An Appraisal of an Artificial Neural Network Model to Predict Improvement of Generating Capacity of a Coal Thermal Power Plant

Abstract

Base load of the South Africa electricity supply utility (Eskom) is primarily generated from the coal thermal power plant. A scheduled routine maintenance is crucial in ensuring that each unit of the power plant continues to operate according to the manufacturer’s specifications. The research focused on the analyses of “before and after” outage data obtained from the unit cards and power meters in one of the Eskom’s “once-through” 600 MW coal boiler, with a mechanical conversion efficiency of 35%. The data set collected from the metering cards and also the power meters installed in the designated unit of the coal thermal power plant were divided into training, validation and testing data set of inputs; which included average air heater temperature, average main stream super heater temperature, average high pressure and temperature heater temperature, the total mass of coal burnt, average of the cold and hot condenser well pressure and temperature and auxiliary power consumption) and the targets (power generated) both “before and after outage” scenarios. An artificial neural network model was developed to predict the desired output using the training data set. Furthermore, the model was validated and tested by the validation and test data set. The train neural network showed that the overall correlation coefficient of the outputs and targets for “before and after outage” was 0.979 and 0.992, respectively.