Ultra-short-term wind power prediction method and system for long-short-term memory network of permutation entropy

Abstract

The invention discloses an ultra-short-term wind power prediction method and system for a long-short-term memory network of permutation entropy, and the method comprises the steps: collecting historical wind power data, and repairing the missing data and abnormal values in the historical wind power data; performing normalization processing on the preprocessed historical wind power data; inputting the normalized data, optimizing and solving the number K of decomposition modes and a penalty factor alpha by using a multi-target grey wolf algorithm, ensuring that fitness functions RMSE and PE are globally minimum, obtaining solved parameters, and further performing decomposition to obtain decomposed subsequences; the decomposed subsequences are used as input, and then prediction model training of the subsequences is carried out; and carrying out rolling decomposition, inputting a prediction model of the subsequences, superposing predicted values of the subsequences, and carrying out reverse normalization to obtain a prediction result, thereby realizing rolling prediction. According to the method, the problem that the prediction precision of data with strong volatility is not high in the existing prediction technology is solved, and the problem that the wind power plant faces assessment due to inaccurate prediction is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wind power prediction, and particularly relates to a permutation entropy long short-term memory network ultra-short-term wind power prediction method and system. BACKGROUND

[0002] Wind power capacity continues to grow and has become an important part of the global power grid. The randomness and volatility of wind power itself have caused certain impact on the safety and stability of the power grid. Wind power prediction can predict the trend of power output, short-term prediction can provide reference for the dispatch center to arrange power generation plan, and ultra-short-term prediction can provide basis for accurate power output of the energy storage system, which provides support for safe grid connection of wind power. Wind power prediction methods are mainly divided into physical methods and artificial intelligence algorithms. Numerical weather prediction, as a common physical method, has limited generalization ability due to the consideration of complex factors such as terrain. In recent years, data-driven artificial intelligence prediction algorithms are still a research hotspot. In this field, common algorithms include autoregressive and moving average models and convolutional neural networks. Long short-term memory network (LSTM) and its variants have become the mainstream choice for wind power prediction due to their strong ability to mine time series dependencies. The main difficulty of wind power prediction lies in the high noise and non-stationary characteristics of sequence data, so researchers introduce signal decomposition techniques such as empirical mode decomposition, ensemble empirical mode decomposition (EEMD) and variational mode decomposition (VMD). VMD overcomes the problem of difficulty in distinguishing frequency characteristics in EEMD, realizes accurate signal decomposition, and has higher overall operation decomposition efficiency. In order to improve the decomposition effect, some researchers also propose to use optimization algorithms to optimize the decomposition parameters, and select sample entropy (SE) as the fitness function. Similarly, these studies select envelope entropy or SE as the fitness function of single-objective optimization, then decompose the power data sequence, and select LSTM or other prediction models for prediction. Single-objective optimization cannot simultaneously consider the authenticity of signal decomposition and the effectiveness of subsequence characteristics, and will inevitably sacrifice one of them, which will affect the accuracy of prediction. At the same time, although SE is more sensitive to sequence time correlation than envelope entropy, it has higher computational complexity. Permutation entropy (PE) evaluates nonlinear dynamic characteristics by analyzing sequence patterns of data points in phase space, and is superior to traditional entropy indicators in low-amplitude chaotic signals. Therefore, PE and root mean square error (RMSE) are used as the fitness function of optimization. The subsequence obtained by introducing them into the fitness function is more conducive to model training. SUMMARY

[0003] The purpose of the present application is to provide a permutation entropy long short-term memory network ultra-short-term wind power prediction method and system, which solves the problem of low prediction accuracy of existing prediction technology for strongly fluctuating data, and reduces the problem of wind farms facing examination due to inaccurate prediction.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: The method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy includes the following steps: Step 1: Data Preprocessing Collect historical wind power data and repair missing data and outliers in the historical wind power data; Step 2: Data normalization processing The preprocessed historical wind power data is normalized. Step 3: VMD parameter optimization Input the normalized data and use the multi-objective gray wolf algorithm to optimize the solution for the number of decomposed modes. K and penalty factor α， To ensure that the fitness functions RMSE and PE are globally minimized, the desired parameters are obtained and then decomposed to obtain the decomposed subsequences; Step 4: Train the prediction model for each subsequence The decomposed subsequences are used as input to train the prediction model for the subsequences. Step 5: Rolling decomposition. The prediction model of the input subsequence is superimposed, and the prediction values ​​of the subsequence are inversely normalized to obtain the prediction result, thereby realizing rolling prediction.

