Premium
Deep Reinforcement Learning Guided Cascade Control for Air Supply of Polymer Exchange Membrane Fuel Cell
Author(s) -
Zhou Jianhao,
Liao Yuhui,
Liu Jun,
Xue Yuan,
Xue Siwu
Publication year - 2021
Publication title -
energy technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.91
H-Index - 44
eISSN - 2194-4296
pISSN - 2194-4288
DOI - 10.1002/ente.202100149
Subject(s) - cascade , reinforcement learning , control theory (sociology) , controller (irrigation) , air compressor , model predictive control , pid controller , computer science , engineering , control engineering , control (management) , temperature control , artificial intelligence , agronomy , chemical engineering , biology , aerospace engineering
The power consumption of an electric air compressor becomes the major parasitic energy waste leading to the deterioration of net efficiency of the fuel cell system (FCS). Optimal control of the oxygen excess ratio (OER) is recognized as a feasible way to resolve the paradox of stack health and system efficiency. Herein, a novel cascade control scheme that leverages model predictive control (MPC), fuzzy Proportion Integral control (FPI), and deep reinforcement learning technique is proposed to regulate the OER for a 75 kW FCS. The master‐slave control structure is adopted, and a reference air mass flow rate and the motor voltage of the air compressor are taken as control targets, respectively. A deep deterministic policy gradient (DDPG) algorithm is used as a parameter tuner for MPCProportion Integral (PI) controller. The results indicate that both MPCFPI and DDPGMPCPI can achieve better control performance indices in comparison to classic controller under step response and a typical driving cycle. The DDPG‐guided cascade controller is also capable of regulating the OER tracking error and the derivative of rotation speed of the air compressor, which reveals the potential for extending the longevity of the FCS.