Photovoltaic power prediction method and device based on multi-source meteorological data and multi-model integration, and storage medium
Through the method of integrating multi-source meteorological data with multi-model, combined with linear interpolation, Spearman correlation analysis and EMD-KPCA feature extraction, the Stacking model was used to solve the problems of poor generalization and feature redundancy in photovoltaic power prediction. Achieve more efficient prediction results.
Patent Information
- Application Number
- CN202411573692.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-06
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When the existing photovoltaic power prediction methods process multi-source meteorological data, the model generalization is poor, and the prediction accuracy is caused by the problem of input feature redundancy and insufficient timing correlation analysis.
The method of integrating multi-source meteorological data with multi-models is adopted to process data loss through linear interpolation. Spearman correlation analysis screens out meteorological factors, and features are performed using empirical modal decomposition algorithm (EMD) and kernel principal component analysis model (KPCA). Extract and perform multi-base learner combination and meta-learner training through Stacking prediction model.
It improves the accuracy and stability of photovoltaic power prediction, enhances the generalization performance of the model, reduces prediction errors, and improves the ability to capture feature changes.
Smart Images

Figure CN119066409A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of photovoltaic prediction of electric power system, and in particular to a photovoltaic power prediction method, device and storage medium based on multi-source meteorological data and multi-model integration. Background Art
[0002] Solar energy is the most abundant and most promising clean energy among renewable energy sources. Its development and utilization has become an important direction for my country's energy structure adjustment. Carrying out short-term prediction of photovoltaic power generation can effectively reduce the impact of photovoltaic output uncertainty on the power grid, improve the reliability of photovoltaic power generation systems, and help the power grid achieve high-quality and stable power supply.
[0003] In recent years, the development and utilization of solar energy has become an important area of global energy transformation. Photovoltaic power generation has entered a stage of large-scale development and has shown good development prospects. The mainstream prediction methods for photovoltaic power generation, such as statistical methods, are simple in principle and easy to implement, but when the sample data is more complex, the prediction effect is average; another type of method, artificial intelligence algorithm, has a poor processing effect on time series and is unsatisfactory when predicting longer time series.
[0004] Among deep learning methods, LSTM is a special recurrent neural network that evolved from RNN and can be applied to photovoltaic power forecasting. LSTM enhances the memory capacity of historical data by optimizing the hidden layer structure, which can effectively handle the long-term dependencies of data and has better applicability in photovoltaic power forecasting. However, when there are many input features, the prediction accuracy of the model will be affected, and the use of a single model for prediction will also result in poor generalization.
[0005] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present disclosure, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art known to those skilled in the art. Summary of the invention
[0006] The present invention provides a photovoltaic power prediction method based on multi-source meteorological data and multi-model integration, which can effectively solve the problems in the background technology.
[0007] In order to achieve the above object, the technical solution adopted by the present invention is: A photovoltaic power prediction method based on multi-source meteorological data and multi-model integration includes the following steps: Collect historical photovoltaic power data and corresponding multi-source meteorological data, and construct power data sets and meteorological data sets respectively; Performing correlation analysis on the meteorological data set and the power data set, and selecting meteorological factor data having a strong correlation with photovoltaic output to form a new meteorological data set; Extracting features of the new meteorological data set based on the empirical mode decomposition algorithm EMD and the kernel principal component analysis model KPCA, and constructing a feature set according to the extracted features; Integrate the feature set with the power data set to obtain a new data set, and divide the new data set into a plurality of sub-data sets and at least one test set; The Stacking prediction model is trained by using the sub-dataset and the test set to obtain a trained meta-learner, multi-source meteorological data is input into the meta-learner, and a prediction result is output.
[0008] Further, the meteorological data set and the power data set are preprocessed by linear interpolation method, including the following steps: First identify missing or low-resolution parts of the dataset; For each missing value, find two known data points before and after it; The missing values are calculated using the linear interpolation formula. The specific formula is as follows: ; in, x and y represent the interpolation nodes and the interpolated functions respectively, x 1 , y 1 and x 2 , y 2 Represent the nodes and data values of the data points before and after the interpolation node respectively.
