Sensitivity analysis on neural network algorithm for primary superheater spray modeling

Abstract

Nonlinear, large inertia with long dead time is always associated with the main steam temperature parameter in coal fired power plant. Successful control of the main steam temperature within ±2°C of its setpoint is the ultimate target for coal-fired power plant operators. Two of the most common main steam temperature circuit are primary superheater spray and secondary superheater spray. Various methods were used to model the primary superheater spray control valve opening, and the neural network remains one of the most popular choices among researchers. It remains inconclusive which neural network algorithm types, setup, number of layers, and training algorithm will give the best result. As such, the paper shows the best setup for the neural network algorithm based on sensitivity analysis methodology for one hidden layer. The inputs selected for the neural network are generator output, main steam flow, total spray flow, and secondary superheater outlet steam temperature, while the output selected is primary spray flow control valve opening.