Wind power prediction method based on Informer combination model

Abstract

The invention relates to a wind power prediction method based on an Informer combination model, and the method specifically comprises the steps: S1, employing a random forest model to screen original meteorological factors, such as wind speed, wind direction, pressure, and the like, and obtaining factors which play a key role in a task; s2, VMD parameters needing to be optimized are determined, and parameters in a VMD algorithm are optimized and dynamically adjusted through a dung beetle optimization algorithm; s3, analyzing an optimization result, and decomposing the wind power time sequence into a plurality of connotation mode components through a VMD algorithm; s4, performing VMD decomposition by using the optimized parameters, and training an Informer wind power prediction model by taking the decomposed IMFs as input features; and S5, performing multi-step prediction on each IMF through Informer, and adding IMF components obtained by prediction to obtain a final wind power sequence prediction result. The method has the advantages that multiple meteorological factors are comprehensively considered based on the combined model, and prediction accuracy and stability are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system safety management, and in particular relates to a wind power prediction method based on an Informer combination model. Background Art

[0002] As global climate change intensifies, wind energy, as an important component of clean energy, continues to increase its proportion in the energy structure. The instability and intermittency of wind power have brought challenges to the stable operation of the power grid, especially under extreme weather conditions. The vulnerability of the wind power system is fully exposed, posing a serious threat to the stability of the power grid and the sustainable development of the social economy. In addition, wind farms are located in diverse geographical locations, from coastal to inland, from plains to mountainous areas. The climatic conditions in different regions pose different challenges to the operation of wind power equipment. For example, wind farms in coastal areas may face the threat of typhoons, while wind farms in inland areas may be affected by sandstorms. These extreme weather events not only damage the physical structure of wind power equipment, but may also cause drastic fluctuations in wind power output, affecting the power grid. stable operation; therefore, in the context of wind power increasingly becoming an important part of power supply, studying and improving the accuracy of wind power prediction is crucial to maintaining the stability and reliability of the power grid. It is urgent to develop advanced wind power prediction models to adapt to changing environmental conditions and improve the disaster resistance and prediction accuracy of wind power systems. This can not only reduce the impact of extreme weather on wind power systems, but also provide grid operators with more accurate power supply plans and ensure the stable operation of power systems. It has important scientific and practical significance for promoting the sustainable development of renewable energy; therefore, it is very necessary to provide a wind power prediction method based on an Informer combined model that comprehensively considers multiple meteorological factors and improves prediction accuracy and stability. Summary of the invention

[0003] 1. Technical issues

[0004] In view of the above-mentioned existing technical status, this application mainly addresses the following technical problems:

[0005] 1. Under extreme weather conditions, the vulnerability of wind power systems is fully exposed, posing a serious threat to the stability of the power grid and the sustainable development of the social economy;

[0006] 2. How to improve the accuracy of wind power forecasting and maintain the stability and reliability of the power grid;

[0007] 3. How to develop advanced wind power prediction models to adapt to changing environmental conditions and improve the disaster resistance and prediction accuracy of wind power systems.

[0008] (II) Technical solution

[0009] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a wind power prediction method based on an Informer combined model that comprehensively considers multiple meteorological factors and improves prediction accuracy and stability.

[0010] The object of the present invention is achieved by: a wind power prediction method based on an Informer combination model, the method comprising the following steps:

[0011] Step 1: Use the random forest model to screen the original meteorological factors of wind speed, wind direction, and pressure to obtain the factors that play a key role in the task;

[0012] Step 2: Determine the VMD parameters that need to be optimized, and optimize and dynamically adjust the parameters in the VMD algorithm through the dung beetle optimization algorithm;

[0013] Step 3: Analyze the optimization results and decompose the wind power time series into multiple intrinsic modal components through the VMD algorithm;

[0014] Step 4: Use the optimized parameters to perform VMD decomposition, use the decomposed IMFs as input features, and train the Informer wind power prediction model;

[0015] Step 5: Perform multi-step forecasts on each IMF through Informer, and add the predicted IMF components to obtain the final wind power series forecast result.

