CEEMD-CNN-RF-based wind-solar power generation output prediction method and system

By decomposing and extracting features from wind and solar power generation data using the CEEMD-CNN-RF model, the problem of universality in wind and solar power generation prediction is solved, achieving high-precision wind and solar power output prediction and supporting the stable operation of multi-energy complementary systems.

CN118940900APending Publication Date: 2024-11-12SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411031725.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Existing technologies are insufficient for efficiently and universally predicting the output of wind and solar power, lacking versatility and practicality, leading to frequent instances of wind and solar power curtailment.

Method used

A CEEMD-CNN-RF-based approach is adopted, which decomposes wind and solar power generation data using CEEMD and combines convolutional neural networks and random forest algorithms for feature extraction and prediction, thereby constructing a wind and solar power generation output prediction model.

Benefits of technology

It improves the accuracy and stability of wind and solar power output prediction, providing a good foundation of data for the optimized scheduling of multi-energy complementary systems and ensuring the safe and stable operation of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940900A_ABST
    Figure CN118940900A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of new energy power generation output prediction, and discloses a CEEMD-CNN-RF-based wind-solar power generation output prediction method and system, and the method comprises the steps: obtaining wind-solar power generation power historical data, and carrying out the preprocessing and normalization processing of the wind-solar power generation power historical data, and obtaining an original time sequence; the method comprises the following steps: processing an original time sequence of wind-solar power generation power by adopting CEEMD, adding a random white noise sequence to the original time sequence, and decomposing to obtain a plurality of IMF components and residual trend terms; the CNN performs local feature extraction on the decomposed IMF components and residual trend terms through a convolutional neural network to obtain a new sequence; dividing a training set and a test set; inputting the training set in the obtained new sequence into the RF model, continuously adjusting the parameters of the RF model, learning and training on the training set, predicting by using the trained model, and verifying the prediction precision on the test set; and carrying out superposition reconstruction and normalized inverse transformation on the predicted data to obtain a final predicted value of the wind and light output.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of new energy prediction, and in particular to a method and system for predicting wind and solar power output based on CEEMD-CNN-RF. Background Art

[0002] Renewable energy, such as wind power and photovoltaics, can alleviate the current problems of increasing exhaustion of traditional fossil energy and continuous deterioration of the natural environment to a certain extent. However, wind power and photovoltaic power generation have the characteristics of volatility, randomness, and intermittency, which bring many challenges to the safe and stable operation of multi-energy complementary systems. In order to minimize the negative impact of wind power and photovoltaics on the cogeneration system and reduce the phenomenon of wind and photovoltaic abandonment, it is necessary to predict the wind and photovoltaic power generation power in advance.

[0003] At present, the methods used for wind and solar power generation prediction mainly include physical methods and statistical methods. The physical method mainly simulates the local wind speed and light intensity through physical models, but it is limited by geographical conditions and is not universal. The statistical method only needs a large amount of historical data and simulated data change rules to predict wind and solar power output, and has low requirements on geographical location. There are algorithms such as artificial neural networks, support vector machines, and fuzzy networks. Through literature search, it is found that the prior art (publication number: CN107392363A) discloses a method of using CEEMD decomposition technology to decompose the original wind power data and calculate the sample entropy of each modal component, and reconstruct the modes with similar sample entropy values ​​into new components, and finally establish a random forest model for each reconstructed component for prediction. The prior art (publication number: CN116151446A) proposes an ultra-short-term wind power prediction model based on CEEMD and CNN-LSTM, which uses CNN sequence to extract effective information and retain longer effective memory information to solve the gradient dispersion problem.

[0004] It can be seen that although the establishment of a combined prediction model of "decomposition-prediction-reconstruction" has improved the prediction accuracy and calculation speed to a certain extent, the data exploration is not sufficient, and the prediction accuracy of processing non-stationary and nonlinear sequence data is not high enough. In addition, the existing model has certain limitations and can only predict a single power generation method. There is a lack of prediction methods applicable to both wind power generation and photovoltaic power generation output. Therefore, it is necessary to design a prediction method that is highly universal and practical and has good results for both wind power generation and photovoltaic power generation. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a wind and solar power generation output prediction method based on CEEMD-CNN-RF to improve the accuracy and stability of new energy output prediction; the prediction results of the method of the present invention can provide good basic data for the optimization scheduling of the multi-energy complementary system, and further ensure the safe and stable operation of the multi-energy complementary system containing renewable energy.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides a method for predicting wind and solar power output based on CEEMD-CNN-RF, comprising the following steps:

