Wind power prediction method based on ICPO optimization VMD coupling deep learning

Abstract

The invention discloses a wind power prediction method based on ICPO optimization VMD coupling deep learning. The wind power prediction method comprises the following steps: 1) collecting a multi-dimensional data set including wind speed, wind direction, air pressure and wind power output power; 2) constructing a VMD decomposition model, and optimizing parameters of the VMD decomposition model by using a preset ICPO algorithm; 3) decomposing the wind power data by using the optimized VMD decomposition model to obtain K sub-components, and forming K pieces of multi-dimensional time sequence data with other original data in sequence; (4) the K pieces of multidimensional time series data are normalized respectively, a BiTCN-BiLSTM-Attention deep learning model is established, and hyper-parameters of the BiTCN-BiLSTM-Attention deep learning model are optimized through a preset ICPO algorithm; and 5) predicting the sub-components by using the optimized IVMD-ICPO-BiTCN-BiLSTM-Attention deep learning model, and adding all prediction results to obtain a wind power predicted value.According to the wind power prediction method disclosed by the invention, the fluctuation characteristics of the original wind power sequence are decomposed and stabilized by using the VMD and the hyper-parameters of the ICPO optimized deep learning model are utilized, so that the wind power prediction method has higher prediction precision and better generalization performance.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power generation output power prediction, and particularly relates to a wind power prediction method based on ICPO-optimized VMD coupled with deep learning. Background Art

[0002] At present, renewable energy power generation led by wind power has become one of the main ways for China to achieve the "dual carbon" goal. However, its inherent characteristics such as volatility, randomness, and intermittency lead to unstable power generation, which not only has a certain impact on the operation efficiency of the unit, but also brings difficulties to the large-scale grid connection of wind power. And high-precision output prediction plays a crucial role in setting the daily wind power generation plan, improving the operation safety and economy of the power system, and reducing wind curtailment.

[0003] Traditional wind power prediction methods are mainly based on statistical models or physical models. Among them, the premise of statistical methods is to assume that the time series satisfies a certain linear relationship, while ignoring the essence of its non-linear relationship, which does not conform to the actual characteristics of wind power generation; the physical method model is mainly based on numerical weather forecasting, geographical information, and unit parameters, etc., and obtains the wind power prediction value through a large amount of calculations, has a strong dependence on numerical weather forecasting data, and needs to establish and solve a large number of partial differential equations, with a large calculation difficulty, and is often difficult to popularize. In recent years, researchers often use deep learning theory combined with signal decomposition methods to build a hybrid prediction model to improve the prediction accuracy and stability and achieve high-precision prediction of wind power output. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a wind power prediction method based on ICPO-optimized VMD coupled with deep learning. By VMD decomposition, the complexity and volatility of wind power output are effectively reduced, the spatial features of time series are extracted by using deep learning theory and the long-term dependence relationship before and after is captured, and the ICPO algorithm is used to optimize the model parameters to enhance the prediction performance of the overall model and improve the prediction accuracy.

[0005] To achieve the above object, the present invention adopts the following technical solutions.

[0006] A wind power prediction method based on ICPO-optimized VMD coupled with deep learning, comprising the following steps:

[0007] Step S1: Collect a multi-dimensional data set including wind speed, wind direction, air pressure, and wind power output;

[0008] Step S2: Construct a VMD decomposition model for the wind power output in the multi-dimensional data set, and optimize the parameters of the VMD decomposition model by using the preset ICPO algorithm;

[0009] Step S3: Decompose the wind power data using the optimized VMD decomposition model to obtain K sub-components, and sequentially form K multi-dimensional time series data with the remaining original data in the multi-dimensional data set;

[0010] Before obtaining the K sub-components, it is necessary to remove or interpolate the abnormal data of the wind farm in the wind power data;

[0011] Step S4: Normalize each of the K multi-dimensional time series data and establish a BiTCN-BiLSTM-Attention deep learning model, and optimize the hyperparameters of the BiTCN-BiLSTM-Attention deep learning model using the preset ICPO algorithm;

[0012] Step S5: Use the optimized IVMD-ICPO-BiTCN-BiLSTM-Attention deep learning model to predict the sub-components, and add all the prediction results to obtain the wind power prediction value.