[0016] Furthermore, the step 1 specifically includes the following steps:

[0017] Step 1.1: Obtain historical meteorological statistics of the area where the wind farm is located;

[0018] Step 1.2: Use the random forest model to screen the initial factors to exclude factors with low correlation. The importance calculation formula of the feature in node m is: Where: is the Gini index for each feature; f Gini,m is the Gini index in node m; f Gini,l is the Gini index in node l; f Gini,r Represents the Gini index in node r.

[0019] Furthermore, the step 2 is specifically as follows: optimizing two parameters, namely the total number of decomposition patterns and the quadratic penalty coefficient in the VMD algorithm by using the dung beetle optimization algorithm.

[0020] Furthermore, the step 3 is specifically: extracting each modal component of the original signal by variational modal decomposition, as follows:

[0021] Step 3.1: Use the envelope spectrum analysis method to obtain the modal functions u of the original signal k (t) envelope spectrum, construct correction coefficient Adjust the mode functions to their respective center frequencies w k On, such as: Where: h(t) is the demodulated signal; δ(t) is the unit pulse function; u k (t) is the kth mode function;

[0022] Step 3.2: Calculate the bandwidth of each modal function through the above formula and construct the corresponding constrained variational problem, such as: Where: k is the center frequency of the kth mode function; is the unit pulse function; s(t) is the original signal;

[0023] Step 3.3: By introducing the Lagrange multiplier operator and the penalty factor τ(t), the constrained variational problem is transformed into an unconstrained variational problem, which can be expressed as: Where: L is the Lagrangian function; α is the penalty factor; τ(t) is the Lagrangian multiplier operator;

[0024] Step 3.4: Use the alternating direction multiplier method to continuously update each component and its center frequency, and finally obtain the saddle point of the unconstrained model, that is, the optimal solution: Where: ω is the frequency; is the frequency domain representation of the kth modal component in the n+1th iteration; is the frequency domain representation of the center frequency of the kth modal component in the n+1th iteration; represents the frequency domain representation of the Lagrange multiplier after the n+1th iteration; is the Fourier transform of the original signal s(t) in the frequency domain; τ is the step size or update coefficient of the Lagrange multiplier; α is the penalty factor; is the frequency domain representation of the Lagrange multiplier τ(t) in the nth iteration.

[0025] Furthermore, the step 4 is specifically as follows: decomposing the original data by DBO-VMD, taking the IMF set obtained by decomposition as the prediction target sequence, and performing multi-step prediction on each IMF by Informer, which is as follows: taking the factor set screened by RF as the input variable, taking the IMF set obtained by DBO-VMD decomposition as the prediction target sequence, and performing multi-step prediction on each IMF by Informer, wherein the Informer model consists of an encoder and a decoder, and the encoder formula is as follows: Where: Q is the query vector; K is the key vector; V is the value vector; d is the scaling factor; the decoder formula is as follows: Where: Decoding of the target sequence; For historical sequence; is the placeholder for the target sequence; C is the sequence connection function.

[0026] Furthermore, the step 5 is specifically as follows: taking the factor set screened by RF as the input variable, the IMF set obtained by DBO-VMD decomposition as the prediction target sequence, and performing multi-step prediction on each IMF through Informer, as follows:

[0027] Step 5.1: Accumulate the independent mode function (IMF) results predicted by the Informer model to construct a complete wind power prediction sequence;

[0028] Step 5.2: By mining the long-term dependence and periodic changes in the wind power series, the prediction accuracy is improved and the effectiveness of the model is verified. Here, the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are used to evaluate the performance of the prediction model: Where: is the predicted value; i is the true value; n is the total number of samples; E RMSE is the root mean square error; E MAE is the mean absolute error; E MAPE is the mean absolute percentage error.

