Abstract: According to the characteristics of the aluminum electrolysis fault, the principal component analysis and improved BP network are used for extracting fault feature, the improved BP neural network can extract the fault feature and can also be used as preliminary diagnosis of the fault, thereby electrolytic method for multiple faults diagnosis of three neural networks aluminum is taken use, this method analyzes the deficiency of single neural network and two stage neural network fault diagnosis and in order to form the decision fusion network, the wavelet analysis and neural network are combined organically. The simulation results show that: the aluminum electrolysis fault diagnosis resrarch based on principal component analysis has the characteristic of large amount of ault forecast.