[0008] Step 1: Obtain historical data of wind and solar power generation power, and preprocess and normalize the historical data of wind and solar power generation power to obtain the original time series;

[0009] Step 2: Use CEEMD to process the original time series of wind and solar power generation, add a random white noise sequence to the original time series, and decompose it to obtain several IMF components and residual trend terms;

[0010] Step 3: CNN uses convolutional neural network to extract local features of the decomposed IMF components and residual trend items to obtain a new sequence; then divide it into training set and test set;

[0011] Step 4: Input the training set in the new sequence into the RF model, and continuously adjust the RF model parameters, learn and train on the training set, use the trained model to make predictions, and verify the prediction accuracy on the test set;

[0012] Step 5: Overlay and reconstruct the predicted data and perform normalized inverse transformation to obtain the final predicted value of wind and solar power output.

[0013] As a further technical solution, the data preprocessing in step 1 includes inputting historical data of wind and solar power generation and cleaning the wind and solar power data, removing any abnormal data, and filling in null values ​​using mean calculation to obtain a complete wind and solar time series; then, the time series is normalized.

[0014] As a further technical solution, the CEEMD decomposition in step 2 includes adding N times of positive and negative opposite random white noise to the wind and solar power generation power P(t) to reduce modal aliasing, that is,

[0015]

[0016]

[0017] in, is the data to be decomposed after adding positive white noise; is the data to be decomposed after adding negative white noise; x(t) is the original signal; μ is the noise amplitude; are the positive and negative opposite random white noises added for the i-th time respectively;

[0018] The EMD method is used to decompose each pair of signals after adding white noise to obtain several IMF components and a residual trend term Re, and each component is averaged, that is,

[0019]

[0020]

[0021] Among them, IMF ij , IMF -ij are the jth IMF components obtained by decomposing the positive and negative white noises after adding them for the i-th time; Re ij , Re -ij are the jth residual trend term obtained by decomposing the positive and negative white noise after adding them for the i-th time.

[0022] As a further technical solution, in step 3, a one-dimensional convolutional neural network is constructed, including a convolutional layer, a maximum pooling layer and a Dropout layer; the convolutional neural network is used to extract features of the IMF sequence after CEEMD decomposition, and more representative new features are extracted through data conversion and dimensionality reduction. Then, the extracted features are merged to construct a new sequence, and the training set and test set are divided.

[0023] As a further technical solution, the RF prediction in step 4 includes inputting the training set in the new sequence constructed in step 3 into the RF model, regressing the data using multiple decision trees, and repeatedly binary dividing the data to reduce computational complexity; adjusting the RF model parameters, using the trained model for prediction through continuous learning and training, and verifying the prediction accuracy on the test set.

[0024] As a further technical solution, the data reconstruction in step 5 includes superimposing and reconstructing the IMF component predicted by the RF model in step 4 and the residual trend term decomposed by the original CEEMD to obtain a complete prediction sequence. The reconstructed prediction sequence is subjected to a normalized inverse transformation to obtain the final prediction value of the wind and solar power generation output.

[0025] In a second aspect, the present invention further provides a wind and solar power generation output prediction system based on CEEMD-CNN-RF, comprising:

[0026] A data preprocessing module is configured to obtain historical data of wind and solar power generation power, and preprocess and normalize the historical data of wind and solar power generation power to obtain an original time series;

[0027] A CEEMD decomposition module, which is configured to process the original time series of wind and solar power generation using CEEMD, add a random white noise sequence to the original time series, and decompose it to obtain several IMF components and residual trend terms;

[0028] CNN feature extraction module: CNN uses convolutional neural network to extract local features of the decomposed IMF components and residual trend items to obtain a new sequence; then divide it into training set and test set;

[0029] RF prediction module: input the training set in the new sequence into the RF model, continuously adjust the RF model parameters, learn and train on the training set, use the trained model to make predictions, and verify the prediction accuracy on the test set;

[0030] Data reconstruction module: The predicted data is superimposed and reconstructed and normalized inverse transformation is performed to obtain the final predicted value of wind and solar power output.