[0029] A wind power multi-step prediction method adopts the wind power prediction method based on the Informer combination model as described above, including a construction method of improved variational mode decomposition and Informer combination model, and a wind power prediction method based on the combination model.

[0030] Furthermore, the construction of the improved variational mode decomposition and Informer combined model is to decompose the wind power prediction problem into three steps: feature screening, signal decomposition and multi-step prediction, and optimize the model parameters through machine learning and deep learning techniques.

[0031] Furthermore, the wind power prediction method includes using a random forest model to screen key meteorological factors, using a variational mode decomposition algorithm improved by a dung beetle optimization algorithm to perform signal decomposition, and applying an Informer model to perform a multi-step prediction calculation method.

[0032] (III) Beneficial effects

[0033] 1. The present invention realizes the evaluation and optimization of distribution network disasters, and provides more accurate theoretical support for the power company's power facility construction tasks and operation and maintenance work;

[0034] 2. While reducing the adverse impact of power accidents caused by extreme weather on the social economy, the present invention improves the safety construction and safety management level of distribution network infrastructure, ensures that the distribution network can operate safely, stably and efficiently, and has very forward-looking scientific research value. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flow chart of the wind power multi-step prediction based on the Informer combination model of the present invention.

[0036] Figure 2 It is a flowchart of the Dung Beetle Optimization Algorithm (DBO) optimization iteration of the present invention. DETAILED DESCRIPTION

[0037] The present invention relates to the technical field of power system security management, and in particular to a wind power multi-step prediction method based on an Informer combination model, which aims to improve wind energy utilization efficiency and sustainable development of power systems, and optimize the configuration and scheduling of power resources by accurately predicting wind power.

[0038] The present invention is further described below in conjunction with embodiments and / or drawings.

[0039] Example 1

[0040] like Figure 1-2 As shown, a wind power prediction method based on an Informer combination model comprises the following steps:

[0041] Step 1: Use the random forest model to screen the original meteorological factors of wind speed, wind direction, and pressure to obtain the factors that play a key role in the task;

[0042] The original meteorological factor data are from the local meteorological bureau, and the main factors include wind speed, wind direction, temperature, humidity, etc.

[0043] The historical data of the wind power prediction model is derived from the power output statistical data of the wind power equipment required to be evaluated provided by the wind farm;

[0044] The modal decomposition parameters are set according to the statistical characteristics of the wind power signal; the annual average power output of the equipment λ avg , as well as the influencing factors Fb and Fm of wind speed and wind direction can be directly obtained through the historical power output data of wind power equipment.

[0045] In the present invention, step 1 is actually a method for calculating the probability of failure of distribution network lines under extreme weather conditions, which determines basic evaluation information and includes the following steps:

[0046] Step 1.1: Obtain historical meteorological statistics of the area where the wind farm is located;

[0047] Step 1.2: Use the random forest model to screen the initial factors to exclude factors with low correlation. The importance calculation formula of the feature in node m is: Where: is the Gini index for each feature; f Gini,m is the Gini index in node m; f Gini,l is the Gini index in node l; f Gini,r Represents the Gini index in node r.

[0048] Step 2: Determine the VMD parameters that need to be optimized, and optimize and dynamically adjust the parameters in the VMD algorithm through the dung beetle optimization algorithm;

[0049] In the present invention, the two parameters of the total number of decomposition modes and the quadratic penalty coefficient in the VMD algorithm are optimized by the dung beetle optimization algorithm, which effectively solves the problem of under-decomposition or over-decomposition caused by unreasonable setting of parameters of the VMD algorithm, reduces noise data, increases the contained characteristic information, and improves the accuracy of wind power sequence prediction.

