Short-term wind power prediction method based on Spearman-ISSA-BiLSTM combined model

Abstract

A short-term wind power prediction method based on a Spearman-ISSA-BiLSTM combination model comprises the following steps: screening important meteorological characteristics influencing wind power through a Spearman correlation analysis method, and taking the meteorological characteristics as input of a BiLSTM model; carrying out hyper-parameter optimization on the BiLSTM model by adopting a sparrow search algorithm improved based on Logistic chaotic mapping, adaptive weight and a local search strategy, finding the most suitable model parameters, and establishing an ISSA-BiLSTM combined model; and inputting the screened meteorological characteristics and historical power data for training to obtain a final prediction result. The method is based on a Spearman correlation analysis method, an improved sparrow search algorithm (ISSA) and a bidirectional long short-term memory network (BiLSTM) combined prediction model, compared with a traditional prediction model, the prediction precision is obviously improved, and the problem that the wind power prediction precision is not high can be well solved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to a short-term wind power prediction method based on a Spearman-ISSA-BiLSTM combined model. Background Art

[0002] As global awareness of environmental protection increases, wind power has been widely used as a clean energy source. However, due to the instability of wind resources, wind power generation also has a certain degree of uncertainty, which also brings challenges to wind power grid connection and power dispatching and consumption. In this case, wind power prediction technology can help balance the relationship between system power supply and load demand, thereby ensuring the safe and stable operation of the system. Therefore, accurate prediction of wind power has important practical value.

[0003] At present, the statistical learning method is the mainstream method for wind power prediction. It uses the historical wind power data of a wind farm and the data of its surrounding wind farms to build a statistical learning model, and then predicts the data on this basis. However, wind speed is uncertain and volatile. Direct training and prediction of the original data will have problems such as difficulty in obtaining data features and low prediction accuracy. Therefore, many combined models used for prediction in the existing technology will use some data decomposition techniques to process the original data, such as wavelet decomposition, variational mode decomposition, and empirical mode decomposition. However, these decomposition methods face problems such as modal aliasing, complex calculations, and poor prediction accuracy. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a flexible DC transmission line fault diagnosis method based on traveling wave characteristics and MIDBO-BiGRU. The method is based on a combined prediction model of Spearman correlation analysis, improved sparrow search algorithm (ISSA), and bidirectional long short-term memory network (BiLSTM). Compared with traditional prediction models, the method has obvious improvement in prediction accuracy and can well solve the problem of low wind power prediction accuracy.

[0005] The technical solution adopted by the present invention is:

[0006] The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model includes the following steps:

[0007] Step 1: Use the Spearman correlation analysis method to select important meteorological features that affect wind power and use them as the input of the BiLSTM model;

[0008] Step 2: Use the sparrow search algorithm improved based on Logistic chaotic mapping, adaptive weights and local search strategy to optimize the hyperparameters of the BiLSTM model, find the most suitable model parameters, and establish the ISSA-BiLSTM combined model; Step 3: Input the screened meteorological characteristics and historical power data for training to obtain the final prediction results.

[0009] In step 1, the Spearman correlation analysis method, the specific steps are as follows:

[0010] Sort the data of each variable, calculate the ranking difference of each pair of data, and then use these ranking differences to calculate the Spearman correlation coefficient. The calculation expression is:

[0011]

[0012] In formula (1): r s represents the Spearman correlation coefficient; d i is the difference between the rankings of the two variables; n is the number of samples.

[0013] The meteorological characteristics that affect wind power are as follows:

[0014] According to the meteorological characteristics and wind power data, the Spearman correlation coefficient between each meteorological characteristic and wind power is calculated. The range of the Spearman correlation coefficient is between [-1,1]. The closer to 1, the stronger the positive correlation, and the closer to -1, the stronger the negative correlation. It is generally believed that the correlation coefficient r s >0.8, it proves that the two variables have a strong positive correlation. Figure 7 It can be seen that the darker the color, the stronger the correlation between the two variables. Combined with the correlation coefficient values ​​in Table 1, it can be seen that the wind speeds at 30 meters and 50 meters have a strong positive correlation with wind power, and their correlation coefficients are 0.8821 and 0.8315, respectively. Therefore, the present invention inputs the wind speeds at 30 meters and 50 meters as features into the BiLSTM model for training, making full use of their relationship with wind power and improving the prediction accuracy of the model.

