Wind speed interval prediction system based on nested shared long short-term memory network
The NSWLSTM-GPR hybrid method addresses long training times and accuracy issues in wind speed prediction by nesting LSTM models with shared gates, achieving efficient and reliable interval predictions.
Patent Information
- Application Number
- CN202411043665.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-12-13
AI Technical Summary
Existing wind speed prediction methods, particularly those using nested LSTM models, face challenges with long training times due to numerous variables and inadequate handling of long-term correlations, leading to suboptimal prediction accuracy.
A wind speed interval prediction system utilizing a nested shared long short-term memory network (NSWLSTM) that combines two SWLSTM models with shared gates and information states, coupled with Gaussian Process Regression (GPR) for high-precision point and interval predictions, reducing training time and improving accuracy by nesting LSTM structures.
The system significantly reduces training time, enhances prediction accuracy, and provides comprehensive evaluation indicators, enabling robust interval predictions with improved deep learning capabilities and reliability.
Smart Images
Figure CN119129647A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind speed interval prediction, and in particular to a wind speed interval prediction system based on a nested shared long short-term memory network. Background Art
[0002] With the development of artificial intelligence, more and more deep learning methods are applied to wind speed prediction. Recurrent Neural Network (RNN) has been proven to be suitable for the prediction of time series data, but when the time series is too long, it is prone to the problems of gradient vanishing and gradient exploding. Therefore, LSTM came into being. LSTM has achieved good results in the field of wind speed prediction. LSTM needs to optimize more variables, which leads to longer training time. The results show that coupling the input gate and the forget gate or deleting the peephole connection will not reduce the prediction accuracy of LSTM, so the SWLSTM method is proposed.
[0003] At present, the common LSTM methods mentioned above are based on simple stacking LSTM, while the nested LSTM (NLSTM) model increases the depth of LSTM by nesting rather than stacking. Although NLSTM can learn more long-term correlations and improve prediction accuracy, it still cannot solve the problem of having many variables to be optimized.
[0004] Therefore, in view of this, the existing structure and deficiencies are studied and improved, and a wind speed range prediction system based on a nested shared long short-term memory network is proposed. Summary of the invention
[0005] In view of the deficiencies in the prior art, the present invention provides a wind speed interval prediction system based on a nested shared long short-term memory network, which solves the problems raised in the above background technology.
[0006] To achieve the above objectives, the present invention is implemented by the following technical scheme: a wind speed interval prediction system based on a nested shared long short-term memory network, the wind speed interval prediction system comprising the following operating steps:
[0007] S1 data preprocessing:
[0008] The wind speeds at the ten moments before the current moment are taken as ten wind speed characteristic variables, and the wind speed characteristic variables at the ten moments before the current moment are found. These characteristic variables have two possibilities: selection and deletion. The genetic algorithm uses the root mean square error to screen various characteristic combinations to obtain the best wind speed characteristic combination.
[0009] S2, NSWLSTM wind speed interval prediction model:
[0010] Two SWLSTM models are nested to form a NSWLSTM model. Each SWLSTM couples input gate, output gate and forget gate based on LSTM. The two SWLSTM models are nested and divided into an inner SWLSTM and an outer SWLSTM. The intermediate variables, shared gates and information states of the inner SWLSTM are as follows:
[0011]
[0012] The cell state of the inner SWLSTM is as follows:
[0013]
[0014] The hidden layer output of the inner SWLSTM is as follows:
[0015]
[0016] The intermediate variables, shared gates, and information states of the outer SWLSTM are as follows:
[0017] net t =w h ·h t-1 +w x ·x t +b
[0018] s t =σ(net t )=σ(w h ·h t-1 +w x ·x t +b)
[0019] a t =tanh(net t )=tanh(w h ·h t-1 +w x ·x t + b);
[0020] The input variables of the inner SWLSTM are as follows:
[0021]
[0022] The hidden layer of the inner SWLSTM at the previous moment is as follows:
[0023]
[0024] The outer SWLSTM output is as follows:
[0025]
[0026] The NSWLSTM model is used to speed up the training of prediction models in the case of massive data.
