CNN-GRU-based wind power prediction method
Through the wind power power prediction method based on CNN-GRU, combined with CEEMD decomposition and feature extraction technology, the shortcomings of traditional prediction models in reducing wind power power volatility and extracting spatiotemporal features are solved, and higher prediction accuracy is achieved.
Patent Information
- Application Number
- CN202411635077.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-15
- Publication Date
- 2025-02-25
AI Technical Summary
Traditional wind power power prediction models cannot effectively reduce wind power power volatility, and it is difficult to extract the spatiotemporal and spatial characteristics and hidden features in the historical data of wind farms, resulting in low prediction accuracy.
The wind power power prediction method based on CNN-GRU is used to decompose the wind power power sequence through CEEMD to reduce volatility; the spatial features are extracted using the CNN model, and the GRU model extracts the temporal features to build a prediction model.
It improves the accuracy of wind power prediction, can more effectively reduce wind power power fluctuations, extract the spatiotemporal characteristics in wind farm data, and thus improve prediction accuracy.
Smart Images
Figure CN119518743A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of wind power prediction. Background Art
[0002] The existing wind power forecasting methods mainly include physical method, statistical method and combined forecasting method. The physical method relies on the digital weather prediction (NWP) system and wind farm information to predict wind power; the statistical method predicts according to the rules of historical power and meteorological data, which can be roughly divided into prediction methods based on time series extrapolation and prediction methods based on artificial intelligence (AI) algorithms. The existing mainstream combined forecasting methods include data processing part, prediction model part and result analysis part. Among them, the prediction models include time series prediction, neural network, nuclear extreme learning machine, etc. However, the traditional prediction model cannot reduce the volatility of wind power when predicting wind power, and it is difficult to fully extract the spatiotemporal characteristics and hidden features in the historical data of wind farms. The prediction accuracy is low. The above problems need to be solved urgently. Summary of the invention
[0003] The purpose of the present invention is to solve the problem that the traditional prediction model cannot reduce the volatility of wind power when predicting wind power, and it is difficult to fully extract the spatiotemporal characteristics and hidden features in the historical data of wind farms, resulting in low prediction accuracy. The present invention provides a wind power prediction method based on CNN-GRU.
[0004] A wind power prediction method based on CNN-GRU, the method includes the following steps:
[0005] The wind power, temperature and wind speed at each sampling time of the wind farm are collected, and the daily wind power sequence is decomposed using CEEMD to obtain a data set. The input data of each sample in the data set is the decomposed wind power sequence, temperature sequence and wind speed sequence corresponding to the historical date, and the output data of the sample is the actual wind power sequence corresponding to the predicted date.
[0006] Construct a CNN-GRU prediction model from the CNN model and the GRU model;
[0007] The CNN-GRU prediction model is trained through each sample in the data set, where the input data in the sample is used as the input of the CNN model, and the input data is used as the output of the GRU model; and the CNN model is used to extract the sequence spatial features of the input data, and then sent to the GRU model for sequence temporal feature extraction;
[0008] The trained CNN-GRU prediction model is used to predict the wind power series on the prediction date.
[0009] Preferably, it is also used to remove outliers in a data set.
[0010] Preferably, the elimination method is: clustering the wind power data using the clustering algorithm DBSCAN, and eliminating abnormal wind power data that is out of the cluster.
[0011] Preferably, the CNN model consists of two convolutional layers and a flattening operation, and each convolutional layer performs convolution and pooling operations.
[0012] Preferably, each convolution layer contains 64 convolution kernels and the convolution kernel size is 2.
[0013] Preferably, the GRU module includes multiple gated recurrent units to extract the temporal features of the sequence, and the outputs of all gated recurrent units are connected to the fully connected layer, and the wind power sequence prediction result is obtained by calculating the output of the fully connected layer.
[0014] Advantages of the present invention:
[0015] The present invention proposes a wind power prediction method based on CNN-GRU, which uses complementary ensemble empirical mode decomposition (CEEMD) to decompose the power sequence and reduce the volatility of the wind power sequence; selects features with strong correlation between wind speed, temperature and wind power as model input, then uses CNN model to extract spatial features, uses GRU model to extract temporal features, and finally completes wind power prediction, thereby improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 It is a schematic diagram of the principle of the wind power prediction method based on CNN-GRU described in the present invention. DETAILED DESCRIPTION
[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0018] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.