[0031] As a further technical solution, the CEEMD decomposition module is specifically configured to add N times of positive and negative opposite random white noise to the wind and solar power generation power P(t) to reduce modal aliasing, that is,

[0032]

[0033]

[0034] in, is the data to be decomposed after adding positive white noise; is the data to be decomposed after adding negative white noise; x(t) is the original signal; μ is the noise amplitude; are the positive and negative opposite random white noises added for the i-th time respectively;

[0035] The EMD method is used to decompose each pair of signals after adding white noise to obtain several IMF components and a residual trend term Re, and each component is averaged, that is,

[0036]

[0037]

[0038] Among them, IMF ij , IMF -ij are the jth IMF components obtained by decomposing the positive and negative white noises after adding them for the i-th time; Re ij, Re -ij are the jth residual trend term obtained by decomposing the positive and negative white noise after adding them for the i-th time.

[0039] As a further technical solution, the CNN feature extraction module is specifically configured to construct a one-dimensional convolutional neural network, including a convolution layer, a maximum pooling layer and a Dropout layer; the convolutional neural network is used to extract features of the IMF sequence after CEEMD decomposition, and more representative new features are extracted through data conversion and dimensionality reduction. Then, the extracted features are merged to construct a new sequence, and the training set and test set are divided.

[0040] As a further technical solution, the RF prediction module is specifically configured to input the training set in the new sequence constructed in step 3 into the RF model, regress the data using multiple decision trees, and repeatedly binary the data to reduce computational complexity; adjust the RF model parameters, use the trained model for prediction through continuous learning and training, and verify the prediction accuracy on the test set.

[0041] The beneficial effects of the present invention are:

[0042] The present invention is a wind and solar power output prediction model based on CEEMD-CNN-RF. First, CEEMD is used to decompose the original data of wind and solar power generation, and the wind and solar time series is stabilized and denoised and simplified. CNN is used to extract features from the decomposed components to obtain a new sequence. Then, the random forest algorithm is used to repeatedly divide the data into two parts for regression, so that the amount of calculation for training prediction is reduced. Finally, the lumped average is performed to obtain the final wind and solar power prediction result. Compared with other traditional models, a combined prediction model of "decomposition-feature extraction-prediction-reconstruction" is proposed, which effectively combines the adaptive decomposition ability of CEEMD and the feature extraction ability of CNN, fully explores the potential characteristics of wind and solar power generation data, and shows higher accuracy and robustness. Through actual wind and solar power generation data, it is verified that the present invention has strong versatility and practicality, and has good prediction effects on both wind power generation and photovoltaic power generation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic flow chart of the wind and solar power generation output prediction method based on CEEMD-CNN-RF of the present invention.

[0044] Figure 2 It is a structural diagram of the RF unit of the present invention.

[0045] Figure 3 The photovoltaic power generation prediction result predicted by the prediction algorithm provided by the embodiment of the present invention.

[0046] Figure 4This is a wind power generation prediction result predicted by the prediction algorithm provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further explained below. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the examples. 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.

[0048] Example 1

[0049] In view of the problems existing in the prior art, the present invention provides a method for predicting wind and solar power generation output based on CEEMD-CNN-RF. The present invention is described in detail below in conjunction with the accompanying drawings.

[0050] like Figure 1 As shown, the wind and solar power generation output prediction method based on CEEMD-CNN-RF provided in an embodiment of the present invention includes the following steps:

[0051] Step 1: Data preprocessing: Input historical data of wind and solar power generation, normalize the data, and divide it into training set and test set.

[0052] Step 2: CEEMD decomposition. CEEMD is used to process the original time series of wind and solar power generation, and a random white noise sequence is added to the original series to decompose it into several IMF components and residual trend terms.

[0053] Step 3: CNN feature extraction. CNN extracts local features of the decomposed sequence through convolutional layers and pooling layers to obtain a new sequence.

[0054] Step 4: RF prediction. Input the new sequence into the RF model, and use multiple decision trees to repeatedly divide the data into two parts for regression, so as to reduce the amount of calculation. Adjust the RF model parameters, and use the trained model for prediction through continuous learning and training to improve the prediction accuracy.

[0055] Step 5: Data reconstruction: The predicted data is reconstructed by superposition and normalized inverse transformation to obtain the final predicted value of wind and solar power output.

[0056] Furthermore, the data preprocessing in step 1 includes inputting historical data of wind and solar power generation and cleaning the wind and solar power data, removing any abnormal data, and using mean calculation to fill in empty values ​​to obtain a complete wind and solar time series. Then, the time series is normalized and divided into a training set and a test set.

[0057] Furthermore, the CEEMD decomposition in step 2 includes adding N times of positive and negative opposite random white noise to the wind and solar power generation power P(t) to reduce modal aliasing, that is,

[0058]

[0059]

[0060] in, is the data to be decomposed after adding positive white noise; is the data to be decomposed after adding negative white noise; x(t) is the original signal; μ is the noise amplitude; are the positive and negative opposite random white noises added for the i-th time respectively.

