzy decision method to find the optimal solution of the dual-objective Pareto non-inferior solution set that meets the root mean square error RMSE and the training error standard deviation evaluation index SD. The accuracy and robustness of the prediction can only be explained if they are met at the same time. The gray wolf dominates A gray wolf, namely:

[0148]

[0149] The dominant gray wolf individuals after each iteration are saved to an external archive to form the Pareto optimal frontier.

[0150] By using the Pareto non-inferior solution selection strategy, the optimal compromise solution of the gray wolf on the Pareto frontier surface can be calculated and saved to the external archive. The gray wolf's satisfactory compromise coefficient is recorded as f fit,1 、f fit,2 The calculation formula is:

[0151]

[0152]

[0153] In the formula, RMSE μ , SD μThe dual objectives of the optimization algorithm for the μth gray wolf are the root mean square error and the standard deviation of the training error, which are used to evaluate the prediction accuracy and robustness of the CNN-BiLSTM prediction model when the μth gray wolf hyperparameters are used. The formula is:

[0154]

[0155]

[0156] In the formula, is the total number of samples in the prediction data set; is the mean training error.

[0157] The optimal compromise solution of the Pareto frontier is selected by the maximum value of the following formula: Grey Wolf compromise satisfaction calculation formula for:

[0158]

[0159] Step 6: Check whether the preset accuracy condition is met. If so, exit and output the result. Otherwise, repeat steps 2 to 5 until the preset accuracy is met or the maximum number of iterations is reached.

[0160] Set model preset accuracy The judgment is as follows:

[0161] when Once the model’s preset accuracy requirements are met and model training is completed, the model can be used for prediction.

[0162] when If the model has not reached the preset accuracy requirement, the iterative process of the improved sparrow search algorithm will be repeated or the number of iterations of the improved sparrow search algorithm exceeds the set maximum number of iterations, the model training will also be stopped.

[0163] In this embodiment, the model convergence accuracy is set For 10 -6 .

[0164] S7: Perform wind power prediction using the trained wind power prediction model.

[0165] The wind speed data and wind power data after wind speed fluctuations are divided in the last 10 days of the four seasons of a certain year to be tested are input into the trained prediction model to obtain the wind power prediction values ​​for each moment in the last 10 days of the four seasons.

[0166] In order to better distinguish the wind power prediction effects of the four seasons before and after the wind speed threshold division, the wind power prediction model outputs the prediction results of the last 10 days of spring, summer, autumn and winter before the wind speed threshold division as follows: Figure 3-Figure 6As shown in the figure, the wind power prediction model outputs the prediction results of the last 10 days of spring, summer, autumn and winter after the wind speed threshold is divided as follows Figure 7-10 As shown in the figure, the forecast results for the last 10 days of the year without dividing into seasons before the wind speed threshold is divided are as follows Fig.11 As shown in the figure, the forecast results for the last 10 days of the year without dividing the seasons after the wind speed threshold is divided are as follows Fig.12 As shown, Figure 3-Figure 11 This is a curve diagram of wind power prediction and actual value with time scale of .

[0167] The wind power prediction results and actual wind power values ​​every 15 minutes starting from 0 o'clock before the wind speed threshold is divided are shown in Table 1:

[0168] Table 1 Wind power prediction results before some wind speed thresholds are divided

[0169]

[0170]

[0171] The wind power prediction results and actual wind power values ​​every 15 minutes starting from 0 o'clock after the wind speed threshold is divided are shown in Table 2:

[0172] Table 2 Wind power prediction results after some wind speed threshold divisions

[0173]

[0174] The wind power prediction results and actual wind power values ​​every 15 minutes starting from 0 o'clock after the wind speed threshold is divided are shown in Table 3:

[0175] Table 3 Wind power prediction results for the last 10 days of a year without seasonal division after some wind speed thresholds are divided

[0176]

[0177] The root mean square error (RMSE), mean absolute error (MAE), and standard deviation (SD) of the wind power prediction results at each moment starting from 0 o'clock before and after the division threshold are calculated respectively. The results are shown in Table 4. The calculation formula of the mean absolute error (MAE) is:

[0178]

[0179] In the formula, The sample parameter is 960; P t is the predicted value at time t; P t is the true value at time t.

[0180] Table 4 Comparison of prediction evaluation indicators

[0181]

[0182]

[0183] As can be seen from Table 4, the prediction effect of considering seasons after dividing the threshold is better than that before dividing the threshold, the prediction effect of not considering different seasons after dividing the threshold is better than that before dividing the threshold, and the prediction effect of considering different seasons after dividing the threshold is better than that before dividing the threshold. It can be clearly seen that the accuracy and robustness of the prediction system considering different seasons after dividing the threshold are improved.

[0184] The wind power prediction method proposed in the present invention has the advantages of high prediction accuracy, high prediction reliability, high efficiency, etc., and meets the needs of practical applications.

[0185] Embodiment 2

[0186] In order to execute the method corresponding to the above-mentioned embodiment 1 to achieve the corresponding functions and technical effects, a wind power prediction system based on an improved dual-objective grey wolf optimization algorithm is provided below.

[0187] The system includes:

[0188] The division threshold calculation module is used to calculate the division thresholds of different wind speed fluctuation processes in the wind farm in each season.

[0189] The fluctuation process partition model construction module is used to construct the fluctuation process partition model based on the partition threshold.

