Short-term wind speed prediction method and system based on gated learning network

Abstract

The invention discloses a short-term wind speed prediction method and system based on a gated learning network, and the method comprises the steps: carrying out the decomposition processing of processed historical wind speed time sequence data and real-time meteorological environment sequence data through a multivariate empirical mode decomposition algorithm, and obtaining an initial sample data set; performing multi-modal feature extraction processing on the initial sample data set to obtain a target sample data set; training a wind speed prediction neural network model by using the target sample data set and the determined model correction parameters, wherein the wind speed prediction neural network model is designed to be constructed by a gating mechanism and a feature learning adaptive algorithm; performing short-term wind speed prediction according to the wind speed prediction neural network model; and the short-term wind speed can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of wind speed prediction, and in particular to a short-term wind speed prediction method and system based on a gated learning network. Background Art

[0002] Wind energy is one of the most promising renewable energy sources and has attracted widespread attention worldwide due to its abundant reserves, eco-friendliness, and economic efficiency. However, due to the intermittent nature of wind speed, wind power access may have an impact on the stability of the power system. In order to maximize the utilization and effectively manage wind power generation, accurate and stable wind speed forecasting is essential. Generally speaking, wind speed forecasting has different time scales, including ultra-short term (<30 minutes), short term (30 minutes-6 hours), medium term (6 hours-1 day), and long term (>1 day). Among them, short-term wind speed forecasting plays a vital role in turbine control, operational safety, and economic load scheduling.

[0003] In order to achieve accurate wind speed prediction, researchers have proposed many strategies. However, since wind speed is affected by many factors such as temperature, air pressure, terrain, etc. and has strong randomness, the existing short-term wind speed prediction methods will affect the output accuracy of the prediction results during the actual prediction process due to their own limitations, and cannot achieve accurate prediction of short-term wind speed.

[0004] It can be seen that how to accurately predict short-term wind speed has become a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention

[0005] The present invention provides a short-term wind speed prediction method and system based on a gated learning network, so as to solve the problem of how to improve the process of wind speed prediction so as to improve the wind speed prediction accuracy.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a short-term wind speed prediction method based on a gated learning network, comprising:

[0007] The processed historical wind speed time series data and real-time meteorological environment series data are decomposed using the multivariate empirical mode decomposition algorithm to obtain the initial sample data set;

[0008] Performing multimodal feature extraction processing on the initial sample data set to obtain a target sample data set;

[0009] A wind speed prediction neural network model is constructed using the target sample data set and the determined model correction parameters, wherein the wind speed prediction neural network model is designed to be constructed by a gating mechanism and a feature learning adaptive algorithm;

[0010] The short-term wind speed is predicted based on the wind speed prediction neural network model.

[0011] Furthermore, the real-time meteorological environment sequence data includes at least wind direction data, temperature and humidity data, air pressure data, rainfall data and radiation data;

[0012] The method of using the multivariate empirical mode decomposition algorithm to decompose the processed historical wind speed time series data and real-time meteorological environment series data includes:

[0013] Decomposing the historical wind speed time series data and the real-time meteorological environment series data respectively by the multivariate empirical mode decomposition algorithm to obtain the intrinsic mode functions and residual components corresponding to the two;

[0014] Furthermore, the performing multimodal feature extraction processing on the initial sample data set to obtain a target sample data set includes:

[0015] Using a partial autocorrelation function to extract features from the decomposed historical wind speed time series data to obtain a first feature set;

[0016] Using a cross-correlation function to extract features from the decomposed real-time meteorological environment sequence data to obtain a second feature set;

[0017] The first feature set and the second feature set are normalized to obtain the target sample data set.

[0018] Furthermore, the process of constructing the wind speed prediction neural network model includes:

[0019] Inputting the training set in the target sample data set into the neural network model for training to construct the wind speed prediction neural network model;

[0020] Predicting the short-term wind speed according to the wind speed prediction neural network model to obtain a wind speed prediction value, and calculating a root mean square error between the wind speed prediction value and the actual wind speed value to obtain a corresponding fitness parameter;

[0021] The fitness parameter is expressed by the following formula:

[0022]

[0023] Among them, γ i is the actual value of wind speed; i ' is the wind speed prediction value; RMSE is the root mean square error;

[0024] Under preset conditions, based on the fitness parameters, optimizing and adjusting the hyperparameters of the wind speed prediction neural network model;

[0025] The optimized wind speed prediction neural network model is trained based on the model correction parameters determined after optimization and adjustment.

