A Demand Forecasting Strategy Based on a Retrofit Architecture for Remote Monitoring of Legacy Building Circuits

Round 1

Reviewer 1 Report

This research innovatively forecasts energy demand in legacy building systems using an IoT retrofit architecture and the SmartLVGrid metamodel. Various learning algorithms were applied with the LSTM neural network and the XGBoost Regressor showing superior performance. The study presents a cost-effective approach to improving energy efficiency, facilitating remote monitoring and predictive analysis in older facilities.

 

 

Overall, the paper is well written and structured. However, it needs to be improved on different aspects:

1) While the scope of the research presented in this paper is extensive, it would be beneficial to enrich Tables 1 - 4 with more information such as the utilized machine learning model, dataset details, associated limitations and advantages, and so forth.

2) Figure 6 should be expanded to increase it width.

3) It would be prudent to incorporate a discussion regarding the limitations and drawbacks of the proposed method within the Conclusion section, prior to outlining future research directions.

4) Kindly ensure that the text undergoes a thorough proofreading process to rectify any existing grammatical issues and typos. Additionally, please aim for clarity and brevity in sentence construction, as some of the current sentences are rather lengthy and challenging to comprehend.

 

The text should undergo a thorough proofreading process to rectify any existing grammatical issues and typos. Additionally, clarity and brevity in sentence construction should be considered, as some of the current sentences are rather lengthy and challenging to comprehend.

Author Response

Replies to reviewer 1

First, we would like to thank you for taking the time to evaluate our work.

Comments: Overall, the paper is well written and structured. However, it needs to be improved on different aspects: 1.

While the scope of the research presented in this paper is extensive, it would be beneficial to enrich Tables 1 - 4 with more information such as the utilized machine learning model, dataset details, associated limitations and advantages, and so forth.
R: Many thanks for the valuable suggestion. We have reformulated Tables 1 - 4 to include the learning models used and the origin of the datasets for each work cited in our state of the art. We thank you for your contribution.

2. Figure 6 should be expanded to increase it width.
R: We appreciate the suggestion to increase the width of Figure 6. We made the modification to improve the readability and understanding of the presented results. Thanks for the valuable contribution.

3. It would be prudent to incorporate a discussion regarding the limitations and drawbacks of the proposed method within the Conclusion section, prior to outlining future research directions.
R: Thanks for the feedback. In response to the suggestion, we include a discussion of some of the limitations and disadvantages of our proposed method in our conclusions. We believe this addition improves the overall scope of the study and provides a more balanced perspective on the results. Thank you for this contribution.

4. Kindly ensure that the text undergoes a thorough proofreading process to rectify any existing grammatical issues and typos. Additionally, please aim for clarity and brevity in sentence construction, as some of the current sentences are rather lengthy and challenging to comprehend.
R: Thanks for your feedback. We apologize for any grammatical problems, typos and long sentences present in the text. We performed a thorough review process to correct these issues and improve the clarity and brevity of sentence construction. We appreciate your attention.

Thank you very much! We remain at your disposal for any other queries.
Best Regards!

Author Response File: Author Response.pdf

Reviewer 2 Report

 

This paper proposes a demand forecasting strategy using an IoT retrofit architecture based on the SmartLVGrid meta model. The strategy employs various regression models, including Linear Regression, Support Vector Regressor, Random Forest Regressor, XGBoost Regressor, and Long Short Term Memory (LSTM), to accurately forecast the energy demand of both the legacy building and its circuits.

While the paper is well-written and addresses the demanding problem, there is room for further improvement. The introduction section, in particular, is excessively long and may cause readers to struggle in reaching the main point. Therefore, it is advisable to make it shorter and more concise, focusing on providing background information, identifying existing gaps, and presenting the solution you have provided. Additionally, it is important to perform a thorough proofreading to identify and correct any typos or formatting errors, such as the one found in line 460. This will greatly enhance the overall language quality.

The related work is well written but authors can enhance it by including some recent studies published on energy forecasting e.g.

https://doi.org/10.1016/j.rineng.2023.100888

https://doi.org/10.1038/s41598-022-26499-y

 

Moreover, there are some subheadings that lack accompanying text. Please include a brief description under each subheading, even if it is just a single line. For example, in section 6.4 "Learning models," provide a concise explanation of the learning models used.

The grammar is fine but there are certain typo errors specially while mentioning dates like 7th July etc. It is recommended to proofread the paper for any possible errors. 

