Novel Fuzzy Control Energy Management Strategy for Fuel Cell Hybrid Electric Vehicles Considering State of Health

Round 1

Reviewer 1 Report

The authors design a fuzzy control strategy to manage the power supply of a hybrid electric vehicle, then they use a genetic algorithm to optimize the parameters of the fuzzy controller. Finally, they train a feed-forward neural network for predictive control.

The paper is well structured and clear, however, I have some major concerns about the GA, the neural network, and the results.

• line 119 "to realizing" 
• Tab1
  The first 4 rows and last 4 rows are the same.
• GA
  First, It looks like all the variables are real numbers, thus I would recommend using other evolutionary algorithms, such as differential evolution, instead of GA. Otherwise, I suggest a better description of the crossover and mutation strategies.
• Line 298 "The code length of 15 decision variables to be optimized is 115 bits." why? what does it mean? please explain
• A small population (n=20) may lead to premature convergence to a local min (see Fig7), moreover a very small mutation probability limits the exploration of the search space (Tab4, the second half of the optimized chromosome is the same as the original). I suggest repeating the experiments with different hyperparameter combinations and different RNG seeds to check the convergence of the GA.
• Line 366 "... (the FFNN) is trained with 1309 sets of data, and 60 sets of data are tested". This kind of train/test set may lead to unreliable results. I suggest performing k-fold cross-validation.
• Fig. 7
  The cost function (13) should be minimized, but in Fig7 the function is increasing. Please explain.
  Can you please describe the boundaries of the cost function? (0,+inf)? What is the baseline (such as a human expert) of the cost function?
• Tab.4
  What is the "original" chromosome? Human expert? Randomly sampled? Why (z1, ...z8) are the same in the "optimized" chromosome
• Tab5
  Please add the definition of UDDS, NEDC, HWFET. 
  What is FL? 
  Are you training a single NN for the UDDS, NEDC, HWFET, or three NNs one for each scenario?
• Tab8.
  Why does GA score change from tab5? Which is the score of the NN?
  Are BPMM, ANN, and NN the same network? 

Author Response

Thank you for your careful and comprehensive review. We have thoroughly checked and revised this manuscript in accordance with the comments. The following is the response to each comment. Please see the attachment. 

Author Response File: Author Response.docx

Reviewer 2 Report

This paper developed a fuzzy control energy management strategy for fuel cell – battery hybrid electric vehicle with online estimation ability of fuel cell heath status, using a genetic algorithm for its optimization and an artificial neural network for predictive control. The paper structure is well organized and the content is persuasive with details. I only have a few comments in below:

1. For citing other’s work, please use author names together with the reference number.
2. Some abbreviations are not specified in the main text, such as UDDS, NEDC, and HWFET.
3. In Section 5, please delete the first sentence

Author Response

Thank you for your careful and comprehensive review. We have thoroughly checked and revised this manuscript in accordance with the comments. The following is the response to each comment.

Point 1: For citing other’s work, please use author names together with the reference number.

Response 1: The style of citizen has been changed in this manuscript, especially in Section 1, Introduction Part, Paragraph 2 and Paragraph 4.

Point 2: Some abbreviations are not specified in the main text, such as UDDS, NEDC, and HWFET.

Response 2: The definitions of acronyms have been added in Section 3.3, Paragraph 2 and in Section 4.1, Paragraph 1. In addition, they are also added in Nomenclature Section, Acronyms Part. They are the standard driving cycles for vehicle performance test.

Point 3: In Section 5, please delete the first sentence.

Response 3: This sentence has been deleted in Section 5 Conclusions.

Reviewer 3 Report

This paper proposes a rule-based fuzzy control strategy and has used a genetic algorithm to optimize the fuzzy controller. Then, an artificial neural network is used to deal with variable driving cycles. Simulation results are provided to validate the performance of the proposed method.

From the point of my view, the organization and quality are both bad and a few major issues need to be addressed before considering publishing.

