| Accurate prediction of photovoltaic(PV)power generation is crucial for the operation of the power grid.However,traditional PV power prediction research often suffers from issues such as inadequate data partitioning and insufficient exploration of data mining algorithms.To address this problem,this paper proposes a PV power prediction method based on clustered ensembles and integrated deep learning algorithms,which improves the accuracy of PV power prediction through the fusion of multiple models.Firstly,this paper preprocesses the photovoltaic(PV)data by using the Pearson correlation coefficient to analyze the correlation between each feature and the PV power generation,and selects features with strong correlation.The data is then examined for missing and abnormal values.The random forest algorithm is used to repair missing and abnormal values,and the entire dataset is standardized.Secondly,this paper proposes a clustered ensemble method based on the Hungarian algorithm.Using meteorological factors as indicators,the four traditional clustering methods are combined using the Hungarian algorithm to construct a clustered ensemble model,which can classify data more accurately.It is found that the clustered ensemble method based on similar days clustering outperforms traditional single clustering methods in terms of silhouette coefficient,Calinski-Harabasz Index,and Davies-Bouldin Index.Furthermore,to better extract feature information from PV data,this paper uses the PSOVMD algorithm to decompose PV power generation time series data.To address the issue of poor reliability of traditional time series prediction and the difficulty of multivariate regression prediction in predicting changes in the target variable in advance,this paper comprehensively considers time series data and various feature data,and proposes a PV power prediction method based on the CNN-GRU model.In the upper layer of the model,the decomposed PV time series data is processed by a multi-head GRU,while various feature data is processed by CNN.In the lower layer of the model,CNN is used to fit the time series data and feature data to obtain the final prediction result.Finally,taking a PV power station in Australia as an example,this paper conducts simulation analysis,and the results show that the PV power prediction model based on clustered ensembles outperforms single-cluster PV power prediction models in terms of prediction accuracy,while the prediction model based on integrated deep learning algorithms generally performs better than traditional single prediction models. |
