A multi-step wind speed prediction method based on variational decomposed deep projection echo state network

Abstract

The application provides a multi-step wind speed prediction method based on a variational decomposition deep echo state network, step 1, a data set is acquired, and data is preprocessed; a wind speed original sequence is decomposed into a plurality of sub-sequences with different characteristics by using a variational mode decomposition VMD; step 2, a DEESN model is established; the DEESN model is composed of a plurality of echo state network ESN modules and extreme learning machine ELM encoders, each ESN module is responsible for different step-ahead prediction, and the ELM encoders between two sub-reservoirs are used for reducing dimensions and optimizing reservoir states; for Q-step ahead wind speed prediction, the DEESN model combines the Qth ESN module and the Q-1th ELM module, and the parameters of the DEESN model are determined by trial and error; step 3, multi-step wind speed prediction is performed by using the DEESN model, and a prediction result is obtained. The VMD-DEESN model is established, the mapping capability and the time dependence between future time steps are improved, and then the modeling precision and efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind speed prediction, in particular to a multi-step wind speed prediction method based on variational decomposition deep echo state network. BACKGROUND

[0002] With the increasing demand for energy, the global non-renewable resources are gradually exhausted, the exploitation of resources is increasing, and the energy crisis is expanding. In order to reduce the consumption of traditional energy and the emission of harmful substances, it is necessary to develop renewable new energy. As a kind of renewable new energy, wind energy has the advantages of clean and environmental protection, easy to obtain, abundant reserves, etc., and the use of wind energy is gradually increasing all over the world. However, due to the influence of time-varying meteorological factors, wind energy has intermittency, instability and volatility, which leads to unstable wind power collection and reduces the conversion rate of wind energy. In addition, the fluctuation of wind speed will also affect the stability of the grid voltage. Therefore, it is of great significance to accurately predict the wind speed.

[0003] At present, there are many methods for wind speed prediction, and patents have been applied, such as application publication number [CN110414045A], the invention name is short-term wind speed prediction method based on VMD-GRU; application publication number [CN111950759A], the invention name is a short-term wind speed prediction method based on two-stage decomposition, LSTM and AT; application publication number [CN114897260A], the invention name is a short-term wind speed prediction model modeling method and prediction method based on LSTM neural network; application publication number [CN115689039A], the invention name is a CNN+GRU fusion super-short-time wind speed prediction method and system; application publication number [CN115859816A], the invention name is a wind power minute-level prediction method and system based on CNN-LSTM algorithm.

[0004] These mainly focus on traditional recurrent neural network modeling methods and single-step prediction tasks. However, in deep learning, with the increase of the number of layers, the traditional recurrent neural network uses gradient descent method to update parameters, which is easy to produce gradient explosion or gradient disappearance problem. At the same time, the increase of the number of layers will also increase the amount of calculation and the training time, which cannot meet the real-time requirement of prediction. SUMMARY

[0005] The purpose of the present application is to provide a multi-step wind speed prediction method based on variational decomposition deep echo state network, which mainly solves the accuracy, efficiency and generalization of the prediction model in wind speed multi-step prediction. Specifically includes:

[0006] 1) Establish a new algorithm of deep echo state network DEESN prediction;

[0007] 2) Investigate the time dependencies between future time steps to reduce the cumulative error in multi-step prediction and improve prediction accuracy; investigate high-dimensional state dimensionality reduction coding methods to improve modeling efficiency.

[0008] 3) The VMD data preprocessing method is used to decompose the raw wind speed data into several low-complexity subsequences to improve the generalization ability of DEESN. Then, a new multi-step wind speed prediction model based on VMD-DEESN is established.

[0009] A multi-step wind speed prediction method based on Variation Decomposition Depth Projection Encoded Echo State Network (VMD-DEESN) specifically includes the following steps:

[0010] 1. Obtain the dataset and preprocess it. Raw wind speed data exhibits high nonlinearity and non-stationarity, which can affect modeling accuracy to some extent. To address this issue, this invention employs VMD data preprocessing technology to decompose the raw data sequence into several subsequences with different characteristics.