[0005] A further improvement of this invention is that, in step 1, historical wind power data is collected, and missing data and outliers in the historical wind power data are repaired, as follows: (1) in, for i The true value of historical wind power data at any given time. For the sample size, for i Correction values ​​for missing data at any given time. This is the average of historical wind power data. , , These are the weighting coefficients.

[0006] A further improvement of this invention is that, in step 2, the preprocessed historical wind power data is normalized, and the normalization formula is as shown in equation (2): (2) in, The minimum value of this type of data. The maximum value of this type of data. This represents the actual data value.

[0007] Further improvement of the present application is that in step 3, the obtained parameters are further decomposed to obtain the decomposed sub-sequences, and when signal decomposition is performed, the number of given decomposition modes K and a penalty factor α , comprising: The expression of the optimization problem is shown as formula (3): (3) (4) Wherein, represents the i-th mode, represents the center frequency, K is the total number of modes, is the Dirac distribution, is the original signal; Solving the optimization expression, introducing a penalty factor α , a Lagrange multiplier operator λ , so as to transform into an unconstrained variational problem, and obtain an augmented Lagrange expression: (5) Wherein, α is a penalty factor parameter, λ is a Lagrange multiplier, represents a convolution operator; Then use the alternating direction multiplier method to solve the variational problem, and obtain the updated mode component and center frequency expression: (6) (7) Wherein represents the conversion of the signal from time domain to frequency domain through Fourier transform, 、 、 、 , into 、 、 、 ; When using variational mode decomposition VMD for signal decomposition, the number of decomposition modes K and the penalty factor α are preset; In order to improve the signal decomposition performance of variational mode decomposition VMD, the grey wolf optimization algorithm GWO is used to optimize the number of decomposition modes K , the penalty factor α and the time step of VMD, and the root mean square error and permutation entropy are used as the fitness function of GWO to obtain the decomposition effect considering consistency and feature effectiveness.

[0008] The further improvement of the present application is that in step 4, the decomposed subsequence is input to further train the prediction model of the subsequence, including: The feedforward calculation process of the long short-term memory network is represented as: (10) (11) (12) (13) (14) (15) wherein, , and are output signals of the forget gate, the input gate and the output gate of the neuron respectively; is the potential state information of the neuron, represents the state information of the neuron, represents the hidden state of the neuron, and tanh represents the hyperbolic tangent activation function, represents the S-shaped activation function, represents the weight of the corresponding gate, b represents the bias term of the gate, is the input sequence value of the neuron.

[0009] The permutation entropy long short-term memory network ultra-short-term wind power prediction system comprises: A data preprocessing unit is configured to collect historical wind power data and repair missing data and abnormal values in the historical wind power data. A data normalization processing unit is configured to perform normalization processing on the preprocessed historical wind power data. A VMD parameter optimization unit is configured to input the normalized data, and use a multi-objective grey wolf algorithm to optimize and solve the number of decomposition modes K and a penalty factor α， to ensure the global minimum of the fitness function RMSE and PE, obtain the solved parameters, and then perform decomposition to obtain the decomposed subsequence. A prediction model training unit is configured to input the decomposed subsequence to further train the prediction model of the subsequence. A prediction unit is configured to perform rolling decomposition, input the prediction model of the subsequence, superimpose the subsequence prediction value, and perform inverse normalization to obtain a prediction result, thereby realizing rolling prediction.

[0010] A further improvement of this invention is that, in the data preprocessing unit, historical wind power data is collected, and missing data and outliers in the historical wind power data are repaired, as follows: (1) in, for i The true value of historical wind power data at any given time. For the sample size, for i Correction values ​​for missing data at any given time. This is the average of historical wind power data. , , These are the weighting coefficients.

