AI Based Autonomous Control for Oil and Gas Using FKDPP Technology

This paper demonstrates the application of the Autonomous Reinforcement learning based AI Controller, based on the Factorial Kernal Dynamic Policy Programming (FKDPP) algorithm in an upstream oil and gas processing facility to enhance process stability and reduce manual operator interventions. FKDPP utilizes reinforcement learning to address operational challenges, such as liquid level and pressure fluctuations in separators caused by unstable fluids and uneven crude distribution. The AI controller analyzes inputs from multiple separators simultaneously, managing final control elements to optimize conditions with minimal reliance on traditional PID control. Trained on dynamic simulators and plant historical data, it adjusts separator inlet level control valves (LCVs) to stabilize levels, gas flow, suction pressure, and inlet pressure. Integrated via OPC communication at Level 3 of the control system, it considers upstream and downstream dynamics to mitigate disturbances. Currently operating in closed loops on 3 of 5 separators, initial performance shows promising potential for scaling across platforms, reducing intervention, and improving operational efficiency. Future phases aim to achieve full autonomous control on all 5 separators.