[0015] In step 2, Logistic chaotic mapping, adaptive weights and local search strategies are introduced to improve the sparrow search algorithm, including:

[0016] When improving the sparrow search algorithm, the present invention first initializes the population through Logistic chaotic mapping to obtain a higher quality initial population; secondly, introduces adaptive weight to update the finder position, which can avoid the problem of inadequate update of the finder position, so as to better obtain the optimal solution; finally, adds a local search strategy to search in the vicinity of the optimal solution, so as to obtain a more accurate result. The specific optimization steps are as follows:

[0017] 1) Logistic chaotic mapping:

[0018] The initialization of the Logistic chaotic mapping uses a chaotic sequence to initialize the population position, which can effectively improve the quality of the initialized population, thereby enhancing the global search ability of the algorithm. At the same time, this population has good ergodicity and randomness, which is more conducive to the population's search and effectively avoids the problem of falling into local optima. Its expression is:

[0019] x n+1 = r·x n ·(1 - x n )(2);

[0020] In the formula: r is the chaotic mapping parameter, x n+1 is the (n + 1)-th chaotic variable, and x n is the n-th chaotic variable.

[0021] 2) Adaptive weight:

[0022] Traditional SSA performs global exploration during the iteration process to update the position of the discoverer. However, the weight of position update is relatively large, and the change of its position update is relatively small, with a high probability of not finding the global optimal solution. Therefore, the present invention introduces an adaptive weight. The adaptive weight strategy dynamically adjusts the size of the weight during the search process to better balance the global search and local search capabilities. The dynamic adjustment of the weight enables the algorithm to be more inclined to global search in the initial stage and gradually turn to local search in the later stage, thereby improving the convergence speed and accuracy. The expression for updating the position of the discoverer after adding the adaptive weight is:

[0023]

[0024] In the formula: t is the iteration number; iter max is the maximum iteration number; is the position of the sparrow at the (t + 1)-th iteration; is the position of the sparrow at the t-th iteration; α is a random number in the range of (0, 1]; R2 ∈ [0, 1] is the warning value; ST ∈ [0.5, 1] is the safety value; Q is a random number obeying the normal distribution; L is a 1×d matrix where each element is 1; w(t) is the weight at the t-th iteration; w max 、w min are the maximum and minimum values of the weight; when R2 < ST, the sparrow population is in a safe state and there is no predator nearby, so global search for foraging can be carried out. If R2 ≥ ST, it indicates that there are predators around the sparrows. At this time, the sparrows issue an alarm and the sparrow population starts to transfer to a safe area.

[0025] 3) Local search strategy:

[0026] In the traditional sparrow search algorithm, a local optimal solution usually appears. The solution at this time is not necessarily the global optimal solution. Therefore, it is necessary to conduct a detailed search in the neighboring area of ​​the current optimal solution in order to obtain a more accurate optimal solution. The expression is:

[0027]

[0028] r search =δ·(ub-lb) (6);

[0029] Where: X local is the optimal solution after local search strategy optimization; r search is the local search radius; γ is a random vector between the interval [0,1]; δ is a smaller proportional factor, which is 0.15 in the present invention; ub and lb are the upper and lower bounds of the variable respectively. If X local <X best , then update X local is the current optimal solution.

[0030] In step 2, the improved sparrow search algorithm is used to optimize the bidirectional long short-term memory network (BiLSTM). The bidirectional long short-term memory network (BiLSTM) model includes a forget gate, an input gate, and an output gate. The operation expression is as follows:

[0031] Forget Gate:

[0032] f t =σ(W f ·[h t-1 ,x t ]+b f ) (7);

[0033] In formula (7): f t represents the output of the forget gate at time t, W f represents the weight matrix of the forget gate; h t-1 represents the hidden state at time t-1; x t represents the input of the forget gate at time t; b f Represents the bias vector of the forget gate; σ() Sigmoid function;

[0034] Input Gate:

[0035] i t =σ(W i ·[h t-1 ,x t ]+b i )(8);

[0036] In formula (8): i t Represents the input of the input gate; Wi Represents the weight matrix of the input gate Sigmoid function; b i Represents the bias vector of the input gate Sigmoid function;

[0037]

[0038] In formula (9): represents the cell state at time t; W C The weight matrix of the activation function in the input gate of C Represents the bias vector of the input gate activation function.

