Short-term wind power prediction method and system based on AVMD and SMA-LSSVM combined model

Abstract

The invention relates to the technical field of a novel power system power grid, in particular to a short-term wind power prediction method and system based on an AVMD and SMA-LSSVM combined model. The method comprises the following steps: firstly, performing multi-scale decomposition on a wind power signal by using an adaptive variational mode decomposition (AVMD) method to obtain a plurality of subsequences; secondly, constructing a prediction model of the subsequences by using a myxomycete algorithm (SMA) and a least square support vector machine (LSSVM), and adaptively determining parameters of the LSSVM; and finally, performing weighted fusion on prediction results of the subsequences to obtain a final wind power prediction value. An AVMD method is adopted to reduce the non-stationarity and noise interference of signals and improve the prediction precision of subsequences; a new intelligent optimization algorithm is constructed in combination with SMA and LSSVM, and errors caused by manual parameter setting are avoided. The prediction results of the subsequences are subjected to weighted fusion, the non-stationarity and nonlinearity of the wind power sequence are effectively processed by fully utilizing the prediction information of each subsequence, and the precision and stability of short-term wind power prediction are improved.

Description

Technical Field

[0001] The invention relates to the technical field of new power system power grid, and in particular to a short-term wind power prediction method and system based on an AVMD and SMA-LSSVM combined model. Background Art

[0002] Wind power generation is a clean energy that converts wind energy into electrical energy. It has the advantages of low carbon and no pollution, and is an important part of future energy. However, wind power generation is also affected by the instability of wind speed and wind force, which brings difficulties to the operation and management of wind farms. Therefore, accurate prediction of wind power generation is of great significance for improving the efficiency and safety of wind farms, reducing the conflict between wind power and power grids, and optimizing the scheduling and maintenance of wind farms. In order to solve the problem of wind power generation prediction, scholars have proposed a variety of methods, which can be roughly divided into physical methods and statistical methods. The physical method is based on numerical weather forecast data and describes the process of converting wind energy into electrical energy by establishing a complex physical model. The statistical method is based on the idea of ​​data-driven and fits the relationship between historical data and wind power generation by establishing a mathematical function. Statistical methods are usually faster and more accurate than physical methods, but they need to consider issues such as data quality and feature selection. In recent years, with the development of artificial intelligence technology, machine learning and deep learning methods have been widely used in wind power generation prediction. In addition, in order to reduce the non-stationarity of the original data, the signal decomposition algorithm is often used to decompose the original data into several sub-modes. Variational mode decomposition, as an adaptive and completely non-recursive decomposition method, is more robust to noise.

[0003] In view of the above analysis, the present invention proposes a wind power prediction model based on adaptive variational mode decomposition and SMA-LSSVM combination. Firstly, the genetic algorithm is used to optimize VMD to perform adaptive multi-scale decomposition of wind power signals to obtain several subsequences; secondly, the slime mold algorithm (SMA) and the least squares support vector machine (LSSVM) are used to construct a prediction model for the subsequences, and the parameters of the LSSVM are adaptively determined; finally, the prediction results of the subsequences are weighted and fused to obtain the final wind power prediction value. The present invention shows through example analysis that the proposed model has higher prediction accuracy and lower prediction error. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that in recent years, with the development of artificial intelligence technology, machine learning and deep learning methods have been widely used in wind power generation prediction. In addition, in order to reduce the non-stationarity of the original data, a signal decomposition algorithm is often used to decompose the original data into several sub-modes. Variational mode decomposition, as an adaptive and completely non-recursive decomposition method, has stronger robustness to noise.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a short-term wind power prediction method based on a combined model of AVMD and SMA-LSSVM, which includes collecting historical wind speed and wind power data of a wind farm to form a wind power time series;

[0008] The wind power signal is decomposed into multiple scales using the adaptive variational mode decomposition method to obtain several subsequences.

[0009] The prediction model of subsequences is constructed by using slime mold algorithm and least squares support vector machine, and the parameters of least squares support vector machine are adaptively determined;

[0010] The prediction results of the subsequences are weighted and integrated to obtain the final wind power prediction value;

[0011] Mean absolute error, mean square error, root mean square error and determination coefficient were selected as evaluation indicators to verify the effectiveness of the model.

