Particle Swarm Optimization Fuzzy Neural Network and its Application in Soft-Sensor Modeling of Acrylonitrile Yield

Firstly, particle swarm optimization fuzzy neural network (PSOFNN) is proposed and the algorithm flow of PSOFNN are given in this paper. Secondly, PSOFNN is applied in soft-sensor modeling of acrylonitrile yield. The new method assumes that fuzzy neural network (FNN) is used to construct the soft-sensor model of acrylonitrile yield and particle swarm optimization algorithm (PSO) is employed to optimize parameters of FNN. Moreover, how to choose the auxiliary variables of soft-sensor is studied carefully. Experiment results show that the model based on PSOFNN has higher precision and better performance than the model based on PSONN. The method proposed by this paper is feasible and effective in soft-sensor of acrylonitrile yield.

1. Introduction

Acrylonitrile is the major raw material of organic chemistry and it is widely applied to fabricate the macromolecule material, such as synthetic fibre, synthetic resin, synthetic rubber and so on. Additionally, both acrylonitrile polymer and acrylonitrile ramification are very useful in many fields [1]. In recent years, with the acrylonitrile downriver products' developing, especially the research and application of new downriver fine chemistry products, the demand of acrylonitrile is increasing rapidly in the world. Moreover acrylonitrile polymer and acrylonitrile ramification are also popular and useful for fabricating the architecture material and commodity.