[0009] Further, data screening is performed according to the correlation analysis between the meteorological data set and the power data set, including the following steps: Pairing the interpolated meteorological data set with the power data set; Calculate the Spearman correlation coefficient for each pair of data using the following formula: ; in, x i and y i Respectively represent variables X and Y No. i The rank of the observations (i.e., the sorted position), and Respectively represent variablesX and Y The average rank of is calculated as the sum of the ranks of all observations divided by the total number of observations n; According to the absolute value of the Spearman correlation coefficient, meteorological factor data having a strong correlation with the photovoltaic output are selected to form the new meteorological data set.
[0010] Furthermore, feature extraction of the new meteorological data set is performed based on the EMD algorithm and the KPCA model, including the following steps: Decomposing the new meteorological data by using an EMD algorithm; The decomposed IMFs are used as new feature variables and KPCA is used to reduce the dimension. The principal component with a contribution rate of 85% is selected to form the feature set after dimensionality reduction processing.
[0011] Furthermore, the feature data in the feature set is normalized to the interval [0,1], specifically using the following formula: ; in, p ' represents the normalized value, p Represents the original value, p min Indicates the minimum value, p max Indicates the maximum value; The prediction results output by the meta-learner are denormalized to restore them to the original data range, and then the prediction results are evaluated.
[0012] Furthermore, during the training process of the meta-learner, the sub-data set and the test set are input into the first layer of the Stacking prediction model for training, and a second training set and a second test set are generated; The second training set and the second test set are input into the second layer of the Stacking prediction model for training to obtain the meta-learner.
[0013] Furthermore, the first layer training operation of the Stacking prediction model includes the following steps: Setting N training algorithms in the first layer of the Stacking prediction model; Based on the M sub-datasets and the 1 test set, select any training algorithm and use a cross-validation method to perform base learner training; Repeat the base learner training for the N-1 training algorithms N-1 times to form a second training set and a second test set based on N base learners.
[0014] Furthermore, during the training process of any base learner, any of the sub-datasets is selected as a validation set, and the remaining M-1 sub-datasets are selected as training sets, and M trainings are performed to obtain M groups of prediction values for the validation set and M groups of prediction values for the test set; Combine the predicted values of the M groups for the validation set to form the first prediction set T 1 =[ t 1 , t 2 ,..., t M ], averaging the predicted values of the M groups for the test set to obtain a set of second predicted values b 1 ; Combining the first prediction sets of N base learners to form the first training set [ T 1 , T 2 ,..., T N ], combining the second prediction sets of N base learners to form the second training set [ b 1 , b 2 ,..., b N ].
[0015] Furthermore, the second-layer training operation of the Stacking prediction model includes the following steps: The first training set [ T 1 , T 2 ,..., T N ] as the second layer training set of the Stacking prediction model, and the second training set [ b 1 , b 2 ,..., b N ] as the second-layer test set of the Stacking prediction model; The second layer of the Stacking prediction model is trained by a Linear Regression training algorithm to obtain the meta-learner.
[0016] Furthermore, the values of the evaluation indicators are calculated based on the predicted values and the true values. There are four evaluation indicators, namely, relative mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R 2 ).
[0017] A photovoltaic power prediction device based on multi-source meteorological data and multi-model integration adopts the above-mentioned photovoltaic power prediction method based on multi-source meteorological data and multi-model integration, comprising: Data acquisition module: used to collect historical photovoltaic power data and corresponding multi-source meteorological data, and construct power data sets and meteorological data sets respectively; Data processing module: generating a plurality of sub-datasets and at least one test set for training a model based on a power data set and a meteorological data set; Model training module: used to train the Stacking prediction model through the sub-dataset and the test set to obtain a trained meta-learner; Prediction output module: used to receive multi-source meteorological data, perform power prediction through the meta-learner, and output the prediction results.
[0018] Furthermore, the data processing module includes: A correlation analyzer, used to perform correlation analysis on the meteorological data set and the power data set, and screen out meteorological factor data that has a strong correlation with photovoltaic output to form a new meteorological data set; A feature extractor, which extracts features from the new meteorological data set based on an empirical mode decomposition algorithm EMD and a kernel principal component analysis model KPCA, and constructs a feature set; A data integrator, used to integrate the feature set with the power data set to form a new data set; A data divider is used to divide the new data set into multiple sub-data sets and at least one test set.
[0019] A computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the photovoltaic power prediction method based on multi-source meteorological data and multi-model integration.