• The major problem of this paper is that the novelty is weak. The key technology of this study is fuzzy controller combining the genetic algorithm and neural networks, they are conventional algorithms and many previous studies have been proposed. Furthermore, the genetic algorithm and fuzzy logic algorithm may not get the global optimal solution in Section 3.2.
• The introduction is not well-organized to establish the research background. For example, the author attempts to provide many related researches on fuel cell, energy management strategies and so on, but most of the information contained is not closely related to the knowledge of fuzzy controller, genetic algorithm and neural networks involved in this article.
• The paper has many careless errors such as using unclear simulation curves (See Figure7~10), using incorrect table title (such as Line 465 “Results comparison between NN-FCEMS and NN-FCEMS ……”), ignoring symbol definition (Line 194: RC model) and using inappropriate sentences (Line 477: This section is not mandatory but can be added to the manuscript if the discussion is unusually long or complex.) etc.
• The weight parameters selected for the various costs of fuel cells have not been found in Reference [41], and the validity of the evaluation deserves further discussion.
• The optimization goal of fuzzy logic algorithm is hydrogen consumption, the optimization goal of genetic algorithm is hydrogen consumption and the SOH, so the latter includes the former. I don't think it's necessary to do it step by step. You should directly combine genetic algorithm and fuzzy logic algorithm to optimize hydrogen consumption and the health state of the battery.
• It is mentioned in Section 3.3 that the neural network algorithm can reduce the amount of calculation, but there is no calculation time comparison curve or comparison table between neural network algorithm and genetic algorithm to illustrate the improvement effect of neural network algorithm on real-time calculation.
• For Table 4: Comparison of membership functions between FCEMS and GA-FCEMS, the original value and optimized value for “zi, i=1, 2 , … , 8” are no different. Since the rules or membership functions of the fuzzy controller are optimized, does “z” not belong to the optimized factor?
• Line 388: Comparing Figure 8(a) and Figure 8(b), there is no obvious difference between the output power of the fuel cell and the power of the battery. Besides, the output power fluctuation of the fuel cell has not been significantly improved.
• In Section 4.2, the MAPEC of the HWFET cycle in the Table 8 is 5.0749%, which seems to be a very ideal prediction accuracy.
• The content expressed in Table 8 is difficult to understand, not only due to the wrong title but also the missing units.

Author Response

Author Response File: Author Response.docx

Reviewer 4 Report

Reviewer’s comments

 Manuscript Number: energies-1300277

Title: Novel fuzzy control energy management strategy for fuel cell hybrid electric vehicles considering state of health

Journal: Energies

The manuscript reports a rule-based fuzzy control strategy of fuel cell hybrid electric vehicles. The work is well organized and written. However, some comments need to be addressed before the publication in Energies, such as:

1. High-resolution figures are needed, where Figs. 7, 8, 9, and 10 are a blur and not readable.
2. The mentioned equations need to be supported by relevant references.
3. The main findings should be compared with those obtained previously to show the importance of using this study.
4. Page 16, lines 477 and 478, what is meant by “This section is not mandatory but can be added to the manuscript if the discussion is unusually long or complex.”
5. More attention is needed for formatting specifically the units.

Author Response

Thank you for your careful and comprehensive review. We have thoroughly checked and revised this manuscript in accordance with the comments. In addition, English language and style have also been edited. The following is the response to each comment.

Point 1: High-resolution figures are needed, where Figs. 7, 8, 9, and 10 are a blur and not readable.

Response 1: We reedited all pictures for a clearer presentation, which include Figure 7, 8, 9, 10. Besides this, the font size of the tick label in figures is magnified.

Point 2: The mentioned equations need to be supported by relevant references.

Response 2: All equations have been added citations, especially in Section 2, modelling and in Section 4.2, result validation.

Point 3: The main findings should be compared with those obtained previously to show the importance of using this study.

Response 3: There are mainly two main findings in this article:

1. A fuzzy control energy management strategy optimized using genetic algorithm (GA-FCESM) is established to comprehensively optimize the economy and the health state of the fuel cell electric vehicle powertrain system. And the simulation results show that compared with the empirical fuzzy controller (FCEMS), the GA-FCEMS can effectively reduce the hydrogen consumption and power loss under different driving conditions, then slow down the reduction of the fuel cell SOH. The effectiveness is demonstrated in Section 4.1, Table 6 and Table 7 (in the revised new manuscript).
2. The neural network algorithm is then used for predictive control of the fuzzy controller (NN-FCEMS) for the onboard application. The results are discussed in Section 4.2. As shown in Table 8, the prediction accuracy of this proposed network is relatively high. And the cost is nearly as low as the cost in GA-FCEMS.
3. Besides this, we added more introductions and citations about the application of fuzzy logic, genetic algorithm and neural network in the establishment of energy management strategies in Section 1 Paragraph 3. These introductions give more research background and findings obtained previously.
4. Finally, we have reedited the last part Section 5 ‘Conclusions’ to make the findings more prominently with accurate numbers. And the further research targets are added in the last sentence of this section.

