Study of the neural network generalized predictive control for the circulating fluidized bed boiler generator

The circulating fluidized bed boiler (CFBB) generator has following characteristics: multi-variable, non-linear, strong coupling, time-varying characteristics. The combustion is still in the fluidized state, so its control is more complex, the traditional PID control effect is not ideal. This paper proposes a good nonlinear function approximation of BP neural network. Because the convergence rate of BP network is slower, a combination method with changing step and inducting momentum is used to improve the convergence rate. Meanwhile this paper achieve generalized predictive control (GPC) with online rolling optimization and real time feedback revision. Simulation results demonstrate the effectiveness of this algorithm.