The Use of Machine Learning in Predicting the Economic Performance of the ASTEP Solar Thermal System

Abstract

A ridge regression model developed in Python was used to predict the economic performance of an innovative solar thermal system called ASTEP. The system was applied to the industrial processes of two end-users, Mandrekas (MAND) and Arcelor Mittal (AMTP). The ASTEP system was designed to supply thermal energy up to 400 °C and consist of three main components: a novel rotary Fresnel Sundial, thermal energy storage (TES) and a control system. The actual levelized cost of energy (LCOE) of the ASTEP system and four other solar thermal plants as presented in the literature were used to evaluate the ability of the ridge regression model to predict their LCOE values. The plant capacity of the ASTEP system is 25 kW, while the capacities of the other plants are 5 MW–50 MW. The model was trained using data from plants with capacities of 5 MW–50 MW as these were the data available in the literature. The actual and predicted LCOE values were compared and the results showed a prediction error of 2.17–4.72 cents/kWh for the four solar thermal plants, 15.64 cents/kWh for AMTP and 17.98 cents/kWh for MAND’s ASTEP system. These findings indicate that the model has lower prediction error for solar thermal plants with capacities of 5 MW–50 MW, but higher prediction error for smaller capacity plants of less than 1 MW. It is recommended that more studies be conducted on the economic performance of small capacity plants, enabling sufficient data to be available to train machine learning (ML) models, resulting in higher prediction accuracy of the LCOE of these plants.