Short-term wind power output prediction method and device considering multi-dimensional factors
Through the improved Transformer model, combined with the multi-head causal attention mechanism and feed-forward neural network, a short-term wind power output prediction model is constructed, which solves the challenge of wind power instability to the power system, achieves accurate prediction of short-term wind power output, and supports electric power Stable operation of the system.
Patent Information
- Application Number
- CN202411370356.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-29
- Publication Date
- 2025-01-21
AI Technical Summary
The instability and discontinuity of wind power pose challenges to the stable operation of power systems, and it is difficult for existing technology to accurately predict short-term wind power output.
An improved Transformer model is used, combined with a multi-head causal attention mechanism and a feedforward neural network, to build a short-term wind power output prediction model. The model processes input vectors through a multi-head causal attention module and utilizes feedforward neural networks for further calculations, finally obtaining prediction results through residual connections and layer normalization.
It achieves accurate prediction of short-term wind power output, helps power production and dispatching agencies understand wind power output in advance, arrange dispatch plans reasonably, and ensure the stable operation of the power system.
Smart Images
Figure CN119340969A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind power prediction, and in particular to a method and a device for short-term wind power output prediction considering multi-dimensional factors. Background Art
[0002] With the continuous increase in global energy consumption, traditional primary energy such as coal and oil is decreasing, and the environmental pollution caused by energy use is becoming more and more serious. Therefore, the development and utilization of clean energy has become an important way to solve the energy crisis and environmental problems. Wind energy, as a clean and renewable energy, has received widespread attention. In recent years, wind power technology has become more mature, the capacity of wind power units and the scale of grid-connected wind farms have continued to expand, and the proportion of wind power in the total power generation of the power system has also increased year by year. However, the instability and intermittency of wind energy have brought new challenges to the stable operation of the power system. With the increase in wind power installed capacity, wind power grid connection has become an important part of the power system. In order to ensure the stable operation of the power system, it is necessary to accurately predict the output of wind power. Accurate wind power output prediction helps power production and dispatching agencies to understand the output of wind power in advance, so as to reasonably arrange the dispatching plan and ensure the stable operation of the power system. At the same time, the prediction results can also provide a basis for the optimal dispatching and resource allocation of the power system and reduce the system operation cost. Summary of the invention
[0003] The short-term wind power output forecasting method considering multi-dimensional factors includes the following steps:
[0004] Obtaining wind power output data corresponding to the historical time series of short-term wind power output and a sequence of related influencing factors; the related influencing factors include: wind speed, wind direction, and temperature data;
[0005] Constructing a short-term wind power output prediction model, and using the acquired wind power output data and a sequence of related influencing factors to train the prediction model, to obtain a trained short-term wind power output prediction model; wherein the short-term wind power output prediction model is obtained by improving the Transformer model;
[0006] The time period to be predicted is input into a pre-trained short-term wind power output prediction model to obtain short-term wind power output data at each time point in the time period to be predicted.
[0007] Furthermore, the short-term wind power output prediction model is obtained by improving the Transformer model, and the improved method is:
[0008] The short-term wind power output prediction model is connected to the feedforward neural network through a multi-head causal attention module, and the multi-head causal attention module is used to process the vector output by the feedforward neural network to obtain the output of the multi-head causal attention mechanism layer; the output of the multi-head causal attention mechanism is obtained through linear change and Softmax function, where the Softmax function is:
[0009]
[0010] Among them, d k is the gradient factor, and mask(·) represents the masking operation. The mask sets the upper triangular elements of the matrix (excluding the diagonal) to -∞, ensuring that the attention of a point can only be calculated through the elements before the point, preventing information leakage, and making the correlation calculation towards the historical direction; at the same time, it enhances / weakens the spatial characteristics of strong / weak correlation, which helps the model better focus on important information. The output of the multi-head causal attention mechanism is:
[0011]
[0012] Where: The matrices Q, K and V are the linear transformation weight matrices of the i-th head respectively; W o is the linear transformation weight matrix after concatenating multiple causal attention matrices; Concat is the concatenation of multiple Attention results; matrices Q, K, and V are the short-term wind power output forecast input matrices, which are mapped to three different vector matrices after three different linear transformations, namely the query matrix Q, the key matrix K, and the value matrix V:
[0013]
[0014] Where X is the sample input matrix; W Q , W K , W ν are the linear transformation weight matrices respectively;
[0015] After the multi-head causal attention mechanism calculates the short-term wind power output forecast sample, it enters a fully connected feedforward neural network layer for calculation, and then the output is:
[0016] F FN (Z) = max(0,ZW1+b1)W2+b2
[0017] Where: b1, b2 represent bias;
[0018] The two sub-layers of multi-head causal attention and feedforward neural network are encoded and normalized using residual connection to obtain the short-term wind power output forecast result:
[0019]
[0020] Where: s Out is the short-term wind power output in the sample; LN is layer normalization; u L , are the mean and variance respectively; α and β represent the gain and bias respectively; ε is the adjustment parameter.
