Offshore wind power prediction method based on extended long-short term memory neural network

Abstract

The invention discloses an offshore wind power prediction method based on an extended long-short term memory neural network, and relates to the technical field of renewable energy prediction, and the method comprises the following steps: S1, extracting trend and seasonal components from original wind power data; s2, predicting the extracted trend and seasonal components by constructing an extended long-short term memory neural network; and S3, integrating the prediction results of the extended long-short term memory neural network under different combinations to obtain a final wind power prediction value. According to the offshore wind power prediction method based on the extended long-short-term memory neural network, the trend and seasonal components are extracted, and the efficient modeling capability of the extended long-short-term memory neural network is combined, so that the precision and stability of wind power prediction are remarkably improved, reliable support is provided for power system dispatching, and the method is suitable for popularization and application. The method reduces the influence of wind power fluctuation on a power grid, and has important engineering application value.

Description

Technical Field

[0001] The invention relates to the technical field of renewable energy prediction, and in particular to an offshore wind power prediction method based on an extended long short-term memory neural network. Background Art

[0002] With the transformation of the global energy structure, offshore wind power is developing rapidly as a clean and renewable form of energy. Offshore wind power has the advantages of abundant resources and high power generation efficiency, but its power output is affected by environmental factors such as wind speed, wind direction, and waves, showing strong volatility and intermittency, which brings great challenges to grid dispatch and stable operation. Existing wind power prediction methods, especially those for offshore wind power, often find it difficult to accurately capture its complex spatiotemporal variation characteristics, resulting in insufficient prediction accuracy and unable to meet the needs of real-time dispatch of power systems. Therefore, developing a prediction method that can effectively handle the volatility and periodicity characteristics of offshore wind power data is of great significance for improving the predictability of offshore wind power, optimizing grid dispatch, and reducing operating costs.

[0003] In recent years, deep learning technology has made significant progress in the field of time series forecasting, especially the long short-term memory network and its variants have performed well in capturing long-term dependencies. However, traditional long short-term memory networks still have problems such as limited memory capacity and low training efficiency when processing wind power data. To this end, the extended long short-term memory neural network significantly improves the model's memory capacity and computational efficiency by introducing an exponential gating mechanism and an enhanced memory structure, providing a new solution for wind power forecasting.

[0004] In view of this, a wind power prediction method based on extended long short-term memory neural network is provided, which can effectively improve the prediction accuracy and stability and has important engineering application value. Summary of the invention

[0005] The purpose of the present invention is to provide an offshore wind power prediction method based on an extended long short-term memory neural network. By combining the extraction of trend and seasonal components with the efficient modeling capability of an extended long short-term memory neural network, the accuracy and stability of offshore wind power prediction can be significantly improved, providing reliable data support for the scheduling and operation of the power system.

[0006] To achieve the above object, the present invention provides an offshore wind power prediction method based on an extended long short-term memory neural network, comprising the following steps: S1. Extract trend and seasonal components from raw wind power data; S2. Predict the extracted trend and seasonal components by constructing an extended long short-term memory neural network, which includes two variants: a stable long short-term memory network and a matrix long short-term memory network; S3. Integrate the prediction results of the extended long short-term memory neural network under different combinations to obtain the final wind power prediction value.

[0007] Preferably, S1 comprises the following steps: S11. Determine the trend component: Use the exponential average method to smooth the original wind power data and extract the trend component of the data. The calculation method is as follows: ; In the formula, and They represent the trend components of wind power data at the current time t and the previous time t-1 respectively; Represents the original wind power data at the current time t; is the smoothing coefficient, and its value range is ; S12. Extract seasonal components: By subtracting the trend component from the original data, the remaining part, i.e., the seasonal component, is obtained. The calculation method is as follows: ; In the formula, represents the seasonal component of wind power data at the current time t; S13, finally output trend component and seasonal component as input of subsequent prediction model, its specific function is as follows: ; In the formula, represents the original signal; Indicates trend component; Indicates seasonal component.

[0008] Preferably, S2 comprises the following steps: S21, stable long short-term memory network uses exponential function instead of traditional Sigmoid function, through normalization state To stabilize the cell state update, by introducing a stabilized state Ensure the numerical stability of the input gate and forget gate; S22, Matrix Long Short-Term Memory Network uses matrices to replace traditional scalar cell states, introduces query, key and value mechanisms, and removes connections between hidden states; S23. Construct an extended long short-term memory neural network by combining stable long short-term memory neural networks and matrix long short-term memory neural network modules with different numbers of layers.