[0039]

[0040] In formula (10): C t-1 represents the cell state at time t-1; Multiply the corresponding vector elements;

[0041] Output Gate:

[0042] o t =σ(W o ·[h t-1 ,x t ]+b o ) (11);

[0043]

[0044] Where: t Represents the input of the output gate; W o represents the weight matrix of the output gate; b o is the bias vector for each response gate;

[0045] h t represents the hidden state at time t; tanh(C t ) represents the activation function of the unit state at time t; Multiply the corresponding vector elements.

[0046] Perform hyperparameter optimization on the BiLSTM model to find the most suitable model parameters; the details are as follows:

[0047] Set the number of neurons to 32, Epochs to 90, Batchsizes to 128, input step size to 2, number of input features to 2, optimization algorithm population to 6, and number of evolutions to 10.

[0048] In the step 3, a Spearman-ISSA-BiLSTM prediction model is established for each subsequence obtained by decomposition, including: S3.1: inputting wind power and characteristic data;

[0049] S3.2: Spearman correlation analysis method to screen meteorological characteristics;

[0050] S3.3: Initialize the BiLSTM model structure;

[0051] S3.4: Improve the sparrow search algorithm to obtain the optimal parameters;

[0052] S3.5: Establish and train the Spearman-ISSA-BiLSTM combined model based on the optimal parameters.

[0053] The present invention provides a short-term wind power prediction method based on the Spearman-ISSA-BiLSTM combined model, and the technical effects are as follows:

[0054] 1) The present invention uses the Spearman correlation analysis method to screen important meteorological characteristic data, which can make full use of the relationship between meteorological characteristics and output power, better assist model training, and improve prediction accuracy.

[0055] 2) By using Logistic chaotic mapping to initialize the population, introducing adaptive weights and adding local search strategies, the optimization ability of the traditional sparrow search algorithm can be effectively improved, avoiding falling into the local optimal solution, and better helping the model to optimize parameters.

[0056] 3) Compared with a single model or other combined models, the Spearman-ISSA-BiLSTM combined model proposed in the present invention shows better model performance on both data sets in the article, has higher prediction accuracy, and can be well adapted to the current short-term wind power prediction tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0058] Figure 1 Schematic diagram of the BiLSTM model structure.

[0059] Figure 2 Flowchart of the Spearman-ISSA-BiLSTM structure.

[0060] Figure 3 is the original wind power curve.

[0061] Figure 4 The result of the F5 test function.

[0062] Figure 5 The result of the F6 test function.

[0063] Figure 6 Prediction results for each optimal model and BiLSTM model.

[0064] Figure 7 This is the Spearman correlation coefficient heat map. DETAILED DESCRIPTION

[0065] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0066] The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model includes the following steps:

[0067] Step 1: Use the Spearman correlation analysis method to select important meteorological features that affect wind power and use them as the input of the BiLSTM model;

[0068] Step 2: Use the sparrow search algorithm based on Logistic chaotic mapping, adaptive weights, and local search strategy to optimize the hyperparameters of the BiLSTM model, find the most suitable model parameters, and establish the ISSA-BiLSTM combination model;

[0069] Step 3: Input the screened meteorological characteristics and historical power data for training to obtain the final prediction results.

[0070] In step 1, the Spearman correlation analysis method evaluates the correlation between two variables by converting the data into rankings. This method mainly measures the monotonic relationship between variables and is not limited to linear relationships. Unlike the Pearson correlation coefficient, the Spearman correlation does not require the data to satisfy the normal distribution. It can be applied to data of various distributions, which makes it more flexible when dealing with non-normally distributed data and is applicable to all monotonic relationships, including non-linear monotonic relationships. When the relationship between variables is not linear but still monotonic, it can provide a more accurate correlation measurement. At the same time, the Spearman correlation analysis uses rankings instead of raw data, so it is insensitive to outliers, which makes its performance more robust in data sets with extreme values. In general, the Spearman correlation analysis has significant advantages in dealing with nonlinear, non-normally distributed and outlier data, and is a flexible and robust correlation analysis method. Its specific steps are:

[0071] Sort the data of each variable, calculate the ranking difference of each pair of data, and then use these ranking differences to calculate the Spearman correlation coefficient. The calculation expression is:

[0072]

[0073] In formula (1): r srepresents the Spearman correlation coefficient; d i is the difference between the rankings of the two variables; n is the number of samples.