Point 4: Page 16, lines 477 and 478, what is meant by “This section is not mandatory but can be added to the manuscript if the discussion is unusually long or complex.”

Response 4: This sentence is an initial sentence in the manuscript template, and has been deleted in Section 5 Conclusions.

Point 5: More attention is needed for formatting specifically the units.

Response 5: All format errors in this manuscript are rechecked and have been corrected. The superscript and subscript of the units are reedited.

Author Response File: Author Response.docx

Reviewer 5 Report

The manuscript presents a rule-based fuzzy control strategy based on the constructed data-driven online estimation model of fuel cell health. The results show that the developed strategy has a good generalization capability for variable driving cycles. The results are interesting for the community. The manuscript is worthy of consideration. However, the authors need to address the comments/questions below before making final recommendation for publication.

1. The models for SOC and power had several assumptions. The author should justify the accuracy of these parameters.
2. The authors should describe the working conditions for the vehicle to justify the comparison.
3. Please delete the sentence “This section is not mandatory but can be added to the manuscript if the discussion is unusually long or complex.” in the conclusion section.
4. All equations should have appropriate citations.
5. There are several format errors in the manuscript. For example, the font size of the tick label in Figure 9 is too small; the unit in Table 1 does not show superscript; LiMn2O4 should have subscript, etc.

Author Response

Thank you for your careful and comprehensive review. We have thoroughly checked and revised this manuscript in accordance with the comments. The following is the response to each comment.

Point 1: The models for SOC and power had several assumptions. The author should justify the accuracy of these parameters.

Response 1:

In the original manuscript, some conditions and assumptions are ignored, which made a lot of misunderstandings about the battery modelling and subsequent calculation process. So, we changed Section 2.3 into a more comprehensive one.

1. For battery modelling, we use Rint model for simplification to characterize the battery's voltage and internal resistance, referring to [47] and [48] in Section 2.3 Paragraph 1. And the change of open circuit voltage and the charge or discharge resistance of a single cell under different SOC are measured at the temperature of 25 ℃. Besides this, the parameters in this part, such as the parameters in Table 1, are all measured at the temperature of 25 ℃.
2. For SOC calculation, we use ampere-hour integral method, and this Equation 6 refers to several studies [50][51][52] in Section 2.3 Paragraph 2, which ignores the influence of the consistency and the temperature.
3. For the calculation of equivalent hydrogen consumption of the battery, we refer to the method proposed by Xu et al. [53] in Section 2.3 Paragraph 3.

Point 2: The authors should describe the working conditions for the vehicle to justify the comparison.

Response 2: In Section 4 ‘Results and discussion’ Paragraph 1, Sentences have been added in this part to describe the working conditions for the vehicle. ‘And simulation conditions are as follows: In all simulation processes, the initial value of battery SOC is set to 0.5. The ambient temperature for vehicle operation is 298.15K. Because of the focus on the effectiveness evaluation of different strategies for the cost function, the amount of on-board hydrogen storage is unlimited and not considered.’

Point 3: Please delete the sentence “This section is not mandatory but can be added to the manuscript if the discussion is unusually long or complex.” in the conclusion section.

Response 3: This sentence has been deleted in Section 5 Conclusions.

Point 4: All equations should have appropriate citations.

Response 4: All equations have been added citations, especially in Section 2, modelling and in Section 4.2, result validation.

Point 5: There are several format errors in the manuscript. For example, the font size of the tick label in Figure 9 is too small; the unit in Table 1 does not show superscript; LiMn2O4 should have subscript, etc.

Response 5: All format errors in this manuscript are rechecked and have been corrected. The font size of the tick label in Figure 9 is magnified. Besides this, we reedited all pictures for a clearer presentation, which include Figure 7, 8, 9, 10. And the figures are still clear when they are magnified.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I think the authors improved the paper, clarifying some aspects and making it more appealing for the readers.
I have no more suggestions to provide.