[0011] 2. Establishing the DEESN prediction algorithm. To improve mapping ability and the ability to learn the temporal dependencies between future time steps, this invention proposes a novel DEESN model, which consists of multiple echo state networks (ESNs).

[0012] The module is cascaded with an Extreme Learning Machine (ELM) encoder. First, the k-th ESN learner is responsible for the prediction in the k-th step, and connects the prediction output of the previous ESN module with the input variables and the encoded state to form a new input signal for the next prediction. Second, an ELM encoder is used to optimize and reduce the dimensionality of the reservoir state information to reduce time consumption.

[0013] 3. Establish and evaluate the performance of the novel VMD-DEESN multi-step wind speed prediction model. Based on the above, a VMD-DEESN multi-step wind speed prediction model is established, and its effectiveness and superiority are verified. First, experiments are conducted on the DEESN model using different decomposition methods and without using any decomposition method to demonstrate the effectiveness of the Variational Mode Decomposition (VMD) method. Second, under the same VMD decomposition method, the DEESN model is experimentally compared with other models to verify the superiority of DEESN. Finally, using wind speed data at different time scales, the same two comparative experiments are conducted to verify that the proposed VMD-DEESN outperforms other models in handling multi-step wind speed prediction.

[0014] The beneficial effects of this invention are:

[0015] 1. Establish a DEESN model to improve mapping capabilities and the ability to learn the temporal dependencies between future time steps, thereby improving modeling accuracy and efficiency.

[0016] 2. The variational mode decomposition (VMD) method is used to decompose the wind speed sequence into several subsequences with different characteristics, thereby improving the accuracy and generalization ability of the prediction model.

[0017] 3. Through wind speed prediction experiments at different time scales, it is demonstrated that the proposed VMD-DEESN prediction model is superior to other methods. Attached Figure Description

[0018] Figure 1 A flowchart of a multi-step wind speed prediction method based on Variation Decomposition Depth Projection Encoded Echo State Network (VMD-DEESN).

[0019] Figure 2 This is a schematic diagram of VMD decomposition.

[0020] Figure 3 Here is a structural diagram of the ESN model;

[0021] Figure 4 This is a structural diagram of the DEESN model. Detailed Implementation

[0022] The present invention will be further described below with reference to specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0023] This embodiment provides a multi-step wind speed prediction method based on Variational Mode Decomposition (VMD-DEESN) to improve modeling accuracy and efficiency. Furthermore, to address the nonlinearity of wind speed data, Variational Mode Decomposition (VMD) is used for data preprocessing to further enhance the model's generalization ability. Finally, a multi-step wind speed prediction model based on VMD-DEESN is established, and the effectiveness and superiority of the prediction results are verified.

[0024] Figure 1 The diagram shows a flowchart of a wind speed prediction method based on VMD-DEESN. The flowchart includes the following steps:

[0025] Step 1. Obtain the dataset. Collect two datasets at different time scales: wind speed records every 10 minutes and every hour, and split them into training and test sets in a 4:1 ratio. Then, preprocess the data, using VMD to decompose the original wind speed sequence into several subsequences with different characteristics.

[0026] Step 2. Establish the DEESN model. The DEESN model consists of multiple Echo State Network (ESN) modules and an Extreme Learning Machine (ELM) encoder. Each ESN module is responsible for asynchronous advance forecasts, and the ELM encoder between two sub-reservoirs is used to reduce dimensionality and optimize reservoir state. For Q-step advance wind speed prediction, the DEESN model combines the Q-th ESN module and the (Q-1)-th ELM module, and determines the parameters of the DEESN model through trial and error.

[0027] Step 3. Use the DEESN model to perform multi-step wind speed prediction and obtain the prediction results.

[0028] Step 4. Quantitative Evaluation. The performance of VMD-DEESN was evaluated using three different metrics: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Normalized Root Mean Square Error (NRMSE), and experiments were conducted. The effectiveness and superiority of introducing VMD decomposition into DEESN were verified by comparing it with EEMD, CEEMDAN decomposition techniques, and models without decomposition. The overall performance of VMD-DEESN was verified by comparing it with LSTM, CBLM, and the classic recursive ESN model. Wind speed prediction experiments at different time scales, such as hourly and deciminute intervals, were used to demonstrate the model's generalization ability.