[0074] The important meteorological characteristics that affect wind power are as follows:

[0075] According to the meteorological characteristics and wind power data of a certain year, the Spearman correlation coefficient between each meteorological characteristic and wind power is calculated using formula (1). The Spearman correlation coefficient heat map and the correlation coefficient table of each characteristic are as follows: Figure 7 And as shown in Table 1.

[0076] Table 1 Spearman correlation coefficients of various meteorological characteristics

[0077]

[0078] The range of the Spearman correlation coefficient is between [-1,1]. The closer it is to 1, the stronger the positive correlation is, and the closer it is to -1, the stronger the negative correlation is. It is generally believed that the correlation coefficient r s >0.8, it proves that the two variables have a strong positive correlation. Figure 7 It can be seen that the darker the color, the stronger the correlation between the two variables. Combined with the correlation coefficient values ​​in Table 1, it can be seen that the wind speeds at 30 meters and 50 meters have a strong positive correlation with wind power, and their correlation coefficients are 0.8821 and 0.8315, respectively. Therefore, the present invention inputs the wind speeds at 30 meters and 50 meters as features into the BiLSTM model for training, making full use of their relationship with wind power and improving the prediction accuracy of the model.

[0079] In step 2, Logistic chaotic mapping, adaptive weights and local search strategies are introduced to improve the sparrow search algorithm, including:

[0080] When improving the sparrow search algorithm, the present invention first initializes the population through Logistic chaotic mapping to obtain a higher quality initial population; secondly, introduces adaptive weight to update the finder position, which can avoid the problem of inadequate update of the finder position, so as to better obtain the optimal solution; finally, adds a local search strategy to search in the vicinity of the optimal solution, so as to obtain a more accurate result. The specific optimization steps are as follows:

[0081] 1) Logistic Chaotic Mapping:

[0082] Logistic chaotic map initialization uses chaotic sequences to initialize the population position, which can effectively improve the quality of the initialized population and thus improve the global search capability of the algorithm. At the same time, the population has good ergodicity and randomness, which is more conducive to the population search and effectively avoids falling into the problem of local optimality. Its expression is:

[0083] x n+1 = r·x n ·(1 - x n )(2);

[0084] Where: r is the chaotic mapping parameter, x n+1 is the (n + 1)-th chaotic variable, x n is the n-th chaotic variable.

[0085] 2) Adaptive weight:

[0086] Traditional SSA performs global exploration during the iteration process to update the position of the discoverer. However, the weight of position update is relatively large, and the change of its position update is small, so there is a high probability that the global optimal solution cannot be found. Therefore, the present invention introduces an adaptive weight, and the adaptive weight strategy dynamically adjusts the size of the weight during the search process to better balance the global search and local search capabilities. The dynamic adjustment of the weight enables the algorithm to be more inclined to global search in the initial stage and gradually turn to local search in the later stage, thereby improving the convergence speed and accuracy. The expression for updating the position of the discoverer after adding the adaptive weight is:

[0087]

[0088] Where: t is the iteration number; iter max is the maximum number of iterations; is the position of the sparrow at the (t + 1)-th iteration; is the position of the sparrow at the t-th iteration; α is a random number in the range of (0, 1]; R2 ∈ [0, 1] is the warning value; ST ∈ [0.5, 1] is the safety value; Q is a random number obeying the normal distribution; L is a 1×d matrix where each element is 1; w(t) is the weight at the t-th iteration; w max 、w min are the maximum and minimum values of the weight; when R2 < ST, the sparrow population is in a safe state and there is no predator nearby, so global search for foraging can be carried out. If R2 ≥ ST, it indicates that there is a predator around the sparrow, and at this time the sparrow sounds an alarm and the sparrow population starts to transfer to a safe area.

