Neural network approached evaluation of burner flaming state of a furnace in thermal power plant

摘要

The control of a boiler in thermal power plant has been evaluated based upon the experience of operators. Therefore, there is a necessity for the evaluation method on the state of burner flame to realize an intelligent control. As a solution to this problem, a method using Multi Layered Neural Network (NN), using Backpropagation Algorithm (BP) has been presented. However, the reliability of NN is low, because BP can get trapped in standstill or local minima of training easily. We have proposed a method to enable improved training convergence. In this paper, we consider an evaluation of the state of burner flame by NN using our proposed training methods. As a result, NN enabled the evaluation of the state of combustion as good as experienced operators. The numerical information given by output value of NN can provide precise information for the control of a boiler.