[0029] Specifically, the variational mode decomposition (VMD) in step 1 is described as follows:

[0030] VMD can capture the complex features of the original sequence and then decompose it into K subsequences with different sparsity features. Figure 2 This is a schematic diagram illustrating the decomposition of the original sequence into three subsequences. Assume each subsequence is u. k The corresponding decomposition process is shown in formula (1):

[0031]

[0032] Where, ω k Indicate u k The center frequency of , δ(t) represents the Dirac distribution, (*) represents the convolution operation, and z is the original signal.

[0033] To solve the above optimization problem, a quadratic penalty function factor *a* and a Lagrange multiplier *λ* are introduced to ensure the accuracy and constraints of the reconstruction. The extended equation is rewritten as follows:

[0034]

[0035] By iteratively employing the Alternating Direction Multiplier Method (ADMM), each subsequence and center frequency are updated to seek the optimal y. k ω kAnd λ. Updated as follows:

[0036]

[0037] Where n represents the number of iterations, and γ is the noise margin. and Let z(t), u(t), and λ(t) represent the Fourier transforms of z(t), u(t), and λ(t), respectively. The iteration terminates when the following condition is met.

[0038]

[0039] Where, ∈ is and The limit error between them.

[0040] Specifically, in step 2, ESN and DEESN are described as follows:

[0041] (1) Echo State Network (ESN)

[0042] ESN is a variant of RNN with a three-layer structure: input layer, hidden layer, and output layer, such as... Figure 3 As shown in the diagram. The hidden layer, also known as the reservoir, has connection weights between neurons that are randomly and sparsely initialized. This reservoir provides a complex nonlinear mapping that encodes the input signal from a low-dimensional input space to a high-dimensional state space.

[0043] Assume that the number of neurons in each layer is K, M, and N, respectively. Therefore, the input signal can be set as follows: The status of the reserve pool is l train The length of the training sequence is represented as; the output weights are represented as... The training process mainly includes state collection and output weight calculation. When the input signal enters the reservoir, the state information can be updated and obtained. Based on the reservoir state, the output weight matrix is ​​adjusted using a regression method. The updated reservoir state equation is shown below:

[0044]

[0045] d(t) represents the reservoir state at time t, and α is the leakage rate in the range [0, 1]. and These represent the connection weights between the input and the reservoir, and between reservoirs, respectively, ranging from [-1, 1]. The final features in the output layer include the input signal and the reservoir state, i.e.

[0046] x(t)=[U(t),d(t)] (8)

[0047] Finally, ridge regression was used to calculate W.out , represented as:

[0048] W out =YX(X T X+δE) -1 (9)

[0049] Where Y is the target value, δ represents the regularization parameter, E is the identity matrix, and X is collected from x(t), as shown below:

[0050] X = [x(1),x(2),…,x(l)] train )] T (10)

[0051] (2) DEESN algorithm

[0052] To improve mapping capabilities and learn the temporal dependencies between future time steps, this invention proposes a novel DEESN model for multi-step wind speed prediction, such as... Figure 4 As shown, DEESN consists of multiple ESN modules and an ELM encoder. Each ESN module is responsible for asynchronous advance forecasts, and the ELM encoder between two sub-reservoirs is used to reduce dimensionality and optimize reservoir state. Starting from the second module, the total input includes not only the input variables but also the prediction output of the previous module and the encoded state information. Therefore, DEESN integrates the prediction output of previous steps in each future step prediction, which can learn time dependencies and reduce accumulated errors. For Q-step advance wind speed prediction, DEESN combines the Q-th ESN module and the (Q-1)-th ELM module. The reservoir state at time step t of the first ESN module is calculated using the following formula.

[0053]

[0054] U1(t) represents the input signal at time t. and These represent the connection weights input to the reservoir and the reservoir-to-reservoir connections, respectively.