[0089] 3) Local search strategy:

[0090] In the traditional sparrow search algorithm, local optimal solutions usually appear, and the solutions at this time are not necessarily the global optimal solutions. Therefore, it is necessary to conduct a detailed search in the neighboring area of the current optimal solution in order to obtain a more accurate optimal solution. The expression is:

[0091]

[0092] r search=δ·(ub-lb) (6);

[0093] Where: X local is the optimal solution after local search strategy optimization; r search is the local search radius; γ is a random vector between the interval [0,1]; δ is a smaller proportional factor, which is 0.15 in the present invention; ub and lb are the upper and lower bounds of the variable respectively. If X local <X best , then update X local is the current optimal solution.

[0094] In step 2, the improved sparrow search algorithm is used to optimize the bidirectional long short-term memory network (BiLSTM). The bidirectional long short-term memory network (BiLSTM) model includes a forget gate, an input gate, and an output gate. The operation expression is as follows:

[0095] Forget Gate:

[0096] f t =σ(W f ·[h t-1 ,x t ]+b f )(7);

[0097] In formula (7): f t represents the output of the forget gate at time t, W f represents the weight matrix of the forget gate; h t-1 represents the hidden state at time t-1; x t represents the input of the forget gate at time t; b f Represents the bias vector of the forget gate; σ() Sigmoid function; input gate:

[0098] i t =σ(W i ·[h t-1 ,x t ]+b i )(8);

[0099] In formula (8): i t Represents the input of the input gate; W i Represents the weight matrix of the input gate Sigmoid function; b i Represents the bias vector of the input gate Sigmoid function;

[0100]

[0101] In formula (9): represents the cell state at time t; W C The weight matrix of the activation function in the input gate ofC Represents the bias vector of the input gate activation function.

[0102]

[0103] In formula (10): C t-1 represents the cell state at time t-1; Multiply the corresponding vector elements;

[0104] Output Gate:

[0105] o t =σ(W o ·[h t-1 ,x t ]+b o )(11);

[0106]

[0107] Where: t Represents the input of the output gate; W o represents the weight matrix of the output gate; b o is the bias vector for each response gate;

[0108] h t represents the hidden state at time t; tanh(C t ) represents the activation function of the unit state at time t; Multiply the corresponding vector elements.

[0109] Perform hyperparameter optimization on the BiLSTM model to find the most suitable model parameters; the details are as follows:

[0110] Hyperparameter optimization refers to the process of adjusting the parameters outside the model in machine learning and deep learning models to optimize the model performance and generalization ability. Hyperparameters are parameters set before training the model. They cannot be learned through training data, but need to be set manually. In order to better compare the training effects of each combined model with the model proposed in the present invention, the present invention uniformly sets the number of neurons to 32, Epochs to 90, Batchsizes to 128, the input step size to 2, the input feature number to 2, the optimization algorithm population number to 6, and the number of evolutions to 10.

[0111] The ISSA-BiLSTM combination model is as follows:

[0112] The ISSA-BiLSTM combination model is a hybrid model that combines the improved sparrow search algorithm (ISSA) and the bidirectional long short-term memory network (BiLSTM). ISSA is a semi-supervised learning method that can effectively use unlabeled data to improve model performance. ISSA combines supervised learning and unsupervised learning to gradually use unlabeled data to improve the model to gradually enhance the generalization ability of the model. BiLSTM is a neural network structure suitable for sequence data modeling. It can consider both past and future information at the same time, so as to better capture long-distance dependencies in the sequence. By combining ISSA and BiLSTM, the ISSA-BiLSTM combination model can fully utilize the advantages of semi-supervised learning, while using Bi-directional LSTM to process sequence data, further improving the performance and generalization ability of the model. The relevant formulas are shown in equations (1) to (12), and the specific implementation process is as follows: Figure 2 shown.

