Chapter 13 - MARS model for prediction of short- and long-term global solar radiation

Abstract

This chapter applies a machine learning approach based on Multivariate Adaptive Regression Splines (MARS) algorithm for developing solar radiation estimation and forecasting models for regional Queensland. First, a short-term (daily) global solar radiation model is constructed using the MARS algorithm considering the nonlinear behavior of surface-level solar radiation with its plausible list of predictor variables. Second, a long-term (monthly) global solar radiation model is built using the MARS algorithm mainly to test the tool that can later be used for solar energy assessment over a long-term period and considering seasonal climatic cycles. The accuracy of the MARS model with respect to an alternative data-driven framework is evaluated using Autoregressive Integrated Moving Average at both time horizons. The results indicate that the MARS-based solar radiation forecast can be applied in Central Queensland in both short-term and long-term forecasting scenarios and is a valuable tool for future solar energy projects.