[0055] Then, the state d1(t) of the above-mentioned reservoir is concatenated with the input signal U1(t) to form the following new state information.

[0056] x1(t)=[U1(t),d1(t)] (12)

[0057] Therefore, output weights It can be represented as follows:

[0058]

[0059] Where Y1 is the actual wind speed value predicted in the first step of the forecast. This indicates the status of the reserve pool as follows.

[0060]

[0061] in This represents the state signal of the j-th node in the i-th reservoir at time t.

[0062] Finally, the first step of the advance prediction value at time t is calculated as follows:

[0063]

[0064] For the second module, the total input includes not only the input variables but also the prediction output and encoded state information from the previous module. The state information of the first reservoir is encoded by ELM according to the following formula:

[0065]

[0066] Where g1(t) represents the encoded information of the first ELM encoder at time t. b1 represents the weights that are randomly and uniformly generated in the first reservoir, and b1 represents the bias of the compensation coded signal.

[0067] Then, the total input signal for the second reservoir is:

[0068]

[0069] Subsequently, the state of the second reserve pool is updated according to the following formula:

[0070]

[0071] in, and These represent the weights of the second reservoir input to the reservoir and the weights of the reservoir input to the reservoir, respectively.

[0072] Then, the state of the reservoir is collected and cascaded with the input signal U1(t), as shown below:

[0073] x2(t)=[U1(t),d2(t)] (19)

[0074] Through iterative steps, the state of each reservoir is calculated as follows:

[0075]

[0076] d q (t) represents the state of the q-th reserve pool at time t, where and These represent the connection weights from the q-th ESN module input to the reservoir and from the reservoir to the reservoir, respectively. q(t) is the input to the q-th reservoir:

[0077]

[0078] Among them, g q-1 (t) represents the state signal encoded by the (q-1)th ELM encoder at time t. This is the advance prediction result at time t, step (q-1). The state information encoded by ELM is as follows:

[0079]

[0080] Among them, W q-1 and b q-1 These are the weights and regularization values ​​of the (q-1)th ELM encoder, respectively, d q-1 (t) represents the final feature that needs to be trained through the (q-1)th ridge regression.

[0081] After each reservoir state update stabilizes, the output weights of each ESN module can be calculated, as shown below:

[0082]

[0083] in Y represents the output weight of the q-th ESN module. q Let X represent the target value at step q. q The final feature matrix obtained through the q-th ESN module is represented as follows:

[0084]

[0085] Where x q (t)=[U1(t),d q (t)].

[0086] Finally, the predicted output for each step can be obtained using the following formula:

[0087]

[0088] in, This represents the advance prediction result at time t, step q.

[0089] Specifically, in step 4:

[0090] (1) Quantitative evaluation indicators

[0091] Three different evaluation metrics are used to assess the performance of VMD-DEESN: mean absolute error (MAE), root mean square error (RMSE), and normalized root mean square error (NRMSE). The calculation formulas are shown below:

[0092]

[0093] Where T is the length of the test data, Y q (t) and Let represent the true value and predicted value at time t, step Q.

[0094] (2) Performance verification

[0095] 1) Comparison with EEMD decomposition, CEEMDan decomposition, and no decomposition method

[0096] The original sequences were decomposed using VMD, EEMD, and CEEMDAN, respectively, and experimental verification was performed on the DEESN model. The experimental results were analyzed to evaluate the effectiveness and superiority of introducing VMD decomposition into DEESN.

[0097] 2) Comparison of DEESN with LSTM, CBLM and classic recursive ESN models.

[0098] To ensure fairness and reasonable parameter settings, the original sequences were decomposed using VMD, and then experiments were conducted to compare the performance of four prediction models: DEESN, LSTM, CBLM, and the classic recursive ESN. The experimental results were used to analyze whether the DEESN model outperformed the other three models.

