Short-term solar power forecasting using random vector functional link (RVFL) network

Abstract

Accurate solar power forecasting greatly influences the planning processes undertaken in operation centres of energy providers that relate to actual solar power generation, distribution, system maintenance and pricing. This paper compares the performance of three networks, namely single shrouded layer feed-forward neural network (SLFN), random weight single shrouded layer feed-forward neural network (RWSLFN) and random vector functional link (RVFL) network on the solar power data of Sydney, Australia, of the year 2015 for day and week ahead solar power forecasting. We show that the introduced scheme may adequately learn hidden patterns and accurately determine the solar power forecast by utilizing a range of heterogeneous sources of input that relate not necessarily with the measurement of solar power itself but also other parameters such as effects of temperature, humidity, time and wind speed. The effect of input–output connections is studied, and it is found out that RWSLFN with direct input–output connections also known as RVFL performs better than other RWSLFNs and SLFNs.