Hybrid Machine Learning Model for Forecasting Solar Power Generation

Solar power generation through photovoltaic technology is one of the most popular renewable energy sources. But solar energy is a non-dispatchable source and it is dynamic in nature. Hence from the power system operation point of view, solar power forecasting becomes imperative for a stable grid operation. In this paper, a novel hybrid machine learning approach is proposed for forecasting solar power generation through a hybrid Ensemble Averager technique which exploits the advantages of different machine learning approaches and incorporates them into a single model. Missing values in insolation have been dealt with using a univariate regression-based imputation technique. The ensemble averager is a weighted average model of five individual models, namely - a non-linear autoregressive neural network (NAR-NN), a non-linear autoregressive neural network with exogenous signal (NARX-NN), a least square boosted decision tree model, a support vector regressor with RBF kernel and an Extreme Learning Machine (ELM). The proposed model is tested on a real-world dataset of a 1 MW solar park situated in Gujarat, India (23 degrees 09'15.1 '' N 72 degrees 40'00.8 '' E). Proposed model shows better performance as compared to other models.