[0020] The beneficial effects of the present invention are: In view of the common defects of the commonly used deep learning algorithms for photovoltaic power prediction in power systems, the present invention proposes a photovoltaic power prediction method based on multi-source meteorological data and multi-model integration. Firstly, the problem of low resolution of original data is solved by data interpolation method, and the problem of insufficient consideration of the time series correlation of data is solved by correlation analysis. The problem of input feature redundancy is solved by EMD and KPCA methods, and the problem of poor model generalization is solved by Stacking model.
[0021] The present invention fully utilizes the advantages of multiple base learners through the training and prediction process of the Stacking model, and further optimizes and integrates them through a meta-learner, thereby improving the accuracy and stability of photovoltaic power prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0023] Figure 1 It is a schematic diagram of the process of the photovoltaic power prediction method in the present invention; Figure 2 This is a structural diagram of the combined prediction model in the present invention. DETAILED DESCRIPTION
[0024] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0025] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only implementation method.
[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0027] like Figure 1 A photovoltaic power prediction method based on multi-source meteorological data and multi-model integration is shown, comprising the following steps: Collect historical photovoltaic power data and the corresponding multi-source meteorological data, and construct a power data set and a meteorological data set respectively; perform correlation analysis on the meteorological data set and the power data set, and screen the meteorological factor data that has a strong correlation with the photovoltaic output to form a new meteorological data set; extract features of the new meteorological data set based on the empirical mode decomposition algorithm EMD and the kernel principal component analysis model KPCA, and construct a feature set based on the extracted features; integrate the feature set with the power data set to obtain a new data set, and divide the new data set into multiple sub-data sets and at least one test set; train the Stacking prediction model through the sub-data sets and the test set, and obtain a trained meta-learner, input multi-source meteorological data into the meta-learner and output the prediction results.
[0028] This application aims to solve the problem of poor generalization when using a single model for prediction. Ensemble learning combines a set of differentiated sub-prediction models to obtain a performance-enhanced ensemble model. Compared with individual learning units, the ensemble model can improve generalization performance and show more superior performance.
[0029] like Figure 2 The Stacking prediction model in ensemble learning shown in the figure is a method of combining multiple base learners by training a meta-learner. First, the original training data is used to train several base learners, and then the outputs of the base learners are combined into a new data set to train the meta-learner. Due to its special model structure, the Stacking method can train multiple models simultaneously through cross-validation. Compared with general combination models, it has better prediction performance and stronger generalization ability.
[0030] In view of the common defects of the commonly used deep learning algorithms for photovoltaic power prediction in power systems, the present invention proposes a photovoltaic power prediction method based on multi-source meteorological data and multi-model integration. First, the problem of low resolution of original data is solved by data interpolation method, and the problem of insufficient consideration of the temporal correlation of data is solved by correlation analysis. The problem of input feature redundancy is solved by EMD and KPCA methods, and the problem of poor model generalization is solved by Stacking model. Therefore, the prediction method disclosed by the present invention has the advantages of efficient processing of multi-source data, consideration of data temporal correlation, accurate capture of feature changes, and excellent model generalization. It also includes analyzing the prediction effect of the model by calculating the value of the evaluation index.
[0031] In the present invention, the meteorological data set and the power data set are preprocessed by linear interpolation method, including the following steps: First identify missing or low-resolution parts of the dataset; For each missing value, find two known data points before and after it; The missing values are calculated using the linear interpolation formula. The specific formula is as follows: ; in, x and y represent the interpolation nodes and the interpolated functions respectively, x 1 , y 1 and x 2 , y 2 Represent the nodes and data values of the data points before and after the interpolation node respectively.
[0032] In the process of preprocessing the original data, data interpolation is used to improve the resolution of the data, enhance the integrity and continuity of the data, and provide a higher quality data basis for subsequent analysis and modeling. This embodiment adopts linear interpolation method, which estimates the value of unknown data points by assuming that the change between two known data points is linear, which is simple and effective.
[0033] Data screening is performed based on the correlation analysis of the meteorological dataset and the power dataset, including the following steps: Pair the interpolated meteorological data set with the power data set; Calculate the Spearman correlation coefficient for each pair of data using the following formula: ; in, x i and y i Respectively represent variables X and Y No. i The rank of the observations (i.e., the sorted position), and Respectively represent variables X and Y The average rank of is calculated as the sum of the ranks of all observations divided by the total number of observations n; According to the absolute value of the Spearman correlation coefficient, meteorological factors with strong correlation with photovoltaic output are selected to form a new meteorological data set. The coefficient reflects the degree of correlation between meteorological factors and photovoltaic output. The larger the absolute value of the coefficient, the stronger the correlation. According to the strength of the correlation, meteorological factors with significant correlation with photovoltaic output are selected as input features for subsequent modeling, while those with weak or irrelevant correlation are excluded.