[0026] Furthermore, the process of constructing the wind speed prediction neural network model also includes:

[0027] According to the target sample data set, a corresponding initial population vector is randomly generated; the initial population vector includes the hyperparameter to be optimized and adjusted;

[0028] Iteratively updating and locally searching the initial population vector according to a weighted average algorithm to determine the corresponding model correction parameters;

[0029] The wind speed prediction neural network model is optimized according to the model correction parameter. Another embodiment of the present invention provides a short-term wind speed prediction system based on a gated learning network, comprising:

[0030] The data decomposition module is used to decompose the processed historical wind speed time series data and real-time meteorological environment series data using the multivariate empirical mode decomposition algorithm to obtain an initial sample data set;

[0031] A feature extraction module, used to perform multimodal feature extraction processing on the initial sample data set to obtain a target sample data set;

[0032] A prediction model construction module, used to construct a wind speed prediction neural network model using the target sample data set and the determined model correction parameters, wherein the wind speed prediction neural network model is designed to be constructed by a gating mechanism and a feature learning adaptive algorithm;

[0033] The wind speed prediction module is used to predict short-term wind speed according to the wind speed prediction neural network model.

[0034] Furthermore, the real-time meteorological environment sequence data includes at least wind direction data, temperature and humidity data, air pressure data, rainfall data and radiation data;

[0035] The data decomposition module is specifically used for:

[0036] Decomposing the historical wind speed time series data and the real-time meteorological environment series data respectively by the multivariate empirical mode decomposition algorithm to obtain the intrinsic mode functions and residual components corresponding to the two;

[0037] Furthermore, the feature extraction module is specifically used for:

[0038] Using a partial autocorrelation function to extract features from the decomposed historical wind speed time series data to obtain a first feature set;

[0039] Using a cross-correlation function to extract features from the decomposed real-time meteorological environment sequence data to obtain a second feature set;

[0040] The first feature set and the second feature set are normalized to obtain the target sample data set.

[0041] Furthermore, the prediction model building module is specifically used to:

[0042] Inputting the training set in the target sample data set into the neural network model for training to construct the wind speed prediction neural network model;

[0043] Predicting the short-term wind speed according to the wind speed prediction neural network model to obtain a wind speed prediction value, and calculating a root mean square error between the wind speed prediction value and the actual wind speed value to obtain a corresponding fitness parameter;

[0044] The fitness parameter is expressed by the following formula:

[0045]

[0046] Among them, γ i is the actual value of wind speed; i ' is the wind speed prediction value; RMSE is the root mean square error;

[0047] Under preset conditions, based on the fitness parameters, optimizing and adjusting the hyperparameters of the wind speed prediction neural network model;

[0048] The optimized wind speed prediction neural network model is trained based on the model correction parameters determined after optimization and adjustment.

[0049] Furthermore, the prediction model building module is also used to:

[0050] According to the target sample data set, a corresponding initial population vector is randomly generated; the initial population vector includes the hyperparameter to be optimized and adjusted;

[0051] Iteratively updating and locally searching the initial population vector according to a weighted average algorithm to determine the corresponding model correction parameters;

[0052] The wind speed prediction neural network model is optimized according to the model correction parameters.

[0053] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0054] (1) The multivariate empirical mode decomposition algorithm is used to concurrently decompose multi-channel data into subsequences with uniform intrinsic mode functions and residual components to ensure the frequency consistency of the subsequences; and partial autocorrelation function and cross-correlation function are used for feature extraction to eliminate redundant components and noise, which can improve the prediction accuracy and reduce the computational complexity.