[0061] The EMD method is used to decompose each pair of signals after adding white noise to obtain several IMF components and a residual trend term Re, and each component is averaged, that is,

[0062]

[0063]

[0064] Among them, IMF ij , IMF -ij are the jth IMF components obtained by decomposing the positive and negative white noises after adding them for the i-th time; Re jj , Re -ij are the jth residual trend term obtained by decomposing the positive and negative white noise after adding them for the i-th time.

[0065] Furthermore, the CNN feature extraction in step 3 includes constructing a one-dimensional convolutional neural network, including a convolution layer, a maximum pooling layer and a Dropout layer. The convolutional neural network is used to extract features from the sequence after CEEMD decomposition, and more representative new features are extracted through data conversion and dimensionality reduction, and then the extracted features are merged to construct a new sequence.

[0066] Furthermore, the RF prediction in step 4 includes inputting the new sequence constructed in step 3 into the RF model, regressing the data using multiple decision trees, and repeatedly binary dividing the data to reduce the computational complexity. The RF model parameters are adjusted, and the trained model is used for prediction through continuous learning and training, and the prediction accuracy is verified on the test set to ensure that the model has good generalization ability.

[0067] Furthermore, the data reconstruction in step 5 includes superimposing and reconstructing the prediction results of the RF model obtained in step 4 and the residual trend terms of the CEEMD decomposition in step 3 to obtain a complete prediction sequence, and performing a normalized inverse transformation on the reconstructed prediction sequence to obtain the final prediction value of the wind and solar power generation output.

[0068] In order to further verify the performance of the proposed method, the error evaluation index is used to characterize the distribution of the prediction error and compare the prediction accuracy of the model. The prediction model evaluation indicators selected in this paper are mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and goodness of fit (R 2 ):

[0069]

[0070]

[0071]

[0072]

[0073] In the formula, y i is the actual power value at the i-th moment; is the power prediction value at the i-th moment; y i,ave is the average value of n power values.

[0074] Example 2

[0075] This embodiment is based on the wind-solar power generation output prediction method based on CEEMD-CNN-RF in Embodiment 1, and further provides a wind-solar power generation output prediction system based on CEEMD-CNN-RF, including:

[0076] A data preprocessing module is configured to obtain historical data of wind and solar power generation power, and preprocess and normalize the historical data of wind and solar power generation power to obtain an original time series;

[0077] A CEEMD decomposition module, which is configured to process the original time series of wind and solar power generation using CEEMD, add a random white noise sequence to the original time series, and decompose it to obtain several IMF components and residual trend terms;

[0078] CNN feature extraction module: CNN uses convolutional neural network to extract local features of the decomposed IMF components and residual trend items to obtain a new sequence; then divide it into training set and test set;

[0079] RF prediction module: input the training set in the new sequence into the RF model, continuously adjust the RF model parameters, learn and train on the training set, use the trained model to make predictions, and verify the prediction accuracy on the test set;

[0080] Data reconstruction module: The predicted data is superimposed and reconstructed and normalized inverse transformation is performed to obtain the final predicted value of wind and solar power output.

[0081] Furthermore, the CEEMD decomposition module is specifically configured to add N times of positive and negative opposite random white noise to the wind and solar power generation power P(t) to reduce modal aliasing, that is,

[0082]

[0083]

[0084] in, is the data to be decomposed after adding positive white noise; is the data to be decomposed after adding negative white noise; x(t) is the original signal; μ is the noise amplitude; are the positive and negative opposite random white noises added for the i-th time respectively;

[0085] The EMD method is used to decompose each pair of signals after adding white noise to obtain several IMF components and a residual trend term Re, and each component is averaged, that is,

[0086]

[0087]

[0088] Among them, IMF ij , IMF -ij are the jth IMF components obtained by decomposing the positive and negative white noises after adding them for the i-th time; Re ij , Re -ij are the jth residual trend term obtained by decomposing the positive and negative white noise after adding them for the i-th time.

[0089] Furthermore, the CNN feature extraction module is specifically configured to construct a one-dimensional convolutional neural network, including a convolution layer, a maximum pooling layer and a Dropout layer; use the convolutional neural network to extract features of the IMF sequence after CEEMD decomposition, extract more representative new features through data conversion and dimensionality reduction, and then merge the extracted features to construct a new sequence, and divide it into a training set and a test set.

