Abstract: Thermostatically controlled loads can provide frequency regulation auxiliary services for power grids through demand response. Aiming at the problems of limited adjustment ability and large temperature variation on traditional control methods, a hybrid control strategy based on the cerebellar model articulation controller（CMAC） is proposed. First, the CMAC is used to optimize the power regulation quantity in real time, so as to assign the regulation task to the load clusters with different control methods. Then, the load cluster completes the assigned adjustment task and defines the energy storage index and discomfort index according to the different needs of users, respectively. Finally, the fuzzy comprehensive evaluation method is used to evaluate user satisfaction, and then the user satisfaction and root mean square error are weighed to establish a comprehensive evaluation index which is fed back to the CMAC, thereby providing a basis for the optimization of power regulation quantity. Simulation results show that the proposed strategy can not only increase the system tracking accuracy, but also improve the user satisfaction degree.