[0055] (2) The adoption of adaptive update rules and local random search strategies can more effectively handle the short-term wind speed prediction optimization problem and ensure the robustness of the complex solution space. The short-term wind speed prediction model is constructed using a gated feature adaptive learning network with better feature representation capabilities, which improves the prediction accuracy of the algorithm and achieves accurate prediction of short-term wind speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic flow chart of a short-term wind speed prediction method based on a gated learning network in one embodiment of the present invention;

[0057] Figure 2 is a schematic diagram of a gated learning network structure in one embodiment of the present invention;

[0058] Figure 3 It is an overall flow chart of the short-term wind speed prediction process in one embodiment of the present invention;

[0059] Figure 4 is a schematic diagram of the structure of a short-term wind speed prediction system based on a gated learning network in one embodiment of the present invention; DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0062] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are only for illustrative purposes, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0063] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0064] In order to achieve accurate wind speed forecasting, researchers have proposed many strategies, which can be divided into physical methods, statistical methods and neural network methods. Physical methods make predictions by simulating the basic meteorological and physical principles that control atmospheric conditions, mainly based on numerical weather prediction models. Numerical weather prediction models usually require a large amount of data for complex calculations and are more suitable for long-term predictions. As for statistical models, although statistical models based on autoregression are very effective for predicting linear univariate data, they are not suitable for nonlinear data with multivariate characteristics.

[0065] Based on this, the embodiment of the present invention provides a short-term wind speed prediction method based on a gated learning network. For details, please refer to Figure 1 , Figure 1 The figure shows a flow chart of a short-term wind speed prediction method based on a gated learning network in one embodiment of the present invention, which includes the following steps:

[0066] S1. Use the multivariate empirical mode decomposition algorithm to decompose the processed historical wind speed time series data and real-time meteorological environment series data to obtain the initial sample data set.

[0067] It can be understood that in an embodiment of the present invention, the constructed historical wind speed time series data and real-time meteorological environment series data are important indicator factors affecting short-term wind speed, wherein the real-time meteorological environment series data at least includes wind direction data, temperature and humidity data, air pressure data, rainfall data and radiation data.

[0068] Before decomposing these indicator sequence data, preprocessing is required. In some embodiments of the present invention, missing value filling and data cleaning are performed on these indicator sequence data. Specifically:

[0069] The mean interpolation method is used to fill the missing values ​​of the original data set. Common missing forms are:

[0070] {K V,1 ,M V,1 ,M V,2 ,…,M V,m ,K V,2}

[0071] Among them, M V is a missing value, K V is a known value, and m is the number of consecutive missing values.

[0072] Then the i-th missing value can be interpolated using the following formula:

[0073]

[0074] The process of data cleaning can be to clean the data by replacing the extreme outliers with the median of the time series of the above-mentioned various data indicators. The above-mentioned preprocessing process can ensure the integrity of the data set and eliminate extreme outliers, which is conducive to improving the prediction accuracy.

[0075] It can be understood that the preprocessed indicator sequence data set is divided into a test set and a training set. Exemplarily, the data set can be divided in a ratio of 2 / 3 training set and 1 / 3 test set.

[0076] Then, the multivariate empirical mode decomposition (MEMD) method is used to decompose the wind speed time series data and real-time meteorological series data in the divided training set, that is, the series data of wind speed, wind direction, maximum temperature, minimum temperature, relative humidity, mean sea level pressure, total rainfall, and total solar radiation are decomposed into their own intrinsic mode functions (IMFs) and residual components. These IMFs and residual components reflect the different frequency components and trends of the data.

[0077] The following is a specific calculation process to describe the above decomposition process in detail:

[0078] First, use a vector to represent the preprocessed historical wind speed time series data and real-time meteorological environment series data:

[0079]

[0080] Among them, WS is wind speed, WD is wind direction, T max is the maximum temperature, T min is the minimum temperature, RH is the relative humidity, Pmsl is the mean sea level pressure, Rain is the total rainfall, and Radn is the total solar radiation. This data set is an eight-element sequence, and T is the sequence length.

[0081] The process of multivariate empirical mode decomposition is as follows:

[0082] 1) Based on the Hammersley sequence sampling method (the number of sampling points N is 500), the direction vector of the n-dimensional space is obtained on the (n-1)-dimensional sphere;

[0083] 2) Get the original signal X(t) in the direction vector Mapping in all directions It means that the direction angle on the n-1 dimensional sphere is A vector set of ;

[0084] 3) Determine the direction vector mapping The instants corresponding to the minimum and maximum points of

[0085] 4) Determine the projection envelope using interpolation and its mean m(t);

[0086] 5) Calculate the intrinsic mode function h(t) (h(t) = X(t) - m(t)); if the intrinsic mode function satisfies the conditions of the multivariate intrinsic mode function, then h(t) is regarded as a multivariate IMF, and X(t) - h(t) is used as a new input variable, and the above steps are repeated to obtain more IMFs; otherwise, h(t) is used as a new input signal and operations (2) to (5) are continued until the stopping criterion is met.