[0012] As a preferred solution of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model of the present invention, wherein: the wind power signal is decomposed at multiple scales using the adaptive variational mode decomposition method to obtain several subsequences, which specifically adopts the following steps:

[0013] Perform VMD decomposition on the wind power time series to obtain the decomposition factor K and penalty factor α;

[0014] The sample entropy is selected as the fitness function of the genetic algorithm, and the optimal decomposition factor K and penalty factor α are output adaptively.

[0015] As a preferred solution of the short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model of the present invention, wherein: the wind power time series is subjected to VMD decomposition to obtain the decomposition factor K and the penalty factor α, which specifically adopts the following steps:

[0016] Establish a variational optimization model; VMD is a non-recursive method that decomposes the time series y(t) into K modes Each mode u k(t) has a center frequency w k , the goal of VMD is to minimize the sum of the estimated bandwidth of each mode. In order to estimate the bandwidth of all modes, the following variational optimization model is established:

[0017]

[0018] In the formula, * represents convolution; represents the decomposed set of modal signals; represents the set of center frequencies; δ(t) is the Dirac function, (δ(t)+j / (πt)*u k (t) is u k The Hilbert transform of (t), δ t represents the partial derivative operator with respect to t, j 2 = -1;

[0019] Construct an augmented Lagrangian function; In order to solve the above optimization problem, first construct an augmented Lagrangian function, and its specific formula is:

[0020]

[0021] Where: α is the quadratic penalty factor and α>0; <.,.> is the inner product operator of the function; λ is the Lagrange factor;

[0022] Use ADMM to update alternately and When{u k}、{w k} and λ are given, first update The Fourier transform of is:

[0023]

[0024] Update center frequency:

[0025]

[0026] Update the Lagrange multipliers:

[0027]

[0028] In the formula, and They are Fourier transform of y(ω) and λ(ω); τ is the noise margin and τ>0;

[0029] Finally, when the iteration termination condition is met, the iteration is terminated. The specific formula is:

[0030]

[0031] In the formula, e represents the judgment constraint condition and e>0.

[0032] As a preferred solution of the short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model of the present invention, wherein: the sample entropy is selected as the fitness function of the genetic algorithm, and the optimal decomposition factor K and the penalty factor α are adaptively output, which specifically adopts the following steps:

[0033] First, define the signal X as a time series of length N. The specific formula is:

[0034] X={x 1 ,x 2 ,...x N}

[0035] The signal X is constructed as an m-dimensional vector, and its specific formula is:

[0036] X(i) = {x i ,x i+1 ,...x i+m-1}

[0037] Where i=1,2,3,...,N-m+1, m is the window length;

[0038] Define the distance parameter d[X i ,X j ] indicates X i and X j The maximum distance difference between them is as follows:

[0039] d[X i ,X j ]=max k∈(0,m-1) |x i+k -x j+k |

[0040] Set the similarity tolerance threshold r, and count d[X i ,X j ]<r, and construct a ratio with the total number of vectors Nm:

[0041]

[0042] For all r, the average of all proportion statistics is:

[0043]

[0044] Increase m (m=m+1), repeat Step 3.2-Step 3.5,

[0045] The sample entropy of the signal sequence is obtained as:

[0046] SampEn=-ln[B m+1 (r) / B m (r)].

[0047] As a preferred solution of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model of the present invention, wherein: the prediction model of the subsequence is constructed by using the slime mold algorithm and the least squares support vector machine, and the parameters of the least squares support vector machine are adaptively determined, which specifically adopts the following steps:

[0048] The SMA is an intelligent optimization algorithm based on natural phenomena, which simulates the movement behavior of slime mold and searches for the optimal solution by continuously updating the position and mass of the slime mold. It can effectively avoid falling into the local optimum and improve the optimization efficiency. The LSSVM is a regression analysis method based on the principle of structural risk minimization. By introducing Lagrange multipliers, the original nonlinear optimization problem is converted into a linear equation system, which reduces the computational complexity and improves the prediction accuracy.

