Abstract: Based on the theory of genetic algorithm（GA）, the parameters of thermostatically controlled loads（TCLs）control system were optimized, and a better control strategy was proposed by analyzing the operation effect of optimized parameters. The equivalent thermal parameter（ETP）model was simple and practical, which was often used to describe the dynamic characteristics of large-scale TCLs. Therefore, this model was used to study the control strategy of large-scale TCLs. According to the predicted outdoor temperature curve, the unbalanced active power, which was the demand load,could be calculated. The difference between the demand load and the response load functions as the control signal, which was adjusted the total power of TCLs through the proportional integral controller. In this process, a GA was introduced to optimize the control parameters after multiple rounds of selection, crossover and mutation. The optimized control parameters were substituted into the calculation to ensure the high performance of the control system. At the same time, the temperature set-point, as the core of thermostat set-point control mechanism（TSCM）, was discretized to reduce the switching frequency. In the temperature control strategy of large-scale TCLs, the tracking effect of large-scale TCLs control system was characterized by the difference between the demand load and the response load, and changes of control parameter optimization based on a GA were discussed. Correspondingly, with the integrated square error（ISE）of the aggregate power of large-scale TCLs, the difference between continuous and discrete variation of the temperature set-point was compared by changing the minimum adjustment quantity. The simulation results illustrated that the parameters optimized by a GA could make the aggregate power of TCLs follow the required curve more smoothly in direct load control（DLC）, which meant the system responds to the changes of the external temperature in a shorter time. In addition, to avoid the adverse effects of frequently switching, the discretized temperature set-point could ensure the accuracy of the control system. The optimization method based on a GA could apparently improve the performance of the control system and bring more benefits.