[0013] Specifically, for the preset ICPO algorithm in Step S2 and Step S4, the ICPO algorithm is improved from the CPO algorithm, and the improvement content is as follows:

[0014] There are four defense measures in the CPO algorithm, namely visual defense strategy, auditory defense strategy, odor defense strategy, and physical attack defense strategy. The ICPO algorithm improves the population initialization, visual defense strategy, and physical attack defense strategy among them:

[0015] Initialize the population based on the Circle chaotic map, which can effectively avoid the phenomenon of uneven individual distribution in the random initialization of the standard CPO algorithm. Its mathematical expression is:

[0016]

[0017] In the above formula, mod is the remainder calculation operator; c 1 、c 2 are coefficients taking 0.5 and 0.2 respectively;

[0018] Improve the visual defense strategy of the CPO algorithm based on the Gaussian random walk model to generate new perturbations, which can accelerate the convergence speed of the algorithm and escape at the local optimum. Its mathematical expression is:

[0019]

[0020] In the above formula, is the position of the i-th individual at the n-th iteration; t max is the maximum number of iterations; is the optimal individual in the population; Z is the current iteration number;

[0021] Improve the physical attack defense strategy of the CPO algorithm based on the sine-cosine algorithm. Utilize the periodicity and volatility of the sine-cosine function, and accelerate the convergence speed of the algorithm by continuously adjusting the function values. Its mathematical expression is:

[0022]

[0023] In the above formula, β 1 is the weight factor, β 2 is the search factor, R 7 and R 8 are random numbers between [0, 2π] and [0, 2] respectively, is the force parameter affected by inelastic collision, is the parameter for controlling the search direction, D n is the defense coefficient, is the average force of the crested porcupine on the predator.

[0024] Specifically, in step S2, the preset ICPO algorithm is used to optimize the parameters of the VMD decomposition model. The optimization process is as follows:

[0025] Step S21: Import the original wind power time series into the VMD decomposition model, and decompose the original wind power using VMD decomposition;

[0026] Step S22: Use the ICPO algorithm with the minimum envelope entropy min{E e} as the objective function to optimize the number of sub-components K and the penalty factor α of the VMD wind power data decomposition model. The mathematical expression of the optimization process is:

[0027]

[0028] In the above formula, e i is the normalized form of a i , a i is the envelope signal of the sub-component after Hilbert transform;

[0029] Step S23: After multiple optimization iterations, until the objective function value no longer changes significantly or reaches the preset maximum number of iterations, output the corresponding optimal number of components K and penalty factor α, and re-enter them into the VMD decomposition model to obtain the optimized VMD decomposition model.

[0030] Specifically, in step S4, the hyperparameters of the BiTCN-BiLSTM-Attention deep learning model are optimized using the preset ICPO algorithm, and the optimization process is as follows:

[0031] Step S41: Normalize each of the K multi-dimensional time series data, and its mathematical expression is:

[0032]

[0033] In the above formula, o is the original value, o′ is the normalized value, o max is the maximum value of the corresponding feature, o min is the minimum value of the corresponding feature;

[0034] Step S42: Divide the K multi-dimensional time series data after normalization into a training set and a validation set according to a ratio of 8:2, and import them into the BiTCN-BiLSTM-Attention deep learning model to predict the original wind power;

[0035] Step S43: Use the ICPO algorithm to take the minimum root mean square error min{Y RMSE} of the prediction results of the BiTCN-BiLSTM-Attention deep learning model as the objective function, and optimize the learning rate, regularization parameter, convolution kernel size, number of convolution kernels, number of BiLSTM hidden units, and number of Attention hidden units of the BiTCN-BiLSTM-Attention deep learning model. Its mathematical expression is:

[0036]

[0037] In the above formula, y i is the actual value of wind power, Y i is the predicted value of wind power, and N is the number of data sets;

