Abstract: Excitation control system of synchronous generator is a strong nonlinearity, multi-variable, strong couple and time-varying control system. It is very difficult for traditional Proportional Integral Derivative (PID) to get good control performance. A new excitation control strategy based on PID controller and Cerebellar Model Articulation Controller (CMAC) is proposed in this study. To solve the problem of PID and CMAC compound controller multi-parameter setting, an Improved Multi-agent Genetic Algorithm (IMAGA) is presented. The PID parameters Kp, Ki, Kd and CMAC parameters η, α are regarded as a agent. Each agent continuously improves its fitness value through competition and cooperation between the other agents according to the objective function of Integral of Time-weighted Absolute value of the Error (ITAE). This algorithm adopts multi-agent coordinate optimization to realize the five parameters of Kp, Ki, Kd, η, αonline tuning. The simulations results show that the compound control scheme based on multi-agent genetic algorithm can improve the precision of excitation control, the speed of responding and has better dynamic and steady-state characteristics.