Neural network based optimization for cascade filling process of on-board hydrogen tank

Compressed hydrogen storage is widely used in hydrogen fuel cell vehicles (HFCVs). Cascade filling systems can provide different pressure levels associated with various source tanks allowing for a variable mass flow rate. To meet refueling performance objectives, safe and fast filling processes must be available to HFCVs. The main objective of this paper is to establish an optimization methodology to determine the initial thermodynamic conditions of the filling system that leads to the lowest final temperature of hydrogen in the on-board storage tank with minimal energy consumption. First, a zero dimensional lumped parameter model is established. This simplified model, implemented in Matlab/Simulink, is then used to simulate the flow of hydrogen from cascade pressure tanks to an on-board hydrogen storage tank. A neural network is then trained with model calculation results and experimental data for multi-objective optimization. It is found to have good prediction, allowing the determination of optimal filling parameters. The study shows that a cascade filling system can well refuel the on-board storage tank with constant average pressure ramp rate (APRR). Furthermore, a strong pre-cooling system can effectively lower the final temperature at a cost of larger energy consumption. By using the proposed neural network, for charging times less than 183s, the optimization procedure predicts that the inlet temperature is 259.99-266.58 K, which can effectively reduce energy consumption by about 2.5%. (c) 2020 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.

Automated summary

A study established an optimization methodology to determine initial thermodynamic conditions for compressed hydrogen storage with minimum energy consumption; training a neural network for predicting optimal filling parameters can effectively reduce energy consumption.