[0027] S3, GPR interval prediction:
[0028] GPR is used for interval prediction, where the regression equation with noise is as follows:
[0029] Y = f(X) + ξ;
[0030] The input variable is X, the function value is f(X), the observed value disturbed by additive noise is Y, and the noise satisfies The prior distribution of the observed value Y and the joint probability distribution of the observed value Y and the predicted value y are obtained as follows:
[0031]
[0032] Where K(x,x)=(κ ij ) is the positive definite covariance matrix, x i With x j The correlation between ij To express, K(x * ,x)=K(x,x * ) T is the test set x * The covariance matrix with the training set x, K(x * ,x * ) is the covariance matrix of the test set, I n is the n-dimensional identity matrix, and the square exponential kernel formula is as follows:
[0033]
[0034] The posterior distribution of the predicted value y is as follows:
[0035]
[0036] Therefore, the point prediction of GPR is The interval prediction corresponding to the 95% confidence level is
[0037] S4, NSWLSTM-GPR wind speed range prediction:
[0038] Combining NSWLSTM and GPR can obtain high-precision point prediction and high-reliability interval prediction. The process of the hybrid method is to first train the NSWLSTM network, and then use the training set and test set as input to obtain high-precision point prediction results. The final result of GPR is the interval prediction;
[0039] S5. Interval prediction evaluation index:
[0040] The interval prediction evaluation indicators include coverage probability and average width percentage, where the formula for coverage probability is defined as follows:
[0041]
[0042] The average width percentage formula is defined as follows:
[0043]
[0044] MC=MWP / CP.
[0045] Furthermore, in step S1, for each feature combination, GPR and SVR are used to calculate the RMSE between the predicted value and the actual value as the fitness, GPR is Gaussian process regression, and SVR is support vector regression.
[0046] Furthermore, in step S2, LSTM is a long short-term memory network, and the superscript symbol "~" represents the parameters of the inner SWLSTM; and w h Represent the hidden layer weights of the inner SWLSTM and outer SWLSTM respectively; and w x Represent the input layer weights of the inner SWLSTM and outer SWLSTM respectively; and b represent the biases of the inner and outer SWLSTM layers, respectively.
[0047] Furthermore, in step S2, and net t They represent the intermediate variables of the inner SWLSTM and the outer SWLSTM respectively; and t They represent the shared gates of the inner SWLSTM and the outer SWLSTM respectively; and a t Respectively represent the information status of the inner SWLSTM and the outer SWLSTM at the current moment; and x t Respectively represent the input of the inner SWLSTM and the outer SWLSTM; and They represent the hidden layer outputs of the inner SWLSTM at the previous moment and the current moment respectively.
[0048] Furthermore, in step S2, and h t Respectively represent the hidden layer output of the outer SWLSTM at the previous moment and the current moment; and Respectively represent the unit states of the inner SWLSTM at the previous moment and the current moment; Ct-1 and C t Respectively represent the unit states of the outer SWLSTM at the previous moment and the current moment;
[0049] The symbol “·” indicates matrix multiplication;
[0050] The symbol “*” indicates multiplication between matrix elements;
[0051] σ(x) and tanh(x) are activation functions.
[0052] Furthermore, in step S3, in the square exponential kernel formula, p1 and p2 are adjustment parameters.
[0053] Furthermore, in the formula of coverage probability in step S5, CP is defined as the probability that the observed value falls within the prediction interval of the α confidence level, c α This is the number of observations that fall within the prediction interval.
[0054] Furthermore, in the step S5, in the average width percentage formula, MWP is defined as the average percentage of the interval width to the observed value, and MC is the metric.
[0055] Furthermore, the wind speed interval prediction system based on the nested shared long short-term memory network is applied to the technical field of wind speed interval prediction.