[0038] Step S44: After multiple optimization iterations, until the objective function value no longer changes significantly or reaches the preset maximum number of iterations, output the corresponding optimal learning rate, regularization parameter, convolution kernel size, number of convolution kernels, number of BiLSTM hidden units, and number of Attention hidden units, and re-enter them into the BiTCN-BiLSTM-Attention deep learning model to obtain the optimized IVMD-ICPO-BiTCN-BiLSTM-Attention deep learning model.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] (1) The present invention takes into account the inherent volatility and randomness of wind power output. Using VMD to decompose the original wind power can more effectively reduce its complexity and volatility, better extract time series features, thereby improving the prediction accuracy of the model. Moreover, by optimizing the number of components K and the penalty factor α of the VMD decomposition through the improved ICPO algorithm, the original wind power sequence can be decomposed more effectively, and the prediction accuracy of the hybrid model can be improved;

[0041] (2) The BiTCN-BiLSTM-Attention hybrid model established in the present invention uses the BiTCN residual block to initially extract features, introduces BiLSTM to further mine the long-term dependence relationship and correlation information in the sequence, and applies the attention mechanism to weight the output of BiLSTM to highlight key features. Through the fully connected layer, multiple important features are integrated and mapped to a one-dimensional space, enabling it to more effectively capture key information in long-term time series prediction and realizing the comprehensive capture and fusion of historical and future features, thereby improving the overall prediction performance of the model;

[0042] (3) The present invention optimizes the learning rate, regularization parameter, convolutional kernel size, number of convolutional kernels, number of BiLSTM hidden units, and number of Attention hidden units of the BiTCN-BiLSTM-Attention deep learning model through the improved ICPO algorithm, enabling the combined model to effectively utilize the advantages of each single model to further mine and extract information from the input sequence, more accurately grasp the hidden relationship between wind power features and predicted values, enhance the adaptability of the hybrid model to different wind farm data, and effectively improve the prediction accuracy, stability, and robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present disclosure, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 It is a schematic flow chart of the wind power prediction method based on ICPO-optimized VMD coupled with deep learning of the present invention;

[0045] Figure 2 It is a wind measurement data diagram of the present invention;

[0046] Figure 3 It is a schematic diagram of the prediction results and true values of the BiTCN-BiLSTM-Attention model;

[0047] Figure 4 Schematic diagram of the prediction results and true values of the ICPO-BiTCN-BiLSTM-Attention model

[0048] Figure 5 Schematic diagram of the prediction results and true values of the VMD-ICPO-BiTCN-BiLSTM-Attention model

[0049] Figure 6 Schematic diagram of the prediction results and true values of the IVMD-ICPO-BiTCN-BiLSTM-Attention model proposed by the present invention Detailed implementation manners

[0050] For the convenience of those of ordinary skill in the art to understand and implement the present invention, the following will detail each step of the method proposed by the present invention. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of this application

[0051] Embodiment

[0052] As Figure 1 shown, the present invention discloses a wind power prediction method based on ICPO-optimized VMD coupled with deep learning, including the following steps

[0053] Step S1: Collect a multi-dimensional data set including wind speed, wind direction, air pressure, and wind power output

[0054] Step S2: Construct a VMD decomposition model for the wind power output in the multi-dimensional data set, and optimize the parameters of the VMD decomposition model using the preset ICPO algorithm

[0055] Step S3: Decompose the wind power data using the optimized VMD decomposition model to obtain K sub-components, and sequentially form K multi-dimensional time series data with the remaining original data in the multi-dimensional data set

[0056] Before obtaining the K sub-components, it is necessary to remove or interpolate the abnormal data of the wind farm in the wind power data

[0057] Step S4: Normalize each of the K multi-dimensional time series data and establish a BiTCN-BiLSTM-Attention deep learning model, and optimize the hyperparameters of the BiTCN-BiLSTM-Attention deep learning model using the preset ICPO algorithm

[0058] Step S5: Use the optimized BiTCN-BiLSTM-Attention deep learning model to predict the sub-components, and add up all the prediction results to obtain the wind power prediction value.

[0059] Specifically, for the preset ICPO algorithm described in Step S2 and Step S4, the ICPO algorithm is improved from the CPO algorithm. Before that, a brief introduction and description of the CPO algorithm are given:

[0060] Adopt random initialization of the population, and the formula is:

[0061]

[0062] In the above formula, is the upper limit of the search range, is the lower limit of the search range, is a randomly initialized vector between 0 and 1.