[0021] Furthermore, since the Transformer model does not have a temporal relationship with the input sample sequence, the sine and cosine functions need to be used to positionally encode the input data. Each dimension of the positional encoding corresponds to a sine signal or a cosine signal, that is:
[0022] PE(p os ,2i)=sin(p os / 10000 2i / d )
[0023] PE(p os ,2i+1)=cos(p os / 10000 2i / d )
[0024] Where: p os is the position of the element in the short-term wind power output prediction sample; i is the dimension of the element in the short-term wind power output prediction sample; d is the dimension of the position vector in the model.
[0025] Furthermore, before training the prediction model using the acquired wind power output data corresponding to the historical time series of short-term wind power output and the relevant influencing factor series, it is necessary to perform data normalization, data smoothing and data set division.
[0026] Furthermore, the method for data normalization processing is:
[0027] Each data is processed using the following formula to convert the data into its logarithmic form in order to better meet the requirements of the model:
[0028]
[0029] Where: x is the original sequence data, and x' is the sequence data after logarithmic normalization.
[0030] Furthermore, the data smoothing method is:
[0031] First, a total of 2n+1 observations before and after the current moment are filtered and fitted with a k-1 order polynomial. For the observations at the current moment, the following formula is used for fitting:
[0032] x t=a0+a1*t+a2*t 2 +…+a k-1 *t k-1
[0033] Among them, x t is the t-th period series data, t is the time variable, a0, a1, a2, a k-1 is the equation parameter, k is the order;
[0034] Secondly, the predicted values of the previous and next moments can also be calculated using the above formula, resulting in a total of 2n+1 formulas, forming a matrix:
[0035]
[0036] Simplify the above matrix:
[0037] X (2n+1)×1 =T (2n+1)×k +A k×1 +E (2n+1)×1
[0038] The individual parameter subscripts indicate their respective dimensions;
[0039] Then, calculate the filtered value of the model:
[0040] P=T·A=T·(T trans ·T) -1 ·T trans ·X=B·X
[0041] Where A=(T trans ·T) -1 ·T trans ·X, superscript trans means transpose;
[0042] Finally, the relationship matrix between the filtered values and the observed values is calculated:
[0043] B=T·(T trans ·T) -1 ·T trans
[0044] Where n is the number of samples, ε is the error term, X is the matrix about x, T is the matrix about t, A is the matrix about a, E is the matrix about ε, P is the filter value, and B is the relationship matrix; the Savitzky-Golay filtering formula can effectively remove the noise in the original data set and make the original data plot curve smooth, which is conducive to avoiding overfitting or underfitting.
[0045] Furthermore, the method for dividing the data set is:
[0046] The k-fold cross-validation method is used to divide the data set into k folds, each fold is used as a test set in turn, and the remaining k-1 folds are used as training sets, and cross-validation is performed multiple times.
[0047] A short-term wind power output prediction device considering multi-dimensional factors includes:
[0048] A data acquisition module is used to acquire wind power output data corresponding to the historical time series of short-term wind power output and a sequence of related influencing factors; the related influencing factors include: wind speed, wind direction, and temperature data;
[0049] A prediction model building module is used to build a short-term wind power output prediction model, and train the prediction model using the acquired wind power output data and a sequence of related influencing factors to obtain a trained short-term wind power output prediction model; wherein the short-term wind power output prediction model is obtained by improving the Transformer model;
[0050] The short-term wind power output data prediction module is used to input the time period to be predicted into a pre-trained short-term wind power output prediction model to obtain the short-term wind power output data at each time point in the time period to be predicted.
[0051] A computing device comprising:
[0052] one or more processing units;
[0053] A storage unit for storing one or more programs,
[0054] Wherein, when the one or more programs are executed by the one or more processing units, the one or more processing units execute the short-term wind power output prediction method considering multi-dimensional factors as described above.