[0087] 6) After multivariate empirical mode decomposition, the original sequence h(t) is decomposed into a series of IMF components and residual components, which are expressed by the following formula:

[0088]

[0089] Where s(t) is the residual component.

[0090] It can be understood that each decomposed IMF component consists of an eight-element subsequence of wind speed, wind direction, maximum temperature, minimum temperature, relative humidity, mean sea level pressure, total rainfall, and total solar radiation.

[0091] The embodiment of the present invention can use a multivariate empirical mode decomposition algorithm to concurrently decompose multi-channel data into subsequences having uniform intrinsic mode functions and residual components, thereby ensuring the frequency division consistency of the subsequences.

[0092] S2. Perform multimodal feature extraction processing on the initial sample data set to obtain a target sample data set.

[0093] Since the decomposition of multivariate data generates many predictor components, some of which are redundant or noisy, reducing the correlation with the target variable, the embodiment of the present invention provides partial autocorrelation function and cross-correlation function to extract features from the data. The partial autocorrelation function is used to extract features from the decomposed historical wind speed time series data to obtain a first feature set; the cross-correlation function is used to extract features from the decomposed real-time meteorological environment series data to obtain a second feature set. By eliminating redundant components and noise, the prediction accuracy can be improved and the computational complexity can be reduced.

[0094] Specifically, the partial autocorrelation function is used to calculate the significant lag of the IMF and residual components obtained by decomposing the historical wind speed time series data, and the cross-correlation function is used to calculate the significant lag of the IMF and residual components obtained by decomposing the real-time meteorological environment series data. The results of feature extraction are shown in Table 1:

[0095] Table 1 Feature selection results

[0096]

[0097] Among them, * represents the significant lag calculated by PACF; NS represents not significant; the number represents the order of significant lag.

[0098] Since the importance of lags greater than 5th order is relatively small, only lags within 5th order are considered. Lags exceeding the 95% confidence interval of PACF and CCF are considered significant lags. In this embodiment, insignificant IMFs are removed, and the remaining IMFs and residual components are re-superimposed to obtain the time series of wind speed and other meteorological input variables after feature extraction. As can be seen from Table 1, all IMFs and residual components of total solar radiation are insignificant, and the input variable of total solar radiation is eliminated. It is worth noting that in this embodiment, since the IMF components and residual components generated by decomposition are multiple components obtained through multiple decompositions, after feature extraction of these components, the components need to be re-summed, that is, the first feature set and the second feature set are represented as the data set to be normalized after the components are re-summed.

[0099] Next, in order to improve the generalization ability and prediction performance of the model, this embodiment normalizes the first feature set and the second feature set, and at the same time, the previously divided test set also needs to be normalized to obtain the target sample data set. The normalization process is expressed as follows:

[0100]

[0101] Among them, X n is the normalized value of the variable; X act is the actual value of the variable; X max is the maximum value of the variable; X min is the minimum value of the variable.

[0102] S3-S4, constructing a wind speed prediction neural network model with the target sample data set and the determined model correction parameters, wherein the wind speed prediction neural network model is designed to be constructed by a gating mechanism and a feature learning adaptive algorithm.

[0103] In this embodiment, the wind speed prediction neural network model is constructed by inputting the training set in the target sample data set into the neural network model for training, and the short-term wind speed is predicted according to the wind speed prediction neural network model to obtain the wind speed prediction value.

[0104] The wind speed prediction neural network model is constructed by a gating mechanism and a feature learning adaptive algorithm. Specifically, Figure 2 As shown, Figure 2 FIG. 1 is a schematic diagram of a gated learning network structure in one embodiment of the present invention. In the embodiments of the present invention, a gated feature learning adaptive network is used to represent the wind speed prediction neural network model.