[0050] Step 3: Analyze the optimization results and decompose the wind power time series into multiple intrinsic modal components through the VMD algorithm;

[0051] In the present invention, variational modal decomposition extracts the modal components of the original signal, as follows:

[0052] Step 3.1: Use the envelope spectrum analysis method to obtain the modal functions u of the original signal k (t) envelope spectrum, construct correction coefficient Adjust the mode functions to their respective center frequencies w k On, such as: Where: h(t) is the demodulated signal; δ(t) is the unit pulse function; u k (t) is the kth mode function;

[0053] Step 3.2: Calculate the bandwidth of each modal function through the above formula and construct the corresponding constrained variational problem, such as: Where: k is the center frequency of the kth mode function; is the unit pulse function; s(t) is the original signal;

[0054] Step 3.3: By introducing the Lagrange multiplier operator and the penalty factor τ(t), the constrained variational problem is transformed into an unconstrained variational problem, which can be expressed as: Where: L is the Lagrangian function; α is the penalty factor; τ(t) is the Lagrangian multiplier operator;

[0055] Step 3.4: Use the alternating direction multiplier method to continuously update each component and its center frequency, and finally obtain the saddle point of the unconstrained model, that is, the optimal solution: Where: ω is the frequency; is the frequency domain representation of the kth modal component in the n+1th iteration; is the frequency domain representation of the center frequency of the kth modal component in the n+1th iteration; represents the frequency domain representation of the Lagrange multiplier after the n+1th iteration; is the Fourier transform of the original signal s(t) in the frequency domain; τ is the step size or update coefficient of the Lagrange multiplier; α is the penalty factor; is the frequency domain representation of the Lagrange multiplier τ(t) in the nth iteration.

[0056] Step 4: Use the optimized parameters to perform VMD decomposition, use the decomposed IMFs as input features, and train the Informer wind power prediction model;

[0057] In the present invention, the original data is decomposed by DBO-VMD, and the IMF set obtained by the decomposition is used as the prediction target sequence. Multi-step prediction is performed on each IMF through Informer, which is as follows: the factor set screened by RF is used as the input variable, the IMF set obtained by DBO-VMD decomposition is used as the prediction target sequence, and multi-step prediction is performed on each IMF through Informer, wherein the Informer model consists of an encoder and a decoder, and the encoder formula is as follows: Where: Q is the query vector; K is the key vector; V is the value vector; d is the scaling factor; the decoder formula is as follows: Where: Decoding of the target sequence; For historical sequence; is the placeholder for the target sequence; C is the sequence connection function.

[0058] Step 5: Perform multi-step forecasts on each IMF through Informer, and add the predicted IMF components to obtain the final wind power series forecast result.

[0059] In this invention, the factor set screened by RF is used as the input variable, the IMF set obtained by DBO-VMD decomposition is used as the prediction target sequence, and each IMF is predicted by Informer in multiple steps, as follows:

[0060] Step 5.1: Accumulate the independent mode function (IMF) results predicted by the Informer model to construct a complete wind power prediction sequence;

[0061] Step 5.2: By mining the long-term dependence and periodic changes in the wind power series, the prediction accuracy is improved and the effectiveness of the model is verified. Here, the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are used to evaluate the performance of the prediction model: Where: is the predicted value; i is the true value; n is the total number of samples; E RMSE is the root mean square error; E MAE is the mean absolute error; E MAPE is the mean absolute percentage error.

[0062] The present invention is a wind power prediction method based on an Informer combination model. In use, the present invention aims to provide a wind power multi-step prediction method. Based on the improved variational mode decomposition and the Informer combination model, the wind power signal is decomposed and predicted by comprehensively considering meteorological factors such as wind speed, wind direction, and pressure, so as to improve the accuracy and stability of wind power prediction. The method of the present invention includes a random forest model for screening original meteorological factors, a variational mode decomposition algorithm based on a dung beetle optimization algorithm, and an Informer model for multi-step prediction of wind power. In the construction of the Informer model, the RF screening The factor set is used as the input variable, the IMF set obtained by the variable mode decomposition improved by the dung beetle optimization algorithm is used as the prediction target sequence, and each IMF is predicted by Informer for multi-step prediction, and the predicted IMF components are added to obtain the final wind power sequence prediction result; the Informer model adopts the self-attention mechanism to deeply mine the historical data patterns in the wind power time series; by identifying and mining the hidden information in the wind power sequence, including long-term dependencies and periodic changes, the wind power prediction is further accurately performed; the present invention has the advantages of being based on a combined model, comprehensively considering multiple meteorological factors, and improving prediction accuracy and stability.

