The Adaptive Forecast Combiner

Abstract

An accurate solar power prediction is vital for the smooth and stable operation of the power grid. A combined forecast is often preferred over an individual model's predictions as it significantly increases the overall forecast accuracy. In this paper, we propose a novel forecast aggregation algorithm called the Adaptive Forecast Combiner. It combines multiple input forecasts with dynamic weight allocation after a multi-horizon performance review. Three machine learning models are trained on the dataset of a 48 MW solar power plant in India, and the corresponding intra-day and day-ahead forecasts are obtained. A comprehensive performance evaluation illustrates that the proposed combiner consistently outperforms all the individual machine learning models irrespective of the season or forecasting methodology.