Performance Evaluation of PID and AI-Based Controllers for Centrifugal Pump Operation using Dynamic System Simulation at PT. Pelindo Energi Logistik (PEL)

Abstract

The operation of centrifugal pump systems driven by AC motors is crucial for industrial fluid transportation, particularly in energy and logistics facilities where stable pressure and flow regulation are essential to maintaining operational efficiency. However, conventional control strategies often face limitations when dealing with nonlinear pump dynamics and fluctuating operating conditions. This study therefore evaluates the performance of proportional–integral–derivative (PID) and artificial intelligence (AI)-based controllers for centrifugal pump operation using dynamic system simulation. The research compares three control approaches: PID, neural-network-based control (NNC), and reinforcement learning (RL) based on the Soft Actor-Critic algorithm. A dynamic simulation model representing the pump system used in PT. Pelindo Energi Logistik was developed using operational parameters and hydraulic calculations to analyze system response over a 1000-s simulation period. The results indicate that the RL controller provides improved pressure regulation performance, achieving lower error metrics with MAE and RMSE values of approximately 140.5 kPa and 141.3 kPa, compared with about 160 kPa obtained using the PID controller. In addition, RL maintains a stable efficiency level of around 64.8%, while the other controllers exhibit negligible efficiency values in the simulation environment. These findings demonstrate that reinforcement learning offers superior adaptability and energy-aware control behavior for centrifugal pump systems. Consequently, AI-based control strategies have strong potential to improve the operational stability and efficiency of industrial pumping infrastructure.