[0055] A computer-readable storage medium having a non-volatile program code executable by a processor, wherein when the computer program is executed by the processor, the steps of the short-term wind power output prediction method considering multi-dimensional factors as described above are implemented.
[0056] The advantages and positive effects of the present invention are:
[0057] The short-term wind power output intelligent prediction method of the present invention fully considers various factors that affect short-term wind power output, normalizes, smoothes and divides the original sequence data, uses a multi-head causal attention mechanism and a feedforward neural network to "self-learn" and "self-adjust" the training samples, and establishes a short-term wind power output prediction model based on an improved Transformer to obtain short-term wind power output prediction data that is as accurate as possible, thereby helping power production and dispatching agencies to understand the wind power output situation in advance, reasonably arrange dispatching plans, and ensure the stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] The technical solution of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments, but it should be understood that these drawings are designed only for explanation purposes and are not intended to limit the scope of the present invention. In addition, unless otherwise specified, these drawings are intended only to conceptually illustrate the structural configurations described herein and are not necessarily drawn to scale.
[0059] Figure 1 A schematic diagram of a process for training a short-term wind power output prediction model provided by an embodiment of the present invention;
[0060] Figure 2 A schematic diagram of a short-term wind power output prediction model based on an improved Transformer model provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0061] In order to make the purpose, technical scheme and advantages of the implementation of the present invention clearer, the technical scheme in the embodiment of the present invention will be described in more detail below in conjunction with the drawings in the embodiment of the present invention. In the drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present invention, not all of the embodiments. The embodiments described below with reference to the drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limitations on the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0062] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0063] Example 1
[0064] like Figure 1 , 2 As shown in FIG. 1 , the short-term wind power output forecasting method considering multi-dimensional factors includes the following steps:
[0065] Obtaining wind power output data corresponding to the historical time series of short-term wind power output and a sequence of related influencing factors; the related influencing factors include: wind speed, wind direction, and temperature data;
[0066] Constructing a short-term wind power output prediction model, and using the acquired wind power output data and a sequence of related influencing factors to train the prediction model, to obtain a trained short-term wind power output prediction model; wherein the short-term wind power output prediction model is obtained by improving the Transformer model;
[0067] The time period to be predicted is input into a pre-trained short-term wind power output prediction model to obtain short-term wind power output data at each time point in the time period to be predicted.
[0068] Specifically, the short-term wind power output prediction model is obtained by improving the Transformer model, and the improvement method is:
[0069] The short-term wind power output prediction model is connected to the feedforward neural network through a multi-head causal attention module, and the multi-head causal attention module is used to process the vector output by the feedforward neural network to obtain the output of the multi-head causal attention mechanism layer; the output of the multi-head causal attention mechanism is obtained through linear change and Softmax function, where the Softmax function is:
[0070]
[0071] Among them, d k is the gradient factor, and mask(·) represents the masking operation. The mask sets the upper triangular elements of the matrix (excluding the diagonal) to -∞, ensuring that the attention of a point can only be calculated through the elements before the point, preventing information leakage, and making the correlation calculation towards the historical direction; at the same time, it enhances / weakens the spatial characteristics of strong / weak correlation, which helps the model better focus on important information. The output of the multi-head causal attention mechanism is:
[0072]
[0073] Where: The matrices Q, K and V are the linear transformation weight matrices of the i-th head respectively; W o is the linear transformation weight matrix after concatenating multiple causal attention matrices; Concat is the concatenation of multiple Attention results; matrices Q, K, and V are the short-term wind power output forecast input matrices, which are mapped to three different vector matrices after three different linear transformations, namely the query matrix Q, the key matrix K, and the value matrix V:
[0074]
[0075] Where X is the sample input matrix; W Q , W K , W ν are the linear transformation weight matrices respectively;
[0076] After the multi-head causal attention mechanism calculates the short-term wind power output forecast sample, it enters a fully connected feedforward neural network layer for calculation, and then the output is:
[0077] F FN (Z) = max(0,ZW1+b1)W2+b2
[0078] Where: b1, b2 represent bias;
[0079] The two sub-layers of multi-head causal attention and feedforward neural network are encoded and normalized using residual connection to obtain the short-term wind power output forecast result:
[0080]
[0081] Where: s Out is the short-term wind power output in the sample; LN is layer normalization; u L , are the mean and variance respectively; α and β represent the gain and bias respectively; ε is the adjustment parameter.