[0049] As a preferred solution of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model of the present invention, wherein: the prediction model of the subsequence is constructed by using the slime mold algorithm and the least squares support vector machine, and the parameters of the least squares support vector machine are adaptively determined, which specifically adopts the following steps:

[0050] A sample of training data can be represented as x i ∈R is the i-th input vector, y i ∈R is the target value of the i-th sample, N is the sample space, and the expression of the LSSVM model in the feature space is:

[0051]

[0052] In the formula, is the dimensionality transformation function that maps the input sample data space to the high-dimensional feature space; b is the bias factor; ω is the weight vector;

[0053] The objective function and constraints of the least squares support vector machine are as follows:

[0054]

[0055] In the formula, γ is the penalty factor; e i is the slack variable;

[0056] Then the optimization problem is transformed into a Lagrangian function to solve it. The specific formula is:

[0057]

[0058] In the formula, λ i is the Lagrange multiplier;

[0059] The partial derivative is simplified to KKT conditions and solved using RBF Gaussian kernel function; the specific solution process is as follows:

[0060]

[0061] In summary, the nonlinear equation of LSSVM is:

[0062]

[0063] In the formula, K(x,x i ) is the radial basis kernel function, which transforms data from low-dimensional space to high-dimensional space; δ is the kernel function width, K(x,x i ) = exp(-||x i || 2 / 2δ 2 ).

[0064] As a preferred solution of the short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model of the present invention, wherein: the mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model, which specifically adopts the following steps:

[0065] The mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model. The specific formula is:

[0066]

[0067] Where: n is the number of sampling points; y i and are the predicted value and true value of the i-th wind power point in the test set, respectively.

[0068] In a second aspect, an embodiment of the present invention provides a short-term wind power prediction system based on a combined model of AVMD and SMA-LSSVM, which includes a collection module for collecting historical wind speed and wind power data of a wind farm to form a wind power time series;

[0069] The output module uses the adaptive variational mode decomposition method to perform multi-scale decomposition on the wind power signal to obtain several subsequences;

[0070] A construction module is used to construct a prediction model of subsequences using the slime mold algorithm and the least squares support vector machine, and the parameters of the least squares support vector machine are adaptively determined;

[0071] The weighting module performs weighted fusion on the prediction results of the subsequences to obtain the final wind power prediction value;

[0072] In the validation module, mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model.

[0073] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model as described in the first aspect of the present invention are implemented.

[0074] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model as described in the first aspect of the present invention are implemented.

[0075] The beneficial effects of the present invention are as follows: the present invention proposes a short-term wind power prediction method based on an AVMD and SMA-LSSVM combined model, which utilizes an adaptive variational mode decomposition method to perform multi-scale decomposition on a wind power signal, which can effectively reduce the non-stationarity and noise interference of the signal and improve the prediction accuracy of each subsequence; a new intelligent optimization algorithm is constructed using a slime mold algorithm and a least squares support vector machine, which can adaptively determine the parameters of the least squares support vector machine, avoid errors caused by artificially setting parameters, and improve the generalization ability of the prediction model; weighted fusion is performed on the prediction results of each subsequence, which can make full use of the prediction information of each subsequence, effectively deal with the non-stationarity and nonlinearity of the wind power sequence, and improve the accuracy and stability of short-term wind power prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0077] Figure 1 It is a flow chart of the short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model;

[0078] Figure 2 It is a computer equipment diagram of the short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model;

[0079] Figure 3 The AVMD-SMA-LSSVM prediction model structure is an embodiment of a short-term wind power prediction method based on a combined AVMD and SMA-LSSVM model;

[0080] Figure 4 The original wind power curve diagram of the embodiment of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model;

[0081] Figure 5 is a fitness function curve of an embodiment of a short-term wind power prediction method based on a combined model of AVMD and SMA-LSSVM;

[0082] Figure 6 It is an AVMD decomposition result diagram of an embodiment of a short-term wind power prediction method based on an AVMD and SMA-LSSVM combined model;

[0083] Figure 7 The error comparison under different prediction models of the embodiment of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model is shown;

[0084] Figure 8 The prediction results of the test sets under different models of the embodiment of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model are compared. DETAILED DESCRIPTION

[0085] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0086] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0087] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0088] Example 1

[0089] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model, comprising:

[0090] S100: Collect historical wind speed and wind power data of the wind farm to form a wind power time series;

[0091] S200: performing multi-scale decomposition of the wind power signal using an adaptive variational mode decomposition method to obtain a plurality of subsequences;

[0092] S201: Using an adaptive variational modal decomposition method to perform multi-scale decomposition on the wind power signal to obtain a number of subsequences, which specifically adopts the following steps:

[0093] Perform VMD decomposition on the wind power time series to obtain the decomposition factor K and penalty factor α;

[0094] The sample entropy is selected as the fitness function of the genetic algorithm, and the optimal decomposition factor K and penalty factor α are output adaptively.

[0095] S202: Perform VMD decomposition on the wind power time series to obtain a decomposition factor K and a penalty factor α, which specifically adopts the following steps:

[0096] Establish a variational optimization model; VMD is a non-recursive method that decomposes the time series y(t) into K modes Each mode u k (t) has a center frequency w k , the goal of VMD is to minimize the sum of the estimated bandwidth of each mode. In order to estimate the bandwidth of all modes, the following variational optimization model is established:

[0097]

[0098] In the formula, * represents convolution; represents the decomposed set of modal signals; represents the set of center frequencies; δ(t) is the Dirac function, (δ(t)+j / (πt)*u k (t) is u k The Hilbert transform of (t), δ t represents the partial derivative operator with respect to t, j 2 = -1;

[0099] Construct an augmented Lagrangian function; In order to solve the above optimization problem, first construct an augmented Lagrangian function, and its specific formula is:

[0100]

[0101] Where: α is the quadratic penalty factor and α>0; <.,.> is the inner product operator of the function; λ is the Lagrange factor;

[0102] Use ADMM to update alternately and λ n+1 , when {u k}、{w k} and λ are given, first update The Fourier transform of is:

[0103]

[0104] Update center frequency:

[0105]

[0106] Update the Lagrange multipliers:

[0107]

[0108] In the formula, and They are Fourier transform of y(ω) and λ(ω); τ is the noise margin and τ>0;

[0109] Finally, when the iteration termination condition is met, the iteration is terminated. The specific formula is:

[0110]

[0111] In the formula, e represents the judgment constraint condition and e>0.

[0112] S203: Select sample entropy as the fitness function of the genetic algorithm, and adaptively output the optimal decomposition factor K and penalty factor α, which specifically adopts the following steps:

[0113] First, define the signal X as a time series of length N. The specific formula is:

[0114] X={x 1 ,x 2 ,...x N}

[0115] The signal X is constructed as an m-dimensional vector, and its specific formula is:

[0116] X(i) = {x i ,x i+1 ,...x i+m-1}

[0117] Where i=1,2,3,...,N-m+1, m is the window length;

[0118] Define the distance parameter d[X i ,X j ] indicates X i and X jThe maximum distance difference between them is as follows:

[0119] d[X i ,X j ]=max k∈(0,m-1) |x i+k -x j+k |

[0120] Set the similarity tolerance threshold r, and count d[X i ,X j ]<r, and construct a ratio with the total number of vectors Nm:

[0121]

[0122] For all r, the average of all proportion statistics is:

[0123]

[0124] Increase m (m=m+1), repeat Step 3.2-Step 3.5,

[0125] The sample entropy of the signal sequence is obtained as:

[0126] SampEn=-ln[B m+1 (r) / B m (r)].

[0127] S300: constructing a prediction model of a subsequence using a slime mold algorithm and a least squares support vector machine, and adaptively determining parameters of the least squares support vector machine;

[0128] S301: construct a prediction model for subsequences using the slime mold algorithm and the least squares support vector machine, and adaptively determine the parameters of the least squares support vector machine, which specifically adopts the following steps:

[0129] SMA is an intelligent optimization algorithm based on natural phenomena. It simulates the movement behavior of slime mold and searches for the optimal solution by continuously updating the position and mass of the slime mold. It can effectively avoid falling into the local optimum and improve the optimization efficiency. LSSVM is a regression analysis method based on the principle of structural risk minimization. By introducing Lagrange multipliers, the original nonlinear optimization problem is transformed into a set of linear equations, which reduces the computational complexity and improves the prediction accuracy.

