A Novel Approach to Performance Prediction of Boiler Parameters in Thermal Power Plants using Soft Computing Techniques

Abstract

Soft computing techniques are being increasingly applied to predict parameters relating to the performance of boilers in thermal power plants. In this paper a combination of Genetic Algortihms (GA) and Artificial Neural Networks (ANN) are used for performance prediction of boilers. An implementation of Genetic algorithms, called Gene Hunter is used for feature selection. A threshold of value 0.1 was identified and features whose importance exceeded 0.1 were taken as the selected features. For the ANN, Superheater spray and Reheater spray which perform the primary role of controlling the main steam/reheat temperature and NOx are chosen as the output parameters. In this paper cascade correlation algorithm has been used to predict parameters of Reheater Spray, Superheater Spray and NOx. It is a novel approach that automatically adds new hidden neurons one by one and fixing the network topology once training is done. Turboprop 2, a variant of the cascade correlation algorithm has been used for this purpose. The overall results were of the order of 99.6%, 98.6% and 91.6% for RH Spray, SH Spray and NOx respectively.