A Photovoltaic Power Generation Data Repair Method Based on Clustering and GAIN

During the production and operation of photovoltaic power plants, it is necessary to constantly collect the power generation data of the plants to monitor their operational status. However, the collected output data often contains varying degrees of anomalies, which greatly hinders tasks such as photovoltaic power prediction and power plant generation planning. Therefore, a cleaning method for photovoltaic data is proposed in this paper. Firstly, this method identifies the anomalies in the photovoltaic power generation data using DBSCAN algorithm, and remove them to obtain the sequence with missing data. Secondly, considering the close relationship between photovoltaic power generation data and weather types, this paper uses K-means++ algorithm to cluster photovoltaic power generation data on a daily basis. Finally, this paper uses the data from each cluster to train the Generative Adversarial Imputation Networks separately and fills in the sequence obtained in the first step. The effectiveness of the proposed method has been demonstrated through experiments.