[0063] Adopt a periodic cyclic population reduction technique to accelerate the convergence speed and maintain the diversity of the search population. The formula for periodically changing the population size is:

[0064]

[0065] In the above formula, t c is the number of cycles; Z is the current iteration number; Z n-1 is the population size before the current iteration; t max is the maximum number of iterations; % is the remainder operation symbol; Z min is the minimum number of individuals in the population;

[0066] When the crested porcupine individual detects a predator, it may take four defense measures, namely, visual defense strategy, auditory defense strategy, odor defense strategy, and physical attack defense strategy:

[0067] In the visual defense strategy, the crested porcupine gives a certain deterrence to the predator visually by raising and waving its feathers. After being deterred, the predator has only two choices, either move towards the crested porcupine or move away. The mathematical expression of this behavior is:

[0068]

[0069] In the above formula, is the position of the i-th individual at the n-th iteration; is the optimal individual in the population; R 1 is a random number generated based on the normal distribution; R 2 is a random number between [0, 1]; is the position of any individual in the population at the n-th iteration;

[0070] In the auditory defense strategy, when the crested porcupine detects a predator at a relatively far distance, it may make noises to threaten the predator, and the closer the predator approaches, the louder the noises it makes. The mathematical expression for this behavior is:

[0071]

[0072] In the above formula, is a random binary vector with element values of 0 or 1; and are the positions of any two individuals in the population at the nth iteration; R 3 is a random number between [0, 1];

[0073] In the odor defense strategy, the crested porcupine drives away the predator by secreting a stench, which spreads in the surrounding area. The mathematical expression for this behavior is:

[0074]

[0075] In the above formula, is the position of any individual in the population at the nth iteration; is a parameter controlling the search direction, D n is the defense coefficient, R 3 is a random number between [0, 1], is the odor diffusion factor, τ is a random vector with element values between 0 and 1; R 4 is a random variable between 0 and 1; F(Y i n ) is the objective function value of the corresponding individual; Δ is a minimum value parameter to avoid division by zero;

[0076] In the physical attack defense strategy, when the predator gets very close to the crested porcupine, the crested porcupine will launch a physical attack on it, and during this physical attack process, the crested porcupine and the predator have a one-dimensional inelastic collision. The mathematical expression for this behavior is:

[0077]

[0078] In the above formula, σ is the convergence factor; R 5 and R 6 are random numbers between [0, 1]; is the average force of the crested porcupine's influence on the predator, is a random vector with element values between 0 and 1.

[0079] In this embodiment, the improvements to the above CPO algorithm are as follows:

[0080] There are four defense measures in the CPO algorithm, namely visual defense strategy, auditory defense strategy, odor defense strategy, and physical attack defense strategy. The ICPO algorithm improves the population initialization, visual defense strategy, and physical attack defense strategy among them:

[0081] Initializing the population based on the Circle chaotic map can effectively avoid the uneven distribution of individuals in the standard CPO algorithm's random population initialization. Its mathematical expression is:

[0082]

[0083] In the above formula, mod is the remainder calculation operator; c 1 、c 2 are coefficients taking 0.5 and 0.2 respectively;

[0084] Improving the visual defense strategy of the CPO algorithm based on the Gaussian random walk model to generate new perturbations, which can accelerate the convergence speed of the algorithm and escape at the local optimum. Its mathematical expression is:

[0085]

[0086] In the above formula, is the position of the i-th individual at the n-th iteration; t max is the maximum number of iterations; is the optimal individual in the population; Z is the current iteration number;

[0087] Improving the physical attack defense strategy of the CPO algorithm based on the sine-cosine algorithm, using the periodicity and volatility of the sine-cosine function, and accelerating the convergence speed of the algorithm by continuously adjusting the function value. Its mathematical expression is:

[0088]

[0089] In the above formula, β 1 is the weight factor, β 2 is the search factor, R 7 、R 8 are random numbers between [0, 2π] and [0, 2] respectively, is the force parameter affected by inelastic collision, is the parameter controlling the search direction, D n is the defense coefficient, is the average force of the crested porcupine on the predator.