[0063] Example 2

[0064] like Figure 1-2 As shown, a wind power multi-step prediction method adopts the wind power prediction method based on the Informer combination model as described above, including a construction method of an improved variational mode decomposition and Informer combination model, and a wind power prediction method based on the combination model.

[0065] The construction of the improved variational mode decomposition and Informer combined model is to decompose the wind power prediction problem into three steps: feature screening, signal decomposition and multi-step prediction, and optimize the model parameters through machine learning and deep learning technology.

[0066] The wind power prediction method includes using a random forest model to screen key meteorological factors, using a variational mode decomposition algorithm improved by a dung beetle optimization algorithm to perform signal decomposition, and applying an informer model to perform a multi-step prediction calculation method.

[0067] The present invention is a wind power prediction method based on an Informer combination model. In use, the present invention aims to provide a wind power multi-step prediction method. Based on the improved variational mode decomposition and the Informer combination model, the wind power signal is decomposed and predicted by comprehensively considering meteorological factors such as wind speed, wind direction, and pressure, so as to improve the accuracy and stability of wind power prediction. The method of the present invention includes a random forest model for screening original meteorological factors, a variational mode decomposition algorithm based on a dung beetle optimization algorithm, and an Informer model for multi-step prediction of wind power. In the construction of the Informer model, the RF screening The factor set is used as the input variable, the IMF set obtained by the variable mode decomposition improved by the dung beetle optimization algorithm is used as the prediction target sequence, and each IMF is predicted by Informer for multi-step prediction, and the predicted IMF components are added to obtain the final wind power sequence prediction result; the Informer model adopts the self-attention mechanism to deeply mine the historical data patterns in the wind power time series; by identifying and mining the hidden information in the wind power sequence, including long-term dependencies and periodic changes, the wind power prediction is further accurately performed; the present invention has the advantages of being based on a combined model, comprehensively considering multiple meteorological factors, and improving prediction accuracy and stability.

Claims

1. A wind power prediction method based on an Informer combination model, characterized in that: The method comprises the following steps: Step 1: Use the random forest model to screen the original meteorological factors of wind speed, wind direction, and pressure to obtain the factors that play a key role in the task; Step 2: Determine the VMD parameters that need to be optimized, and optimize and dynamically adjust the parameters in the VMD algorithm through the dung beetle optimization algorithm; Step 3: Analyze the optimization results and decompose the wind power time series into multiple intrinsic modal components through the VMD algorithm; Step 4: Use the optimized parameters to perform VMD decomposition, use the decomposed IMFs as input features, and train the Informer wind power prediction model; Step 5: Perform multi-step forecasts on each IMF through Informer, and add the predicted IMF components to obtain the final wind power series forecast result.

2. The wind power prediction method based on the Informer combination model according to claim 1, characterized in that: The step 1 specifically comprises the following steps: Step 1.1: Obtain historical meteorological statistics of the area where the wind farm is located; Step 1.2: Use the random forest model to screen the initial factors to exclude factors with low correlation. The importance calculation formula of the feature in node m is: Where: is the Gini index for each feature; f Gini,m is the Gini index in node m; f Gini,l is the Gini index in node l; f Gini,r Represents the Gini index in node r.

3. The wind power prediction method based on the Informer combination model according to claim 1, characterized in that: The step 2 specifically includes: optimizing two parameters, the total number of decomposition patterns and the quadratic penalty coefficient in the VMD algorithm by using the dung beetle optimization algorithm.

