The AGC system of hot strip mill is very complex,and it has the characteristics of high nonlinearity,time variation and time delay.The optimal control strategy was adopted for the above mentioned problems.The predicted extrapolation algorithm was used to solve the pure lag problem that caused by Mornitor-AGC detection,the remaining part of the lag time was improved by using Smith estimation algorithm and Fuzzy-PID parameters self-tuning feedback control.At the same time,the introduction of GM-AGC feedforward control,reduced the burden of feedback regulation.Simulation results show that in the case of 20% interference,the dynamic response of the system is fast,the adjusting time is short,and almost no overshoot.It give practical value to improve the strip quality in nonferrous and metallurgy industry.