Wind speed short-term multi-step prediction method integrating direct prediction and recursive prediction

Abstract

The invention discloses a wind speed short-term multi-step prediction method fusing direct prediction and recursive prediction, and the method comprises the steps: obtaining an original data set of a wind speed short-term multi-step prediction task, and carrying out the normalization processing, and obtaining a training data set; constructing a multi-channel LSTM coding layer based on a global encoder and a single-step encoder; inputting the training data set into a multi-channel LSTM coding layer for coding to obtain coding state features; constructing a two-way decoder, and inputting the coding state features into the two-way decoder for global recursive decoding and single-step direct decoding to obtain global decoding features and single-step decoding features; splicing and fusing the global decoding features and the single-step decoding features to obtain fused features; and performing training prediction on the fusion features based on a residual network and a filling strategy to obtain a multi-step prediction result. According to the invention, the prediction precision can be improved, and the error accumulation during the prediction period can be reduced at the same time. The method can be widely applied to the technical field of wind speed short-term multi-step prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of short-term multi-step wind speed prediction, and in particular to a short-term multi-step wind speed prediction method integrating direct prediction and recursive prediction. Background Art

[0002] The short-term multi-step wind speed prediction technology refers to learning a period of historical window data, and then predicting the changes in wind speed in the future. The historical window data includes the wind speed values ​​and other related variables at historical moments. For example, factors such as temperature, air pressure, humidity, and rainfall. The short-term multi-step wind speed prediction faces many challenges, including external reasons such as the randomness and uncertainty of wind speed, changes in weather conditions, and the influence of terrain; there are also internal technical difficulties such as considering the information association between multiple prediction moments and reducing the error propagation of multi-step predictions. In addition, since the short-term changes in wind speed are usually small, the accuracy of the prediction is required to be high. At the same time, this prediction is of great significance to the operation and management of wind farms, the dispatch of power systems, and the trading of energy markets. Therefore, it is necessary to continuously improve and optimize the prediction method to solve and meet the relevant needs.

[0003] In traditional methods, wind speed prediction is mainly based on physical models and statistical models. Physical models are represented by CFD models and Kalman filter methods. Although these methods can take into account the influence of multiple physical factors, they require more time to correct wind speed, so they are more suitable for long-term predictions, but not for short-term predictions. Statistical models are represented by ARIMA models, which can only capture linear relationships and may not accurately capture nonlinear trends and changes in wind speed, resulting in reduced prediction accuracy.

[0004] Compared with traditional methods, machine learning and deep learning models have better universality and flexibility in wind speed forecasting. In the model training stage, it can adapt to different meteorological conditions and time periods for learning and make corresponding optimizations. In the forecasting stage, the existing forecasting schemes can be divided into direct forecasting and recursive forecasting; among them, direct forecasting does not consider the correlation between the forecasting moments and cannot predict the changing trend well. Recursive forecasting is prone to error accumulation, and the accuracy of subsequent step predictions is not high. Therefore, the existing technology cannot take into account both improving the forecasting accuracy and reducing the error accumulation during the forecasting period. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a short-term multi-step wind speed prediction method that integrates direct prediction and recursive prediction. By integrating the advantages of recursive prediction and direct prediction, the prediction accuracy can be improved while reducing the error accumulation during the prediction period.

[0006] The technical solution adopted by the present invention is: a short-term multi-step wind speed prediction method integrating direct prediction and recursive prediction, comprising the following steps:

[0007] The original data set of the short-term multi-step wind speed prediction task is obtained and normalized to obtain a training data set;

[0008] Construct a multi-channel LSTM encoding layer based on the global encoder and the single-step encoder;

[0009] Inputting the training data set into the multi-channel LSTM encoding layer for encoding to obtain encoding state features;

[0010] Constructing a two-way decoder, and inputting the encoding state feature into the two-way decoder for global recursive decoding and single-step direct decoding to obtain a global decoding feature and a single-step decoding feature;

[0011] Concatenating and fusing the global decoding features and the single-step decoding features to obtain a fused feature;

[0012] The fusion features are trained and predicted based on the residual network and the filling strategy to obtain a multi-step prediction result.

[0013] Further, the multi-channel LSTM encoding layer is composed of a first LSTM channel and other LSTM channels, wherein:

[0014] The first LSTM channel adopts a global encoder;

[0015] The other LSTM channels adopt a single-step encoder.

[0016] Furthermore, the encoding state feature is expressed as follows:

[0017]

[0018] Among them, x t represents the tth sample in the training data set; represents the first LSTM channel; Represents the global encoding state; represents the nth LSTM channel; Represents the single-step encoding state of the n-th LSTM channel.

[0019] Furthermore, the global decoding feature is expressed as follows:

[0020]

[0021] in, Represents the global encoding state; represents the first LSTM channel decoder; Represents the global recursive decoding feature at the nth step.