[0099] 3) Wind speed prediction experiments at different time scales

[0100] To demonstrate the generalization ability of the VMD-DEESN model, wind speed data at different time scales were used for validation, including hourly wind speed prediction and ten-minute wind speed prediction. The two comparative experiments were repeated. The experimental results were then used to analyze whether the VMD-DEESN model exhibits good performance.

[0101] (3) Experimental Results and Analysis

[0102] To validate the performance of the VMD-DEESN model, wind speed datasets at two time scales were collected from three sites in Tanghe County, Henan Province, China, including hourly and deciminute wind speed records from October 2016 to September 2017. The datasets were divided into training and test sets in a 4:1 ratio, and corresponding experiments were conducted.

[0103] 1) Tables 1 and 2 list the average statistical results of MAE, RMSE, and NRMSE for each ten minutes and hour for multi-step predictions, respectively, under the conditions of no decomposition tool and with three different decomposition tools, with a prediction step size of 12. The results show that the model using the decomposition tool outperforms the model without the decomposition tool in terms of prediction performance. When using the VMD decomposition tool, the DEESN model has the best prediction accuracy.

[0104] Table 1. Average statistical results of wind speed prediction evaluation indicators per ten minutes based on different decomposition tools of the DEESN model.

[0105]

[0106] Table 2. Average statistical results of hourly wind speed prediction evaluation indicators based on different decomposition tools of the DEESN model.

[0107]

[0108]

[0109] 2) Tables 3 and 4 list the average statistical results of MAE, RMSE, and NRMSE for each ten-minute and hourly interval for the four models calculating multi-step predictions, with a prediction step size of 12. The results show that DEESN outperforms the other three models, with lower prediction error and less computation time.

[0110] Table 3. Average statistical results of wind speed prediction evaluation indicators for the four models every ten minutes.

[0111]

[0112]

[0113] Table 4. Average statistical results of hourly wind speed prediction evaluation indicators for the four models.

[0114]

[0115] 3) The results of the comparative experiments above show that the VMD-DEESN model performs well on both the ten-minute and hourly timescales. The VMD-DEESN model also exhibits good generalization ability.

[0116] The above description is merely a preferred embodiment of the present invention, used to illustrate the technical solution of the present invention, and not to limit the present invention. It should be noted that for those skilled in the art, modifications and substitutions to some technical features without departing from the concept of the present invention will not depart from the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A multi-step wind speed prediction method based on variational decomposition depth projection coded echo state network, characterized in that, Includes the following steps: Step 1: Obtain the dataset and preprocess the data; Two datasets at different time scales were collected, namely wind speed records every 10 minutes and every hour, and were split into training and test sets in a 4:1 ratio. Then, the data were preprocessed, and variational mode decomposition (VMD) was used to decompose the original wind speed sequence into several subsequences with different features. Step 2, establish the DEESN model; the DEESN model consists of multiple Echo State Network (ESN) modules and an Extreme Learning Machine (ELM) encoder. Each ESN module is responsible for asynchronous early prediction, and the ELM encoder between two sub-reservoirs is used to reduce dimensionality and optimize reservoir states; for The DEESN model will predict wind speeds ahead of time. The first ESN module and the first The DEESN model is determined through a trial-and-error process, combining multiple ESN modules. The DEESN model consists of multiple ESN modules and an ELM encoder. Each ESN module is responsible for asynchronous early predictions, and the ELM encoder between two sub-reservoirs is used to reduce dimensionality and optimize reservoir states. Starting from the second module, the total input includes not only the input variables but also the prediction output and encoded state information from the previous module. DEESN will use advanced wind speed forecasting to... The first ESN module and the first The ELM modules are combined; the reservoir state of the first ESN module at time step t is calculated using the following formula: ； This represents the input signal at time t. The rate of omission integration in the range [0, 1] and These represent the connection weights input to the reservoir and the reservoir-to-reservoir connections, respectively. The above reservoir state With input signal Cascaded, forming the following new state information; ； Output weights It is expressed as follows: ； in, This is the first step in predicting the actual wind speed. Represents the regularization parameter. The following indicates the status of the reserve pool; ； ； in Indicates time The The first reservoir The status signals of each node; Finally, the first step of the advance prediction value at time t is calculated as follows: ； For the second module, the total input includes not only the input variables but also the prediction output and encoded state information from the previous module; the state information of the first reservoir is encoded by ELM according to the following formula: ； in, Indicates the first ELM encoder in Encoded information at any given time The weights are randomly and uniformly generated in the first reserve pool. This indicates the bias of the compensated coded signal. This represents the activation function of the ELM encoder; Then, the total input signal for the second reservoir is: ； Subsequently, the state of the second reservoir is updated according to the following formula: ； in, and These represent the weights of the second reservoir input to the reservoir and the weights of reservoir-to-reservoir inputs, respectively. Then, the reservoir state is collected and compared with the input signal. Cascaded, as shown below: ； Through iterative steps, each reservoir state is calculated as follows: ； express Time of the first One reservoir state, among which , and They represent the first The weights of each reservoir input to the reservoir and the weights of each reservoir input to the reservoir; It is the first Input to each reservoir: ； in, express Time by the first The status signal encoded by the ELM encoder; Time of the first The step-by-step prediction result; the state information encoded by ELM is as follows: ； in, and They are the first The weights and regularization values ​​of each ELM encoder. This indicates that it needs to be passed through the first... The ultimate characteristic of individual ridge regression training; After each reservoir state update stabilizes, the output weights of each ESN module can be calculated, as shown below: ； in Indicates the first The output weights of each ESN module Represents the regularization parameter. Indicates the first The target value of the step, Through the first The final feature matrix of each ESN module is represented as follows: ； ； in ; Finally, the predicted output for each step is obtained using the following formula: ； in, express Time of the first Step-by-step prediction results; Step 3: Use the DEESN model to perform multi-step wind speed prediction and obtain the prediction results.