[0056] The present invention provides a wind speed interval prediction system based on a nested shared long short-term memory network, which has the following beneficial effects:
[0057] Compared with the traditional LSTM, the wind speed interval prediction system based on the nested shared long short-term memory network reduces the training time of the model by sharing the gate structure of LSTM. By nesting LSTM, the prediction accuracy of the model is improved, and the evaluation indicators are more systematic and comprehensive, covering the suitability of interval prediction. By nesting SWLSTM, it can process historical data with a longer time scale and improve the accuracy of the model to a certain extent. The introduction of NSWLSTM-GPR enables the model to have strong deep learning capabilities and interval prediction characteristics. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] Figure 1 It is a schematic diagram of the characteristic combination flow of wind speed of the present invention;
[0059] Figure 2 It is a schematic diagram of the NSWLSTM structure of the present invention;
[0060] Figure 3 It is a schematic diagram of the NSWLSTM-GPR process of the present invention;
[0061] Figure 4It is a schematic diagram of the NSWLSTM-GPR graphical framework of the present invention;
[0062] Figure 5 Schematic diagram of statistical information of four data sets according to an embodiment of the present invention;
[0063] Figure 6 Schematic diagram of optimal feature combination of four data sets according to an embodiment of the present invention;
[0064] Figure 7 A schematic diagram of detailed parameter information of a model according to an embodiment of the present invention;
[0065] Figure 8 It is a schematic diagram of the 95% confidence level interval prediction results of the three methods in the embodiments of the present invention in data set 1;
[0066] Fig. 9 It is a schematic diagram of the 95% confidence level interval prediction results of the three methods in the embodiments of the present invention in data set 2;
[0067] Fig.10 It is a schematic diagram of the 95% confidence level interval prediction results of the three methods in the embodiments of the present invention in data set 3;
[0068] Fig.11 It is a schematic diagram of the 95% confidence level interval prediction results of the three methods in the embodiments of the present invention in data set 4;
[0069] Fig.12 It is a schematic diagram of interval prediction evaluation index at 95% confidence level in an embodiment of the present invention. DETAILED DESCRIPTION
[0070] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0071] like Figure 1-Figure 4 As shown, the present invention provides a technical solution: a wind speed interval prediction system based on a nested shared long short-term memory network, and the wind speed interval prediction system includes the following operating steps:
[0072] S1 data preprocessing:
[0073] The wind speeds at the ten moments before the current moment are taken as ten wind speed characteristic variables. The wind speed characteristic variables at the ten moments before the current moment are found. These characteristic variables have two possibilities: selection and deletion. The genetic algorithm uses the root mean square error to screen various feature combinations to obtain the best wind speed feature combination. For each feature combination, GPR and SVR are used to calculate the RMSE between the predicted value and the actual value as the fitness. GPR is Gaussian process regression, SVR is support vector regression, and data preprocessing is as follows: Figure 1 As shown;
[0074] S2, NSWLSTM wind speed interval prediction model:
[0075] Two SWLSTM models are nested to form a NSWLSTM model. Each SWLSTM couples input gate, output gate and forget gate based on LSTM. The two SWLSTM models are nested and divided into an inner SWLSTM and an outer SWLSTM. The intermediate variables, shared gates and information states of the inner SWLSTM are as follows:
[0076]
[0077] The cell state of the inner SWLSTM is as follows:
[0078]
[0079] The hidden layer output of the inner SWLSTM is as follows:
[0080]
[0081] The intermediate variables, shared gates, and information states of the outer SWLSTM are as follows:
[0082] net t =w h ·h t-1 +w x ·x t +b
[0083] s t =σ(net t )=σ(w h ·h t-1 +w x ·x t +b)
[0084] a t =tanh(net t )=tanh(w h ·h t-1 +w x ·x t + b);
[0085] The input variables of the inner SWLSTM are as follows:
[0086]
[0087] The hidden layer of the inner SWLSTM at the previous moment is as follows:
[0088]
[0089] The outer SWLSTM output is as follows:
[0090]
[0091] The NSWLSTM model is used to speed up the training of prediction models in the case of massive data.