[0105] The following is a detailed description of the process of building the above wind speed model and wind speed prediction:

[0106] In some implementations of the present invention, the training set in the target sample data set is set to D = (χ, γ)

[0107] Among them, γ is the wind speed time series after feature extraction and normalization, and χ is the time series of other input variables except wind speed after feature extraction and normalization.

[0108] Next, the corresponding learning mask M is constructed based on the gated feature learning adaptive network n , that is, the learnable mask of the gated feature learning unit is calculated using the following formula:

[0109] M n =t-softmax(F n ,t)

[0110] And understandably,

[0111] t-softmax(F n ,t)=softmax(F n ,w t )

[0112] w t =ReL U(F n,i +t-max(F n ))

[0113]

[0114] Among them, F n is the learnable parameter vector, F n,i is the i-th item in the vector, w i is the corresponding weight; t is a non-learnable parameter, t>0; Re LU is a linear rectification function.

[0115] The meteorological sequence-related features of the input model are calculated based on the obtained learning mask, which is expressed as follows:

[0116] χ n =M n ⊙x

[0117] Among them, ⊙ is the element-by-element multiplication operation.

[0118] And calculate the hidden features H of the data n , which is the previous feature representation H n-1 and the current candidate feature representation The linear interpolation between is expressed as follows:

[0119]

[0120] Among them, z n To update the gate, it decides how much information to use to update its internal feature representation.

[0121] Then the update gate is defined as follows:

[0122]

[0123] Among them, [H n-1 ;x n ] is H n-1 and χ n The series operation between them; σ is the sigmoid activation function; are learnable parameters.

[0124] The candidate features It is expressed by the following formula:

[0125]

[0126] Among them, r n To reset the gate, decide how much information to forget from the previous feature representation; are learnable parameters.

[0127] The reset gate r n is defined as follows:

[0128]

[0129] In this embodiment, the gated feature learning adaptation network consists of a stack of N gated feature learning units arranged in sequence, and for each stage N, a feature subset is selected and feature representation is learned.

[0130] The output result of the gated feature learning adaptive network, i.e., the wind speed prediction neural network model, is expressed as follows:

[0131] γ ′ =f K (H N ; W1,b1,…,W K ,b K )

[0132] Among them, H N is the feature representation of the last gated feature learning unit; f K is a K-layer multilayer perceptron; W1,…,W K is the weight of the K-layer multilayer perceptron; b1,…,b K is the bias of the K-layer multilayer perceptron; γ′ is the predicted wind speed value.

[0133] Then, the root mean square error between the wind speed prediction value and the actual value is calculated to obtain the corresponding fitness parameter, and then the fitness parameter is optimized and adjusted to construct the wind speed prediction neural network model; the fitness parameter is expressed by the following formula:

[0134]

[0135] Among them, γ i is the actual value of wind speed; RMSE is the root mean square error.

[0136] Next, under preset conditions, based on the fitness parameters, the hyperparameters of the wind speed prediction neural network model are optimized and adjusted, and according to the model correction parameters determined after the optimization and adjustment, the optimized wind speed prediction neural network model is trained.

[0137] It is understandable that during the model training process, the model parameters need to be optimized and adjusted so that the final model output results are more accurate. The calculated fitness parameter is expressed as the error between the actual value and the predicted value. In order to make this error result smaller or to make the error meet the preset condition threshold, the hyperparameters of the model will be optimized, that is, the optimal parameters of a gated feature learning stage will be determined. Then, based on this parameter, the model is optimized and trained to obtain an optimized wind speed prediction neural network model for wind speed prediction. Specifically:

[0138] In some embodiments of the present invention, Figure 3 As shown, Figure 3 FIG. 1 is a flowchart of the overall process of short-term wind speed prediction in one embodiment of the present invention. Figure 3 It can be seen that after the feature extraction and normalization of the data set, an initial population vector is also generated for iterative update and local search of the vector.

[0139] Specifically, first, according to the target sample data set, a corresponding initial population vector is randomly generated, and the initial population vector includes the hyperparameters to be optimized and adjusted. In this embodiment, the initial population vector can be randomly generated by a random number, then the vector Y g It is expressed as follows:

[0140]

[0141] Among them, N p is the number of individuals in the population, take N p =100; g represents the current generation, g = 1, 2, ..., g max ; g max is the maximum algebra, take g max =100.