[0113] In the step 3, a Spearman-ISSA-BiLSTM prediction model is established for each subsequence obtained by decomposition, including: S3.1: inputting wind power and characteristic data;

[0114] S3.2: Spearman correlation analysis method to screen meteorological characteristics;

[0115] S3.3: Initialize the BiLSTM model structure;

[0116] S3.4: Improve the sparrow search algorithm to obtain the optimal parameters;

[0117] S3.5: Establish and train the Spearman-ISSA-BiLSTM combined model based on the optimal parameters.

[0118] Example:

[0119] The present invention selects the open source wind power data of the State Grid, and selects the wind power and meteorological data of the No. 6 wind farm in March 2019. First, the data is preprocessed, and the missing values ​​and outliers are processed by the mean interpolation method. Then the data set is divided into a training set, a validation set, and a test set according to a ratio of 7:2:1. This can better improve the model training effect and improve the prediction accuracy. The wind power curve is as follows Figure 3 shown.

[0120] In order to test the improvement effect of the present invention on the sparrow search algorithm, the present invention selects the F5 and F8 functions in the CEC test function to test them and compares them with the traditional SSA. Here, the population number is uniformly set to 30, the maximum number of iterations is 1000, the test function is shown in Table 2, and the experimental results are shown in Table 2. Figure 4 , Figure 5 As shown in Table 2.

[0121] Table 2F5, F8 test functions

[0122]

[0123] Depend on Figure 4 , Figure 5 It can be seen that compared with the traditional sparrow search algorithm, the improved sparrow search algorithm has the characteristics of faster convergence speed and higher optimization efficiency. In the F5 and F8 test functions, the improved sparrow algorithm can obtain lower function values ​​with fewer iterations. This also shows that the introduction of Logistic chaotic mapping to initialize the sparrow population, the introduction of adaptive weights to update the discoverer position, and the use of local search strategies to further find the global optimal solution can effectively improve the algorithm's optimization ability and help the model find the optimal parameters.

[0124] According to the meteorological characteristics and wind power data of a certain year, the Spearman correlation coefficient between each meteorological characteristic and wind power is calculated using formula (1). The Spearman correlation coefficients of each characteristic are shown in Table 1.

[0125] The range of the Spearman correlation coefficient is between [-1,1]. The closer it is to 1, the stronger the positive correlation is, and the closer it is to -1, the stronger the negative correlation is. It is generally believed that when the correlation coefficient rs>0.8, it proves that the two variables have a strong positive correlation. Figure 7 It can be seen that the darker the color, the stronger the correlation between the two variables. Combined with the correlation coefficient values ​​in Table 1, it can be seen that the wind speeds at 30 meters and 50 meters have a strong positive correlation with wind power, and their correlation coefficients are 0.8821 and 0.8315, respectively. Therefore, the present invention inputs the wind speeds at 30 meters and 50 meters as features into the model for training, making full use of their relationship with wind power to improve the prediction accuracy of the model.

[0126] In addition, in order to better verify the performance of the model proposed in this invention, LSTM, TCN, BiLSTM and other combined models are selected to conduct comparative experiments on the above data sets. The experimental results are as follows: Figure 6 , as shown in Table 3.

[0127] Table 3 Error indicators of each combination model

[0128]

[0129] Depend on Figure 6It can be seen that compared with a single model, the prediction results of the combined model with an optimization algorithm for parameter optimization have a better fit with the original data and a smaller error. At the same time, the Spearman-ISSA-BiLSTM proposed in the present invention is also more advantageous than other combined models, and the prediction results are closer to the original data.

[0130] Combined with the evaluation index results in Table 3, it can be seen that after adding the optimization algorithm, the prediction error of the model is effectively reduced. For example, compared with the LSTM model, the MAE of the PSO-LSTM and ISSA-LSTM models is reduced by 17.55% and 24.34%, the MAPE is reduced by 44.96% and 69.80%, the RMSE is reduced by 14.68% and 18.09%, and the MSE is reduced by 27.21% and 32.92%. Similarly, adding the optimization algorithm to the TCN as the basic model can also reduce the prediction error. For example, compared with the TCN model, the MAE of the SSA-TCN and GWO-TCN models is reduced by 9.13% and 13.26%, the MAPE is reduced by 62.36% and 44.50%, the RMSE is reduced by 5.78% and 9.24%, and the MSE is reduced by 8.29% and 17.63%.

