The Comparison of Regression Models and Artificial Neural Networks in Predicting Power Generation in a Thermal Coal Power Plant

Abstract

A significant Eskom’s grid electricity is generated from the thermal coal-fired plants. The study focused on modelling the generated electricity during the “before and after” outage of a typical unit, in one of the Eskom Benson’s thermal coal power plants rated at 600 MW and mechanical conversion efficiency of 35%. The dataset for the chosen input parameters are collected from the metering cards and the generated electrical power are obtained from the installed power meters to the designated unit in the power plant. Multiple linear regression models (MLR) and Artificial Neural Networks (ANN) for both the “before and after outage” power generated are developed, tested and validated with the input parameters as the average air heater temperature, average main stream super-heater temperature, average high pressure heater’s temperature, the total mass of coal burnt, average of the condenser well pressure and temperature and the auxiliary power consumed. The MLR models and the ANNs for both the “before and after” outage power generated gave excellent correlation coefficients of over 0.950. Furthermore, it can be concluded that the ANNs gave better predictions over the counterparts MLRs model based on the correlation coefficients and the mean square errors derived from the models.