Multidimensional Minimization Training Algorithms for Steam Boiler Drum Level Trip Using Artificial Intelligence Monitoring System

This paper deals with the Fault Detection and Diagnosis of steam boiler drum low level by artificial Neural Networks using two different interpretation algorithms. The BFGS quasi-Newton (TRAINBFG) and Levenberg-Marquart (TRAINLV) have been adopted as training algorithms of the neural network model. Real site data was captured from a 3x700MWatt coal-fired thermal power plant in Perak state - Malaysia. Among three power units in the plant, the boiler drum data of unit3 was considered. The selection of the relevant variables for the neural networks is based on merging between theoretical analysis base and the plant operator experience. The merging procedure is described in the paper. Results are obtained from one hidden layer neural network and two hidden layers neural network structures, for both adopted algorithms. Comparisons have been made between the results from the two algorithms based on the Root Mean Square Error. The one hidden layer with one neuron using BFG training algorithm provides the best optimum neural network structure.

Keywords Steam Boiler, Drum Level Trip, Fault Detection and Diagnosis(FDD), Artificial Neural Networks (ANN).