[0092] Among them, LSTM is a long short-term memory network, and the superscript symbol “~” represents the parameters of the inner SWLSTM; and w h Represent the hidden layer weights of the inner SWLSTM and outer SWLSTM respectively; and w x Represent the input layer weights of the inner SWLSTM and outer SWLSTM respectively; and b represent the bias of the inner SWLSTM and outer SWLSTM, respectively; and net t They represent the intermediate variables of the inner SWLSTM and the outer SWLSTM respectively; and t They represent the shared gates of the inner SWLSTM and the outer SWLSTM respectively; and a t Respectively represent the information status of the inner SWLSTM and the outer SWLSTM at the current moment; and x t Respectively represent the input of the inner SWLSTM and the outer SWLSTM; and Respectively represent the hidden layer output of the inner SWLSTM at the previous moment and the current moment; and h t Respectively represent the hidden layer output of the outer SWLSTM at the previous moment and the current moment; and Respectively represent the unit states of the inner SWLSTM at the previous moment and the current moment; C t-1 and C t Respectively represent the unit states of the outer SWLSTM at the previous moment and the current moment;
[0093] The symbol “·” indicates matrix multiplication;
[0094] The symbol “*” indicates multiplication between matrix elements;
[0095] σ(x) and tanh(x) are activation functions;
[0096] The structure of NSWLSTM is as follows Figure 2 As shown, n i Indicates the number of nodes in the input layer, nh Indicates the number of nodes in the hidden layer, n o Indicates the number of nodes in the output layer. In time series regression, n o is equal to 1, so the matrix w x , w h The dimensions of and b are [n i ×n h ],[n h ×n h ] and [n h ×n o ], the number of variables that NSWLSTM needs to optimize is n h (n i +n h +n o ), however, the input gate, forget gate, output gate, and information state of LSTM need to calculate similar weight matrices, so the number of variables that LSTM needs to optimize is 4n h (n i +n h +n o ), therefore, NSWLSTM shares the variables that need to be optimized, greatly reducing the number of variables that need to be optimized, and can greatly speed up the training of the prediction model in the case of massive data;
[0097] S3, GPR interval prediction:
[0098] GPR is used for interval prediction, where the regression equation with noise is as follows:
[0099] Y = f(X) + ξ;
[0100] The input variable is X, the function value is f(X), the observed value disturbed by additive noise is Y, and the noise satisfies The prior distribution of the observed value Y and the joint probability distribution of the observed value Y and the predicted value y are obtained as follows:
[0101]
[0102] Where K(x,x)=(κ ij ) is the positive definite covariance matrix, x i With x j The correlation between ij To express, K(x * ,x)=K(x,x * ) T is the test set x * The covariance matrix with the training set x, K(x * ,x * ) is the covariance matrix of the test set, I nis the n-dimensional identity matrix, and the square exponential kernel formula is as follows:
[0103]
[0104] Among them, in the square exponential kernel formula, p1 and p2 are adjustment parameters, which can be adjusted according to actual conditions;
[0105] The posterior distribution of the predicted value y is as follows:
[0106]
[0107] Therefore, the point prediction of GPR is The interval prediction corresponding to the 95% confidence level is
[0108] S4, NSWLSTM-GPR wind speed range prediction:
[0109] Combining NSWLSTM and GPR can obtain high-precision point prediction and high-reliability interval prediction. The process of the hybrid method is to first train the NSWLSTM network, and then use the training set and test set as input to obtain high-precision point prediction results. The final result of GPR is the interval prediction;
[0110] Details of NSWLSTM-GPR Figure 3 and Figure 4 As shown, x t For network input, y 1t is the point prediction result of NSWLSTM, y 2t is the result obtained by using NSWLSTM-GPR, which includes the results of point prediction and interval prediction. t is the observed value, with superscript t a With t e are the identifiers of the training set and the test set, respectively. Ta and Te are the number of training sets and the test set, respectively. First, the NSWLSTM network needs to be and To train, then the trained NSWLSTM network needs and The input calculation point prediction result and and Is input into GPR to produce interval prediction results at last, and Used to evaluate the performance of the model;
[0111] S5. Interval prediction evaluation index:
[0112] The interval prediction evaluation indicators include coverage probability and average width percentage, where the formula for coverage probability is defined as follows:
[0113]
[0114] The average width percentage formula is defined as follows:
[0115]
[0116] MC = MWP / CP;
[0117] In the formula of coverage probability, CP is defined as the probability that the observed value falls within the prediction interval at the α confidence level, c α That is, the number of observations that fall into the prediction interval; in the average width percentage formula, MWP is defined as the average percentage of interval width and observations, and MC is the metric.
[0118] The wind speed interval prediction system based on the nested shared long short-term memory network is applied to the technical field of wind speed interval prediction.
