A Hybrid Approach of Solar Power Forecasting Using Machine Learning

Photovoltaic (PV) power generation is prone to fluctuations and it is affected by different weather conditions. Therefore, accurate forecasting becomes vital for grid operators to manage grid operations. In this study, multiple machine learning models are used, and different weather parameters are analyzed for PV forecasting. The performance of different models is compared. The proposed forecasting models are tested with 4 kWp PV device installed at our institute, fortiss GmbH and the comparison of the models is done through error analysis. The accuracy is evaluated using historical weather data. The study proposes a novel approach of PV forecasting where a model is built based on the principle of pipelining clustering, classification and regression algorithms. Clustering and classification algorithms group the data-points with similar weather conditions. Subsequently, segmented regression is applied to these groups. Weather forecast of the next day is utilized to determine the models used for forecasting. The results show that the proposed forecasting model for PV systems is effective and promising.