To mitigate the demand for ultrahigh precision thermal control in the field of semiconductor lithography, a thermal control method based on model prediction is proposed in this paper. The step and pseudorandom thermal disturbance input signals of a lens system are generated. The temperature response data of the lens system are obtained using a high precision temperature-measurement system, and the thermal response model of the lens system is obtained via the model identification method. The rolling optimization strategy is used to derive the model’s predictive thermal control law, and the precision lens system’s thermal control experimental platform is established. The thermal control experiment results show that the thermal control method has a fast convergence speed, strong anti-interference ability, high precision, and the temperature control error is controlled within ±8 mK. The thermal-phase-difference test results show that the control method successfully suppresses the lens system’s focal plane drift. Further, the method is suitable for application in the field of high-end equipment with strict temperature control accuracy requirements.