[0009] Preferably, in S21, the normalized state To stabilize the update of cell state, the specific function is as follows: ; ; In the formula, represents the hidden state at the current time t; , Respectively represent the normalized states at the current time t and the previous time t-1; Indicates cell status; , , They correspond to the three gating units: input gate, forget gate, and output gate.

[0010] Preferably, in S21, a stabilized state is introduced , to ensure the numerical stability of the input gate and the forget gate, the specific function is as follows: ; ; ; In the formula, , Indicates the stable state at the current time t and the previous time t-1; , Corresponding to the original input gate and forget gate; , Represents the input gate and forget gate after scaling in the stabilized state; Represents an exponential function.

[0011] Preferably, S22 uses a matrix to replace the traditional scalar cell state, and the relationship between the input and the cell state is as follows: ; In the formula, and Respectively represent the matrix cell states at the current moment and the previous moment; , They are forget gate and input gate respectively; , are the key vector and value vector at the current time t respectively.

[0012] Preferably, in the matrix long short-term memory neural network in S22, a query, key, and value mechanism is introduced to improve the memory capacity of the model, and its specific calculation function is as follows: ; ; ; In the formula, , , Represent query vector, key vector, and value vector respectively; , , The weight matrices representing query, key, and value respectively; , , Represents the bias items of query, key, and value respectively; represents the input vector; Represents the model dimension used to scale the values ​​of the key vector.

[0013] Preferably, the input gate, forget gate, and output gate in S22 remove the connection between the hidden states, and the specific function is as follows: ; ; ; In the formula, , , They correspond to the three gating units: input gate, forget gate, and output gate; represents the exponential function; Represents the Sigmoid function; , , Respectively represent the weight vectors corresponding to the inputs of the three gating units; , , Respectively represent the biases of the three gating units; Represents the current input vector.

[0014] Preferably, the model in S23 stacks i layers of stable long short-term memory neural network and j layers of matrix long short-term memory neural network, and the calculation process can be expressed as follows: ; ; ; In the formula, , represents the hidden state of the k-th layer and the k-1-th layer of the stable long short-term memory neural network at time t, ; represents the hidden state of the k-th layer of the stable long short-term memory neural network at the previous time t-1; , represents the hidden state of the l-th layer and l-1-th layer matrix long short-term memory neural network at time t, ; Represents the hidden state of the l-th layer matrix long short-term memory neural network at the previous time t-1; Represents the final output of the model at time t; , Represents the weight matrix and bias term of the output layer; , It represents the calculation process of the stable long short-term memory neural network and the matrix long short-term memory neural network, as shown in step S21 and step S22.

[0015] Preferably, in S3, the outputs of these combinations are weighted and integrated to obtain the final wind power forecast value, and the specific calculation formula is as follows: ; In the formula, represents the final wind power forecast value, which is the integrated output at the current time t; It represents the predicted output under the mth combination. There are 4 combinations in total.

[0016] Therefore, the present invention adopts the above-mentioned offshore wind power prediction method based on extended long short-term memory neural network, which has the following beneficial effects: (1) The present invention significantly improves the accuracy and stability of offshore wind power prediction by extracting the trend and seasonal components of offshore wind power data and combining it with the efficient modeling capability of the extended long short-term memory neural network.

[0017] (2) The present invention can more accurately capture the complex changing characteristics of offshore wind power, effectively handle long-term dependencies, and improve the stability and robustness of the model.

[0018] (3) The present invention provides the power system with more accurate offshore wind power prediction results, helping dispatchers to adjust power generation plans in advance to ensure stable grid operation, and helping power generation companies optimize the operation and maintenance strategies of offshore wind farms, improve equipment utilization, and reduce operating costs.

[0019] (4) The present invention promotes the sustainable development of the offshore wind power industry and assists in the transition to clean energy by improving the predictability and economy of offshore wind power.

[0020] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the working process of the offshore wind power prediction method based on the extended long short-term memory neural network of the present invention; Figure 2 A diagram of an extended long short-term memory neural network prediction model constructed for the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0024] Example See also Figure 1-2 The present invention provides an offshore wind power prediction method based on an extended long short-term memory neural network, comprising the following steps: S1. With the goal of improving forecast accuracy, the exponential averaging method is used to extract trend and seasonal components from raw wind power data.