[0130] S302: construct a prediction model for the subsequence using the slime mold algorithm and the least squares support vector machine, and adaptively determine the parameters of the least squares support vector machine, which specifically adopts the following steps:

[0131] A sample of training data can be represented as x i ∈R is the i-th input vector, yi ∈R is the target value of the i-th sample, N is the sample space, and the expression of the LSSVM model in the feature space is:

[0132]

[0133] In the formula, is the dimensionality transformation function that maps the input sample data space to the high-dimensional feature space; b is the bias factor; ω is the weight vector;

[0134] The objective function and constraints of the least squares support vector machine are as follows:

[0135]

[0136] In the formula, γ is the penalty factor; e i is the slack variable;

[0137] Then the optimization problem is transformed into a Lagrangian function to solve it. The specific formula is:

[0138]

[0139] In the formula, λ i is the Lagrange multiplier;

[0140] The partial derivative is simplified to KKT conditions and solved using RBF Gaussian kernel function; the specific solution process is as follows:

[0141]

[0142] In summary, the nonlinear equation of LSSVM is:

[0143]

[0144] In the formula, K(x,x i ) is the radial basis kernel function, which transforms data from low-dimensional space to high-dimensional space; δ is the kernel function width, K(x,x i ) = exp(-||x i || 2 / 2δ 2 ).

[0145] S400: performing weighted fusion on the prediction results of the subsequences to obtain a final wind power prediction value;

[0146] S500: Select mean absolute error, mean square error, root mean square error and determination coefficient as evaluation indicators to verify the effectiveness of the model.

[0147] S501: Select mean absolute error, mean square error, root mean square error and determination coefficient as evaluation indicators to verify the effectiveness of the model. The specific steps are as follows:

[0148] The mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model. The specific formula is:

[0149]

[0150]

[0151] Where: n is the number of sampling points; y i and are the predicted value and true value of the i-th wind power point in the test set, respectively.

[0152] Furthermore, this embodiment also provides a short-term wind power prediction system based on the AVMD and SMA-LSSVM combined model, comprising:

[0153] The acquisition module collects historical wind speed and wind power data of the wind farm to form a wind power time series;

[0154] The output module uses the adaptive variational mode decomposition method to perform multi-scale decomposition on the wind power signal to obtain several subsequences;

[0155] A construction module is used to construct a prediction model of subsequences using the slime mold algorithm and the least squares support vector machine, and the parameters of the least squares support vector machine are adaptively determined;

[0156] The weighting module performs weighted fusion on the prediction results of the subsequences to obtain the final wind power prediction value;

[0157] In the validation module, mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model.

[0158] This embodiment also provides a computer device, which is suitable for the case of a short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model proposed in the above embodiment.

[0159] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0160] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model proposed in the above embodiment is implemented.

[0161] In summary, the present invention proposes a short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model. The method uses the adaptive variational mode decomposition method to perform multi-scale decomposition on the wind power signal, which can effectively reduce the non-stationarity and noise interference of the signal and improve the prediction accuracy of each subsequence; a new intelligent optimization algorithm is constructed using the slime mold algorithm and the least squares support vector machine, which can adaptively determine the parameters of the least squares support vector machine, avoid the errors caused by artificially set parameters, and improve the generalization ability of the prediction model; the prediction results of each subsequence are weighted and fused, which can make full use of the prediction information of each subsequence, effectively deal with the non-stationarity and nonlinearity of the wind power sequence, and improve the accuracy and stability of short-term wind power prediction.

[0162] Example 2

[0163] Reference Figure 1 - Figure 8 , which is the second embodiment of the present invention, and this embodiment provides a short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0164] The present invention uses the power generation data of a wind farm in the north in 2020 as a sample. The original wind power sample data is as follows: Figure 4 shown. Figure 3This is the AVMD-SMA-LSSVM prediction model structure of the embodiment, and the AVMD-SMA-LSSVM combined model is used for prediction and compared with the SMA-LSSVM model. In order to accurately evaluate the model performance, the annual data is divided into several quarters, the first 75% of the data in each quarter is the training set, and the last 25% of the data is the test set. The experimental results show that the combined model proposed in the present invention has advantages in wind power generation prediction. When processing data, the time dimension is selected as 15 minutes, which not only takes into account the influence of historical data, but also avoids the influence of historical data that is too far away and too small. In addition, the 15-minute time dimension retains more data features, which is conducive to modal decomposition to map the historical relationship between training data and test data, thereby improving prediction accuracy. The experimental data selected the first 15 days of data from the typical months of February, May, August and November in the four seasons. The prediction results of the test set under different models are compared. Figure 8 shown.

