Small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion

Abstract

The invention provides a small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion. According to the method, firstly, photovoltaic historical power, weather and power station operation data are comprehensively collected, and data quality is ensured through fine data preprocessing including theoretical power calculation auxiliary abnormal value detection and replacement and data normalization. Then, adaptive feature selection is performed by using an XGBoost algorithm, weather types are classified in combination with K-Means clustering, and key influence factors are effectively extracted. And on the aspect of prediction model construction, GRU and Informer are creatively fused as base learners, and SVR is used as a meta learner, so that an integrated prediction model is formed, and the prediction precision is remarkably improved. Through multi-index error analysis, the performance of the model is comprehensively evaluated, and different prediction scenes are set to verify the generalization ability. According to the method, the problem of photovoltaic power prediction under the small sample condition is solved, and powerful support is provided for optimal scheduling and operation and maintenance management of a photovoltaic power generation system.

Description

Technical Field

[0001] The present invention relates to the field of photovoltaic power generation technology, and in particular to a small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning. Background Art

[0002] As the world's reliance on renewable energy increases, photovoltaic power generation, as one of the important clean energy sources, has an important role to play in the stable operation of the power system, optimal resource allocation, and market scheduling. However, the prediction of photovoltaic power is affected by many factors, including historical photovoltaic power data, meteorological conditions (such as temperature, humidity, wind speed, radiation intensity, etc.), and the operating status of the power station. The complex relationship between these factors makes photovoltaic power prediction a challenging task.

[0003] Existing photovoltaic power prediction methods are mainly divided into two categories: statistical methods and machine learning methods. Statistical methods mainly rely on time series analysis of historical data, which makes it difficult to capture complex nonlinear relationships; while traditional machine learning methods, such as support vector machines (SVM) and random forests, can improve prediction accuracy to a certain extent, but are still insufficient in processing large-scale, high-dimensional data and small samples. In particular, when faced with changing meteorological conditions and power station operating conditions, how to effectively select and utilize features becomes the key to improving prediction accuracy. Existing photovoltaic power prediction methods also have the following shortcomings in specific use:

[0004] (1) Traditional methods often rely on manual experience or fixed rules to select features and cannot adaptively adjust feature sets according to data characteristics, resulting in limited model generalization capabilities;

[0005] (2) It is difficult for a single model to fully capture the complex laws of photovoltaic power changes, especially in small samples and variable environments, and the prediction accuracy is difficult to guarantee;

[0006] (3) Although some deep learning models have excellent performance, they have high computational complexity and are not suitable for real-time prediction and online learning;

[0007] (4) Using only a single error indicator for evaluation is difficult to fully reflect the predictive performance of the model. Summary of the invention

[0008] The technical problem to be solved by the present invention is to overcome the above technical defects and provide a small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning.

[0009] In order to solve the above problems, the technical solution of the present invention comprises the following steps:

[0010] S1. Collecting original data sets that affect photovoltaic power generation, wherein the original data sets include historical photovoltaic power data, meteorological data, and power station operation data;

[0011] S2. Performing data preprocessing on the original data set, wherein the data preprocessing includes data cleaning, outlier detection and replacement, and data standardization;

[0012] S3, perform adaptive feature selection, calculate the contribution of climate factors to the change of photovoltaic power through the XGBoost algorithm, screen out features that are strongly correlated with photovoltaic power, and use the K-Means clustering algorithm to classify weather data;

[0013] S4, build an integrated prediction model, which combines GRU and Informer as base learners and SVR as meta-learner, inputs the preprocessed data set into the model for training, and outputs the prediction results;

[0014] S5. Conduct comparative analysis and error analysis on the prediction results, and introduce error indicators such as MAE, RMSE, and R2 for performance evaluation.

[0015] Further, the adaptive feature selection in step S3 includes:

[0016] S3.1. Use the XGBoost algorithm to calculate the importance of each climate feature, and select the strongly correlated features according to the importance ranking;

[0017] S3.2. Use K-Means clustering algorithm to divide daily data into three weather types: sunny, cloudy and sudden weather.

[0018] Furthermore, the GRU and Informer model parameter settings in step S4.2 include:

[0019] S4.2.1, set the hyperparameters of the GRU neural network, such as the number of input layer nodes corresponding to the number of filtered weather features;

[0020] S4.2.2. Adjust the parameters of the Informer model to adapt to the characteristics of small sample data and improve prediction accuracy.

[0021] Furthermore, the theoretical photovoltaic power calculation formula in step S2.1 is:

[0022] P=etatimesStimesItimes(1+alphatimes(tt ref ))

[0023] Where P is photovoltaic power, η is conversion efficiency, S is area, Ι is total solar radiation intensity, t is temperature, α is temperature coefficient, t refis the reference temperature.

[0024] Furthermore, the SVR model in step S4.3 uses a radial basis function as a kernel function, and selects optimal parameters through cross-validation to improve the generalization ability of the model.