[0082] In addition, since the Transformer model does not have a temporal relationship with the input sample sequence, the sine and cosine functions need to be used to positionally encode the input data. Each dimension of the positional encoding corresponds to a sine signal or a cosine signal, that is:
[0083] PE(p os ,2i)=sin(p os / 10000 2i / d )
[0084] PE(p os ,2i+1)=cos(p os / 10000 2i / d )
[0085] Where: p os is the position of the element in the short-term wind power output prediction sample; i is the dimension of the element in the short-term wind power output prediction sample; d is the dimension of the position vector in the model.
[0086] Before training the prediction model using the wind power output data corresponding to the historical time series of the acquired short-term wind power output and the relevant influencing factor series, it is necessary to perform data normalization, data smoothing and data set division; respectively as follows:
[0087] The method for data normalization processing is:
[0088] The data ranges of relevant influencing factors are different in scale. If there is a data with a very large scale, the impact of the data with a small scale on the model training is easily ignored, so normalization is required; data normalization is to process each data using the following formula to convert the data into its logarithmic form in order to better meet the requirements of the model:
[0089]
[0090] Where: x is the original sequence data, and x' is the sequence data after logarithmic normalization.
[0091] The method for data smoothing is:
[0092] The collected data often has burrs and noise, and the effect of directly using it for prediction and analysis is not good. The original sample data needs to be smoothed and preprocessed. The Savitzky-Golay filter formula can be used for data smoothing. Its main idea is to fit a polynomial near the data point and use pre-calculated coefficients to weighted average the adjacent data points. This can smooth the data while maintaining the shape of the data without introducing too much phase delay. The workflow is: First, select some parameters to configure the working mode of the Savitzky-Golay filter. These parameters include the number of fitting polynomials (usually 2 to 4) and the window size (usually an odd number). Secondly, according to the selected parameters, calculate the coefficients required for the Savitzky-Golay filter. These coefficients are pre-calculated and used to weighted average adjacent data points. Then, for each data point, the data points in the window (the number is the window size) are weighted averaged with the corresponding coefficients to generate a smoothed data sequence. Finally, boundary processing is performed to ensure that the boundary points can also be smoothed. The specific mathematical principles are as follows:
[0093] First, a total of 2n+1 observations before and after the current moment are filtered and fitted with a k-1 order polynomial. For the observations at the current moment, the following formula is used for fitting:
[0094] x t =a0+a1*t+a2*t 2 +…+a k-1 *t k-1
[0095] Among them, x t is the t-th period series data, t is the time variable, a0, a1, a2, a k-1 is the equation parameter and k is the order.
[0096] Secondly, the predicted values of the previous and next moments (such as t-1, t+1, t-2, t+2, etc.) can also be calculated using the above formula, resulting in a total of 2n+1 formulas, forming a matrix:
[0097]
[0098] Where n is the sample size and ε is the error term;
[0099] Simplify the above matrix:
[0100] X (2n+1)×1 =T(2n+1)×k +A k×1 +E (2n+1)×1
[0101] Where X is the matrix about x, T is the matrix about t, A is the matrix about a, E is the matrix about ε, and the subscripts of each parameter represent their respective dimensions;
[0102] Then, calculate the filtered value of the model:
[0103] P=T·A=T·(T trans ·T) -1 ·T trans ·X=B·X
[0104] Where A=(T trans ·T) -1 ·T trans ·X, the superscript trans indicates transposition, P is the filter value, and B is the relationship matrix.
[0105] Finally, the relationship matrix between the filtered values and the observed values is calculated:
[0106] B=T·(T trans ·T) -1 ·T trans
[0107] The Savitzky-Golay filtering formula can effectively remove the noise in the original data set and make the original data plot curve smooth, which is helpful to avoid overfitting or underfitting.
[0108] After data normalization and data smoothing, the data set needs to be divided into a training set and a test set before model training. The k-fold cross-validation method is used to divide the data set into k folds, and each fold is used as a test set in turn, and the remaining k-1 folds are used as training sets, and cross-validation is performed multiple times. The specific steps are: first, the original data set is divided into k non-overlapping subsets, each subset is called a fold. Secondly, for each fold i, the i-th fold is used as a test set, and the remaining k-1 folds are used as training sets. Then the model is trained on the training set and the performance is evaluated on the test set. Then, the performance indicators of the model in the i-th cross-validation, such as accuracy, precision, recall, etc., are calculated. Finally, the above process is repeated until each fold is used as a test set, thereby completing k cross-validations, and the performance indicators of each cross-validation are averaged to obtain the average performance estimate of the model.
