Normally， the hydro and thermal automatic generation control（AGC） units exist simultaneously in a multi-area interconnected system. Owing to the inherent non-linearity and complexity of the system， it's hard for an AGC system based on classical linear control theory to meet the requirements for smart grid development. In light of the non-linear characteristics of the hydro unit system， the reinforcement learning method is employed to transform AGC environment knowledge into prior knowledge to develop the AGC controller of a hydro-unit. A detailed analysis is made of the AGC controller design for the multi-area hybrid system of a hydro-power unit and thermal-power unit. In order to generalize the model proposed， a typical three-area hybrid system model is developed with the feasibility of the design method shown by simulation results.