[0034] Spearman correlation analysis is a non-parametric correlation analysis method that can evaluate the monotonic relationship between two variables. It does not require the data to follow a specific distribution, so it is suitable for correlation analysis between meteorological data and photovoltaic power. Through correlation analysis, meteorological factors that have an important impact on photovoltaic power prediction are identified, and those factors with less correlation or irrelevant influencing factors are excluded to simplify the model and improve efficiency.
[0035] In the present invention, feature extraction is performed on the new meteorological data set based on the EMD algorithm and the KPCA model, including the following steps: The new meteorological data were decomposed by EMD algorithm. The IMFs obtained by decomposition were used as new feature variables and KPCA was used for dimensionality reduction. The principal components with a contribution rate of 85% were selected to form the feature set after dimensionality reduction.
[0036] Decomposing meteorological characteristic data through the EMD (Empirical Mode Decomposition) algorithm can enrich the diversity of input variables and highlight the local characteristics of environmental sequences at different time scales. EMD can decompose complex data sets into a series of simple intrinsic mode functions (IMFs) with different frequency characteristics, thereby helping to reveal the inherent structure of the data.
[0037] Since a large number of features may be generated after EMD decomposition, directly using these features for modeling may lead to problems such as high computational complexity and model overfitting. Therefore, KPCA (kernel principal component analysis) is used for dimensionality reduction to extract the most important features and reduce the dimension of the feature space.
[0038] In this embodiment, EMD decomposition is performed on the meteorological feature data that has been interpolated and screened, and each IMF obtained by decomposition is used as a new feature variable, which together with the original meteorological features constitutes a richer feature set. KPCA processing is performed on the meteorological feature data after EMD decomposition, and the principal component with a contribution rate of 85% is selected as the feature set after dimensionality reduction, so as to retain most of the original information while reducing the number of features.
[0039] In order to eliminate the dimensional differences between different features and improve the convergence speed and prediction accuracy of the model, the feature data in the feature set is normalized to the [0,1] interval. The following formula is used: ; in, p ' represents the normalized value, p Represents the original value, p min Indicates the minimum value, p maxIndicates the maximum value; the prediction results output by the meta-learner are denormalized to restore them to the original data range, and then the prediction results are evaluated. The prediction model is trained on normalized data, so the prediction results are also normalized values; in order to convert the prediction results back to the actual numerical range, a denormalization operation is required.
[0040] In the present invention, during the training process of the meta-learner, the sub-data set and the test set are input into the first layer of the Stacking prediction model for training, and a second training set and a second test set are generated; the second training set and the second test set are input into the second layer of the Stacking prediction model for training to obtain the meta-learner.
[0041] Specifically, the first-layer training operation of the Stacking prediction model includes the following steps: Set N training algorithms in the first layer of the Stacking prediction model; based on M sub-data sets and 1 test set, select any training algorithm and use cross-validation to train the base learner; then repeat N-1 base learner training for N-1 training algorithms N-1 times to form a second training set and a second test set based on N base learners.
[0042] In the training process of any base learner, any sub-dataset is selected as the validation set, and the remaining M-1 sub-datasets are used as the training set. M trainings are performed to obtain M groups of prediction values for the validation set and M groups of prediction values for the test set. The prediction values of the M groups for the validation set are combined to form the first prediction set. T 1 =[ t 1 , t 2 ,..., t M ], average the predicted values of the M groups for the test set to obtain a set of second predicted values b 1 ; Combine the first prediction sets of N base learners to form the first training set [ T 1 , T 2 ,..., T N ], combining the second prediction sets of N base learners to form the second training set [ b 1 , b 2 ,..., b N ].
[0043] Furthermore, the second-layer training operation of the Stacking prediction model includes the following steps: The first training set [ T 1 , T 2 ,..., T N ] as the second-layer training set of the Stacking prediction model, and the second training set [ b 1 , b 2 ,..., b N ] is used as the second-layer test set of the Stacking prediction model; the second layer of the Stacking prediction model is trained by the Linear Regression training algorithm to obtain a meta-learner.