[0019] A preliminary analysis of the factors affecting wind power generation is conducted, and input features are determined. In order to determine the specific impact of input features on wind power, the Pearson correlation coefficient is used for analysis. This method accurately calculates the correlation between two variables X and Y, and finally selects features with strong correlation between wind speed, temperature and wind power as model inputs. The following implementation methods are provided, specifically:
[0020] Specific implementation method 1. Figure 1 The present embodiment describes a wind power prediction method based on CNN-GRU, which comprises the following steps:
[0021] The wind power, temperature and wind speed at each sampling time of the wind farm are collected, and the daily wind power sequence is decomposed using CEEMD to obtain a data set. The input data of each sample in the data set is the decomposed wind power sequence, temperature sequence and wind speed sequence corresponding to the historical date, and the output data of the sample is the actual wind power sequence corresponding to the predicted date.
[0022] Construct a CNN-GRU prediction model from the CNN model and the GRU model;
[0023] The CNN-GRU prediction model is trained through each sample in the data set, where the input data in the sample is used as the input of the CNN model, and the input data is used as the output of the GRU model; and the CNN model is used to extract the sequence spatial features of the input data, and then sent to the GRU model for sequence temporal feature extraction;
[0024] The trained CNN-GRU prediction model is used to predict the wind power series on the prediction date.
[0025] In this preferred embodiment, complementary set empirical mode decomposition (CEEMD) is used to decompose the power sequence to reduce the volatility of the wind power sequence; features with strong correlation between wind speed, temperature and wind power are selected as model inputs, and then the CNN model is used to extract spatial features, and the GRU model is used to extract temporal features. Finally, the wind power prediction is completed, thereby improving the prediction accuracy.
[0026] In specific applications, three common error evaluation indicators are used to evaluate the performance of the model, namely, mean absolute percentage error (MAPE), root mean square error (RMSE) and mean absolute error (MAE).
[0027] Furthermore, it is also used to remove abnormal values in the data set. Specifically, the removal method is: clustering the wind power data using the clustering algorithm DBSCAN, and removing abnormal wind power data that is out of the cluster.
[0028] In this preferred implementation manner, the abnormal data mainly includes: zero-value data, such as wind power less than 0; and missing values of data.
[0029] Furthermore, the CNN model consists of two convolutional layers and a flattening operation, and each convolutional layer performs convolution and pooling operations. Each convolutional layer contains 64 convolution kernels with a convolution kernel size of 2.
[0030] Furthermore, the GRU module contains multiple gated recurrent units to extract the temporal features of the sequence. The outputs of all gated recurrent units are connected to the fully connected layer. By calculating the output of the fully connected layer, the wind power sequence prediction result is obtained.
[0031] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.
Claims
1. A wind power prediction method based on CNN-GRU, characterized in that: The method comprises the following steps: The wind power, temperature and wind speed at each sampling time of the wind farm are collected, and the daily wind power sequence is decomposed using CEEMD to obtain a data set. The input data of each sample in the data set is the decomposed wind power sequence, temperature sequence and wind speed sequence corresponding to the historical date, and the output data of the sample is the actual wind power sequence corresponding to the predicted date. Construct a CNN-GRU prediction model from the CNN model and the GRU model; The CNN-GRU prediction model is trained through each sample in the data set, where the input data in the sample is used as the input of the CNN model, and the input data is used as the output of the GRU model; and the CNN model is used to extract the sequence spatial features of the input data, and then sent to the GRU model for sequence temporal feature extraction; The trained CNN-GRU prediction model is used to predict the wind power series on the prediction date.
2. The wind power prediction method based on CNN-GRU according to claim 1 is characterized in that: It is also used to remove outliers in the data set.
3. The wind power prediction method based on CNN-GRU according to claim 2 is characterized in that: The elimination method is: clustering the wind power data using the clustering algorithm DBSCAN, and eliminating the abnormal wind power data that is out of the cluster.
4. The wind power prediction method based on CNN-GRU according to claim 1 is characterized in that: The CNN model consists of two convolutional layers and a flattening operation, and each convolutional layer performs convolution and pooling operations.
5. The wind power prediction method based on CNN-GRU according to claim 4 is characterized in that: Each convolutional layer contains 64 convolution kernels with a kernel size of 2.
6. The wind power prediction method based on CNN-GRU according to claim 1, characterized in that: The GRU module contains multiple gated recurrent units to extract the temporal features of the sequence. The outputs of all gated recurrent units are connected to the fully connected layer. By calculating the output of the fully connected layer, the wind power sequence prediction result is obtained.