An Efficient Machine Learning Approach for Accurate Short Term Solar Power Prediction

Abstract: Solar based electricity generations have experienced strong and impactful growth in recent years. The regulation, scheduling, dispatching, and unit commitment of intermittent solar power is dependent on the accuracy of the forecasting methods. In this paper, a robust Expanded Extreme Learning Machine (EELM) is proposed to accurately predict solar power for different time horizons and weather conditions. The proposed EELM technique has no randomness due to the absence of random input layer weights and takes very less time to predict the solar power efficiently. The performance of the proposed EELM is validated through historical data collected from the National Renewable Energy Laboratory (NREL) through various performance metrics. The efficacy of the proposed EELM method is evaluated against basic ELM and Functional Link Neural Network (FLNN) for 5 minutes and 1 hour ahead time horizon.