[0109] Through the above-mentioned data normalization, data smoothing and data set division methods, the data is converted into a format suitable for model input, a Transformer model suitable for the task is selected, the preprocessed wind speed, wind direction and temperature data are passed to the model as input variables, and the short-term wind power output is used as the output variable. The model is used to predict the short-term wind power output.
[0110] Example 2
[0111] This embodiment provides a short-term wind power output prediction device that considers multi-dimensional factors, including:
[0112] A data acquisition module is used to acquire wind power output data corresponding to the historical time series of short-term wind power output and a sequence of related influencing factors; the related influencing factors include: wind speed, wind direction, and temperature data;
[0113] A prediction model building module is used to build a short-term wind power output prediction model, and train the prediction model using the acquired wind power output data and a sequence of related influencing factors to obtain a trained short-term wind power output prediction model; wherein the short-term wind power output prediction model is obtained by improving the Transformer model;
[0114] The short-term wind power output data prediction module is used to input the time period to be predicted into a pre-trained short-term wind power output prediction model to obtain the short-term wind power output data at each time point in the time period to be predicted.
[0115] A computing device is also provided, comprising:
[0116] one or more processing units;
[0117] A storage unit for storing one or more programs,
[0118] Among them, when the one or more programs are executed by the one or more processing units, the one or more processing units execute the short-term wind power output prediction method considering multi-dimensional factors in this embodiment; it should be noted that the computing device may include but is not limited to a processing unit and a storage unit; those skilled in the art can understand that the computing device including a processing unit and a storage unit does not constitute a limitation on the computing device, and may include more components, or a combination of certain components, or different components. For example, the computing device may also include input and output devices, network access devices, buses, etc.
[0119] A computer-readable storage medium having a non-volatile program code executable by a processor, wherein when the computer program is executed by the processor, the steps of the short-term wind power output prediction method considering multi-dimensional factors in this embodiment are implemented; it should be noted that the readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above; the program contained on the readable medium can be transmitted using any appropriate medium, including, but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above. For example, the program code for performing the operation of the present invention can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also conventional procedural programming languages such as C language or similar programming languages. The program code can be executed entirely on a user computing device, partially on a user device, as an independent software package, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).
[0120] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A short-term wind power output forecasting method considering multi-dimensional factors, characterized in that: The steps include: Obtaining wind power output data corresponding to the historical time series of short-term wind power output and a sequence of related influencing factors; the related influencing factors include: wind speed, wind direction, and temperature data; Constructing a short-term wind power output prediction model, and using the acquired wind power output data and a sequence of related influencing factors to train the prediction model, to obtain a trained short-term wind power output prediction model; wherein the short-term wind power output prediction model is obtained by improving the Transformer model; The time period to be predicted is input into a pre-trained short-term wind power output prediction model to obtain short-term wind power output data at each time point in the time period to be predicted.
2. The short-term wind power output prediction method considering multi-dimensional factors according to claim 1 is characterized in that: The short-term wind power output prediction model is obtained by improving the Transformer model. The improvement method is: The short-term wind power output prediction model is connected to the feedforward neural network through a multi-head causal attention module, and the multi-head causal attention module is used to process the vector output by the feedforward neural network to obtain the output of the multi-head causal attention mechanism layer; the output of the multi-head causal attention mechanism is obtained through linear change and Softmax function, where the Softmax function is: Among them, d k is the gradient factor, and mask(·) represents the masking operation. The mask sets the upper triangular elements of the matrix (excluding the diagonal) to -∞, ensuring that the attention of a point can only be calculated through the elements before the point, preventing information leakage, and making the correlation calculation towards the historical direction; at the same time, it enhances / weakens the spatial characteristics of strong / weak correlation, which helps the model better focus on important information. The output of the multi-head causal attention mechanism is: Where: W i Q , W i K , W i V The matrices Q, K and V are the linear transformation weight matrices of the i-th head respectively; W o is the linear transformation weight matrix after concatenating multiple causal attention matrices; Concat is the concatenation of multiple Attention results; matrices Q, K, and V are the short-term wind power output forecast input matrices, which are mapped to three different vector matrices after three different linear transformations, namely the query matrix Q, the key matrix K, and the value matrix V: Among them, X is the sample input matrix; W Q , W K , W ν are the linear transformation weight matrices respectively; After the multi-head causal attention mechanism calculates the short-term wind power output forecast sample, it enters a fully connected feedforward neural network layer for calculation, and then the output is: F FN (Z)=max(0,ZW1+b1)W2+b2 Where: b1, b2 represent bias; The two sub-layers of multi-head causal attention and feedforward neural network are encoded and normalized using residual connection to obtain the short-term wind power output forecast result: Where: s Out is the short-term wind power output in the sample; LN is layer normalization; u L , are the mean and variance respectively; α and β represent the gain and bias respectively; ε is the adjustment parameter.
