Assessment of hydrogen vehicle fuel economy using MRAC based on deep learning

Han, Jaesu; Yi, Sun; Yu, Sangseok · 2025 · DOAJ

DOI: 10.1038/s41598-025-97082-4

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of optimizing fuel economy and motor control in Fuel Cell Electric Vehicles (FCEVs). Unlike Battery Electric Vehicles (BEVs) or internal combustion engine vehicles, FCEVs utilize a hybrid power system combining proton exchange membrane fuel cells (PEMFCs) and batteries. This configuration requires delicate power management to ensure durability and efficiency, as the power generation is disconnected from traction. The authors argue that conventional logic-based controllers, such as Proportional-Integral (PI) control, struggle with the nonlinear dynamics and uncertainties inherent in these systems. To overcome these limitations, the paper investigates the application of Deep Learning-based Model Reference Adaptive Control (DL-MRAC) for motor control, aiming to achieve superior fuel economy and robustness compared to traditional PI and standard MRAC methods. The researchers constructed a comprehensive simulation model of a hydrogen vehicle using MATLAB/Simulink. The system included a fuel cell stack (828 cells, 380 cm² active area), a lithium-ion battery (750 V, 6.5 Ah), DC–DC converters, a 3-phase inverter, and a Permanent Magnet Synchronous Motor (PMSM). The vehicle specifications were based on commercial hydrogen vehicles, with a total mass of 19,000 kg. Three control strategies were implemented and compared: PI control, standard MRAC, and DL-MRAC. The DL-MRAC architecture utilized neural networks to identify the plant model and train a controller to track reference outputs, employing normalization to handle nonlinear behaviors across varying input ranges. Performance was evaluated through step inputs of vehicle speed references (5–40 km/h) and disturbance tests involving a 30-degree incline to simulate rolling resistance changes. Metrics included speed tracking accuracy, convergence speed, oscillation, undershoot, and fuel economy calculated via battery State of Charge (SOC) simulations. The results demonstrated that DL-MRAC provided the best overall performance. In speed response tests, standard MRAC exhibited instability at high speeds due to its reliance on a single adaptive constant, whereas DL-MRAC maintained stable tracking despite initial overshoot. Under disturbance conditions, PI control showed slower recovery and abrupt oscillations, while MRAC suffered from pronounced undershoot. DL-MRAC effectively balanced responsiveness and stability. Crucially, DL-MRAC achieved the highest fuel economy among the three controllers, with a value of 19.903, compared to 19.896 for MRAC and 19.872 for PI control. Although the current margin of improvement is slight, the authors note that DL-MRAC’s performance is expected to improve with additional training data. The significance of this work lies in demonstrating the viability of deep learning-enhanced adaptive control for FCEV motor management. By leveraging data-driven optimization rather than fixed logic, DL-MRAC offers a more robust solution for handling the complex, nonlinear dynamics of hybrid power sources. This approach not only enhances fuel efficiency but also contributes to the durable operation of PEMFCs by smoothing power demands. The study suggests that integrating AI-based controllers into vehicle powertrains can overcome the limitations of conventional methods, paving the way for more efficient and reliable hydrogen vehicle technologies.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• physical impairment