[0165] Figure 1 This is a flowchart of the adaptive variational mode decomposition implementation. VMD is a non-recursive method that decomposes the time series y(t) into K modes. Each mode u k (t) has a center frequency w k , the goal of VMD is to minimize the sum of the estimated bandwidths of each mode. In order to estimate the bandwidths of all modes, the following variational optimization model is established:

[0166]

[0167] In the formula, * represents convolution; represents the decomposed set of modal signals; represents the set of center frequencies; δ(t) is the Dirac function, (δ(t)+j / (πt)*u k (t) is u k The Hilbert transform of (t), δ t represents the partial derivative operator with respect to t, j 2 =-1.

[0168] Construct an augmented Lagrangian function; In order to solve the above optimization problem, first construct an augmented Lagrangian function, and its specific formula is:

[0169]

[0170] Where: α is the quadratic penalty factor and α>0; <.,.> is the function inner product operator; λ is the Lagrange factor.

[0171] Use ADMM to update alternately and λ n+1 . When{u k}、{w k} and λ are given, first update The Fourier transform of is:

[0172]

[0173] Update center frequency:

[0174]

[0175] Update the Lagrange multipliers:

[0176]

[0177] In the formula, and They are Fourier transform of y(ω) and λ(ω); τ is the noise margin and τ>0.

[0178] Finally, when the iteration termination condition is met, the iteration is terminated. The specific formula is:

[0179]

[0180] In the formula, e represents the judgment constraint condition and e>0.

[0181] The values ​​of the decomposition factor K and the penalty factor α in the VMD algorithm directly affect the final signal processing results. When the K value is too large, the signal will be over-decomposed and the signal details will be lost; when the K value is too small, the signal will be under-decomposed and the signal-to-noise separation cannot be improved. At the same time, if α is too large, the frequency band signal will be lost; on the contrary, the information will be redundant. In the present invention, the sample entropy (SampEn) is selected as the fitness function of the genetic algorithm. Figure 4 is the fitness function curve of the embodiment. It is an improved algorithm derived from approximate entropy to measure the complexity of time series. The smaller the sample entropy value, the more concentrated the spectrum, the higher the sequence self-similarity, the more complex the sequence, and the wider the spectrum. The algorithm of SampEn is as follows:

[0182] First, define the signal X as a time series of length N. The specific formula is:

[0183] X={x 1 ,x 2 ,...x N}

[0184] The signal X is constructed as an m-dimensional vector, and its specific formula is:

[0185] X(i) = {x i ,x i+1 ,...x i+m-1}

[0186] Wherein, i=1,2,3,...,N-m+1, and m is the window length.

[0187] Define the distance parameter d[X i ,X j ] indicates X i and X j The maximum distance difference between them is as follows:

[0188] d[X i ,X j ]=max k∈(0,m-1) |x i+k -x j+k |

[0189] Set the similarity tolerance threshold r, and count d[X i ,X j ]<r, and construct a ratio with the total number of vectors Nm.

[0190]

[0191] For all r, the average of all proportion statistics is:

[0192]

[0193] Increase m (m=m+1) and repeat equations (8) to (10).

[0194] The sample entropy of the signal sequence is obtained as:

[0195] SampEn=-ln[B m+1 (r) / B m (r)]

[0196] In summary, we can get the sample entropy of each component sequence decomposed by the VMD algorithm under different K and α. Finally, we seek the decomposition method with the smallest chromosome and the smallest average fitness value. At the same time, we get the optimal K and α by comparison. The AVMD decomposition result is shown in the figure below: Figure 6 shown.