[0011] A further improvement of this invention is that, in the data normalization processing unit, the preprocessed historical wind power data is normalized, and the normalization formula is as shown in equation (2): (2) in, The minimum value of this type of data. The maximum value of this type of data. This represents the actual data value.

[0012] A further improvement of this invention lies in that, in the VMD parameter optimization unit, the desired parameters are obtained and then decomposed to obtain the decomposed subsequences. When performing signal decomposition, the number of decomposition modes is given. K and penalty factor α ,include: The expression for the optimization problem is shown in equation (3): (3) (4) in, Represents the i-th mode. Indicates the center frequency. K It is the total number of modes. It is the Dirac distribution. It is the original signal; Solving for the optimization expression involves introducing a penalty factor based on the construction of a variational problem. α Lagrange multiplication operator λ This transforms the problem into an unconstrained variational problem, leading to the augmented Lagrange expression: (5) in, α It is the penalty factor parameter. λ It is a Lagrange multiplication operator. denotes a convolution operator; Then the alternating direction multiplier method is used to solve the variational problem, and the updated modal component and center frequency expression are obtained: (6) (7) Wherein denotes the conversion of the signal from the time domain to the frequency domain through the Fourier transform, 、 、 、 , into 、 、 、 ; When using variational mode decomposition VMD for signal decomposition, the decomposition mode number K and the penalty factor α are set in advance; In order to improve the signal decomposition performance of variational mode decomposition VMD, the grey wolf optimization algorithm GWO is used to optimize the decomposition mode number K , the penalty factor α and the time step of VMD, and the root mean square error and the permutation entropy are used as the fitness function of GWO to obtain the decomposition effect considering consistency and feature effectiveness.

[0013] A computer readable storage medium stores a computer program, and the computer program realizes the steps of the permutation entropy long short-term memory network ultra-short-term wind power prediction method when executed by a processor.

[0014] Compared with the prior art, the present application has at least the following beneficial technical effects: For the scene of strong wind speed change, randomness, large fluctuation amplitude and time sequence characteristics, VMD is used for power decomposition and LSTM is used for effective power prediction. The permutation entropy and its root mean square error are introduced as the fitness function of VMD parameter optimization. The root mean square error is a global measurement criterion, which ensures the strict consistency of the signal before and after decomposition. PE algorithm is a method that can effectively enhance the weak change of time series and detect the randomness of time series. It has the advantages of simplicity, strong anti-interference ability and good robustness. In this study, PE algorithm is used to measure the chaos degree of each subsequence. Smaller PE value will be more conducive to deep learning training and prediction. This goal is achieved through multi-objective grey wolf optimization algorithm. GWO algorithm is a common swarm intelligence optimization algorithm, which seeks the optimal solution by simulating the cooperation and hunting strategy of wolves. It has the advantages of simplicity, easy implementation and fast convergence. In order to realize the multi-objective optimization of GWO, this study improves GWO by adding a new external archive for storing and retrieving Pareto optimal solutions, and supplements the leader selection strategy to help select new leader groups during the hunting process, thus forming MOGWO algorithm. The MOGWO-VMD-LSTM ultra-short-term wind power prediction method is obtained. The effectiveness of the method is verified through single-step and multi-step prediction which meets the actual engineering requirements. Compared with other algorithms without enhanced power processing, this algorithm has better prediction performance, which has key theoretical significance and practical engineering value. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application. Those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0016] Figure 1 is the total flowchart of the prediction method of the present application.

[0017] Figure 2 is the result schematic diagram of the prediction method embodiment of the present application, in which the 16th point is selected as the evaluation point for multi-step prediction, corresponding to the 4th hour in actual prediction.

[0018] Figure 3 is the structural block diagram of the prediction system of the present application. DETAILED DESCRIPTION

[0019] In the following, certain exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present application. Therefore, the drawings and description are to be regarded as illustrative in nature rather than restrictive.

[0020] In the description of the present application, it is to be understood that the terms "including", "comprising", "having" and "with" when used in this specification and in the following claims indicate the presence of the stated features, integers, steps, operations, elements, and / or components but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0021] It should also be understood that the terms used in the present specification and the appended claims are merely for the purpose of describing particular embodiments and do not intend to limit the present application. As used in the specification and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0022] It should further be understood that the term "and / or" used in the present specification and the appended claims, means one or more of the associated listed items as well as all possible combinations of the items.