[0142] It is worth noting that this vector contains the hyperparameters of the gated feature learning adaptive network model (the number of feature learning stages N and the parameter t).

[0143] Next, the initial population vector is iteratively updated and locally searched according to a weighted average algorithm to obtain the corresponding model correction parameters.

[0144] The specific update iteration process is as follows:

[0145] First, in this embodiment, the weighted average of the vector is calculated by randomly generating the initial population vector. In some embodiments of the present invention, the position of the vector is updated by a mean rule (Mean Rule), and the MeanRule value is expressed by the following formula:

[0146]

[0147] but:

[0148]

[0149] Among them, a1≠a2≠a3≠1, which is from [1, N p ]; are the optimal, suboptimal and worst solution vectors in the g-th generation population respectively; ω1, ω2, ω3, ω4, ω5, ω6 are weighting functions used to calculate the weighted average of the vector; δ is the weighted average factor; ε is a small constant; rand is a random number between (0, 1); r′ is a random number between [0, 0.5].

[0150] The weighting function can be obtained by the above formula. As an example, in some embodiments of the present invention, the weighting functions ω1, ω2, and ω3 are respectively expressed by the following formulas:

[0151]

[0152] in, for The fitness of the solution vector The root mean square error.

[0153] The iteratively updated position update vector is calculated according to the above weighted process, and the position update vector is combined with the initial population vector to generate a new vector. Based on the combined new vector, a corresponding new solution vector is generated around the optimal solution vector.

[0154] Specifically, the new vector generated after combining It is expressed by the following formula:

[0155]

[0156] in, is the new vector obtained by merging the vectors in the g-th generation; μ = 0.05 × rand, where rand is a standard normal distribution random value; is the lth population vector of the gth generation; and and It is represented as the position update vector of the g-th iteration.

[0157] Understandably, the position update vector and They are expressed by the following formulas:

[0158]

[0159]

[0160] Where λ=δ and randn is a standard normal distribution random value.

[0161] Understandably, when all After all new vectors are updated, in order to prevent falling into a local optimum, the embodiment of the present invention creates a random number rand.

[0162] Correspondingly, if rand < 0.5, then around the optimal solution vector Generate a new solution vector, then the new solution vector is expressed as follows:

[0163]

[0164] in,

[0165]

[0166] in, is a random number between (0, 1); v1, v2 are two random numbers.

[0167] like Figure 3 As shown, a local search is performed around the optimal population vector to update the hyperparameters of the optimal population vector and to determine whether the number of iterations g reaches the preset maximum number of iterations.

[0168] When the maximum number of iterations is reached, the target model correction parameters are determined based on the currently updated hyperparameters, that is, the gated feature learning adaptive network is optimized and trained according to the number of feature learning stages N and parameter t contained in each individual in the generated new population to construct a wind speed prediction neural network model. The optimized wind speed prediction neural network model is used to predict short-term wind speed. It is worth noting that after the wind speed prediction neural network model is trained, the previously divided test set is used as the input of the model to output a wind speed prediction value, and the fitness of this wind speed prediction value with the actual value is calculated to judge the prediction performance of the model.

[0169] This embodiment uses adaptive update rules and local random search strategies to adjust model parameters, which can more effectively handle the problem of short-term wind speed prediction optimization and ensure the robustness of complex solution space.

[0170] In summary, the embodiment of the present invention adopts a multivariate empirical mode decomposition algorithm to concurrently decompose multi-channel sequence data into uniform intrinsic mode functions and residual components, and uses partial autocorrelation functions and cross-correlation functions to extract features of the decomposed data, eliminate redundant components and noise, so as to improve prediction accuracy and reduce computational complexity; and constructs a prediction model through a gated feature learning adaptive network to predict short-term wind speed.

