Active Disturbance Rejection Control of Output Voltage of Solid Oxide Fuel Cell Based on Reinforcement Learning

ObjectivesIn order to improve the performance and lifetime of solid oxide fuel cell (SOFC) systems, the 100 kW SOFC system was taken as the research object. The continuous adjustment of the controller coefficients was explored through reinforcement learning to realize the best comprehensive performance, while ensuring the output voltage tracking performance.MethodsA mechanism-based SOFC output voltage system model was established, an improved nonlinear active disturbance rejection controller (NLADRC) was used to make the output voltage track the reference value well by controlling the input gas flow. Conventional single-channel controllers can only satisfy one objective at a time, and dual-channel controllers will increase system complexity, cost and risk of failure. An improved NLADRC controller based on the twin delayed deep deterministic policy gradient (TD3) was proposed to optimize the coefficients of nonlinear error feedback control law.ResultsThe designed controller can improve SOFC output voltage tracking performance without violating fuel utilization constraints.ConclusionsThe designed controller has the advantages of strong adaptability, high stability, and the ability to overcome uncertainty, providing theoretical reference for designing output voltage controllers in practical SOFC systems.