[0025] Furthermore, the data preprocessing in step S2 further includes:

[0026] S2.1. Calculate theoretical photovoltaic power through temperature and radiation intensity, and detect and replace abnormal values ​​of power data accordingly;

[0027] S2.2. Use the min-max normalization method to standardize the data set to the specified interval to eliminate the differences in magnitude and dimension between the indicator data.

[0028] Furthermore, the integrated prediction model construction in step S4 includes:

[0029] S4.1. Select the data after feature screening and divide it into training set, validation set and test set according to a certain ratio;

[0030] S4.2, use GRU and Informer as the first-layer base learners, input the training set and validation set for training, and output the first-layer results;

[0031] S4.3, use SVR as the meta-learner of the second layer, input the output of the first layer as the new training set for training, and obtain the final prediction result;

[0032] S4.4. De-normalize the prediction results and output the results.

[0033] Further, the error analysis in step S5 includes:

[0034] S5.1. The output set of the prediction results after de-normalization is used as the final prediction result;

[0035] S5.2, introduce error indicators MAE, RMSE, and R2 for error analysis and result visualization;

[0036] S5.3. Set up different prediction scenario models and compare the prediction performance of the models under different scenarios.

[0037] Further, the objective function of the XGBoost algorithm in step S3.1 is defined as:

[0038]

[0039] Among them, l is the loss function, and are the actual value and the predicted value respectively, Ω is the regular term, and Τ is the number of regression trees.

[0040] The advantages of the present invention compared with the prior art are:

[0041] 1. The present invention provides a small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning. The contribution of each climate feature to the photovoltaic power change is calculated by the XGBoost algorithm, and strongly correlated features are screened out. The weather data is classified using the K-Means clustering algorithm, which realizes the adaptive selection of features and the accurate classification of weather types, thereby improving the adaptability and prediction accuracy of the model.

[0042] 2. The present invention provides a small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning, and constructs an integrated prediction model that integrates GRU and Informer as base learners and SVR as meta-learner. GRU is good at processing time series data, while Informer can capture long-distance dependencies. The combination of the two can more comprehensively capture the law of photovoltaic power changes. As a meta-learner, SVR can further integrate the output of the base learner to improve the overall prediction performance;

[0043] 3. The present invention provides a small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning. By adjusting the parameters of the Informer model to adapt it to the characteristics of small sample data, the prediction accuracy of the model in the case of small samples is improved;

[0044] 4. The present invention provides a small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning, introduces multiple error indicators such as MAE, RMSE, R2 for performance evaluation, and sets different prediction scenario models for comparison to comprehensively reflect the prediction performance of the model. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of 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.

[0046] The small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning includes the following steps:

[0047] S1. Collecting original data sets that affect photovoltaic power generation, wherein the original data sets include historical photovoltaic power data, meteorological data, and power station operation data;

[0048] S2. Performing data preprocessing on the original data set, wherein the data preprocessing includes data cleaning, outlier detection and replacement, and data standardization;

[0049] S3, perform adaptive feature selection, calculate the contribution of climate factors to the change of photovoltaic power through the XGBoost algorithm, screen out features that are strongly correlated with photovoltaic power, and use the K-Means clustering algorithm to classify weather data;

[0050] S4, build an integrated prediction model, which integrates GRU and Informer as base learners and SVR as meta-learner, inputs the preprocessed data set into the model for training, and outputs the prediction results;

[0051] S5. Conduct comparative analysis and error analysis on the prediction results, and introduce error indicators such as MAE, RMSE, and R2 for performance evaluation.

[0052] Furthermore, the data preprocessing in step S2 also includes:

[0053] S2.1. Calculate theoretical photovoltaic power through temperature and radiation intensity, and detect and replace abnormal values ​​of power data accordingly;

[0054] S2.2. Use the min-max normalization method to standardize the data set to the specified interval to eliminate the differences in magnitude and dimension between the indicator data.

[0055] Furthermore, the adaptive feature selection in step S3 includes:

[0056] S3.1. Use the XGBoost algorithm to calculate the importance of each climate feature, and select the strongly correlated features according to the importance ranking;

[0057] S3.2. Use K-Means clustering algorithm to divide daily data into three weather types: sunny, cloudy and sudden changes.

[0058] Furthermore, the integrated prediction model construction in step S4 includes:

[0059] S4.1. Select the data after feature screening and divide it into training set, validation set and test set according to a certain ratio;

[0060] S4.2, GRU and Informer are used as the first-layer base learners, the training set and the validation set are input for training, and the first-layer results are output;

[0061] S4.3, use SVR as the meta-learner of the second layer, input the output of the first layer as the new training set for training, and obtain the final prediction result;

[0062] S4.4. De-normalize the prediction results and output the results.

[0063] Furthermore, the GRU and Informer model parameter settings in step S4.2 include:

[0064] S4.2.1, set the hyperparameters of the GRU neural network, such as the number of input layer nodes corresponding to the number of filtered weather features;

[0065] S4.2.2. Adjust the parameters of the Informer model to adapt to the characteristics of small sample data and improve prediction accuracy.