2. The multi-step wind speed prediction method based on variational decomposition depth projection coded echo state network according to claim 1, characterized in that, In step 1, VMD is used to capture the complex features of the original sequence, and then it is decomposed into K subsequences with different sparsity features. Each decomposed subsequence is... The corresponding decomposition process is shown in formula (1): ； in, express The center frequency, The asterisk (*) represents the Dirac distribution, and the asterisk (*) represents the convolution operation. It is the original signal. Indicates time The partial derivative operator; Introducing a quadratic penalty function factor and Lagrange multipliers To ensure the accuracy and constraints of the reconstruction, the extended equation is rewritten as follows: ； ； in, Indicates time The original signal, Indicates time Lagrange multipliers, Indicates time The modal functions; The optimal solution is found by iterating through the Alternating Direction Method (ADMM) using multipliers to update each subsequence and its center frequency. , and Updated as follows: ； ； ； Where n represents the number of iterations, It is noise margin. , and They represent , The Fourier transform of , the iteration terminates when the following condition is met; ； in, Yes = and The limit error between them.

3. The multi-step wind speed prediction method based on variational decomposition depth projection coded echo state network according to claim 2, characterized in that, In step 2, ESN is a variant of RNN with a three-layer structure, namely input layer, hidden layer and output layer; wherein, the hidden layer is also called the reservoir, and the connection weights between neurons are randomly and sparsely initialized; the reservoir provides a complex nonlinear mapping that encodes the input signal from a low-dimensional input space to a high-dimensional state space; The number of neurons in each layer are respectively , , The input signal is set to The status of the reserve pool is as follows: , The length of the training sequence is represented as; the output weights are represented as... The training process includes state collection and output weight calculation; when the input signal enters the reservoir, the state information is updated and obtained; based on the reservoir state, the output weight matrix is ​​adjusted using a regression method; the updated reservoir state equation is shown below: ； Indicates time The reservoir state at that time, The rate of omission integration in the range [0, 1] and These represent the connection weights between the input and the reservoir, and between reservoirs, respectively, ranging from [-1, 1]. The final features in the output layer include the input signal and the reservoir state, i.e. ； Finally, the ridge regression method was used to obtain... , is represented as: ； in It is the target value. Represents the regularization parameter. It is the identity matrix. from Collect, as shown below: 。

Images