Author Response

Replies to reviewer 2

First, we would like to thank you for taking the time to evaluate our work.

Comments: While the paper is well-written and addresses the demanding problem, there is room for further improvement.

1. The introduction section, in particular, is excessively long and may cause readers to struggle in reaching the main point. Therefore, it is advisable to make it shorter and more concise, focusing on providing background information, identifying existing gaps, and presenting the solution you have provided.

R: Thank you for your valuable contribution. I agree that the introduction section should be concise. We take steps to shorten it by focusing on presenting background information and highlighting the solution we propose. His guidance certainly improved the quality of our work.

2. Additionally, it is important to perform a thorough proofreading to identify and correct any typos or formatting errors, such as the one found in line 460. This will greatly enhance the overall language quality.

R: Thanks for the feedback. We apologize for any typos or formatting errors that may have occurred in the text, including the one found on line 460. We understand the importance of conducting a thorough review process to ensure the highest quality language possible. Thank you for your attention.

3. The related work is well written but authors can enhance it by including some recent studies published on energy forecasting e.g.:

• https://doi.org/10.1016/j.rineng.2023.100888

• https://doi.org/10.1038/s41598-022-26499-y

R: Many thanks for the suggestion. We add the valuable references in our state-of-the-art. Certainly, great contributions to energy forecasting with the TFT and N-BEATS models, respectively. These models will be objects of future studies in our research and these works could be great points of reference.

 

4. Moreover, there are some subheadings that lack accompanying text. Please include a brief description under each subheading, even if it is just a single line. For example, section 6.4 "Learning models," provide a concise explanation of the learning models used.

R: Again, thank you very much for your contribution. We have inserted a brief description for the sections referring to the following subheadings: Data pre-processing, Evaluation metrics and Learning models.

It is important to mention that we carried out proofreading to correct possible textual errors throughout the work.

Thank you very much! We remain at your disposal for any other queries.
Best Regards!

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors made some good contribution on Demand Forecasting Strategy; however, I have few concerns and suggestions.

1. Abstract should be revised to illustrate the core contribution of the work numerically.

2. The authors can use Mean Bias Error (MBE) metrics to assess the tendency of a measurement process to overestimate or underestimate the value of a parameter.

3. Clarify Why Retrofit Architecture for Remote Monitoring is required?

na

Author Response

Replies to reviewer 3

First, we would like to thank you for taking the time to evaluate our work.

Comments: The authors made some good contribution on Demand Forecasting Strategy. However, I have few concerns and suggestions.

1.Abstract should be revised to illustrate the core contribution of the work numerically.

R: Many thanks for the suggestion. We added to the abstract some of the numerical results obtained by verifying and comparing the number of demand overruns of the LSTM, XGBoost and RFR models compared to the number of overtakes that occurred during the models' testing period. 

2. The authors can use Mean Bias Error (MBE) metrics to assess the tendency of a measurement process to overestimate or underestimate the value of a parameter.

R: We appreciate your insightful comment suggesting the use of Mean Bias Error (MBE) as an evaluation metric. Undoubtedly, the MBE provides valuable information about the tendency of a model to overestimate or underestimate parameter values. However, in our specific context of forecasting energy demand for the next 15 minutes, we chose to use RMSE, MAE and R² as the main evaluation metrics. The reasons for this choice are as follows:
• RMSE effectively measures model performance by penalizing larger errors due to squared residuals.
• MAE provides a clear and straightforward interpretation of the mean error magnitude, which is particularly valuable when comparing different models.
• R² illustrates how much of the variance in the dependent variable our model can explain, serving as a robust model performance indicator.
Together, these metrics provide a comprehensive understanding of how our models are performing and address our specific research needs. Adding MBE to this mix could potentially complicate interpretation without offering significant additional insights, given the already comprehensive nature of our chosen metrics. Furthermore, unlike the MBE, both the RMSE and the MAE consider the magnitude of the error, regardless of the direction, providing a more complete view of the model's performance.
We hope this explanation responds to your comment, and we sincerely value your constructive feedback to improve our work.

3. Clarify why Retrofit Architecture for Remote Monitoring is required.

R: Again, many thanks for the valuable suggestion. We rectify this issue by valuing the why to use our architecture in the Introduction and Conclusion sections. 

Thank you very much! We remain at your disposal for any other queries.
Best Regards!

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have carefully addressed all my concerns, I have no further suggestions.

The language has been improved.

Reviewer 3 Report

Thanks for the answers relating the raised concerns.