[0171] An embodiment of the present invention provides a short-term wind speed prediction system based on a gated learning network. For details, see Figure 4 , Figure 4 The structure block diagram of a short-term wind speed prediction system based on a gated learning network in one embodiment of the present invention is shown, including:

[0172] The data decomposition module M1 is used to decompose the processed historical wind speed time series data and real-time meteorological environment series data using a multivariate empirical mode decomposition algorithm to obtain an initial sample data set;

[0173] A feature extraction module M2 is used to perform multimodal feature extraction processing on the initial sample data set to obtain a target sample data set;

[0174] A prediction model construction module M3, used to construct a wind speed prediction neural network model using the target sample data set and the determined model correction parameters, wherein the wind speed prediction neural network model is designed to be constructed by a gating mechanism and a feature learning adaptive algorithm;

[0175] The wind speed prediction module M4 is used to predict short-term wind speed according to the wind speed prediction neural network model.

[0176] Specifically, in this embodiment, the real-time meteorological environment sequence data at least includes wind direction data, temperature and humidity data, air pressure data, rainfall data and radiation data.

[0177] Exemplarily, the embodiment of the present invention uses the data decomposition module M1 to decompose the historical wind speed time series data and the real-time meteorological environment series data respectively by using the multivariate empirical mode decomposition algorithm to obtain the intrinsic mode functions and residual components corresponding to the two.

[0178] Exemplarily, the embodiment of the present invention uses the feature extraction module M2 to extract features from the decomposed historical wind speed time series data using a partial autocorrelation function to obtain a first feature set;

[0179] Using a cross-correlation function to extract features from the decomposed real-time meteorological environment sequence data to obtain a second feature set;

[0180] The first feature set and the second feature set are normalized to obtain the target sample data set.

[0181] Exemplarily, the embodiment of the present invention uses the prediction model construction module M3 to input the training set in the target sample data set into the neural network model for training, so as to construct the wind speed prediction neural network model;

[0182] Predicting the short-term wind speed according to the wind speed prediction neural network model to obtain a wind speed prediction value, and calculating a root mean square error between the wind speed prediction value and the actual wind speed value to obtain a corresponding fitness parameter;

[0183] The fitness parameter is expressed by the following formula:

[0184]

[0185] Among them, γ i is the actual value of wind speed; i ' is the wind speed prediction value; RMSE is the root mean square error;

[0186] Under preset conditions, based on the fitness parameters, optimizing and adjusting the hyperparameters of the wind speed prediction neural network model;

[0187] The optimized wind speed prediction neural network model is trained based on the model correction parameters determined after optimization and adjustment.

[0188] Furthermore, the prediction model building module M4 is also used to randomly generate a corresponding initial population vector according to the target sample data set; the initial population vector includes the hyperparameter to be optimized and adjusted;

[0189] The initial population vector is iteratively updated and locally searched according to a weighted average algorithm to determine the corresponding model correction parameters; the neural network model is retrained according to the model correction parameters to obtain the wind speed prediction neural network model.

[0190] The technical features and technical effects of the short-term wind speed prediction system based on the gated learning network proposed in the embodiment of the present invention are the same as the technical features and technical effects of the short-term wind speed prediction method based on the gated learning network proposed in the embodiment of the present invention, and will not be repeated here.

[0191] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A short-term wind speed prediction method based on gated learning network, characterized in that: include: The processed historical wind speed time series data and real-time meteorological environment series data are decomposed using the multivariate empirical mode decomposition algorithm to obtain the initial sample data set; Performing multimodal feature extraction processing on the initial sample data set to obtain a target sample data set; A wind speed prediction neural network model is constructed using the target sample data set and the determined model correction parameters, wherein the wind speed prediction neural network model is designed to be constructed by a gating mechanism and a feature learning adaptive algorithm; The short-term wind speed is predicted based on the wind speed prediction neural network model.

2. The short-term wind speed prediction method based on gated learning network according to claim 1, characterized in that: The real-time meteorological environment sequence data includes at least wind direction data, temperature and humidity data, air pressure data, rainfall data and radiation data; The method of using the multivariate empirical mode decomposition algorithm to decompose the processed historical wind speed time series data and real-time meteorological environment series data includes: The historical wind speed time series data and the real-time meteorological environment series data are decomposed respectively by the multivariate empirical mode decomposition algorithm to obtain the corresponding intrinsic mode functions and residual components of the two.

3. The short-term wind speed prediction method based on gated learning network according to claim 1, characterized in that: The performing multimodal feature extraction processing on the initial sample data set to obtain a target sample data set includes: Using a partial autocorrelation function to extract features from the decomposed historical wind speed time series data to obtain a first feature set; Using a cross-correlation function to extract features from the decomposed real-time meteorological environment sequence data to obtain a second feature set; The first feature set and the second feature set are normalized to obtain the target sample data set.

