FORECASTING THERMAL ENERGY DEMANDS FOR VARIOUS PROCESS INDUSTRIES USING MACHINE LEARNING TECHNIQUES

Abstract

Despite the fact that India has vast solar energy potential, the process industries are dependent on conventional fossil fuels for their thermal energy needs. Solar thermal energy is a viable option for industrial process heating applications to mitigate the utilization of conventional fossil fuels. The utilization of solar energy in the process industries to meet their energy demands helps to reduce the carbon footprint, and eventually will help India balance its energy needs by reducing the import of crude oil. The exploitation of solar energy in the process industries requires the assessment of future thermal energy demands. In this study, since the data points were nonlinear in nature, support vector regression (SVR) and long short-term memory (LSTM) algorithms were applied to forecast the thermal energy demands of various process industries. The process industries considered in this study were textile, food processing, leather and footwear, chemical and pharmaceutical, dairy, iron and steel, and automotive. Data from 1998 to 2014 were used for training and data from 2015 to 2017 were used for testing. The mean absolute percentage error (MAPE) was used as a performance measure metric to measure the performance of the SVR and LSTM algorithms. The average MAPEs obtained for given industries by the SVR with the genetic algorithm (SVRGA), SVR with grid search (SVRGS), and LSTM algorithm were 7.56%, 8.34%, and 11.10% respectively. The SVRGA outperformed the SVRGS and LSTM algorithm for the given training and testing data.