[0025] S11. Determine the trend component: Use the exponential averaging method to smooth the original wind power data and extract the trend component of the data. The exponential averaging method can effectively capture the long-term trend of the data by weighted averaging the historical data. The calculation method is as follows: ; In the formula, and They represent the trend components of wind power data at the current time t and the previous time t-1 respectively; Represents the original wind power data at the current time t; is the smoothing coefficient, and its value range is .

[0026] S12. Extract seasonal components: By subtracting the trend component from the original data, the remaining part, namely the seasonal component, is obtained. The seasonal component reflects the cyclical fluctuation characteristics of the data. The calculation method is as follows: ; In the formula, Represents the seasonal component of wind power data at the current time t.

[0027] S13. Output trend and seasonal components: Finally, output trend components and seasonal components as inputs for subsequent prediction models. Trend components are used to capture long-term changes in data, and seasonal components are used to capture cyclical fluctuations in data. The specific functions are as follows: ; In the formula, represents the original signal; Indicates trend component; Indicates seasonal component.

[0028] S2. The extracted trend and seasonal components are predicted by constructing an extended long short-term memory neural network. The extended long short-term memory neural network is an improved long short-term memory network. By introducing an exponential gating mechanism and an enhanced memory structure, it can better handle long-term dependencies and improve the memory capacity of the model. The extended long short-term memory neural network includes two variants: stable long short-term memory network and matrix long short-term memory network, which are suitable for different task requirements.

[0029] S21. Stable LSTM Network: Stable LSTM Network enhances the stability and memory capacity of traditional LSTM network by introducing exponential gating and normalization mechanism. Its core improvements include using exponential function to replace traditional Sigmoid function, enhancing the nonlinear expression ability of gating mechanism; To stabilize the update of cell state and avoid gradient disappearance or explosion problems; by introducing a stable state , ensuring the numerical stability of the input gate and the forget gate.

[0030] In a stable LSTM neural network, the relationship between input and cell state is as follows: ; ; In the formula, and Respectively represent the cell status at the current moment and the previous moment; , They are forget gate and input gate respectively; For cell input, it controls what is input to the cell state; is the input vector; is the tanh activation function; is the weight matrix corresponding to the input of the cell state; is the weight matrix of the cell state corresponding to the hidden state at the previous moment; The hidden state at the last moment; For bias.

[0031] The specific functions of the input gate, forget gate, and output gate are as follows: ; ; ; In the formula, , , They correspond to the three gating units: input gate, forget gate, and output gate; represents the exponential function; Represents the Sigmoid function; , , Respectively represent the weight vectors corresponding to the inputs of the three gating units; , , Respectively represent the weight vectors of the three gating units corresponding to the hidden state of the previous moment; , , Respectively represent the biases of the three gating units; Represents the current input vector; Indicates the hidden state at the last moment.

[0032] For the hidden state, in order to avoid the numerical overflow problem of the exponential function, the normalized state is introduced The specific function of updating the stable hidden state is as follows: ; ; In the formula, represents the hidden state at the current time t; , Respectively represent the normalized states at the current time t and the previous time t-1; Indicates cell status; , , They correspond to the three gating units: input gate, forget gate, and output gate.

[0033] In order to further stabilize the exponential gating function, the input gate and the forget gate are combined into a stabilized state, and its specific function is as follows: ; ; ; In the formula, , Indicates the stable state at the current time t and the previous time t-1; , Corresponding to the original input gate and forget gate; , Represents the input gate and forget gate after scaling to the stabilized state.

[0034] S22. Matrix Long Short-Term Memory Network: The matrix long short-term memory network significantly improves the memory capacity and parallel computing capability of the model by introducing matrix memory and covariance update mechanism. Its core improvements include using matrices to replace traditional scalar cell states, which can store more complex state information; efficiently updating and retrieving information in memory through query, key and value mechanisms; removing the connection between hidden states, so that the calculation of the matrix long short-term memory network can be parallelized, significantly improving the training and reasoning efficiency.

[0035] In the matrix long short-term memory neural network, the relationship between input and cell state is as follows: ; In the formula, and Respectively represent the matrix cell states at the current moment and the previous moment; , They are forget gate and input gate respectively; , are the key vector and value vector at the current time t respectively.