4. The short-term wind speed prediction method based on gated learning network according to claim 1, characterized in that: The construction process of the wind speed prediction neural network model includes: Inputting the training set in the target sample data set into the neural network model for training to construct the wind speed prediction neural network model; Predicting the short-term wind speed according to the wind speed prediction neural network model to obtain a wind speed prediction value, and calculating a root mean square error between the wind speed prediction value and the actual wind speed value to obtain a corresponding fitness parameter; The fitness parameter is expressed by the following formula: Among them, γ i is the actual value of wind speed; i ′ is the wind speed prediction value; RMSE is the root mean square error; Under preset conditions, based on the fitness parameters, optimizing and adjusting the hyperparameters of the wind speed prediction neural network model; The optimized wind speed prediction neural network model is trained based on the model correction parameters determined after optimization and adjustment.

5. The short-term wind speed prediction method based on gated learning network according to claim 4, characterized in that: The process of constructing the wind speed prediction neural network model also includes: According to the target sample data set, a corresponding initial population vector is randomly generated; the initial population vector includes the hyperparameter to be optimized and adjusted; Iteratively updating and locally searching the initial population vector according to a weighted average algorithm to determine the corresponding model correction parameters; The wind speed prediction neural network model is optimized according to the model correction parameters.

6. A short-term wind speed prediction system based on a gated learning network, characterized in that: include: The data decomposition module is used to decompose the processed historical wind speed time series data and real-time meteorological environment series data using the multivariate empirical mode decomposition algorithm to obtain an initial sample data set; A feature extraction module, used to perform multimodal feature extraction processing on the initial sample data set to obtain a target sample data set; A prediction model construction module, used to construct a wind speed prediction neural network model using the target sample data set and the determined model correction parameters, wherein the wind speed prediction neural network model is designed to be constructed by a gating mechanism and a feature learning adaptive algorithm; The wind speed prediction module is used to predict short-term wind speed according to the wind speed prediction neural network model.

7. The short-term wind speed prediction system based on gated learning network according to claim 6, characterized in that: The real-time meteorological environment sequence data includes at least wind direction data, temperature and humidity data, air pressure data, rainfall data and radiation data; The data decomposition module is specifically used for: The historical wind speed time series data and the real-time meteorological environment series data are decomposed respectively by the multivariate empirical mode decomposition algorithm to obtain the corresponding intrinsic mode functions and residual components of the two.

8. The short-term wind speed prediction system based on gated learning network according to claim 6, characterized in that: The feature extraction module is specifically used for: Using a partial autocorrelation function to extract features from the decomposed historical wind speed time series data to obtain a first feature set; Using a cross-correlation function to extract features from the decomposed real-time meteorological environment sequence data to obtain a second feature set; The first feature set and the second feature set are normalized to obtain the target sample data set.

9. The short-term wind speed prediction system based on gated learning network according to claim 6, characterized in that: The prediction model building module is specifically used for: Inputting the training set in the target sample data set into the neural network model for training to construct the wind speed prediction neural network model; Predicting the short-term wind speed according to the wind speed prediction neural network model to obtain a wind speed prediction value, and calculating a root mean square error between the wind speed prediction value and the actual wind speed value to obtain a corresponding fitness parameter; The fitness parameter is expressed by the following formula: Among them, γ i is the actual value of wind speed; i ′ is the wind speed prediction value; RMSE is the root mean square error; Under preset conditions, based on the fitness parameters, optimizing and adjusting the hyperparameters of the wind speed prediction neural network model; The optimized wind speed prediction neural network model is trained based on the model correction parameters determined after optimization and adjustment.

10. The short-term wind speed prediction system based on gated learning network according to claim 9, characterized in that: The prediction model building module is also used to: According to the target sample data set, a corresponding initial population vector is randomly generated; the initial population vector includes the hyperparameter to be optimized and adjusted; Iteratively updating and locally searching the initial population vector according to a weighted average algorithm to determine the corresponding model correction parameters; The wind speed prediction neural network model is optimized according to the model correction parameters.

Images