[0023] Various structural diagrams according to the disclosed embodiments of the present application are shown in the drawings. These diagrams are not drawn to scale, in which certain details are exaggerated for the purpose of clarity and certain details can be omitted. The shapes of various regions, layers shown in the diagrams and their relative sizes, positional relationships are merely exemplary, and in actuality, they can be deviated due to manufacturing tolerances or technical limitations, and regions / layers with different shapes, sizes, relative positions can be additionally designed by those skilled in the art according to actual needs.

[0024] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0025] Embodiment 1 With reference to Figure 1 The permutation entropy long short-term memory network ultra-short-term wind power prediction method provided by the present application is implemented according to the following steps: Step 1, data preprocessing Collect data and supplement missing data in the historical wind power data of the electricity selling user. For the missing data, use formula (1) for correction.

[0026] (1) Wherein, is ithe true value of the historical power selling data at the time, the number of samples, the mean value of the historical power selling data, i the corrected value of the missing data at the time, the mean value of the historical power selling data, , , the weight coefficient.

[0027] Step 2, data normalization In the deep learning model for wind power prediction, the historical wind power data input needs to be normalized. Wind power is affected by wind speed, equipment status and other factors, and historical wind power data often has large fluctuations. If the model is directly input, the difference in numerical range may cause the model to be overly sensitive to large value features, ignoring the key fluctuation patterns contained in small value features, interfering with the stability of the gradient descent process, delaying the convergence speed and even causing training oscillation. After normalization, the power features at different times are mapped to the same order of magnitude, making the model more efficient in learning the inherent rules of the data, improving the parameter optimization efficiency and prediction accuracy, reducing the interference of extreme values on model training, and enhancing the generalization ability of the model. The normalization formula is as formula (2): (2) wherein, the minimum value of the data, the maximum value of the data, the true value of the data.

[0028] Step 3, VMD parameter optimization The non-stationary characteristics of wind power reduce the efficiency of deep learning, especially when the wind speed increases or decreases rapidly. It is difficult to accurately predict by deep learning alone. Signal decomposition can clearly distinguish between trend items and turbulent items, thereby reducing the difficulty of subsequent deep learning. Variational mode decomposition (VMD) is a widely used signal decomposition technique that can decompose signals into a number of intrinsic mode functions (IMF) with limited bandwidth by variational optimization. The core idea is to minimize the estimated bandwidth of each mode by solving an optimization problem. This decomposition method ensures the extraction of different characteristics of wind power.

[0029] The expression of the optimization problem is shown in formula (3): (3) (4) wherein, the i-th mode, the center frequency, K the total number of modes, the Dirac distribution, is the original signal.

[0030] Solving the optimization expression, introducing a penalty factor on the basis of constructing a variation problem α , Lagrange multiplier operator λ , so as to be converted into an unconstrained variation problem, and a augmented Lagrange expression is obtained: (5) Wherein, α is a penalty factor parameter, λ is a Lagrange multiplier, represent the convolution operator.

[0031] Then the alternating direction multiplier method is used to solve the variation problem, and the updated modal component and center frequency expression is obtained: (6) (7) Wherein indicates that the signal is converted from the time domain to the frequency domain through the Fourier transform, 、 、 、 , is converted into 、 、 、 .

[0032] When using the variation mode decomposition (VMD) for signal decomposition, the decomposition mode number K and the penalty factor α need to be set in advance, and the selection of the two parameters has an important influence on the decomposition result, and then affects the accuracy of wind power prediction. When the decomposition mode number K is small, it is easy to cause under-decomposition, so that important information is not fully extracted, and when K is large, over-decomposition phenomenon occurs, and the center frequency of the decomposed modal component is similar, thereby forming modal aliasing. The penalty factor α mainly affects the detail retention degree of the modal component and the denoising effect of the decomposition process: when α is large, the bandwidth of the decomposed modal component is narrow, and vice versa, when α is small, the bandwidth of the modal component is wide. In addition, although the time step has a small influence on the decomposition result, it also needs to be considered in the optimization process.