[0090] Specifically, in step S2, the preset ICPO algorithm is used to optimize the VMD decomposition model parameters, and the optimization process is as follows:

[0091] Step S21: Import the original wind power time series into the VMD decomposition model, and decompose the original wind power using VMD decomposition;

[0092] Step S22: Use the ICPO algorithm with the minimum envelope entropy min{E e} as the objective function to optimize the number of sub-components K and the penalty factor α of the VMD wind power data decomposition model. The mathematical expression of the optimization process is:

[0093]

[0094] In the above formula, e i is the normalized form of a i , a i is the envelope signal of the sub-component after Hilbert transform;

[0095] Step S23: After multiple optimization iterations, until the objective function value no longer changes significantly or reaches the preset maximum number of iterations, output the corresponding optimal number of components K and the penalty factor α, and re-enter them into the VMD decomposition model to obtain the optimized VMD decomposition model.

[0096] Furthermore, use the optimized VMD decomposition model to decompose the wind power data to obtain K sub-components. The process is as follows:

[0097] Use the VMD decomposition model to decompose the original wind power output signal f into K sub-components, and each sub-component has its own unique center frequency ω and finite bandwidth. The mathematical expression is:

[0098]

[0099] In the above formula, K is the number of sub-components decomposed by the mode; {u k} is the k components obtained by the mode decomposition, {ω k} is the center frequency corresponding to the k components; * is the convolution operator; f(t) is the original signal; is the gradient operation;

[0100] Introducing the Lagrange operator λ and the quadratic penalty factor α can transform the constrained variational problem into an unconstrained variational problem. The transformed mathematical expression is:

[0101]

[0102] In the above formula, α is the quadratic penalty parameter; λ is the Lagrange multiplier operator;

[0103] Finally, the optimal solution of the equation is obtained by the method of alternating direction multiplication and Fourier isometric transformation and λ n+1 , and the specific formula is as follows:

[0104]

[0105] In the above formula, γ is the noise tolerance; u i (ω), f(ω), are respectively u i (t), f(t), the Fourier transform of λ(t).

[0106] Specifically, in step S4, the hyperparameters of the BiTCN-BiLSTM-Attention deep learning model are optimized by using the preset ICPO algorithm, and the optimization process is as follows:

[0107] Step S41: Normalize each of the K multi-dimensional time series data, and its mathematical expression is:

[0108]

[0109] In the above formula, o is the original value, o′ is the normalized value, o max is the maximum value of the corresponding feature, o min is the minimum value of the corresponding feature;

[0110] Step S42: Divide the K multi-dimensional time series data after normalization into a training set and a validation set according to a ratio of 8:2, and import them into the BiTCN-BiLSTM-Attention deep learning model to predict the original wind power;

[0111] Step S43: Use the ICPO algorithm to use the minimum root mean square error min{Y RMSE} of the prediction result of the BiTCN-BiLSTM-Attention deep learning model as the objective function, and optimize the learning rate, regularization parameter, convolution kernel size, number of convolution kernels, number of BiLSTM hidden units, and number of Attention hidden units of the BiTCN-BiLSTM-Attention deep learning model. Its mathematical expression is:

[0112]

[0113] In the above formula, y i is the actual value of the wind power, Y iis the predicted value of wind power, and N is the number of data sets;

[0114] Step S44: After multiple optimization iterations, when the objective function value no longer changes significantly or reaches the preset maximum number of iterations, output the corresponding optimal learning rate, regularization parameter, convolutional kernel size, number of convolutional kernels, number of BiLSTM hidden units, and number of Attention hidden units, and re-enter them into the BiTCN-BiLSTM-Attention deep learning model to obtain the optimized BiTCN-BiLSTM-Attention deep learning model.

[0115] The following is a further illustration of the wind power prediction method based on ICPO-optimized VMD coupled with deep learning of the present invention through an actual example analysis.