[0066] Furthermore, the error analysis in step S5 includes:

[0067] S5.1. The output set of the prediction results after de-normalization is used as the final prediction result;

[0068] S5.2, introduce error indicators MAE, RMSE, and R2 for error analysis and result visualization;

[0069] S5.3. Set up different prediction scenario models and compare the prediction performance of the models under different scenarios.

[0070] Furthermore, the theoretical photovoltaic power calculation formula in step S2.1 is:

[0071] P=etatimesStimesItimes(1+alphatimes(tt ref ))

[0072] Where P is photovoltaic power, η is conversion efficiency, S is area, Ι is total solar radiation intensity, t is temperature, α is temperature coefficient, t ref is the reference temperature.

[0073] Furthermore, the objective function of the XGBoost algorithm in step S3.1 is defined as:

[0074]

[0075] Among them, l is the loss function, and are the actual value and the predicted value respectively, Ω is the regular term, and Τ is the number of regression trees.

[0076] Furthermore, the SVR model in step S4.3 uses the radial basis function as the kernel function and selects the best parameters through cross-validation to improve the generalization ability of the model.

[0077] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0078] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0079] The present invention and its implementation methods are described above, and such description is not restrictive, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design structures and embodiments similar to the technical solution without creativity, which should all fall within the protection scope of the present invention.

Claims

1. A small sample photovoltaic power prediction method based on the fusion of adaptive feature selection and deep learning, characterized in that: The following steps are involved: S1. Collecting original data sets that affect photovoltaic power generation, wherein the original data sets include historical photovoltaic power data, meteorological data, and power station operation data; S2. Performing data preprocessing on the original data set, wherein the data preprocessing includes data cleaning, outlier detection and replacement, and data standardization; S3, perform adaptive feature selection, calculate the contribution of climate factors to the change of photovoltaic power through the XGBoost algorithm, screen out features that are strongly correlated with photovoltaic power, and use the K-Means clustering algorithm to classify weather data; S4, build an integrated prediction model, which integrates GRU and Informer as base learners and SVR as meta-learner, inputs the preprocessed data set into the model for training, and outputs the prediction results; S5. Conduct comparative analysis and error analysis on the prediction results, and introduce error indicators such as MAE, RMSE, and R2 for performance evaluation.

2. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 1 is characterized in that: The data preprocessing in step S2 also includes: S2.

1. Calculate theoretical photovoltaic power through temperature and radiation intensity, and detect and replace abnormal values ​​of power data accordingly; S2.

2. Use the min-max normalization method to standardize the data set to the specified interval to eliminate the differences in magnitude and dimension between the indicator data.

3. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 1 is characterized in that: The adaptive feature selection in step S3 includes: S3.

1. Use the XGBoost algorithm to calculate the importance of each climate feature, and select the strongly correlated features according to the importance ranking; S3.

2. Use K-Means clustering algorithm to divide daily data into three weather types: sunny, cloudy and sudden changes.

4. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 1 is characterized in that: The integrated prediction model construction in step S4 includes: S4.

1. Select the data after feature screening and divide it into training set, validation set and test set according to a certain ratio; S4.2, use GRU and Informer as the first-layer base learners, input the training set and validation set for training, and output the first-layer results; S4.3, use SVR as the meta-learner of the second layer, input the output of the first layer as the new training set for training, and obtain the final prediction result; S4.

4. De-normalize the prediction results and output the results.

5. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 4 is characterized in that: The GRU and Informer model parameter settings in step S4.2 include: S4.2.1, set the hyperparameters of the GRU neural network, such as the number of input layer nodes corresponding to the number of filtered weather features; S4.2.

2. Adjust the parameters of the Informer model to adapt to the characteristics of small sample data and improve prediction accuracy.

6. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 1 is characterized in that: The error analysis in step S5 includes: S5.

1. The output set of the prediction results after de-normalization is used as the final prediction result; S5.2, introduce error indicators MAE, RMSE, and R2 for error analysis and result visualization; S5.

3. Set up different prediction scenario models and compare the prediction performance of the models under different scenarios.

7. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 2 is characterized in that: The theoretical photovoltaic power calculation formula in step S2.1 is: P=ηtimesStimesItimes(1+alphatimes(t-t ref )) Where P is photovoltaic power, η is conversion efficiency, S is area, Ι is total solar radiation intensity, t is temperature, α is temperature coefficient, t ref is the reference temperature.

8. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 3 is characterized in that , the XGBoost algorithm objective function in step S3.1 is defined as: Among them, l is the loss function, and are the actual value and the predicted value respectively, Ω is the regular term, and Τ is the number of regression trees.

9. The small sample photovoltaic power prediction method based on adaptive feature selection and deep learning fusion according to claim 1 is characterized by: The SVR model in step S4.3 uses the radial basis function as the kernel function and selects the best parameters through cross-validation to improve the generalization ability of the model.

Images