[0090] Furthermore, the RF prediction module is specifically configured to input the training set in the new sequence constructed in step 3 into the RF model, regress the data using multiple decision trees, and repeatedly binary divide the data to reduce computational complexity; adjust the RF model parameters, perform predictions using the trained model through continuous learning and training, and verify the prediction accuracy on the test set.

[0091] Furthermore, the data reconstruction module is specifically configured to superimpose and reconstruct the IMF component predicted by the RF model in step 4 and the residual trend term decomposed by the original CEEMD to obtain a complete prediction sequence. The reconstructed prediction sequence is normalized and inversely transformed to obtain the final prediction value of wind and solar power generation output.

[0092] Example:

[0093] The main purpose of this embodiment is to verify the effectiveness of the prediction method of the present invention and the accuracy of the results. Taking the measured data of wind and solar power generation in a certain area of ​​Shandong Province as an example, the sampling time is 24 hours a day, the sampling interval is 1 hour, and a total of 2880 data from 120 days are selected as the sample data set. Divide the training set and the test set, and use the first 80% of the data as the training set and the remaining 20% ​​as the test set.

[0094] In order to further verify the rationality and applicability of the CEEMD-CNN-RF model proposed in this invention in wind and solar power generation prediction, four other models, CNN-LSTM model, RF model, CEEMD-CNN-LSTM model, and CEEMD-SVR model, are used here to make predictions respectively, and the prediction results are compared and analyzed. Tables 1 and 2 show the prediction errors of wind and solar power output of different models.

[0095] Table 1 Wind power prediction error

[0096]

[0097] Table 2 Photoelectric power prediction error

[0098]

[0099] It can be seen from Table 1 and Table 2 that the error MSE, RMSE, MAE, R 2 The values ​​are 821.69, 28.66, 19.48, and 0.96, respectively, while the MSE, RMSE, MAE, and R 2 The values ​​are 31.11, 5.57, 3.51, and 0.99 respectively. All evaluation indicators are better than the other four models, with better prediction performance and better approximation to the true value. In order to more intuitively compare the prediction effects of different methods, Figure 3 , Figure 4The fitting of wind power prediction curves and actual wind power output curves using different methods is shown. Figure 3 and Figure 4 Comparison of the predicted value curve and the actual operation value curve shows that in most time periods, the CEEMD-CNN-RF prediction model is in good agreement with the actual operation value, and better reflects the actual operation situation than the other four prediction models. In summary, the prediction results of the method of the present invention can provide good basic data for the optimization and scheduling of the multi-energy complementary system, and further ensure the safe and stable operation of the multi-energy complementary system containing renewable energy.

[0100] The wind-solar power generation output prediction method based on CEEMD-CNN-RF of the present invention can also include adopting GUI design to provide a user operation interface to facilitate users to use the present invention to perform wind-solar power generation prediction.

[0101] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A wind and solar power generation output prediction method based on CEEMD-CNN-RF, It is characterized in that The steps include: Step 1: Obtain historical data of wind and solar power generation power, and preprocess and normalize the historical data of wind and solar power generation power to obtain the original time series; Step 2: Use CEEMD to process the original time series of wind and solar power generation, add a random white noise sequence to the original time series, and decompose it to obtain several IMF components and residual trend terms; Step 3: CNN uses convolutional neural network to extract local features of the decomposed IMF components and residual trend items to obtain a new sequence; then divide it into training set and test set; Step 4: Input the training set in the new sequence into the RF model, and continuously adjust the RF model parameters, learn and train on the training set, use the trained model to make predictions, and verify the prediction accuracy on the test set; Step 5: Overlay and reconstruct the predicted data and perform normalized inverse transformation to obtain the final predicted value of wind and solar power output.

2. The wind and solar power generation output prediction method based on CEEMD-CNN-RF as claimed in claim 1, It is characterized in that The data preprocessing in step 1 includes inputting historical data of wind and solar power generation and cleaning the wind and solar power data, removing any abnormal data, and filling in empty values ​​using mean calculation to obtain a complete wind and solar time series; then, the time series is normalized.