[0119] Example:
[0120] like Figure 5-Figure 12 As shown in Figure 2, the time step of wind speed data is 15 minutes, and four data sets from different wind fields are used to test the performance of the model. The statistical information is shown in Figure 2. Figure 5 As shown, T, Ta and Te represent the total number of samples, the number of samples in the training set and the number of samples in the test set, respectively. Min and std represent the mean and variance of the total number of samples, respectively. The four data sets use different data lengths of days, namely 5 days, 7 days, 7 days and 10 days, and the sizes of the data sets are 481, 673, 673 and 961, respectively. Four data sets with different time and data lengths were tested. For the four wind speed data sets, the first 80% of the data is used as the training set, and the remaining 20% of the data is used as the test set;
[0121] The wind speed at each moment is regarded as the characteristic variable of wind speed, and the historical wind speed characteristic variables of the previous ten moments are used to combine the wind speed characteristics. According to the RMSE results calculated by GPR and SVR, the best feature combinations of the four data sets are as follows: Figure 6 As shown in the figure, √ indicates the selected wind speed characteristic variable, and × indicates the deleted wind speed characteristic variable;
[0122] GPR and SWLSTM-GPR are selected as the interval prediction comparison models of NSWLSTM-GPR. In these models, the same parameter values are equal to ensure the fairness of the comparison. The parameter details of the NSWLSTM-GPR model are as follows: Figure 7As shown in the figure, since the results of each run of GPR are the same, it is sufficient to run GPR once. When optimizing the weights, NSWLSTM-GPR and SWLSTM need to be iterated, and there are random numbers in the model. The results of each run are different, so the algorithm calculates the average value of 10 times as the final result. The purpose of wind speed interval prediction is to obtain the coverage probability of the prediction interval and the average width of the prediction interval to verify whether the interval prediction is appropriate. The simulation results are shown in Figures 8 to 11 As shown;
[0123] from Figures 8 to 11 It can be seen that most of the observation points fall within the prediction interval, and the prediction interval is as narrow as possible, which ultimately illustrates the effectiveness of the NSWLSTM-GPR model. Fig.12 The results of the 95% confidence level interval prediction indicators of four data sets and three models are shown. In terms of interval prediction at the 95% confidence level, the CP values of NSWLSTM-GPR in data set 1 are close to 95%, but the CP values of NSWLSTM-GPR and GPR models have not reached the corresponding targets, that is, they have not reached the preset confidence level conditions and do not meet the reliability conditions. Therefore, the comprehensive performance of NSWLSTM-GPR interval prediction in data set 1 is higher than that of NSWLSTM-GPR and GPR models.
[0124] In terms of interval prediction at the 95% confidence level, the MC value of the NSWLSTM-GPR model in dataset 2 is 0.391, which is lower than that of the NSWLSTM-GPR and GPR models. In dataset 2, the NSWLSTM-GPR model has the best prediction interval. The MC values of the NSWLSTM-GPR model in dataset 3 are 0.336, which are lower than those of the NSWLSTM-GPR and GPR models. In dataset 3, the NSWLSTM-GPR model has the best prediction interval. The CP value of the NSWLSTM-GPR model in dataset 4 is 0.959, which is closest to the 95% confidence level prediction interval, and the MC value is higher than that of the NSWLSTM-GPR model. In dataset 4, the NSWLSTM-GPR model has the best prediction interval.
[0125] Through the above analysis, it can be seen from the prediction interval of 95% confidence level that the interval prediction of NSWLSTM-GPR model is the most feasible. All observations are within the prediction interval formed by NSWLSTM-GPR. Compared with other models, it contains more observations and obtains the narrowest interval width.
[0126] Therefore, on the basis of the traditional LSTM method, the problem of long training time is solved, the gating structure of LSTM is simplified, and the training time of the model is shortened. By nesting SWLSTM, historical data with a longer time scale can be processed and the accuracy of the model can be improved to a certain extent. The introduction of NSWLSTM-GPR enables the model to have strong deep learning capabilities and interval prediction characteristics.
[0127] The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.