[0033] In order to improve the signal decomposition performance of the variation mode decomposition VMD, the grey wolf optimization algorithm (GWO) is used to optimize the decomposition mode number K of VMD αand time step. The best fitness function needs to be found in order for the GWO algorithm to determine the relevant optimal parameter values of VMD through the comparison of objective function values. In this paper, the root mean square error (RMSE) and permutation entropy (PE) are used as the fitness function of GWO to obtain the decomposition effect that takes into account both consistency and feature effectiveness.

[0034] PE algorithm is a method that can effectively enhance the weak changes in time series and detect the randomness of time series. This algorithm has the advantages of simple principle, strong anti-interference ability and good robustness. In this study, the permutation entropy algorithm is used to measure the complexity of each sub-modal component, which is used as one of the fitness functions. The permutation entropy of the time series is calculated as follows: (8) where, is the number of different permutation patterns, is the probability of the i-th permutation, which is obtained by dividing the number of times the permutation pattern appears by the total number of embedded vectors.

[0035] GWO is a common swarm intelligence optimization algorithm that finds the optimal solution by simulating the cooperation and hunting strategy of wolf packs. It has the advantages of simple principle, easy implementation and fast convergence speed. In the optimization of the key parameters of VMD (penalty factor and mode number), GWO algorithm has an important influence on the decomposition result. The update strategy formula of GWO algorithm in the optimization process is as follows: (9) To achieve multi-objective optimization of GWO, the invention improves GWO by adding an external archive to store and retrieve Pareto optimal solutions, and supplements the leader selection strategy to assist in screening new leader groups during the hunting process, finally forming a multi-objective grey wolf optimization algorithm (MOGWO).

[0036] Step 4, power decomposition and LSTM model training After completing the subsequence decomposition, a prediction model needs to be constructed for each subsequence. Long short-term memory network (LSTM) is an advanced form of recurrent neural network (RNN), which was originally designed to solve the problems faced by traditional RNN in dealing with long-term dependencies. LSTM introduces a cell state structure and uses a unique gating mechanism to fine-tune the memory and forgetting process of information, which can effectively capture and encode long-term dependencies in time series data. This ability significantly improves the accuracy of wind power prediction. The feedforward calculation process of the LSTM model is represented as: (10) (11) (12) (13) (14) (15) wherein, , and are the output signals of the forget gate, input gate and output gate of the neuron, respectively; is the potential state information of the neuron, denotes the state information of the neuron, denotes the hidden state of the neuron, tanh denotes the hyperbolic tangent activation function, represents the S-shaped activation function, represents the weight of the corresponding gate, b represents the bias term of the gate, is the input sequence value of the neuron.

[0037] Step 5, prediction is respectively performed by using the trained sub-sequence model, and the final prediction result is obtained after superposition and inverse normalization.

[0038] Embodiment 2 With reference to Figure 2 , the permutation entropy long short-term memory network ultra-short-term wind power prediction method provided by the application is implemented according to the following steps: Step 1, data preprocessing. Collect historical hourly power generation data of a wind farm from May 1, 2024 to May 31, 2024, and supplement the missing data in the historical wind power data. For the missing data, use formula (1) for correction.

[0039] Step 2, normalize the input historical wind power data, and use formula (2) to map the data to the [0, 1] uniform interval.

[0040] Step 3, VMD parameter optimization. Input the historical wind power data, use the MOGWO algorithm to optimize the VMD parameters, and obtain the parameters that minimize the sum of sub-sequence permutation entropy and the minimum RMSE.

[0041] Step 4, power decomposition and model training. Input the optimized parameters into the VMD decomposition, and train the LSTM model for the decomposed sub-sequences.

[0042] Step 5, by inputting new actual wind power, real-time decomposition and rolling prediction are performed, the sub-sequence prediction results are superimposed, and the final prediction result is obtained after inverse normalization.