3. The short-term wind power output prediction method considering multi-dimensional factors according to claim 2 is characterized in that: Since the Transformer model has no temporal relationship with the input sample sequence, the sine and cosine functions are used to positionally encode the input data. Each dimension of the positional encoding corresponds to a sine signal or a cosine signal, that is: PE(p os ,2i)=sin(p os / 10000 2i / d ) PE(p os ,2i+1)=cos(p os / 10000 2i / d ) Where: p os is the position of the element in the short-term wind power output prediction sample; i is the dimension of the element in the short-term wind power output prediction sample; d is the dimension of the position vector in the model.
4. The short-term wind power output prediction method considering multi-dimensional factors according to claim 1 is characterized in that: Before training the prediction model using the acquired wind power output data corresponding to the historical time series of short-term wind power output and the relevant influencing factor series, it is necessary to perform data normalization, data smoothing and data set division.
5. The short-term wind power output prediction method considering multi-dimensional factors according to claim 4 is characterized in that: The method for data normalization processing is: Each data is processed using the following formula to convert the data into its logarithmic form in order to better meet the requirements of the model: Where: x is the original sequence data, and x' is the sequence data after logarithmic normalization.
6. The short-term wind power output prediction method considering multi-dimensional factors according to claim 4 is characterized in that: The method for data smoothing is: First, a total of 2n+1 observations before and after the current moment are filtered and fitted with a k-1 order polynomial. For the observations at the current moment, the following formula is used for fitting: x t =a0+a1*t+a2*t 2 +…+a k-1 *t k-1 Among them, x t is the t-th period series data, t is the time variable, a0, a1, a2, a k-1 is the equation parameter, k is the order; Secondly, the predicted values of the previous and next moments can also be calculated using the above formula, resulting in a total of 2n+1 formulas, forming a matrix: Simplify the above matrix: X (2n+1)×1 =T (2n+1)×k +A k×1 +E (2n+1)×1 The individual parameter subscripts indicate their respective dimensions; Then, calculate the filtered value of the model: P=T·A=T·(T trans ·T) -1 ·T trans ·X=B·X Where A=(T trans ·T) -1 ·T trans ·X, superscript trans means transpose; Finally, the relationship matrix between the filtered values and the observed values is calculated: B=T·(T trans ·T) -1 ·T trans Where n is the number of samples, ε is the error term, X is the matrix about x, T is the matrix about t, A is the matrix about a, E is the matrix about ε, P is the filter value, and B is the relationship matrix; the Savitzky-Golay filtering formula can effectively remove the noise in the original data set and make the original data plot curve smooth, which is conducive to avoiding overfitting or underfitting.
7. The short-term wind power output prediction method considering multi-dimensional factors according to claim 4 is characterized in that: The method of dividing the data set is: The k-fold cross-validation method is used to divide the data set into k folds, each fold is used as a test set in turn, and the remaining k-1 folds are used as training sets, and cross-validation is performed multiple times.
8. A short-term wind power output prediction device considering multi-dimensional factors, characterized in that: include: A data acquisition module is used to obtain the wind power output data corresponding to the short-term wind power output in the historical time series and the related influencing factor series; The relevant influencing factors include: wind speed, wind direction, and temperature data; A prediction model building module is used to build a short-term wind power output prediction model, and train the prediction model using the acquired wind power output data and a sequence of related influencing factors to obtain a trained short-term wind power output prediction model; wherein the short-term wind power output prediction model is obtained by improving the Transformer model; The short-term wind power output data prediction module is used to input the time period to be predicted into a pre-trained short-term wind power output prediction model to obtain the short-term wind power output data at each time point in the time period to be predicted.
9. A computing device, characterized in that: include: one or more processing units; A storage unit for storing one or more programs, Wherein, when the one or more programs are executed by the one or more processing units, the one or more processing units execute the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a non-volatile program code executable by a processor, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.