Power generation today is an increasingly demanding task, worldwide, because of emphasis on efficient ways of generation, energy conservation, environmental friendliness etc. Boiler is an integral component of steam power plant, whose efficient operation is always desired. Furnace Exit Gas Temperature (FEGT) is a very important design/operating parameter in boiler. The knowledge of FEGT is not only useful for design of convective heating surface but it also helpful for decision making and operating actions and ensure improvement in the economic benefit of the power plant. Nonavailability of FEGT on the operator desk put great limitations on efficient operation.

A novel method of estimating FEGT using neural network is presented in this study, in which training data is first generated by calculating FEGT using heat balances through various heat exchanger. Two types of feed forward neural modeling networks, radial basis function (RBF) and backpropagation (BP) network, were applied and compared based on their network simplicity, model building and prediction accuracy. Results are verified on practical data obtained from the boiler of 210 MW thermal power plant. For the validation of results the high velocity thermocouple test (HVT) is employed.

This work also presents an application of very recent development in neural network, “Kohonen Self Organizing Map Technique” (SOM). In the present work SOM is applied to recognize the correctness, validity and reconstruction in case of failure of sensory inputs to the neural network estimator. This is achieved by examining the residual error, which is called quantization error, of the winning unit on the output layer of the network.