[0197] The least squares support vector machine is a machine learning method based on statistical learning theory and can be used for classification and regression. It improves on the traditional support vector machine by replacing the inequality constraints with equality constraints that can be solved by a system of linear equations. The sample of training data can be represented as x i ∈R is the i-th input vector, y i ∈R is the target value of the i-th sample, and N is the sample space. The expression of LSSVM in the feature space is:

[0198]

[0199] In the formula, is the dimensionality transformation function that maps the input sample data space to the high-dimensional feature space; b is the bias factor; ω is the weight vector.

[0200] The objective function and constraints of the least squares support vector machine are as follows:

[0201]

[0202] In the formula, γ is the penalty factor; e i is the slack variable.

[0203] Then the optimization problem is transformed into a Lagrangian function to solve it. The specific formula is:

[0204] L(ω,b,e,λ)=J(ω,b,e)-

[0205]

[0206] In the formula, λ i is the Lagrange multiplier.

[0207] The partial derivatives of the Lagrangian function above are simplified to KKT conditions and solved using the RBF Gaussian kernel function. The specific solution process is as follows:

[0208]

[0209] In summary, the nonlinear equation of LSSVM is:

[0210]

[0211] In the formula, K(x,x i ) is the radial basis kernel function, which transforms data from low-dimensional space to high-dimensional space; δ is the kernel function width, K(x,x i ) = exp(-||x i || 2 / 2δ 2 ).

[0212] Figure 7 This is the error comparison under different prediction models in the embodiment. In order to evaluate the prediction accuracy of the algorithm proposed in the present invention and the comparison algorithm, the present invention will select four commonly used error metrics for analysis and verify the effectiveness of the model by comparing the size of the error. They are mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE) and determination coefficient (R 2)is the evaluation index. The calculation formulas of each evaluation index are:

[0213]

[0214] Where: n is the number of sampling points; y i and are the predicted value and true value of the i-th wind power point in the test set, respectively.

[0215] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model, characterized by: include, Collect historical wind speed and wind power data of wind farms to form wind power time series; The wind power signal is decomposed into multiple scales using the adaptive variational mode decomposition method to obtain several subsequences. The prediction model of subsequences is constructed by using slime mold algorithm and least squares support vector machine, and the parameters of least squares support vector machine are adaptively determined; The prediction results of the subsequences are weighted and integrated to obtain the final wind power prediction value; Mean absolute error, mean square error, root mean square error and determination coefficient were selected as evaluation indicators to verify the effectiveness of the model.

2. The short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model according to claim 1 is characterized in that: The adaptive variational modal decomposition method is used to perform multi-scale decomposition on the wind power signal to obtain several subsequences, which specifically adopts the following steps: Perform VMD decomposition on the wind power time series to obtain the decomposition factor K and penalty factor α; The sample entropy is selected as the fitness function of the genetic algorithm, and the optimal decomposition factor K and penalty factor α are output adaptively.

3. The short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model as claimed in claim 2 is characterized by: The VMD decomposition of the wind power time series is performed to obtain the decomposition factor K and the penalty factor α, which specifically adopts the following steps: Establish a variational optimization model; VMD is a non-recursive method that decomposes the time series y(t) into K modes Each mode u k (t) has a center frequency w k , the goal of VMD is to minimize the sum of the estimated bandwidth of each mode. In order to estimate the bandwidth of all modes, the following variational optimization model is established: In the formula, * represents convolution; represents the decomposed set of modal signals; represents the set of center frequencies; δ(t) is the Dirac function, (δ(t)+j / (πt)*u k (t) is u k The Hilbert transform of (t), δ t represents the partial derivative operator with respect to t, j 2 = -1; Construct an augmented Lagrangian function; In order to solve the above optimization problem, first construct an augmented Lagrangian function, and its specific formula is: Where: α is the quadratic penalty factor and α>0; <.,.> is the inner product operator of the function; λ is the Lagrange factor; Use ADMM to update alternately and λ n+1 , when {u k }、{w k } and λ are given, first update The Fourier transform of is: Update center frequency: Update the Lagrange multipliers: In the formula, and They are Fourier transform of y(ω) and λ(ω); τ is the noise margin and τ>0; Finally, when the iteration termination condition is met, the iteration is terminated. The specific formula is: In the formula, e represents the judgment constraint condition and e>0.

