Next Article in Journal
Quantum Computing and Machine Learning on an Integrated Photonics Platform
Next Article in Special Issue
Resource Allocation and Pricing in Energy Harvesting Serverless Computing Internet of Things Networks
Previous Article in Journal
Chinese Cyberbullying Detection Using XLNet and Deep Bi-LSTM Hybrid Model
Previous Article in Special Issue
Literature Review: Clinical Data Interoperability Models
 
 
Article
Peer-Review Record

Design of a Meaningful Framework for Time Series Forecasting in Smart Buildings

Information 2024, 15(2), 94; https://doi.org/10.3390/info15020094
by Louis Closson 1,*, Christophe Cérin 2,*, Didier Donsez 3 and Jean-Luc Baudouin 1
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Information 2024, 15(2), 94; https://doi.org/10.3390/info15020094
Submission received: 18 January 2024 / Revised: 1 February 2024 / Accepted: 5 February 2024 / Published: 7 February 2024
(This article belongs to the Special Issue Internet of Things and Cloud-Fog-Edge Computing)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper is well presented, well written and it covers an important regard.

The minor concern that I have are:

1) Why did you use LSTM? There are other models that are state of the art that could be applied, such as LSTM with an attention mechanism.

2) Why did you not use EMD, which you covered in your review on line 157? This technique is very promising for signal denoising in time series.

I think if you address these two questions, that give a direction about why you apply your methodology and you include more related works, I'd say that at least 40 references would be fine, the paper would be acceptable.

 

Comments on the Quality of English Language

The paper is well presented, well written and it covers an important regard.

Author Response

Thanks for the relevant comments.

The number of references has been increased from 25 to 35. Instead of more papers that would not bring more relevant new information, we chose to explain the content of most articles in the related work section.

They now cover attentionEMD and quote Hybrid methods, and concrete cases from the field of smart-building. The end of the section now explains our positionning, compared to these papers.

For 1): as briefly explained, the quantity of data we used is not large enough to exploit an attention mechanism designed to bring focus on the most relevant part of the data or the most relevant model for model fusion.

For 2): EMD seems efficient in finding seasonalities with different granularity. Our encoding of hours/days/holidays might make it redundant. Furthermore, noise is not supposed to be information. However, we expect most of the unknown variations of high frequency to be "real" data despite not being explainable. They may be due to occupants' behaviors. Finally, we decided not to discuss this matter, as it constitutes a plain issue of analyzing the "real" data. This topic should come with the issue of controlling upper/lower bounds of satisfaction or consumption to compromise on building control, which is currently out of our scope.

Reviewer 2 Report

Comments and Suggestions for Authors

The Abstract is well written and provides all crucial information. In particular, an information about original contribution of the paper is provided and explained shortly. 

However, the original contribution is very important element of the scientific paper, therefore it should be not only mentioned in the Abstract, but better, wider described in Introduction. It is not in this manuscript – the authors should provide more, clear information about their contributions within the Introduction section. 

Section 3 related work is very short, and references are only briefly described. The authors should provide more detailed insights, descriptions for analysed publications. Moreover, based on them they should clearly identify main research, technical gaps, covered by their original contribution. 

In my opinion more literature positions, references should be discussed. For example at the beginning of the section 4/subsection 4.1 the authors describe “the literature in the field … RNN … linear models, … ” and there is no references for these literature. All the external solutions, methods included within the research described in the paper should be supported/backgrounded by literature references. 

There is no consistency in acronyms using/expansions. For example acronym LSTM is expanded within the Abstract (that is OK) and should be expanded at first using within the text – that is in line 154 (section related work). But it is not – instead it is expanded in line 185. Moreover, it is additionally expanded in line 361. Other example: acronym MLP for Multi-Layer Perceptions is expanded at least two times in line 187 and 208 and 244. The authors should verify all the acronyms and their extensions within the text, providing only one expansion at the first appearing in the text. 

- in lines 190-195 – the authors provide information that “we introduce an evaluation metric… ”- it is not clear – is it original contribution? Or is it based on other work – references? That should be always clear what is a source of the idea, solution … It is suggested to verify all parts of the text considering this remark. 

In the title of section 6 the authors declare coming work, but it is not described in this section at all. However, there is additional – not numbered (not well formatted) subsection Future works within the Section 7. The authors should consider where they are going to explain/describe coming work/future works or divide descriptions between those two parts of the text. Rearrangement of those parts is suggested.

General remark – the authors very often use personal form of sentences, phrases – for example “we propose” “our experiences” etc. It is better to use non-personal forms within the scientific papers – like “it is proposed” “the authors’ experiences” etc. It is suggested to consider this approach and to format most of the sentences in the text according to this rule. 

Author Response

Thanks for the time in reviewing our paper.

Contributions have been recalled at the end of the introduction while announcing the plan, consequently to the review.

We improved the section Related work. The number of cited papers went from 25 to 35, and we expanded the content of most of them. The section covers the justification using LSTM and other methods, comprised of Attention and EMD methods and "real" cases of AI for smart-buildings. The section concludes with a discussion to effectively distinguish our work and data from related work.

On LSTM acronyms: considering the few other acronyms used, it is easy to remember, so only the first appearance is now expanded to the full length.

The sentence “we introduce an evaluation metric” has been changed to “We also propose a new evaluation metric”. This point paid attention to keeping it clear in the introduction as well.

The confusion with the term “coming work” in the section title has been cancelled since it was not relevant to the section.

For general remarks, we modified many personal forms, but the ones we kept are related to implementation choices, our contributions, conclusions, and possibly heavy paragraphs where any passive form would complicate reading.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The paper is acceptable.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for addressing all of my comments and suggestions. 

Paper looks better, it is consistent and interesting. 

Back to TopTop