3. The wind and solar power generation output prediction method based on CEEMD-CNN-RF as claimed in claim 1, It is characterized in that The CEEMD decomposition in step 2 includes adding N times of positive and negative opposite random white noise to the wind and solar power generation power P(t) to reduce modal aliasing, that is, in, is the data to be decomposed after adding positive white noise; is the data to be decomposed after adding negative white noise; x(t) is the original signal; μ is the noise amplitude; are the positive and negative opposite random white noises added for the i-th time respectively; The EMD method is used to decompose each pair of signals after adding white noise to obtain several IMF components and a residual trend term Re, and each component is averaged, that is, Among them, IMF ij , IMF -ij are the jth IMF components obtained by decomposing the positive and negative white noises after adding them for the i-th time; Re ij , Re -ij are the jth residual trend term obtained by decomposing the positive and negative white noise after adding them for the i-th time.

4. The wind and solar power generation output prediction method based on CEEMD-CNN-RF as claimed in claim 1, It is characterized in that In the step 3, a one-dimensional convolutional neural network is constructed, including a convolution layer, a maximum pooling layer and a Dropout layer; the convolutional neural network is used to extract features of the IMF sequence after CEEMD decomposition, and more representative new features are extracted through data conversion and dimensionality reduction. Then, the extracted features are merged to construct a new sequence, and the training set and the test set are divided.

5. The wind and solar power generation output prediction method based on CEEMD-CNN-RF as claimed in claim 1, It is characterized in that The RF prediction in step 4 includes inputting the training set in the new sequence constructed in step 3 into the RF model, regressing the data using multiple decision trees, and repeatedly binary dividing the data to reduce the computational complexity; Adjust the RF model parameters, use the trained model for prediction through continuous learning and training, and verify the prediction accuracy on the test set.

6. The wind and solar power generation output prediction method based on CEEMD-CNN-RF as claimed in claim 1, It is characterized in that The data reconstruction in step 5 includes superimposing and reconstructing the IMF component predicted by the RF model in step 4 and the residual trend term decomposed by the original CEEMD to obtain a complete prediction sequence. The reconstructed prediction sequence is normalized and inversely transformed to obtain the final prediction value of wind and solar power generation output.

7. A wind and solar power generation output prediction system based on CEEMD-CNN-RF, It is characterized in that include: A data preprocessing module is configured to obtain historical data of wind and solar power generation power, and preprocess and normalize the historical data of wind and solar power generation power to obtain an original time series; A CEEMD decomposition module, which is configured to process the original time series of wind and solar power generation using CEEMD, add a random white noise sequence to the original time series, and decompose it to obtain several IMF components and residual trend terms; CNN feature extraction module: CNN uses convolutional neural network to extract local features of the decomposed IMF components and residual trend items to obtain a new sequence; then divide it into training set and test set; RF prediction module: input the training set in the new sequence into the RF model, continuously adjust the RF model parameters, learn and train on the training set, use the trained model to make predictions, and verify the prediction accuracy on the test set; Data reconstruction module: The predicted data is superimposed and reconstructed and normalized inverse transformation is performed to obtain the final predicted value of wind and solar power output.

8. The wind and solar power generation output prediction system based on CEEMD-CNN-RF as claimed in claim 7, It is characterized in that The CEEMD decomposition module is specifically configured to add N times of positive and negative opposite random white noise to the wind and solar power generation power P(t) to reduce modal aliasing, that is, in, is the data to be decomposed after adding positive white noise; is the data to be decomposed after adding negative white noise; x(t) is the original signal; μ is the noise amplitude; are the positive and negative opposite random white noises added for the i-th time respectively; The EMD method is used to decompose each pair of signals after adding white noise to obtain several IMF components and a residual trend term Re, and each component is averaged, that is, Among them, IMF ij , IMF -ij are the jth IMF components obtained by decomposing the positive and negative white noises after adding them for the i-th time; Re ij , Re -ij are the jth residual trend term obtained by decomposing the positive and negative white noise after adding them for the i-th time.

9. The wind and solar power generation output prediction system based on CEEMD-CNN-RF as claimed in claim 7, It is characterized in that The CNN feature extraction module is specifically configured to construct a one-dimensional convolutional neural network, including a convolution layer, a maximum pooling layer and a Dropout layer; use the convolutional neural network to extract features of the IMF sequence after CEEMD decomposition, extract more representative new features through data conversion and dimensionality reduction, then merge the extracted features to construct a new sequence, and divide it into a training set and a test set.

10. The wind and solar power generation output prediction system based on CEEMD-CNN-RF as claimed in claim 7, It is characterized in that The RF prediction module is specifically configured to input the training set in the new sequence constructed in step 3 into the RF model, regress the data using multiple decision trees, and repeatedly divide the data into two parts to reduce the computational complexity; Adjust the RF model parameters, use the trained model for prediction through continuous learning and training, and verify the prediction accuracy on the test set.