Abstract: Aluminum electrolysis is a complex industrial process with difficult to control and strong interference, where is difficult to make the aluminum electrolytic cell at the best working condition. Consequently, in this study adopted a fault-tolerant control strategy based on extension neural network according to the characteristics of aluminum electrolysis process. The control principle is adopted different control strategies which combined detection and control, made a diagnosis according to the electrolytic cell condition. It was made the aluminum electrolytic cell in the best working state, difficult to control and strong interference optimized the capability of the system such as real-time ability, stability and precision, real-time ability, stability and precision.