4. The wind power prediction method based on the Informer combination model according to claim 1, characterized in that: The step 3 is specifically: extracting each modal component of the original signal by variational modal decomposition, as follows: Step 3.1: Use the envelope spectrum analysis method to obtain the modal functions u of the original signal k (t) envelope spectrum, construct correction coefficient Adjust the mode functions to their respective center frequencies w k On, such as: Where: h(t) is the demodulated signal; δ(t) is the unit pulse function; u k (t) is the kth mode function; Step 3.2: Calculate the bandwidth of each modal function through the above formula and construct the corresponding constrained variational problem, such as: Where: k is the center frequency of the kth mode function; is the unit pulse function; s(t) is the original signal; Step 3.3: By introducing the Lagrange multiplier operator and the penalty factor τ(t), the constrained variational problem is transformed into an unconstrained variational problem, which can be expressed as: Where: L is the Lagrangian function; α is the penalty factor; τ(t) is the Lagrangian multiplier operator; Step 3.4: Use the alternating direction multiplier method to continuously update each component and its center frequency, and finally obtain the saddle point of the unconstrained model, that is, the optimal solution: Where: ω is the frequency; is the frequency domain representation of the kth modal component in the n+1th iteration; is the frequency domain representation of the center frequency of the kth modal component in the n+1th iteration; represents the frequency domain representation of the Lagrange multiplier after the n+1th iteration; is the Fourier transform of the original signal s(t) in the frequency domain; τ is the step size or update coefficient of the Lagrange multiplier; α is the penalty factor; is the frequency domain representation of the Lagrange multiplier τ(t) in the nth iteration.

5. The wind power prediction method based on the Informer combination model according to claim 1, characterized in that: The step 4 is specifically as follows: decomposing the original data by DBO-VMD, taking the IMF set obtained by decomposition as the prediction target sequence, and performing multi-step prediction on each IMF by Informer, which is as follows: taking the factor set screened by RF as the input variable, taking the IMF set obtained by DBO-VMD decomposition as the prediction target sequence, and performing multi-step prediction on each IMF by Informer, wherein the Informer model consists of an encoder and a decoder, and the encoder formula is as follows: Where: Q is the query vector; K is the key vector; V is the value vector; d is the scaling factor; the decoder formula is as follows: Where: Decoding of the target sequence; For historical sequence; is the placeholder for the target sequence; C is the sequence connection function.

6. The wind power prediction method based on the Informer combination model according to claim 1, characterized in that: The step 5 is specifically as follows: taking the factor set screened by RF as the input variable, the IMF set obtained by DBO-VMD decomposition as the prediction target sequence, and performing multi-step prediction on each IMF through Informer, as follows: Step 5.1: Accumulate the independent mode function (IMF) results predicted by the Informer model to construct a complete wind power prediction sequence; Step 5.2: By mining the long-term dependence and periodic changes in the wind power series, the prediction accuracy is improved and the effectiveness of the model is verified. Here, the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are used to evaluate the performance of the prediction model: Where: is the predicted value; i is the true value; n is the total number of samples; E RMSE is the root mean square error; E MAE is the mean absolute error; E MAPE is the mean absolute percentage error.

7. A wind power multi-step prediction method, using the wind power prediction method based on the Informer combination model as claimed in any one of claims 1 to 6, characterized in that: It includes a method for constructing an improved variational mode decomposition and Informer combined model, and a wind power prediction method based on the combined model.

8. A wind power multi-step prediction method according to claim 7, characterized in that: The construction of the improved variational mode decomposition and Informer combined model is to decompose the wind power prediction problem into three steps: feature screening, signal decomposition and multi-step prediction, and optimize the model parameters through machine learning and deep learning technology.

9. A wind power multi-step prediction method according to claim 7, characterized in that: The wind power prediction method includes using a random forest model to screen key meteorological factors, using a variational mode decomposition algorithm improved by a dung beetle optimization algorithm to perform signal decomposition, and applying an informer model to perform a multi-step prediction calculation method.

Images