Application of RBF Neural Network and Nonlinear Particle Filter in the Synthetic Ammonia Decarbornization

Abstract

The synthetic ammonia decarbornization industrial process is a complex production process with strong nonlinearity, large delay and strong coupling. It is difficult to set up the on-line control model of the process. The drawback of the conventional BP neural network algorithm used to building system modeling is easily falling into the minimum value. This paper is concerned with the use of a RBF (Radial Basis Function) neural network control based on particle filter algorithm to solve above problems. The RBF neural network can approximate any continuous function and the particle filter can deal with nonlinear problems. This approach could deal with a complex multi-phase system. The method introduced in the paper is to set up a RBF neural network control model firstly, and then, the weights of RBF neural network are optimized by the particle filter algorithm. Compared to the fuzzy neural network which is applied, the simulation result of the method in this paper demonstrates that the control accuracy and system response speed are improved significantly.