[0043] The method of the present application is based on the method of multi-objective grey wolf optimization, variational mode decomposition and long short-term memory network for ultra-short-term wind power prediction. For the characteristics of strong randomness, large fluctuation amplitude and time sequence of wind power in the scene of strong wind speed change, variational mode decomposition is selected for power sequence decomposition and combined with long short-term memory network for effective power prediction. The parameter setting of variational mode decomposition directly affects the decomposition effect, and the traditional parameter selection method is difficult to balance the signal consistency and the regularity of the subsequence. When dealing with high fluctuation and strong random wind power sequence, long short-term memory network is prone to cause prediction accuracy to decrease due to high input sequence confusion degree.

[0044] The permutation entropy algorithm is a method that can effectively enhance the weak change of time series and detect the randomness of time series, has the advantages of simplicity, strong anti-interference ability and good robustness, and smaller permutation entropy value is more conducive to deep learning training and prediction. The root mean square error is a global measurement criterion, which can strictly ensure the consistency of the signal before and after decomposition. By mixing the two as the fitness function of VMD parameter optimization, the signal consistency guarantee and the regularity of the subsequence can be realized at the same time. In view of the above problems, the multi-objective grey wolf optimization algorithm is used to optimize the VMD parameters. The algorithm seeks the optimal solution by simulating the cooperation and predation strategy of wolves, has the advantages of simplicity, easy implementation and fast convergence, and by increasing the external archive to store the Pareto optimal solution and supplementing the leader selection strategy to update the leader group, the multi-objective grey wolf optimization algorithm is formed. The ultra-short-term wind power prediction model based on MOGWO-VMD-LSTM is established to improve the effectiveness of wind power prediction. The effectiveness of the method is verified through single-step and multi-step prediction, and the method has better prediction performance than other algorithms without enhanced power processing, which has key theoretical significance and practical engineering value.

[0045] Embodiment 3 With reference to Figure 3 The permutation entropy long short-term memory network ultra-short-term wind power prediction system provided by the present application has the characteristics that it comprises: A data preprocessing unit is configured to collect historical wind power data and repair missing data and abnormal values in the historical wind power data. A data normalization processing unit is configured to perform normalization processing on the preprocessed historical wind power data. A VMD parameter optimization unit is configured to input the normalized data, and use a multi-objective grey wolf algorithm to optimize and solve the number of decomposition modes. K and a penalty factor α， The RMSE and PE global minimum of the fitness function are ensured, the parameters are obtained, and the subsequence after decomposition is obtained. A prediction model training unit is configured to input the decomposed subsequence and train the prediction model of the subsequence. The prediction unit is used for rolling decomposition. It takes the prediction model of the input subsequence, superimposes the predicted values ​​of the subsequence, and inversely normalizes them to obtain the prediction result, thereby realizing rolling prediction.

[0046] In the data preprocessing unit of this embodiment, historical wind power data is collected, and missing data and outliers in the historical wind power data are repaired, as follows: (1) in, for i The true value of historical wind power data at any given time. For the sample size, for i Correction values ​​for data that is missing at any given time. This is the average of historical wind power data. , , These are the weighting coefficients.

[0047] In the data normalization processing unit of this embodiment, the preprocessed historical wind power data is normalized, and the normalization formula is as shown in equation (2): (2) in, The minimum value of this type of data. The maximum value of this type of data. This represents the actual data value.

[0048] In the VMD parameter optimization unit of this embodiment, the desired parameters are obtained and then decomposed to obtain the decomposed subsequences. When performing signal decomposition, the number of decomposed modes is given. K and penalty factor α ,include: The expression for the optimization problem is shown in equation (3): (3) (4) in, Represents the i-th mode. Indicates the center frequency. K It is the total number of modes. It is the Dirac distribution. It is the original signal; Solving for the optimization expression involves introducing a penalty factor based on the construction of a variational problem. α Lagrange multiplication operator λ This transforms the problem into an unconstrained variational problem, leading to the augmented Lagrange expression: (5) wherein, α is a penalty factor parameter, λ is a Lagrange multiplier, represents a convolution operator; Then the variational problem is solved using the alternating direction multiplier method to obtain the updated modal component and center frequency expression: (6) (7) wherein represents the signal conversion from time domain to frequency domain through Fourier transform, 、 、 、 is converted into 、 、 、 ; When using the variational mode decomposition VMD for signal decomposition, the decomposition mode number K and the penalty factor α are preset; In order to improve the signal decomposition performance of the variational mode decomposition VMD, the grey wolf optimization algorithm GWO is used to optimize the decomposition mode number K of the VMD, the penalty factor α and the time step, and the root mean square error and the permutation entropy are used as the fitness function of the GWO to obtain a decomposition effect considering consistency and feature effectiveness.

