Lithium-Ion Battery Prognostics through Reinforcement Learning Based on Entropy Measures

Round 1

Reviewer 1 Report

This paper presents the development of a prognostic approach to predict battery SOH and RUL. The research topic and results from this paper are interesting but before it can be considered for publication, some issues should be addressed first.

- The authors need to double-check the English grammar, figure quality, and formatting throughout the paper, in order to make it more legible for readers (random hyphens throughout, inconsistent formatting, etc.)

- The literature review of this paper is quite weak. The authors need to elaborate more on why SOH and RUL prediction is important for the BMS. Also, why would machine learning algorithms be needed if simpler models can be used with just as good accuracy? The authors should check https://doi.org/10.3390/batteries8020019

- The authors need to explain why it would be acceptable to have 90% training and 10% validation. Also, there are only 3 cells? Normally, for SOH and RUL prediction, you want to be able to predict the SOH and RUL much earlier in the battery life (within the first 50-100 cycles or so), not at the 10% at the end of its life. Can the authors prove that the proposed algorithm can achieve this?

Author Response

see attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The charts are difficult to read and can be improved by increasing letter size.

In Algorithm 3, it states the input is PE of voltage but output is capacity. But in this case, shouldn't ARIMA model be applied to capacity and also used for predict it? In addition what's the rational on setting AR model to the order of 5. Please clarify.

In Section 5.2 and 5.3, it's not explained what's driving the larger error in LSTM model in comparison to ARIMA. The latter is in general a mean-reverting process and therefore it's easy to see the prediction is following trend. However LSTM is involves more complicated parameterization and therefore the authors should identify what's driving the larger fluctuation in prediction.

Author Response

please see attachment 

Author Response File: Author Response.pdf

Reviewer 3 Report

The novelty and contribution of this paper is very marginal and unclear.
the methods have been around for quite some time and there is no modification applied on them. so its not ckear what is the novelty?
comparing a mumber of methods is not considered as novelty by itself.

There are some typos in the text, ex: page 3, line 89.

There is no justification why the three methods of LSTM,RL and ARIMA
are used in the study.

the literature review on prognostic and health estimation of Li-ion Batteries is nt enough, there are a large number
of recently developed methods for this problem that have not been mentioned at all.
 
it is necessary toinvestigate the impact of the data point orders on the results, shudffling the data and repeating the training and test procedure and reporting a concatination of accurecy metrices is required.

t is not clear how the models have been tued for their hyperparameters anh what guarantees a fair and reliable comparision.

with this level of details and conserns over presentations I wont recommend the paper for pubmication in that journal

Author Response

please see attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have improved the paper based on the comments

Reviewer 2 Report

Accept in present form

Reviewer 3 Report

I am happy with the modifications