4. The short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model as claimed in claim 3 is characterized by: The sample entropy is selected as the fitness function of the genetic algorithm, and the optimal decomposition factor K and penalty factor α are outputted adaptively. The specific steps are as follows: First, define the signal X as a time series of length N. The specific formula is: X={x1,x2,...x N } The signal X is constructed as an m-dimensional vector, and its specific formula is: X(i)={x i ,x i+1 ,...x i+m-1 } Where i=1,2,3,...,N-m+1, m is the window length; Define the distance parameter d[X i ,X j ] indicates X i and X j The maximum distance difference between them is as follows: d[X i ,X j ]=max k∈(0,m-1) |x i+k -x j+k | Set the similarity tolerance threshold r, and count d[X i ,X j ]<r, and construct a ratio with the total number of vectors Nm: For all r, the average of all proportion statistics is: Increase m (m=m+1), repeat Step 3.2-Step 3.5, The sample entropy of the signal sequence is obtained as: SampEn6-ln[B m+1 (r) / B m (r)] 5. The short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model according to claim 4 is characterized in that: The method of using the slime mold algorithm and the least squares support vector machine to construct a prediction model for a subsequence and adaptively determining the parameters of the least squares support vector machine specifically adopts the following steps: The SMA is an intelligent optimization algorithm based on natural phenomena, which simulates the movement behavior of slime mold and searches for the optimal solution by continuously updating the position and mass of the slime mold. It can effectively avoid falling into the local optimum and improve the optimization efficiency. The LSSVM is a regression analysis method based on the principle of structural risk minimization. By introducing Lagrange multipliers, the original nonlinear optimization problem is converted into a linear equation system, which reduces the computational complexity and improves the prediction accuracy.

6. The short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model according to claim 5 is characterized in that: The method of using the slime mold algorithm and the least squares support vector machine to construct a prediction model for a subsequence and adaptively determining the parameters of the least squares support vector machine specifically adopts the following steps: A sample of training data can be represented as x i ∈R is the i-th input vector, y i ∈R is the target value of the i-th sample, N is the sample space, and the expression of the LSSVM model in the feature space is: In the formula, is the dimensionality transformation function that maps the input sample data space to the high-dimensional feature space; b is the bias factor; ω is the weight vector; The objective function and constraints of the least squares support vector machine are as follows: In the formula, γ is the penalty factor; e i is the slack variable; Then the optimization problem is transformed into a Lagrangian function to solve it. The specific formula is: In the formula, λ i is the Lagrange multiplier; The partial derivative is simplified to KKT conditions and solved using RBF Gaussian kernel function; the specific solution process is as follows: In summary, the nonlinear equation of LSSVM is: In the formula, K(x,x i ) is the radial basis kernel function, which transforms data from low-dimensional space to high-dimensional space; δ is the kernel function width, K(x,x i ) = exp(-||x i || 2 / 2δ 2 ).

7. The short-term wind power forecasting method based on the AVMD and SMA-LSSVM combined model according to claim 6 is characterized in that: The mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model, which specifically adopts the following steps: The mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model. The specific formula is: Where: n is the number of sampling points; y i and are the predicted value and true value of the i-th wind power point in the test set, respectively.

8. A short-term wind power prediction system based on a combined model of AVMD and SMA-LSSVM, based on the short-term wind power prediction method based on a combined model of AVMD and SMA-LSSVM according to any one of claims 1 to 7, characterized in that: Also includes, The acquisition module collects historical wind speed and wind power data of the wind farm to form a wind power time series; The output module uses the adaptive variational mode decomposition method to perform multi-scale decomposition on the wind power signal to obtain several subsequences; A construction module is used to construct a prediction model of subsequences using the slime mold algorithm and the least squares support vector machine, and the parameters of the least squares support vector machine are adaptively determined; The weighting module performs weighted fusion on the prediction results of the subsequences to obtain the final wind power prediction value; In the validation module, mean absolute error, mean square error, root mean square error and determination coefficient are selected as evaluation indicators to verify the effectiveness of the model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the short-term wind power prediction method based on the AVMD and SMA-LSSVM combined model described in any one of claims 1 to 7 are implemented.

Images