Abstract: The thin film thermoelectric cooler has problems such as excessive energy consumption, low temperature control accuracy and chip overtemperature due to the mismatch between the current and the hot spot heat flux density of the chip in the process of integrated cooling chip. To solve these problems, a PPO-PID temperature control algorithm based on power stratification is proposed. The power stratification strategy is used to control the hierarchical current of thin film thermoelectric cooler with different cooling capacity requirements, and the proximal policy optimization (PPO) algorithm is used as the main control algorithm. By means of deep reinforcement learning, the parameters of PID controller are automatically adjusted according to real-time temperature deviation and rate of change, and the current is fine-tuned, which almost eliminates the large temperature fluctuations in the PID temperature control process. The simulation results show that in the process of realizing the temperature control target of thin film thermoelectric cooler cold end, the temperature control accuracy of PPO-PID temperature control algorithm can reach ±0.95 ℃, which is 79.35% higher than that of traditional PID control. In addition, filtering the output current of the hierarchical PPO-PID can improve the stability of the output current, but the average temperature difference between the cold end temperature and the target temperature rises to ±1.15 ℃, so whether to add filtering can be considered according to the requirements of accuracy and current stability.