[0044] In this embodiment, the meteorological data and photovoltaic output data after feature extraction by EMD and KPCA are integrated together to form a complete data set; this complete data set is then divided into N sub-data sets of the same size and a test set. The sub-data sets are used to train the first-layer base learner of the Stacking model, and the test set is used to evaluate the performance of the entire Stacking model.
[0045] In the first layer of the Stacking model, three algorithms, RF (Random Forest), SVM (Support Vector Machine), and KNN (K Nearest Neighbor), are selected as the training algorithms for the base learners. These three algorithms have their own advantages and can capture different patterns in the data.
[0046] For each training algorithm, the base learner is trained using cross-validation. Specifically, each sub-dataset is used as a validation set, and the remaining M-1 sub-datasets are used as training sets, and training is performed M times. In this way, each sub-dataset is used as a validation set, thereby obtaining M sets of prediction values for the validation set (denoted as t 1 , t 2 ,..., t M ) and the predicted values of the M groups on the test set.
[0047] Combine the predicted values of the M groups for the validation set to form a new training set T 1 =[ t 1 , t 2 ,..., t M]. This new training set contains the prediction results of all base learners on the validation set and is used to train the meta-learner of the second layer. At the same time, the prediction values of the M groups of test sets are averaged to obtain a set of prediction values b 1 This average prediction value represents the prediction result of the first training algorithm (such as RF) on the test set. Repeat the above steps of cross-validation training, prediction value combination and averaging for the training algorithms SVM and KNN, and finally obtain the prediction values of the three base learners [ T 1 , T 2 , T 3 ] and the predicted values of the test set [ b 1 , b 2 , b 3 ].
[0048] The prediction value of the first layer base learner [ T 1 , T 2 , T 3 ] is used as the second-layer training set, and the predicted value of the test set obtained in the first layer is [ b 1 , b 2 , b 3 ] as the second-layer test set; therefore, both the training set and the test set of the second layer contain the prediction results of all base learners. In the second layer of the Stacking model, Linear Regression is selected as the training algorithm of the meta-learner, which can learn the linear relationship between the prediction results of the base learners, thereby further improving the accuracy of the prediction. The second-layer training set is trained using the Linear Regression algorithm to obtain a meta-learner. This meta-learner can further optimize and integrate the final prediction results based on the prediction results of the base learners. Use the trained meta-learner to predict the second-layer test set to obtain the final prediction results; the prediction results combine the advantages of all base learners and are further optimized and integrated by the meta-learner, so they have higher accuracy and stability.
[0049] In the present invention, the values of the evaluation indicators are calculated based on the predicted values and the true values. There are four evaluation indicators, namely, relative mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and coefficient of determination (R 2 ).
[0050] The relative mean square error (RMSE) reflects the degree of dispersion of the predicted value. The smaller the RMSE value, the closer the predicted value is to the true value. The mean absolute error (MAE) reflects the average error of the predicted value. The smaller the MAE value, the more accurate the predicted value. The mean absolute percentage error (MAPE) eliminates the influence of the dimension and expresses the error of the predicted value in the form of a percentage. The smaller the MAPE value, the closer the predicted value is to the true value. The coefficient of determination (R 2 ) ranges from 0 to 1. The closer the value is to 1, the stronger the linear relationship between the predicted value and the true value is, and the better the prediction effect is.
[0051] In this embodiment, the data set in the 2014 Global Energy Forecast Competition is used for simulation verification. The data set includes solar energy data and meteorological data from a certain place from 2012 to 2014, with a sampling interval of 1h. The linear interpolation method is used to convert the original data with a resolution of 1 hour into a resolution of 15 minutes. The Spearman correlation coefficient of photovoltaic power and 12 meteorological characteristic parameters is calculated in pairs, and humidity, 2-meter temperature, surface solar radiation, surface thermal radiation and top net solar radiation are finally selected as meteorological characteristic data. The three algorithms of RF, SVM and KNN are used to train the base learner, and the LinearRegression is used to train the meta-learner to form a prediction model based on Stacking. The data of April 2012 are selected for comparative experiments, and the training set and the test set are divided according to the ratio of 7:3. In order to ensure the accuracy of the experimental results, each experiment is carried out 5 times, and finally the prediction results ranked in the median of prediction accuracy are taken for subsequent analysis and comparison. The prediction result indicators are summarized in Table 1, as shown below: Table 1 As shown in Table 1, the RMSE value is 0.022275MW, which is 11.6%, 13.1%, and 60.99% lower than the other three models respectively; the MAPE value is 0.014598NW, which is 35.3%, 37.8%, and 75% lower than the other three models respectively; the MAPE value is 5.4039%, which is 1.62%, 1.87%, and 6.41% lower than the other three models respectively; the determination coefficient R2 is also the highest among the four. In summary, the EMD-KPCA-Stacking model of the present invention has a significant increase in prediction accuracy and a significant decrease in prediction error compared to other models, and the prediction results have a higher fit with the true value, indicating that the photovoltaic power prediction model proposed in the present invention has a significant improvement in performance compared to other comparative prediction models.
