GGAP-RBF Nueral Network Control Strategy For Superheated Steam Temperature

Large delay systems in the control process are very common and difficult to control, such as the majority thermal objects in the coal-fired power plant are with a large time delay. In the actual production, The traditional way of adjusting thermal objects is using the cascade PID controller. For the need of an object mathematical model with high requirements for accuracy, it makes the PID controller can not meet the control needs of objects characteristics variation.In this thesis, the principle of intelligent control and conventional control methods are combined to play their respective advantages. This strategy has a good adaptability and meet satisfied control effect when facing such a large delay complex object in the coal-fired power plants.Neural network control strategy is a branch of intelligent control. Neural network has a strong ability of nonlinear function approximation, adaptive learning, Parallel distributed processing capacity, and great robustness and fault tolerance. It provides an effective way to solve the unknown uncertain time delay nonlinear system modeling and controlling. RBF neural network can approximate any nonlinear mapping with a simple network structure, in which the output and the connection weights is linear thus linear optimization algorithm can be used. In recent years, it has become a research focus.This thesis give research on the existing radial basis function neural network (RBF) learning algorithm, analyzing the fault tolerance of the RBF neural network and giving improvement, improving the structure of RBF neural network, and putting forward a new learing algorithm of the RBF neural network-GGAP. And pay more attention on the application of the GGAP-RBF neural network in the modeling and controlling of the boiler superheated steam temperature control system and carry out simulation experiment. The simulation results show that the GGAP-RBF neural network is better than the traditional cascade PID controller, and proves that the proposed methods has strong effectiveness and feasibility.