[0131] It can be seen from the above data that after adding the optimization algorithm, the model parameters can be better set and the performance of the model can be effectively improved. The Spearman-ISSA-BiLSTM model proposed in the present invention can also effectively improve the prediction accuracy of the model compared with the traditional sparrow search algorithm after adding Logistic chaotic mapping, adaptive weights and local search strategies. For example, compared with the SSA-LSTM, SSA-TCN and SSA-BiLSTM models, the MAE of ISSA-BiLSTM is reduced by 15.39%, 11.56% and 5.27% respectively, and the MSE is reduced by 23.20%, 18.52% and 7.46% respectively. It can be seen that compared with other optimization algorithms, the improved sparrow search algorithm can effectively improve the prediction accuracy of the model. The sparrow search algorithm can obtain a higher quality initialization population through Logistc chaotic mapping, avoiding the problems of low quality and uneven distribution of the traditional random initialization population, and the introduced adaptive weight can also help the discoverer to better update the position in order to better obtain the optimal solution. The local search strategy can further search in the vicinity of the current optimal solution and can also effectively avoid missing the global optimal solution. In summary, the improved sparrow search algorithm has better optimization ability, and compared with a single model, the introduction of the improved sparrow search algorithm to optimize the model parameters can significantly improve the prediction accuracy of the model, and can be better used in wind power prediction tasks.

[0132] In order to solve the problem of large fluctuations in wind power and low prediction accuracy, the present invention proposes a combination model based on Spearman correlation analysis, improved sparrow search algorithm (ISSA), and bidirectional long short-term memory network (BiLSTM). The results show that the test results of the present invention based on the Spearman-ISSA-BiLSTM combination model show that its MAE, MAPE, RMSE, and MSE are 2.6755MW, 1.5784%, 4.5279MW, and 20.5014MW, respectively. Compared with LSTM, TCN (Temporal Convolutional Networks), BiLSTM and other combination models, there is a certain improvement in prediction accuracy, which can well solve the problem of large fluctuations in wind power and low prediction accuracy.

Claims

1. A short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model is characterized by The following steps are involved: Step 1: Use the Spearman correlation analysis method to screen out meteorological features that affect wind power and use them as input to the BiLSTM model; Step 2: Use the sparrow search algorithm based on Logistic chaotic mapping, adaptive weights, and local search strategy to optimize the hyperparameters of the BiLSTM model, find the most suitable model parameters, and establish the ISSA-BiLSTM combination model; Step 3: Input the screened meteorological characteristics and historical power data for training to obtain the final prediction results.

2. The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model according to claim 1 is characterized by: In step 1, the Spearman correlation analysis method, the specific steps are as follows: Sort the data of each variable, calculate the ranking difference of each pair of data, and then use these ranking differences to calculate the Spearman correlation coefficient. The calculation expression is: In formula (1): r s represents the Spearman correlation coefficient; d i is the difference between the rankings of the two variables; n is the number of samples.

3. The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model according to claim 2 is characterized by: The meteorological characteristics that affect wind power are as follows: According to the meteorological characteristics and wind power data, the Spearman correlation coefficient between each meteorological characteristic and wind power is calculated. The range of the Spearman correlation coefficient is between [-1,1]. The closer to 1, the stronger the positive correlation, and the closer to -1, the stronger the negative correlation. The correlation coefficient r s When >0.8, it proves that the two variables have a strong positive correlation; the wind speeds at 30 meters and 50 meters have a strong positive correlation with wind power, and their correlation coefficients are 0.8821 and 0.8315 respectively. Therefore, the wind speeds at 30 meters and 50 meters are input as features into the BiLSTM model for training.

