Robust Optimisation of an Online Energy and Power Management Strategy for a Hybrid Fuel Cell Battery Shunting Locomotive

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for your contribution in the Hydrogen. This paper presents a robust optimization methodology for tuning the parameters of an online energy and power management strategy (EPMS) in a hybrid fuel-cell/battery shunting locomotive. Because shunting locomotives face highly variable mission profiles, traditional offline or predictive optimization is impractical. The authors propose a process that combines Latin Hypercube Sampling to select representative design and scenario spaces, response-surface modeling and neural-network surrogate models to predict performance, and a Differential Evolution metaheuristic to identify optimal parameters. The optimized control minimizes fuel-cell start/stop events, power variation, battery over/under-charge, and total cost of ownership while ensuring robustness across a wide range of operating conditions. Addressing these following comments will improve the clarity, scientific rigor, and clinical relevance of the work and make it publishable in the Hydrogen.

 

Abstract and Introduction

1. The abstract clearly states the problem and proposed methodology, but it is dense with technical terms such as “metaheuristic algorithms,” “response surface model” without enough context for non-specialists. A short statement on the practical benefit like expected savings or performance gain would improve readability.

1. The introduction provides good motivation using cost comparisons of hydrogen vs. diesel/battery trains, but it would benefit from a clearer gap statement highlighting how the proposed method differs from prior robust optimization or surrogate modeling approaches.
2. References [1] and [2] are used for cost assumptions, but more recent cost data or sensitivity analysis would strengthen the argument given the rapid change in hydrogen and fuel-cell pricing.

Modeling Section

1. The longitudinal dynamics and traction drive models are described in detail, yet validation of these models against experimental or field data is missing. Even a brief comparison with measured locomotive performance would increase confidence.

1. Several equations such as modified Davis formula, resistance coefficients are presented without units or parameter values in the main text. Adding a table of key parameters would help readers reproduce the work.
2. The fuel-cell and battery are modeled as black boxes with supplier data, but assumptions on degradation mechanisms and auxiliary consumption are not explained properly.

Optimization Methodology

1. The stepwise methodology is well explained, but the choice of only 100 parameter combinations for training the surrogate model appears somewhat arbitrary without any justifications. The discussion later admits that a larger dataset (150 combinations) might improve performance. This limitation should be acknowledged earlier and quantified mor elaborately.

1. The paper combines four objectives into a single scalar using beta coefficients. The selection and sensitivity of these weights is critical yet only briefly justified. A sensitivity analysis or Pareto front comparison would clarify trade-offs between start/stop minimization and cost reduction.
2. The surrogate model employs a neural network with gradient boosting. Although high accuracy is reported, overfitting concerns remain because no independent validation set beyond the 20% test split is used.

Results and Discussion

1. The reported improvements (6% fewer start-stops, 15% lower power variation) are modest. The authors should discuss the practical significance of these gains relative to the computational effort and operational cost savings.

1. Validation scenarios show small differences between reference and optimized solutions. The discussion attributes this to the missions chosen, but it would be useful to quantify hydrogen savings or maintenance cost reduction to highlight industrial relevance.
2. The analysis identifies paired behaviors between objectives; FC stops vs. ESU range, yet no physical interpretation is offered for why these couplings occur.

Conclusion

1. The conclusion correctly emphasizes robustness, but it could provide a clearer roadmap for future work, such as integrating real-time mission recognition, scaling to different locomotive types, or testing with field data.

1. Mentioning potential industrial deployment steps like pilot tests at Alstom would reinforce the practical value of the research.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper addresses a practical and current problem: the robust optimization of the energy management strategy (EPMS) for hybrid shunting locomotives (fuel cells + batteries). The authors propose a methodology based on reducing the parameter and scenario space (through Latin Hypercube Sampling), utilising surrogate models (Response Surface Model and Neural Networks with boosting), and metaheuristics (Differential Evolution) to identify the optimal parameters. The topic is very relevant, given the pressure to reduce TCO and increase component durability in railway applications with highly variable profiles. The manuscript is well structured, methodologically sound and detailed, but has some limitations regarding the validation and clarity of the discussions. Overall, the contribution is valuable, but major revisions are required before publication.
1. Please clarify how the proposed methodology could be validated experimentally (e.g. hardware-in-the-loop or prototype testing).
2. The optimization strongly depends on the chosen β coefficients. Please provide more justification or sensitivity analysis on these weights.
3. Figures and readability – Figures 7–10 (PDF/CDF distributions and scenario space) are dense and difficult to read. Please provide clearer visualizations. 
4. Some sections (§3.2–3.3) are dense and schematic summaries could be useful.
5. Conclusions should focus on 2–3 key findings and practical implications.
6. References are current and varied, but could include more experimental applications in rail transport (not just automotive).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have clearly addressed all the comments, and the paper is ready for publication.

Reviewer 2 Report

Comments and Suggestions for Authors

We appreciate the thorough revisions. Clarifications regarding methodology, figures, and conclusions have substantially improved the manuscript. Responses adequately address all concerns, and the paper is now well-structured and clear.