[0190] The wind speed data division module is used to divide the historical wind speed data based on the fluctuation process division model to obtain the wind speed data after wind speed fluctuation division.

[0191] The wind power historical power data acquisition module is used to acquire the wind power historical power data corresponding to the wind speed data after the wind speed fluctuation is divided.

[0192] The wind power prediction model building module is used to build a wind power prediction model based on CNN-BiLSTM; the input of the wind power prediction model includes wind speed data after wind speed fluctuation division and wind power historical power data, and the output is wind power prediction power.

[0193] The model training module is used to train the wind power prediction model by improving the dual-objective grey wolf optimization algorithm.

[0194] The wind power prediction module is used to predict wind power using a trained wind power prediction model.

[0195] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0196] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A wind power prediction method based on an improved dual-objective grey wolf optimization algorithm, characterized in that: include: Calculate the division thresholds for different wind speed fluctuation processes in each season; Constructing a fluctuation process partition model based on the partition threshold; Dividing the historical wind speed data based on the fluctuation process division model to obtain wind speed data after wind speed fluctuation division; Obtain historical wind power data corresponding to the wind speed data after wind speed fluctuation division; Constructing a wind power prediction model based on CNN-BiLSTM; the input of the wind power prediction model includes wind speed data after wind speed fluctuation division and wind power historical power data, and the output is wind power prediction power; The wind power prediction model is trained by improving the dual-objective grey wolf optimization algorithm; Wind power prediction is performed using the trained wind power prediction model.

2. The wind power prediction method based on the improved dual-objective grey wolf optimization algorithm according to claim 1 is characterized in that: Calculate the division thresholds of different wind speed fluctuation processes in each season, including: Obtain the peak wind speed data of each wind speed fluctuation process of the wind farm in each season; Based on the peak wind speed data, the kernel density estimation method and the cumulative distribution function are used to calculate the division thresholds of different wind speed fluctuation processes in each season of the wind farm; the division thresholds include a zero output threshold, a medium fluctuation threshold and a large fluctuation threshold.

3. The wind power prediction method based on the improved dual-objective grey wolf optimization algorithm according to claim 2 is characterized in that: If the zero output threshold calculated by the kernel density estimation method and the cumulative distribution function is greater than the set range, the threshold is corrected.

4. The wind power prediction method based on the improved dual-objective grey wolf optimization algorithm according to claim 1 is characterized in that: The expression of the wind speed fluctuation process division model is: In the formula, Z a,0 , Z a,1 , Z a,2 , Z a,3 are the wind speed data column vectors for the wind speed zero output process, small fluctuation process, medium fluctuation process and large fluctuation process respectively; v max is the maximum value of the wind speed sequence during each wind speed fluctuation process; Zero output threshold, is the medium fluctuation threshold, is the large fluctuation threshold.

5. The wind power prediction method based on the improved dual-objective grey wolf optimization algorithm according to claim 1 is characterized in that: The wind power prediction model is trained by improving the dual-objective grey wolf optimization algorithm, specifically including: Initialize the gray wolf population and set the maximum number of iterations, and use the gray wolf population optimized by the elite reverse learning method to perform iterative optimization and search. If the prediction accuracy standard of the wind power prediction model is met, the training is completed, or the training is stopped when the maximum number of iterations is reached.

6. The wind power prediction method based on the improved dual-objective grey wolf optimization algorithm according to claim 5 is characterized in that: The initialization of the gray wolf population specifically includes: A gray wolf population is generated by randomly initializing i gray wolf individuals, each of which has d-dimensional position information; the positions of the gray wolf individuals represent various hyperparameters in the wind power prediction model; The gray wolf optimization algorithm is improved by the elite reverse learning method, and the reverse gray wolf population with a gray wolf individuals and d-dimensional position information is obtained through the dynamic reverse point of the elite reverse learning method. Compare the fitness values ​​of the position information of each gray wolf individual in the initial gray wolf population and the reverse gray wolf population, and select a gray wolf individuals with high fitness from high to low to form the final initialized gray wolf population.

7. The wind power prediction method based on the improved dual-objective grey wolf optimization algorithm according to claim 6 is characterized in that: The calculation formula of the fitness is as follows: In the formula, is the fitness of the μth gray wolf, W μ is the location information of the μth gray wolf individual, The number of power samples for wind power prediction; is the real value of wind power at time t, P t is the predicted wind power at time t.

8. A wind power prediction system based on an improved dual-objective grey wolf optimization algorithm, characterized in that: include: A division threshold calculation module is used to calculate the division thresholds of different wind speed fluctuation processes in each season of the wind farm; A fluctuation process partition model construction module, used to construct a fluctuation process partition model based on the partition threshold; A wind speed data division module, used to divide the historical wind speed data based on the fluctuation process division model to obtain wind speed data after wind speed fluctuation division; A wind power historical power data acquisition module is used to acquire wind power historical power data corresponding to wind speed data after wind speed fluctuation division; A wind power prediction model building module is used to build a wind power prediction model based on CNN-BiLSTM; the input of the wind power prediction model includes wind speed data after wind speed fluctuation division and wind power historical power data, and the output is wind power prediction power; A model training module, used for training the wind power prediction model by improving the dual-objective grey wolf optimization algorithm; The wind power prediction module is used to predict wind power using a trained wind power prediction model.

Images