[0052] The present invention fully utilizes the advantages of multiple base learners through the training and prediction process of the Stacking model, and further optimizes and integrates them through a meta-learner, thereby improving the accuracy and stability of photovoltaic power prediction.
[0053] A photovoltaic power prediction device based on multi-source meteorological data and multi-model integration adopts the above-mentioned photovoltaic power prediction method based on multi-source meteorological data and multi-model integration, comprising: Data acquisition module: used to collect historical photovoltaic power data and corresponding multi-source meteorological data, and construct power data sets and meteorological data sets respectively; Data processing module: generating a plurality of sub-datasets and at least one test set for training a model based on a power data set and a meteorological data set; Model training module: used to train the Stacking prediction model through the sub-dataset and the test set to obtain a trained meta-learner; Prediction output module: used to receive multi-source meteorological data, perform power prediction through the meta-learner, and output the prediction results.
[0054] Furthermore, the data processing module includes: A correlation analyzer, used to perform correlation analysis on the meteorological data set and the power data set, and screen out meteorological factor data that has a strong correlation with photovoltaic output to form a new meteorological data set; A feature extractor, which extracts features from the new meteorological data set based on an empirical mode decomposition algorithm EMD and a kernel principal component analysis model KPCA, and constructs a feature set; A data integrator, used to integrate the feature set with the power data set to form a new data set; A data divider is used to divide the new data set into multiple sub-data sets and at least one test set.
[0055] A computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the photovoltaic power prediction method based on multi-source meteorological data and multi-model integration.
[0056] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A photovoltaic power prediction method based on multi-source meteorological data and multi-model integration, characterized in that: The following steps are involved: Collect historical photovoltaic power data and corresponding multi-source meteorological data, and construct power data sets and meteorological data sets respectively; Performing correlation analysis on the meteorological data set and the power data set, and selecting meteorological factor data having a strong correlation with photovoltaic output to form a new meteorological data set; Extracting features of the new meteorological data set based on the empirical mode decomposition algorithm EMD and the kernel principal component analysis model KPCA, and constructing a feature set according to the extracted features; Integrate the feature set with the power data set to obtain a new data set, and divide the new data set into a plurality of sub-data sets and at least one test set; The Stacking prediction model is trained by using the sub-dataset and the test set to obtain a trained meta-learner, multi-source meteorological data is input into the meta-learner, and a prediction result is output.
2. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 1 is characterized in that: The constructed meteorological data set and the power data set are preprocessed by linear interpolation method, and then correlation analysis is performed, wherein the preprocessing includes the following steps: First identify missing or low-resolution parts of the dataset; For each missing value, find two known data points before and after it; The missing values are calculated using the linear interpolation formula. The specific formula is as follows: ; in, x and y represent the interpolation nodes and the interpolated functions respectively, x 1. y 1 and x 2. y 2 represent the nodes and data values of the data points before and after the interpolation node, respectively.
3. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 2 is characterized in that: Data screening is performed according to the correlation analysis between the meteorological data set and the power data set, comprising the following steps: Pairing the interpolated meteorological data set with the power data set; Calculate the Spearman correlation coefficient for each pair of data using the following formula: ; in, x i and y i Respectively represent variables X and Y No. i The rank of the observations, that is, the sorted position, and Respectively represent variables X and Y The average rank of is calculated as the sum of the ranks of all observations divided by the total number of observations n; According to the absolute value of the Spearman correlation coefficient, meteorological factor data having a strong correlation with the photovoltaic output are selected to form the new meteorological data set.
4. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 1 is characterized in that: The feature extraction of the new meteorological data set based on the EMD algorithm and KPCA model includes the following steps: Decomposing the new meteorological data by using an EMD algorithm; The decomposed IMFs are used as new feature variables and KPCA is used to reduce the dimension. The principal component with a contribution rate of 85% is selected to form the feature set after dimensionality reduction processing.
5. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 4 is characterized in that: The feature data in the feature set is normalized to the interval [0,1], specifically using the following formula: ; in, p ' represents the normalized value, p Represents the original value, p min Indicates the minimum value, p max Indicates the maximum value; The prediction results output by the meta-learner are denormalized to restore them to the original data range, and then the prediction results are evaluated.
6. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 1 is characterized in that: During the training process of the meta-learner, the sub-data set and the test set are input into the first layer of the Stacking prediction model for training, and a second training set and a second test set are generated; The second training set and the second test set are input into the second layer of the Stacking prediction model for training to obtain the meta-learner.
7. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 6 is characterized in that: The first layer training operation of the Stacking prediction model includes the following steps: Setting N training algorithms in the first layer of the Stacking prediction model; Based on the M sub-datasets and the 1 test set, select any training algorithm and use a cross-validation method to perform base learner training; Repeat the base learner training for the N-1 training algorithms N-1 times to form a second training set and a second test set based on N base learners.
8. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 7 is characterized in that: During the training process of any base learner, any of the sub-datasets is selected as a validation set, and the remaining M-1 sub-datasets are selected as training sets, and M trainings are performed to obtain M groups of prediction values for the validation set and M groups of prediction values for the test set; Combine the predicted values of the M groups for the validation set to form the first prediction set T 1=[ t 1, t 2,..., t M ], averaging the predicted values of the M groups for the test set to obtain a set of second predicted values b 1; Combining the first prediction sets of N base learners to form the first training set [ T 1, T 2,..., T N ], combining the second prediction sets of N base learners to form the second training set [ b 1, b 2,..., b N ].
9. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 8 is characterized in that: The second-layer training operation of the Stacking prediction model includes the following steps: The first training set [ T 1, T 2,..., T N ] as the second layer training set of the Stacking prediction model, and the second training set [ b 1, b 2,..., b N ] as the second-layer test set of the Stacking prediction model; The second layer of the Stacking prediction model is trained by a Linear Regression training algorithm to obtain the meta-learner.
10. The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration according to claim 1, characterized in that: The prediction results output by the meta-learner are evaluated. Specifically, the values of the evaluation indicators are calculated based on the predicted values and the true values. There are four evaluation indicators, namely, relative mean square error RMSE, mean absolute error MAE, mean absolute percentage error MAPE and determination coefficient R 2 .
11. A photovoltaic power prediction device based on multi-source meteorological data and multi-model integration, characterized in that: The photovoltaic power prediction method based on multi-source meteorological data and multi-model integration as described in any one of claims 1 to 10 above comprises: Data acquisition module: used to collect historical photovoltaic power data and corresponding multi-source meteorological data, and construct power data sets and meteorological data sets respectively; Data processing module: generating a plurality of sub-datasets and at least one test set for training a model based on a power data set and a meteorological data set; Model training module: used to train the Stacking prediction model through the sub-dataset and the test set to obtain a trained meta-learner; Prediction output module: used to receive multi-source meteorological data, perform power prediction through the meta-learner, and output the prediction results.
12. The photovoltaic power prediction device based on multi-source meteorological data and multi-model integration according to claim 11, characterized in that: The data processing module comprises: A correlation analyzer, used to perform correlation analysis on the meteorological data set and the power data set, and screen out meteorological factor data that has a strong correlation with photovoltaic output to form a new meteorological data set; A feature extractor, which extracts features from the new meteorological data set based on an empirical mode decomposition algorithm EMD and a kernel principal component analysis model KPCA, and constructs a feature set; A data integrator, used to integrate the feature set with the power data set to form a new data set; A data divider is used to divide the new data set into multiple sub-data sets and at least one test set.
13. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the photovoltaic power prediction method based on the integration of multi-source meteorological data and multiple models as described in any one of claims 1 to 10.
Citation Information
Patent Citations
Multivariable input photovoltaic power prediction method based on EMD-PCA-LSTM
CN110516844A
Short-term photovoltaic power generation power prediction method
CN111369045A
Photovoltaic power generation power prediction method based on IMVO-DELM model
CN116579251A
Photovoltaic power generation prediction model construction method, photovoltaic energy supply regulation and control method and equipment
CN117634737A
Photovoltaic power station regulation and operation optimization method, device, equipment and medium
CN118748538A
Cited By
Photovoltaic power short-term prediction method and device based on multi-level reverse error correction, equipment and medium
CN119482452A