[0022] Furthermore, the single-step decoding feature is expressed as follows:

[0023]

[0024] in, represents the nth LSTM channel decoder; Represents the single-step encoding state of the nth LSTM channel; Represents the n-th single-step decoding feature.

[0025] Furthermore, the fusion feature is expressed as follows:

[0026]

[0027] in, represents the global recursive decoding feature of the nth step; Represents the nth single-step decoding feature; Represents the nth fused feature.

[0028] Furthermore, the padding strategy is to pad the left side of the fused feature sequence with a left boundary value to ensure that the sequence length remains unchanged.

[0029] The beneficial effects of the method of the present invention are as follows: the present invention designs a multi-channel LSTM encoding and decoding network, which can extract more comprehensive hidden information by integrating the advantages of recursive prediction and direct prediction in the encoding and decoding process; the final prediction result conversion is performed through the residual network, which fully considers the information association during the prediction period, further improves the prediction accuracy and reduces the error accumulation during the prediction period; the strategy of left padding boundary value is adopted to maintain the sequence length to avoid interference with the learning of sequence information; and finally, the prediction accuracy is improved while reducing the error accumulation during the prediction period. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of steps of a method for short-term multi-step wind speed prediction integrating direct prediction and recursive prediction according to the present invention;

[0031] Figure 2 It is a training prediction schematic diagram of a wind speed short-term multi-step prediction method integrating direct prediction and recursive prediction according to the present invention;

[0032] Figure 3 It is a flow chart of an encoding and decoding network based on LSTM of a wind speed short-term multi-step prediction method integrating direct prediction and recursive prediction of the present invention;

[0033] Figure 4 The present invention is a schematic diagram of a residual network prediction process of a wind speed short-term multi-step prediction method integrating direct prediction and recursive prediction. DETAILED DESCRIPTION

[0034] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0035] Reference Figure 1 and Figure 2 The present invention provides a short-term multi-step wind speed prediction method integrating direct prediction and recursive prediction, the method comprising the following steps:

[0036] S1. Obtain the original data set of the wind speed short-term multi-step prediction task and perform normalization processing to obtain a training data set. Specifically, the original data set of the wind speed short-term multi-step prediction task is normalized, and its expression is as follows:

[0037]

[0038] Among them, x represents the original data set; x min Represents the minimum value in the original data set; x max Represents the maximum value in the original data set; x n Represents the result after normalization.

[0039] Then, according to the maximum step size n required for prediction and the historical window length t of prediction, the training set and the test set are divided into 8:2. t ,x t+1 ,x t+2 …x t+n ] as a training sample, [x2,x3…x t+1 ,x t+2 ,x t+3 …x t+n+1 ] as the next training sample.

[0040] S2, construct a multi-channel LSTM encoding layer based on the global encoder and the single-step encoder;

[0041] Specifically, the multi-channel LSTM encoding layer is composed of a first LSTM channel and other LSTM channels, wherein:

[0042] The first LSTM channel adopts a global encoder;

[0043] The other LSTM channels adopt a single-step encoder.

[0044] S3, inputting the training data set into the multi-channel LSTM encoding layer for encoding to obtain encoding state features;

[0045] Specifically, refer to Figure 3 , use the multi-channel LSTM encoding layer to encode the input training data set and fully explore the hidden information of the sequence. Assume that the input training data set sequence is (x1, x2…x t ), input it into the first LSTM channel, take the output of the last moment of LSTM as the encoding result of the first LSTM channel. In multi-step prediction, this encoding result can provide reference information for the prediction of each subsequent step, so it is called the global encoding state, and its expression is as follows:

[0046]

[0047] Among them, x t represents the tth sample in the training data set; represents the first LSTM channel; Represents the global encoding status.

[0048] The present invention opens a new channel for the second step (the first step is global encoding, and the second step is single-step encoding of the second LSTM channel) and subsequent wind speed predictions to directly learn the prediction information, and performs encoding and decoding separately to obtain the corresponding single-step encoding state. In order to keep the same time interval between the historical data and the prediction time, the input training data set sequence needs to be properly trimmed. Taking the second step prediction as an example, by trimming the historical data, the time interval is expanded to twice the original sequence, and the length becomes half of the original, and a new sequence (x t-2i …x t-2 ,x t ); then input the new sequence into the second LSTM channel network and output the single-step encoding state of the second step prediction By analogy, the subsequent LSTM channel network cuts the corresponding historical data input according to the predicted time step, and outputs the corresponding encoding state. The single-step encoding state is expressed as follows:

[0049]

[0050] Among them, x t represents the tth sample in the training data set; represents the nth LSTM channel; Represents the single-step encoding state of the n-th LSTM channel.

[0051] In order to effectively learn historical data information when performing single-step state encoding, the present invention requires at least two historical data inputs to the LSTM channel network for training, so the prediction step length n should not exceed half of the original sequence window length t.