Claims
1. A wind speed interval prediction system based on a nested shared long short-term memory network, characterized by: The wind speed interval prediction system comprises the following operating steps: S1 data preprocessing: The wind speeds at the ten moments before the current moment are taken as ten wind speed characteristic variables, and the wind speed characteristic variables at the ten moments before the current moment are found. These characteristic variables have two possibilities: selection and deletion. The genetic algorithm uses the root mean square error to screen various characteristic combinations to obtain the best wind speed characteristic combination. S2, NSWLSTM wind speed interval prediction model: Two SWLSTM models are nested to form a NSWLSTM model. Each SWLSTM couples input gate, output gate and forget gate based on LSTM. The two SWLSTM models are nested and divided into an inner SWLSTM and an outer SWLSTM. The intermediate variables, shared gates and information states of the inner SWLSTM are as follows: The cell state of the inner SWLSTM is as follows: The hidden layer output of the inner SWLSTM is as follows: The intermediate variables, shared gates, and information states of the outer SWLSTM are as follows: net t =w h ·h t-1 +w x ·x t +b s t =σ(net t )=σ(w h ·h t-1 +w x ·x t +b) a t =tanh(net t )=tanh(w h ·h t-1 +w x ·x t +b); The input variables of the inner SWLSTM are as follows: The hidden layer of the inner SWLSTM at the previous moment is as follows: The outer SWLSTM output is as follows: The NSWLSTM model is used to speed up the training of prediction models in the case of massive data. S3, GPR interval prediction: GPR is used for interval prediction, where the regression equation with noise is as follows: Y = f(X) + ξ; The input variable is X, the function value is f(X), the observed value disturbed by additive noise is Y, and the noise satisfies The prior distribution of the observed value Y and the joint probability distribution of the observed value Y and the predicted value y are obtained as follows: Where K(x,x)=(κ ij ) is the positive definite covariance matrix, x i With x j The correlation between ij To express, K(x * ,x)=K(x,x * ) T is the test set x * The covariance matrix with the training set x, K(x * ,x * ) is the covariance matrix of the test set, I n is the n-dimensional identity matrix, and the square exponential kernel formula is as follows: The posterior distribution of the predicted value y is as follows: Therefore, the point prediction of GPR is The interval prediction corresponding to the 95% confidence level is S4, NSWLSTM-GPR wind speed range prediction: Combining NSWLSTM and GPR can obtain high-precision point prediction and high-reliability interval prediction. The process of the hybrid method is to first train the NSWLSTM network, and then use the training set and test set as input to obtain high-precision point prediction results. The final result of GPR is the interval prediction; S5. Interval prediction evaluation index: The interval prediction evaluation indicators include coverage probability and average width percentage, where the formula for coverage probability is defined as follows: The average width percentage formula is defined as follows: MC=MWP / CP.
2. The wind speed interval prediction system based on nested shared long short-term memory network according to claim 1 is characterized in that: In step S1, for each feature combination, GPR and SVR are used to calculate the RMSE between the predicted value and the actual value as the fitness, GPR is Gaussian process regression, and SVR is support vector regression.
3. The wind speed interval prediction system based on nested shared long short-term memory network according to claim 1, characterized in that: In step S2, LSTM is a long short-term memory network, and the superscript symbol "~" represents the parameters of the inner SWLSTM; and w h Represent the hidden layer weights of the inner SWLSTM and outer SWLSTM respectively; and w x Represent the input layer weights of the inner SWLSTM and outer SWLSTM respectively; and b represent the biases of the inner and outer SWLSTM layers, respectively.
4. The wind speed interval prediction system based on nested shared long short-term memory network according to claim 1, characterized in that: In the step S2, and net t They represent the intermediate variables of the inner SWLSTM and the outer SWLSTM respectively; and t They represent the shared gates of the inner SWLSTM and the outer SWLSTM respectively; and a t Respectively represent the information status of the inner SWLSTM and the outer SWLSTM at the current moment; and x t Respectively represent the input of the inner SWLSTM and the outer SWLSTM; and They represent the hidden layer outputs of the inner SWLSTM at the previous moment and the current moment respectively.
5. The wind speed interval prediction system based on nested shared long short-term memory network according to claim 1, characterized in that: In the step S2, and h t Respectively represent the hidden layer output of the outer SWLSTM at the previous moment and the current moment; and Respectively represent the unit states of the inner SWLSTM at the previous moment and the current moment; C t-1 and C t Respectively represent the unit states of the outer SWLSTM at the previous moment and the current moment; The symbol "·" represents matrix multiplication; The symbol "*" indicates multiplication between matrix elements; σ(x) and tanh(x) are activation functions.
6. The wind speed interval prediction system based on nested shared long short-term memory network according to claim 1, characterized in that: In step S3, in the square exponential kernel formula, p1 and p2 are adjustment parameters.
7. The wind speed interval prediction system based on nested shared long short-term memory network according to claim 1 is characterized in that: In step S5, in the formula of coverage probability, CP is defined as the probability that the observed value falls within the prediction interval of the α confidence level, c α This is the number of observations that fall within the prediction interval.
8. The wind speed interval prediction system based on nested shared long short-term memory network according to claim 1, characterized in that: In step S5, in the average width percentage formula, MWP is defined as the average percentage of the interval width and the observed value, and MC is the metric.
9. The wind speed interval prediction system based on nested shared long short-term memory network according to any one of claims 1 to 8, characterized in that: The wind speed interval prediction system based on the nested shared long short-term memory network is applied to the technical field of wind speed interval prediction.