4. The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model according to claim 1 is characterized by: In step 2, Logistic chaotic mapping, adaptive weights and local search strategies are introduced to improve the sparrow search algorithm, including: When improving the sparrow search algorithm, the population is first initialized by Logistic chaotic mapping to obtain a higher quality initial population; secondly, the adaptive weight is introduced to update the finder position, which can avoid the problem of insufficient finder position update, so as to better obtain the optimal solution; finally, a local search strategy is added to search in the vicinity of the optimal solution, so as to obtain more accurate results. The specific optimization steps are as follows: 1) Logistic Chaotic Mapping: Logistic chaotic map initialization uses chaotic sequences to initialize the population position, effectively improving the quality of the initialized population. Its expression is: x n+1 =r·x n ·(1-x n )(2); Where: r is the chaotic mapping parameter, x n+1 is the n+1th chaotic variable, x n is the nth chaotic variable; 2) Adaptive weight: Adaptive weights are introduced. The adaptive weight strategy dynamically adjusts the size of the weight during the search process to better balance the capabilities of global search and local search. The update expression of the finder position after adding adaptive weights is: where: t is the number of iterations; iter max is the maximum number of iterations; is the position of the sparrow at the (t + 1)-th iteration; is the position of the sparrow at the t-th iteration; α is a random number in the interval (0, 1]; R2 ∈ [0, 1] is the warning value; ST ∈ [0.5, 1] is the safety value; Q is a random number following a normal distribution; L is a 1×d matrix where each element is 1; w(t) is the weight at the t-th iteration; w max and w min are the maximum and minimum values of the weight; when R2 < ST, the sparrow population is in a safe state and there are no predators nearby, and global search for food can be carried out. If R2 ≥ ST, it indicates that there are predators around the sparrows. At this time, the sparrows issue an alarm and the sparrow population begins to transfer to a safe area; 3) Local search strategy: A detailed search is performed in the neighborhood of the current optimal solution to obtain a more accurate optimal solution. The expression is: rsearch = δ·(ub-lb) (6); Where: X local is the optimal solution after local search strategy optimization; r search is the local search radius; γ is a random vector between the interval [0,1]; δ is a smaller scaling factor; ub and lb are the upper and lower bounds of the variable respectively. If X local <X best , then update X local is the current optimal solution.

5. The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model according to claim 1 is characterized by: In step 2, the improved sparrow search algorithm is used to optimize the bidirectional long short-term memory network (BiLSTM), the bidirectional long short-term memory network (BiLSTM) model includes a forget gate, an input gate, and an output gate. The operation expression is as follows: Forget Gate: f t =σ(W f ·[h t-1 ,x t ]+b f ) (7); In formula (7): f t represents the output of the forget gate at time t, W f represents the weight matrix of the forget gate; h t-1 represents the hidden state at time t-1; x t represents the input of the forget gate at time t; b f Represents the bias vector of the forget gate; σ() Sigmoid function; Input Gate: i t =σ(W i ·[h t-1 ,x t ]+b i ) (8); In formula (8): i t Represents the input of the input gate; W i Represents the weight matrix of the input gate Sigmoid function; b i Represents the bias vector of the input gate Sigmoid function; In formula (9): represents the cell state at time t; W C The weight matrix of the activation function in the input gate of C Represents the bias vector of the input gate activation function; In formula (10): C t-1 represents the cell state at time t-1; Multiply the corresponding vector elements; Output Gate: the t =σ(W o ·[h t-1 ,x t ]+b o ) (11); Where: t Represents the input of the output gate; W o represents the weight matrix of the output gate; b o is the bias vector for each response gate; h t represents the hidden state at time t; tanh(C t ) represents the activation function of the unit state at time t; ο is the corresponding multiplication of each vector element.

6. The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model according to claim 5 is characterized by: Perform hyperparameter optimization on the BiLSTM model to find the most suitable model parameters; the details are as follows: Set the number of neurons to 32, Epochs to 90, Batchsizes to 128, input step size to 2, number of input features to 2, optimization algorithm population to 6, and number of evolutions to 10.

7. The short-term wind power forecasting method based on the Spearman-ISSA-BiLSTM combined model according to claim 6 is characterized by: In step 3, a Spearman-ISSA-BiLSTM prediction model is established for each subsequence obtained by decomposition, including: S3.1: Input wind power and characteristic data; S3.2: Spearman correlation analysis method to screen meteorological characteristics; S3.3: Initialize the BiLSTM model structure; S3.4: Improve the sparrow search algorithm to obtain the optimal parameters; S3.5: Establish and train the Spearman-ISSA-BiLSTM combined model based on the optimal parameters.

Images