Distributed Intelligence in Autonomous PEM Fuel Cell Control

Abstract

A combination of perceptive and deliberative processes is necessary to ensure the efficient and autonomous control of proton exchange membrane fuel cells (PEMFCs) under optimal humidification conditions. These processes enable monitoring and control tasks across various application scenarios and operating conditions. Consequently, it becomes crucial to adjust parameter values corresponding to different states of the PEMFC during its operation. In this context, this work presents the design and development of an architecture for the control and management of a PEMFC with a maximum power output of 500 [W] based on intelligent agents operating under optimal conditions (membrane humidification). The proposed architecture integrates perception and action algorithms that leverage sensory and contextual information using heuristic algorithms. It adopts a hierarchical structure with distinct layers, each featuring varying time windows and levels of abstraction. Notably, this architecture demonstrates its effectiveness in achieving the desired energy efficiency objective, as evidenced by successful validation tests conducted with different electrical power values delivered by the fuel cell, encompassing three distinct operating states (dry, normal, and flooded). An exemplary application of this scheme is the dynamic control of the humidification of the polymeric membrane, which further highlights the capabilities of this architecture.