[0049] Embodiment 4 The computer readable storage medium provided by the application stores a computer program, and the computer program realizes the steps of the permutation entropy long short-term memory network ultra-short-term wind power prediction method when executed by a processor.

[0050] Those skilled in the art should understand that the embodiments of the application can be provided as methods, systems, or computer program products. Therefore, the application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0051] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 system to perform the function specified in the flowchart block or blocks.

[0052] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 system to perform the function specified in the flowchart block or blocks.

[0053] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 system to perform the function specified in the flowchart block or blocks.

[0054] The principles and advantages of the present application have been described above with the illustrated embodiments. It is apparent, however, to one skilled in the art that numerous changes, modifications and substitutions can be made in the specific embodiments which have been illustrated and described without departing from the spirit and scope of the present application. Therefore, it is the intent that the scope of the present application be limited solely by the scope of the claims set forth below. Any reference numbers in the claims are intended to refer to the illustrative embodiments, and not to limit the claims in any way.

[0055] Furthermore, it should be understood that although the specification is described in terms of embodiments, each of which contains only one independent technical solution, the specification is described in this way only for the sake of clarity, and the skilled person should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that the skilled person can understand. The above is only to illustrate the technical idea of the present application, and cannot limit the protection scope of the present application. Any modification made on the basis of the technical solutions according to the technical idea of the present application falls within the protection scope of the claims of the present application.

Claims

1. A method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy, characterized in that, Includes the following steps: Step 1: Data Preprocessing Collect historical wind power data and repair missing data and outliers in the historical wind power data; Step 2: Data normalization processing The preprocessed historical wind power data is normalized. Step 3: VMD parameter optimization Input the normalized data and use the multi-objective gray wolf algorithm to optimize the solution for the number of decomposed modes. K and penalty factor α， To ensure that the fitness functions RMSE and PE are globally minimized, the desired parameters are obtained and then decomposed to obtain the decomposed subsequences; Step 4: Train the prediction model for each subsequence The decomposed subsequences are used as input to train the prediction model for the subsequences. Step 5: Rolling decomposition. The prediction model of the input subsequence is superimposed, and the prediction values ​​of the subsequence are inversely normalized to obtain the prediction result, thereby realizing rolling prediction.

2. The method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy according to claim 1, characterized in that, In step 1, historical wind power data is collected, and missing data and outliers in the historical wind power data are repaired, as follows: (1) in, for i The true value of historical wind power data at any given time. For the sample size, for i Correction values ​​for data that is missing at any given time. This is the average of historical wind power data. , , These are the weighting coefficients.

3. The method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy according to claim 1, characterized in that, In step 2, the preprocessed historical wind power data is normalized using the formula (2): (2) in, The minimum value of this type of data. The maximum value of this type of data. This represents the actual data value.

4. The method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy according to claim 1, characterized in that, In step 3, the desired parameters are obtained and then decomposed to obtain the decomposed subsequences. When performing signal decomposition, the number of decomposition modes is given. K and penalty factor α ,include: The expression for the optimization problem is shown in equation (3): (3) (4) in, Represents the i-th mode. Indicates the center frequency. K It is the total number of modes. It is the Dirac distribution. It is the original signal; Solving for the optimization expression involves introducing a penalty factor based on the construction of a variational problem. α Lagrange multiplication operator λ This transforms the problem into an unconstrained variational problem, leading to the augmented Lagrange expression: (5) in, α It is the penalty factor parameter. λ It is a Lagrange multiplication operator. Represents the convolution operator; Next, the variational problem is solved using the alternating direction multiplier method, yielding updated expressions for the modal components and center frequencies: (6) (7) in This indicates that the signal has been transformed from the time domain to the frequency domain, after undergoing a Fourier transform. , , , It was transformed into , , , ; When using Variational Mode Decomposition (VMD) for signal decomposition, the number of decomposition modes is preset. K and penalty factor α ; To improve the signal decomposition performance of Variational Mode Decomposition (VMD), the Grey Wolf Optimization (GWO) algorithm is used to optimize the decomposition of the VMD modes. K Punishment factor α The time step is optimized, and the root mean square error and permutation entropy are used as the fitness function of GWO to achieve a decomposition effect that balances consistency and feature effectiveness.

