Multi-Fault Diagnosis of Aluminum Electrolysis Based on Modular Fuzzy Neural Networks

Abstract

Aluminum electrolysis is a nonlinear process with the characteristics of multi-variable, strong coupling, time-varying and large time-delay. There are different types of faults that occur frequently in it. According to the fault characteristics of the aluminum electrolysis, a multi-fault diagnosis method of aluminum electrolysis which is based on modular integrated fuzzy neural network is proposed. Considering the shortages of a single network applied in multi-fault diagnosis, a multi-fault diagnosis platform with two layers of sub-network and decision fusion network is constructed in multi-fault diagnosis of aluminum electrolysis, combining fuzzy logic and neural network by the application of the concept of modular integration. Mixed particle swarm optimization algorithm is adopted in the paper so that the convergence speed and accuracy of the network can be increased to some extent. Simulation results show that the proposed method can improve the accuracy rate of fault prediction and give the prediction advance.