Research on Ultra-Short Term Distributed PV Power Forecasting Method Considering Weather Similarity based on Satellite Cloud Map

High-precision meteorological data usually are not available for most distributed PVs, leading to weak generalization ability in power forecasting models and low prediction accuracy. This paper introduces a prediction framework consist by similar time period matching framework (SPMF) to realize ultra-short term distributed PV power forecasting. SPMF aims to match power generation data under similar weather conditions to the current weather from station historical data, and provides the data as a reference to the power forecasting model to address the problem of insufficient generalization caused by lacking of high-precision data. In contrast to most recent approaches that use NWP data and Pearson correlation coefficients to evaluate the similarity of weather conditions, the proposed approach operates on similarity index consisting of SSIM evaluated by satellite cloud map cloud coverage images along with Pearson correlation coefficients evaluated by power generation data. Notable, a power prediction model is developed to selectively measure the reference value of similar period data by similarity index. Results show that the SPMF and the proposed prediction model both improve the accuracy of power prediction, and the utilize of both improvements can increase the accuracy of power forecasting in ultra-short term period by up to 5.91% compared to traditional forecasting model.