5. The method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy according to claim 1, characterized in that, In step 4, the decomposed subsequences are used as input to train the prediction model for the subsequences, including: The feedforward computation process of a Long Short-Term Memory (LSTM) network is represented as follows: (10) (11) (12) (13) (14) (15) in, , and These are the output signals of the neuron's forget gate, input gate, and output gate, respectively. It is the latent state information of neurons. This represents the state information of neurons. represents the hidden state of a neuron, and tanh represents the hyperbolic tangent activation function. Represents the sigmoid activation function. Represents the weight of the corresponding gate. b The term representing the door's bias. It is the input sequence value of the neuron.

6. A long short-term memory network-based ultra-short-term wind power prediction system based on permutation entropy, characterized in that, include: The data preprocessing unit is used to collect historical wind power data and repair missing data and outliers in the historical wind power data; The data normalization processing unit is used to normalize the preprocessed historical wind power data; The VMD parameter optimization unit is used to input normalized data and optimize the number of decomposed modes using the multi-objective gray wolf algorithm. K and penalty factor α， To ensure that the fitness functions RMSE and PE are globally minimized, the desired parameters are obtained and then decomposed to obtain the decomposed subsequences; The prediction model training unit is used to train the prediction model of the subsequence by taking the decomposed subsequence as input. The prediction unit is used for rolling decomposition. It takes the prediction model of the input subsequence, superimposes the predicted values ​​of the subsequence, and inversely normalizes them to obtain the prediction result, thereby realizing rolling prediction.

7. The long short-term memory network ultra-short-term wind power prediction system based on permutation entropy according to claim 6, characterized in that, In the data preprocessing unit, historical wind power data is collected, and missing data and outliers in the historical wind power data are repaired, as follows: (1) in, for i The true value of historical wind power data at any given time. For the sample size, for i Correction values ​​for data that is missing at any given time. This is the average of historical wind power data. , , These are the weighting coefficients.

8. The method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy according to claim 6, characterized in that, In the data normalization processing unit, the preprocessed historical wind power data is normalized using the normalization formula (2): (2) in, The minimum value of this type of data. The maximum value of this type of data. This represents the actual data value.

9. The method for predicting ultra-short-term wind power using long short-term memory networks based on permutation entropy according to claim 6, characterized in that, In the VMD parameter optimization unit, the desired parameters are obtained and then decomposed to obtain the decomposed subsequences. When performing signal decomposition, the number of decomposition modes is given. K and penalty factor α ,include: The expression for the optimization problem is shown in equation (3): (3) (4) in, Represents the i-th mode. Indicates the center frequency. K It is the total number of modes. It is the Dirac distribution. It is the original signal; Solving for the optimization expression involves introducing a penalty factor based on the construction of a variational problem. α Lagrange multiplication operator λ This transforms the problem into an unconstrained variational problem, leading to the augmented Lagrange expression: (5) in, α It is the penalty factor parameter. λ It is a Lagrange multiplication operator. Represents the convolution operator; Next, the variational problem is solved using the alternating direction multiplier method, yielding updated expressions for the modal components and center frequencies: (6) (7) in This indicates that the signal has been transformed from the time domain to the frequency domain, after undergoing a Fourier transform. , , , It was transformed into , , , ; When using Variational Mode Decomposition (VMD) for signal decomposition, the number of decomposition modes is preset. K and penalty factor α ; To improve the signal decomposition performance of Variational Mode Decomposition (VMD), the Grey Wolf Optimization (GWO) algorithm is used to optimize the decomposition of the VMD modes. K Punishment factor α The time step is optimized, and the root mean square error and permutation entropy are used as the fitness function of GWO to achieve a decomposition effect that balances consistency and feature effectiveness.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the long short-term memory network ultra-short-term wind power prediction method based on permutation entropy as described in any one of claims 1-5.

Images