Abstract: In response to the strong dynamic coupling between the steam supply and power generation systems and the difficulty in optimizing multivariable continuous control under deep commissioning conditions of thermal power units, research on dynamic coordinated optimization control of steam supply and power generation under deep commissioning conditions of units based on DDPG algorithm is carried out. Firstly, an analysis was conducted on the operating characteristics of the steam supply power generation system under deep commissioning conditions of the unit. A multidimensional state space covering main steam temperature, pressure, steam supply flow rate, power generation, and turbine speed was constructed, and a deep reinforcement learning controller based on the actor critic framework was designed to achieve continuous and collaborative decision generation of multiple variables such as boiler combustion, turbine inlet, and generator excitation. The DDPG algorithm was used to design a control instruction execution mechanism to accurately execute the generated decisions into the steam supply power generation system, achieving coordinated optimization control objectives. Comparative experiments show that the proposed method exhibits significant advantages in dynamic tracking accuracy, response rate, and operational flexibility compared to existing control methods under rapid load fluctuations and deep peak shaving boundary conditions. It effectively improves the multi energy flow collaborative regulation capability and energy comprehensive utilization efficiency of the unit under harsh operating conditions.