Abstract: In thermal power plants, compressed air systems provide necessary air sources for numerous equipment and process flows. Without graded control, long-term operation at excessively high or low pressures can accelerate equipment wear and tear. To study the graded control of compressed air pressure in thermal power plants under DCS. The DCS system adopts a multi-level hierarchical structure, deploying controllers in the control cabinet and establishing connections with each controller through a low-speed bus. Using an improved particle swarm optimization (PSO) algorithm for PID controllers, the optimal PID parameters are tuned through dynamic inertia weights and local average mechanisms. Furthermore, according to the process requirements, the pressure is divided into three zones: high, medium, and low, and non-linear PI regulation is used for graded control of compressed air pressure. The experimental results show that the fluctuation trend and amplitude of the internal pressure P and the pressure drop ΔP of the regulating valve remain consistent; The steady-state error of the system remains within the specified range, and there has been no overshoot phenomenon. Significantly improved the system's adaptability to complex loads and pressure disturbances.