[0116] To verify the practicability of the wind power prediction method based on ICPO-optimized VMD coupled with deep learning proposed by the present invention; in this embodiment, experiments are carried out based on actual data sets. The experimental data is selected from the data set of a certain area in Jiangsu. The selected data set includes the wind speed and direction at 10m, 30m, 50m, and 70m of the wind measurement tower, and also includes the temperature, air pressure, and humidity during measurement. Finally, the accurate power generation of each period is given. The total data set has 2774 samples, and the ratio of the training set to the validation set is 8:2.

[0117] The software used in the experiment is MATLAB2023a. The simulation environment is based on the Windows10, 64-bit operating system, and the processor is Intel(R) Core(TM) i7-10700 CPU with a main frequency of 2.90GHz. As Figure 2 shown is the wind measurement data graph.

[0118] Furthermore, in order to verify the effectiveness of the proposed IVMD-ICPO-BiTCN-BiLSTM-Attention prediction model, multiple different prediction models (BiTCN-BiLSTM-Attention, ICPO-BiTCN-BiLSTM-Attention, VMD-ICPO-BiTCN-BiLSTM-Attention, IVMD-ICPO-BiTCN-BiLSTM-Attention) are built for comparison, and one-step-ahead wind power predictions are made for them respectively.

[0119] The wind power prediction results of 2774 samples of this wind farm using the IVMD-ICPO-BiTCN-BiLSTM-Attention model and other comparison models are as follows. Figures 3 - 6As shown in the figure. The prediction results adopt the root mean square error RMSE, mean absolute error MAE, relative percentage difference RPD, and coefficient of determination R 2 As evaluation indicators, their calculation formulas are as follows:

[0120]

[0121] Where: yi is the actual value of wind power, and Y i is the predicted value of wind power; is the actual average value of wind power, N is the number of data sets, and Y SD is the standard deviation.

[0122] The evaluation index values of each prediction model are shown in Table 1 below:

[0123] Table 1. Comparison of prediction accuracies of different models

[0124] Model RMSE / MW MAE / MW RPD / % <![CDATA[R 2 > BiTCN - BiLSTM - Attention 3.326 2.609 4.236 0.944 ICPO - BiTCN - BiLSTM - Attention 2.952 2.250 4.771 0.956 VMD - ICPO - BiTCN - BiLSTM - Attention 2.134 1.632 7.065 0.977 IVMD - ICPO - BiTCN - BiLSTM - Attention 1.364 1.867 7.437 0.982

[0125] From the comparison results in Table 1 above, it can be seen that under the four evaluation systems of root mean square error, mean absolute error, relative percentage difference, and coefficient of determination, the values of the IVMD-ICPO-BiTCN-BiLSTM-Attention model proposed by the present invention are the best, which can significantly improve the prediction accuracy of wind power output prediction, assist the staff in the early adjustment effect of wind power generation equipment for the predicted values, and greatly increase the power generation efficiency of wind power.

[0126] The above is only a preferred embodiment of the present invention, and it is not a limitation of the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. A wind power prediction method based on ICPO optimization VMD coupled deep learning, characterized in that: The following steps are involved: Step S1: Collect a multidimensional data set including wind speed, wind direction, air pressure and wind power output power; Step S2: constructing a VMD decomposition model for the wind power output power in the multidimensional data set, and optimizing the VMD decomposition model parameters using a preset ICPO algorithm; Step S3: Decompose the wind power data using the optimized VMD decomposition model to obtain K sub-components, and sequentially combine them with the remaining original data in the multidimensional data set to form K multidimensional time series data; Step S4: Normalize the K multi-dimensional time series data and establish a BiTCN-BiLSTM-Attention deep learning model, and use the preset ICPO algorithm to optimize the hyperparameters of the BiTCN-BiLSTM-Attention deep learning model; Step S5: Use the optimized IVMD-ICPO-BiTCN-BiLSTM-Attention deep learning model to predict the sub-components, and add all the prediction results to obtain the wind power prediction value.