[0036] In the matrix long short-term memory neural network, the query, key, and value mechanisms are introduced to improve the memory capacity of the model. The specific calculation function is as follows: ; ; ; In the formula, , , Represent query vector, key vector, and value vector respectively; , , The weight matrices representing query, key, and value respectively; , , Represents the bias items of query, key, and value respectively; represents the input vector; Represents the model dimension used to scale the values ​​of the key vector.

[0037] In the matrix LSTM neural network, the calculation method of the input gate, forget gate, and output gate is slightly different from that of the stable LSTM neural network. The connection between the hidden states is removed to make it parallelized. The specific function is as follows: ; ; ; In the formula, , , They correspond to the three gating units: input gate, forget gate, and output gate; represents the exponential function; Represents the Sigmoid function; , , Respectively represent the weight vectors corresponding to the inputs of the three gating units; , , Respectively represent the biases of the three gating units; Represents the current input vector.

[0038] Similar to the stable long short-term memory neural network, the matrix long short-term memory neural network also uses the value of the standardized stable exponential gating function. The specific function is as follows: ; ; ; In the formula, represents the hidden state at the current time t; , Respectively represent the normalized states at the current time t and the previous time t; Indicates cell status; , , They correspond to the three gating units: input gate, forget gate, and output gate; represents the normalized hidden state.

[0039] S23. Constructing an extended LSTM neural network: The extended LSTM neural network combines different layers of stable LSTM neural network and matrix LSTM neural network modules, making full use of the advantages of both, so that the model can process local features and global features at the same time, improving the prediction performance of the model. Assuming that the model stacks i layers of stable LSTM neural network and j layers of matrix LSTM neural network, the calculation process can be expressed as follows: ; ; ; In the formula, , represents the hidden state of the k-th layer and the k-1-th layer of the stable long short-term memory neural network at time t, ; represents the hidden state of the k-th layer of the stable long short-term memory neural network at the previous time t-1; , represents the hidden state of the l-th layer and l-1-th layer matrix long short-term memory neural network at time t, ; Represents the hidden state of the l-th layer matrix long short-term memory neural network at the previous time t-1; Represents the final output of the model at time t; , Represents the weight matrix and bias term of the output layer; , It represents the calculation process of the stable long short-term memory neural network and the matrix long short-term memory neural network, as shown in step S21 and step S22.

[0040] S3. Integrate the prediction results of the extended long short-term memory neural network under different combinations to obtain the final wind power prediction value.

[0041] Integrate the prediction results of the extended long short-term memory neural network under different combinations: In order to further improve the accuracy and robustness of wind power prediction, this method integrates the prediction results of four different combinations of extended long short-term memory neural networks, and obtains the final wind power prediction value by weighted integration of the outputs of these combinations. The specific calculation formula is as follows: ; In the formula, represents the final wind power forecast value, which is the integrated output at the current time t; Indicates the predicted output under the mth combination mode, there are 4 combinations in total: among them, when m=1, it means that the model is composed entirely of stable long short-term memory neural networks; when m=2, it means that the model is composed entirely of matrix long short-term memory neural networks; when m=3, it means that the first half of the model is a stable long short-term memory neural network, and the second half is a matrix long short-term memory neural network; when m=4, it means that the first half of the model is a matrix long short-term memory neural network, and the second half is a stable long short-term memory neural network; Represents the weight coefficient of the mth combination, satisfying .

[0042] Therefore, the present invention adopts the above-mentioned offshore wind power prediction method based on extended long short-term memory neural network. The present invention significantly improves the accuracy and stability of offshore wind power prediction by extracting the trend and seasonal components of offshore wind power data and combining the efficient modeling ability of extended long short-term memory neural network. The present invention can more accurately capture the complex changing characteristics of offshore wind power, effectively handle long-term dependencies, and improve the stability and robustness of the model. At the same time, it provides the power system with more accurate offshore wind power prediction results, helps dispatchers adjust power generation plans in advance, ensures stable power grid operation, and helps power generation companies optimize the operation and maintenance strategies of offshore wind farms, improve equipment utilization, and reduce operating costs. In addition, the present invention promotes the sustainable development of the offshore wind power industry and helps the transition to clean energy by improving the predictability and economy of offshore wind power.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An offshore wind power prediction method based on an extended long short-term memory neural network is characterized in that: The following steps are involved: S1. Extract trend and seasonal components from raw wind power data; S2. Predict the extracted trend and seasonal components by constructing an extended long short-term memory neural network, which includes two variants: a stable long short-term memory network and a matrix long short-term memory network; S3. Integrate the prediction results of the extended long short-term memory neural network under different combinations to obtain the final wind power prediction value.

2. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 1 is characterized in that: S1 includes the following steps: S11. Determine the trend component: Use the exponential average method to smooth the original wind power data and extract the trend component of the data. The calculation method is as follows: ； In the formula, and They represent the trend components of wind power data at the current time t and the previous time t-1 respectively; Represents the original wind power data at the current time t; is the smoothing coefficient, and its value range is ; S12. Extract seasonal components: By subtracting the trend component from the original data, the remaining part, i.e., the seasonal component, is obtained. The calculation method is as follows: ； In the formula, represents the seasonal component of wind power data at the current time t; S13, finally output trend component and seasonal component as input of subsequent prediction model, the specific function is as follows: ； In the formula, represents the original signal; Indicates trend component; Indicates seasonal component.

3. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 2 is characterized in that: S2 includes the following steps: S21, stable long short-term memory network uses exponential function instead of traditional Sigmoid function, through normalization state To stabilize the cell state update, by introducing a stabilized state Ensure the numerical stability of the input gate and forget gate; S22, Matrix Long Short-Term Memory Network uses matrices to replace traditional scalar cell states, introduces query, key and value mechanisms, and removes connections between hidden states; S23. Construct an extended long short-term memory neural network by combining stable long short-term memory neural networks and matrix long short-term memory neural network modules with different numbers of layers.

4. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 3 is characterized in that: In S21, the normalized state To stabilize the update of cell state, the specific function is as follows: ； ； In the formula, represents the hidden state at the current time t; , Respectively represent the normalized states at the current time t and the previous time t-1; Indicates cell status; , , They correspond to the three gating units: input gate, forget gate, and output gate.

5. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 4 is characterized in that: In S21, by introducing the stable state , to ensure the numerical stability of the input gate and the forget gate, the specific function is as follows: ； ； ； In the formula, , Indicates the stable state at the current time t and the previous time t-1; , Corresponding to the original input gate and forget gate; , Represents the input gate and forget gate after scaling in the stabilized state; Represents an exponential function.

6. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 5 is characterized in that: S22 uses matrices to replace traditional scalar cell states. The relationship between input and cell state is as follows: ； In the formula, and Respectively represent the matrix cell states at the current moment and the previous moment; , They are forget gate and input gate respectively; , are the key vector and value vector at the current time t respectively.

7. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 6 is characterized in that: In the matrix long short-term memory neural network in S22, the query, key, and value mechanisms are introduced to improve the memory capacity of the model. The specific calculation function is as follows: ； ； ； In the formula, , , Represent query vector, key vector, and value vector respectively; , , The weight matrices representing query, key, and value respectively; , , Represents the bias items of query, key, and value respectively; represents the input vector; Represents the model dimension used to scale the values ​​of the key vector.

8. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 7 is characterized in that: In S22, the input gate, forget gate, and output gate remove the connection between hidden states. The specific functions are as follows: ； ； ； In the formula, , , They correspond to the three gating units: input gate, forget gate, and output gate; represents the exponential function; Represents the Sigmoid function; , , Respectively represent the weight vectors corresponding to the inputs of the three gating units; , , Respectively represent the biases of the three gating units; Represents the current input vector.

9. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 8 is characterized in that: The model in S23 stacks i layers of stable long short-term memory neural network and j layers of matrix long short-term memory neural network, and the calculation process can be expressed as follows: ； ； ； In the formula, , represents the hidden state of the k-th layer and the k-1-th layer of the stable long short-term memory neural network at time t, ; represents the hidden state of the k-th layer of the stable long short-term memory neural network at the previous time t-1; , represents the hidden state of the l-th layer and l-1-th layer matrix long short-term memory neural network at time t, ; Represents the hidden state of the l-th layer matrix long short-term memory neural network at the previous time t-1; Represents the final output of the model at time t; , Represents the weight matrix and bias term of the output layer; , It represents the calculation process of the stable long short-term memory neural network and the matrix long short-term memory neural network, as shown in step S21 and step S22.

10. The offshore wind power prediction method based on extended long short-term memory neural network according to claim 9 is characterized in that: In S3, the final wind power forecast value is obtained by weighted integration of the outputs of these combinations. The specific calculation formula is as follows: ； In the formula, represents the final wind power forecast value, which is the integrated output at the current time t; It represents the predicted output under the mth combination. There are 4 combinations in total.

Images