[0052] S4, constructing a two-way decoder, and inputting the encoding state feature into the two-way decoder for global recursive decoding and single-step direct decoding to obtain global decoding features and single-step decoding features;

[0053] Specifically, the dual-path decoder is composed of two LSTM networks, which respectively perform global recursive decoding operation and single-step direct decoding operation. The global recursive decoding operation input is the global encoding state feature Output the corresponding n-step recursive decoding features Recursion means that the decoding result of the previous moment is used in the calculation during LSTM decoding, and the global decoding feature is obtained recursively. Its expression is as follows:

[0054]

[0055] in, Represents the global encoding state; represents the first LSTM channel decoder; Represents the global recursive decoding feature at the nth step.

[0056] The single-step direct decoding operation outputs the decoding features of the corresponding time step according to the input single-step encoding features. Since the first step prediction does not perform single-step encoding, the present invention uses global encoding features instead. Since the single-step encoding of the first step prediction does not require a cropped sequence, its effect is the same as global encoding. The single-step decoding feature is expressed as follows:

[0057]

[0058] in, represents the nth LSTM channel decoder; Represents the single-step encoding state of the nth LSTM channel; Represents the n-th single-step decoding feature.

[0059] S5, concatenating and fusing the global decoding features and the single-step decoding features to obtain a fused feature;

[0060] Specifically, the splicing method is adopted to merge the splicing methods, and its expression is as follows:

[0061]

[0062] in, represents the global recursive decoding feature of the nth step; Represents the nth single-step decoding feature; Represents the nth fused feature.

[0063] S6. Based on the residual network and the filling strategy, the fusion features are trained and predicted to obtain multi-step prediction results.

[0064] Specifically, refer to Figure 4 , the residual network consists of a convolution layer and a ReLU activation function. Since the time series convolution needs to add padding to ensure that the time series length remains unchanged, the common all-zero padding will interfere with the learning of sequence information. Therefore, the specific embodiment of the present invention sets the convolution kernel size of the convolution layer in the residual network to 2, and adopts a padding strategy to pad the sequence of the fusion features. The padding strategy is to pad the left edge value on the left side of the fusion feature sequence to ensure that the sequence length remains unchanged. The residual network, its expression is as follows:

[0065] y=f(F(x)+x)

[0066] Among them, x represents the padded fused feature sequence; F(·) represents the residual function composed of one convolution layer and one ReLU activation function; f(·) represents the fully connected operation; y represents the multi-step prediction result.

[0067] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A short-term multi-step wind speed prediction method integrating direct prediction and recursive prediction, It is characterized in that The following steps are involved: The original data set of the short-term multi-step wind speed prediction task is obtained and normalized to obtain a training data set; Construct a multi-channel LSTM encoding layer based on the global encoder and the single-step encoder; Inputting the training data set into the multi-channel LSTM encoding layer for encoding to obtain encoding state features; Constructing a two-way decoder, and inputting the encoding state feature into the two-way decoder for global recursive decoding and single-step direct decoding to obtain a global decoding feature and a single-step decoding feature; Concatenating and fusing the global decoding features and the single-step decoding features to obtain a fused feature; The fusion features are trained and predicted based on the residual network and the filling strategy to obtain a multi-step prediction result.

2. According to claim 1, a method for short-term multi-step wind speed prediction integrating direct prediction and recursive prediction, It is characterized in that The multi-channel LSTM encoding layer is composed of a first LSTM channel and other LSTM channels, wherein: The first LSTM channel adopts a global encoder; The other LSTM channels adopt a single-step encoder.

3. According to the method for short-term multi-step wind speed prediction integrating direct prediction and recursive prediction as described in claim 1, It is characterized in that The encoding state feature is expressed as follows: Among them, x t represents the tth sample in the training data set; represents the first LSTM channel; Represents the global encoding state; represents the nth LSTM channel; Represents the single-step encoding state of the n-th LSTM channel.

4. According to claim 1, a method for short-term multi-step wind speed prediction integrating direct prediction and recursive prediction, It is characterized in that The global decoding feature is expressed as follows: in, Represents the global encoding state; represents the first LSTM channel decoder; Represents the global recursive decoding feature at the nth step.

5. According to claim 1, a method for short-term multi-step wind speed prediction integrating direct prediction and recursive prediction, It is characterized in that The single-step decoding feature is expressed as follows: in, represents the nth LSTM channel decoder; Represents the single-step encoding state of the nth LSTM channel; Represents the n-th single-step decoding feature.

6. According to claim 1, a method for short-term multi-step wind speed prediction integrating direct prediction and recursive prediction, It is characterized in that The fusion feature is expressed as follows: in, Represents the global recursive decoding feature of the nth step; Represents the nth single-step decoding feature; Represents the nth fused feature.

7. A method for short-term multi-step wind speed prediction integrating direct prediction and recursive prediction according to claim 1, It is characterized in that The filling strategy is to fill the left side of the fused feature sequence with the left boundary value to ensure that the sequence length remains unchanged.

Images