Data-Driven Fuel Cell Stack Modeling for Real-Time Control Applications

点击翻译按钮获取中文摘要

A fuel cell vehicle (FCV), considered by many as the ultimate clean electric vehicle, depends on the fuel cell stack for its performance. Its drivability and maneuverability rely on control-oriented models of fuel cell stacks for real-time control and optimal management of energy and power. A typical automotive fuel cell stack consists of several hundred individual cells; traditional empirical correlations-based models are unable to represent all states for this high degree of freedom system with sufficient accuracy. Furthermore, first-principles based 2D and 3D computational fluid dynamics models are too computationally intensive for real-time control applications, given the resource limitations of market engine control units. The objective of the research described in this proposal is to develop sufficiently accurate and computationally efficient predictive computer models of fuel cell stacks that can be deployed for real-time dynamic control of FCVs. Two approaches will be taken in this study : the first approach will be first-principles based physics modelling, with a 0-dimensional (0D) model for the reactant distribution in the stack manifold, 1D along the flow channel neighboring the individual fuel cells in the stack, and 1D across the individual cells (hence referred to as 0 + 1 + 1 model). The second approach will be based on a neural network with machine learning (artificial intelligence), hence referred to as the NN/AI model. Both models will be validated and evaluated against real stack performance data, and their viability for real-time dynamic control of FCVs will be evaluated in collaboration with the project supporting organization, Toyota, the manufacturer of the world's first commercial FCVs. The results of the proposed research will be more accurate and faster control-oriented models, accelerating the development of FCV controllers and improving the powertrain performance of FCVs.