2. According to claim 1, a wind power prediction method based on ICPO optimization VMD coupling deep learning is characterized in that: The preset ICPO algorithm described in step S2 and step S4 is obtained by improving the CPO algorithm, and the specific contents are as follows: There are four defense measures in the CPO algorithm, namely visual defense strategy, auditory defense strategy, odor defense strategy and physical attack defense strategy. The ICPO algorithm improves the population initialization, visual defense strategy and physical attack defense strategy: Initializing the population based on the Circle chaotic map can effectively avoid the phenomenon of uneven individual distribution in the CPO algorithm's randomly initialized population. Its mathematical expression is: In the above formula, mod is the remainder operator; c1 and c2 are coefficients of 0.5 and 0.2 respectively; The visual defense strategy of the CPO algorithm is improved based on the Gaussian random walk model to generate new disturbances, which can accelerate the convergence of the algorithm and escape from the local optimum. Its mathematical expression is: In the above formula, is the position of the ith individual in the nth iteration; t max is the maximum number of iterations; is the optimal individual in the population; Z is the current number of iterations; The physical attack defense strategy of the CPO algorithm is improved based on the sine and cosine algorithm. The periodicity and volatility of the sine and cosine functions are used to accelerate the convergence speed of the algorithm by continuously adjusting the function value. The mathematical expression is: In the above formula, β1 is the weight factor, β2 is the search factor, R7 and R8 are random numbers between [0,2π] and [0,2] respectively. is the force parameter affected by inelastic collision, D is the parameter that controls the search direction. n is the defense coefficient, is the average force of the crested porcupine's impact on the predator.

3. The wind power prediction method based on ICPO optimization and VMD coupling deep learning according to claim 1 is characterized in that: In step S2, the preset ICPO algorithm is used to optimize the VMD decomposition model parameters, and the optimization process is as follows: Step S21: importing the original wind power time series into the VMD decomposition model, and decomposing the original wind power by using VMD decomposition; Step S22: Using ICPO algorithm to obtain the minimum envelope entropy min{E e } As the objective function, the number of sub-components K and the penalty factor α of the VMD wind power data decomposition model are optimized. The mathematical expression of the optimization process is: In the above formula, e i for a i The normalized form of a i is the envelope signal of the subcomponent after Hilbert transform; Step S23: After multiple optimization iterations, until the objective function value no longer changes significantly or reaches the preset maximum number of iterations, the corresponding optimal number of components K and penalty factor α are output and re-input into the VMD decomposition model to obtain the optimized VMD decomposition model.

4. The wind power prediction method based on ICPO optimization and VMD coupling deep learning according to claim 1 is characterized in that: In step S4, the preset ICPO algorithm is used to optimize the hyperparameters of the BiTCN-BiLSTM-Attention deep learning model. The optimization process is as follows: Step S41: Normalize the K multi-dimensional time series data respectively, and the mathematical expression is: In the above formula, o is the original value, o′ is the normalized value, and o max is the maximum value of the corresponding feature, o min is the minimum value of the corresponding feature; Step S42: divide the normalized K multi-dimensional time series data into a training set and a validation set in a ratio of 8:2, and import the BiTCN-BiLSTM-Attention deep learning model to predict the original wind power; Step S43: Use the ICPO algorithm to calculate the minimum root mean square error min{Y RMSE } As the objective function, the learning rate, regularization parameter, convolution kernel size, number of convolution kernels, number of BiLSTM hidden units, and number of Attention hidden units of the BiTCN-BiLSTM-Attention deep learning model are optimized. The mathematical expression is: In the above formula, y i is the actual value of wind power, Y i is the wind power prediction value, N is the number of data sets; Step S44: After multiple optimization iterations, until the objective function value no longer changes significantly or reaches the preset maximum number of iterations, the corresponding optimal learning rate, regularization parameter, convolution kernel size, number of convolution kernels, number of BiLSTM hidden units, and number of Attention hidden units are output, and re-input into the BiTCN-BiLSTM-Attention deep learning model to obtain the optimized IVMD-ICPO-BiTCN-BiLSTM-Attention deep learning model.

5. The wind power prediction method based on ICPO optimization VMD coupling deep learning according to claim 1 is characterized in that: In step S3, the optimized VMD decomposition model is used to decompose the wind power data to obtain K sub-components. Before obtaining the K sub-components, abnormal wind farm data in the wind power data is removed or interpolated.

Images