Machine Learning-Based Node Characterization for Smart Grid Demand Response Flexibility Assessment

Round 1

Reviewer 1 Report

Reviewer’s Comments to the Authors

The authors propose a machine learning approach for the classification of historic demand data from several network substations. The manuscript is well-written, but some places need further clarifications. Therefore, I recommend moderate revision.

Line 99. Please remove “as has been demonstrated e.g. in”. So after “…requirements” just mention “[2, 15]”. Also have “According to given scale…” right after that sentence not in a new paragraph. Please note that a paragraph usually has multiple sentences, not just one.

Lines 144 & 148 & 222. Please avoid using “e.g.”, instead use “for example” or simply remove it.  

Please add a paragraph to the end of the Introduction to mention the goal of this research and how you structured the paper.

Line 181. Please elaborate more on k-means classification and why you selected three clusters.

There are several recent techniques for time series classification and forecasting. For example, the Least-Squares Wavelet (LSWAVE) software package contains several statistical tools including the least-squares spectral and wavelet and cross-wavelet analyses, which proved to be very useful tools for estimating trends and seasonal components of any time series and identifying their patterns in the time-frequency domain without any need for preprocessing of the raw measurements:

https://doi.org/10.1007/s10291-019-0841-3

From the spectrum of the time series, you could for example select an appropriate number of clusters, or using the Elbow method? Please include the reference above and/or at least make a short discussion at the end of the paper (e.g., see Line 244).   

Line 202. Have you tried CNN for another set of substations as for the training data sets? If so, have the results changed significantly?  

Line 214. Please give a short example or describe the bootstrap further because this method is applied differently (with thresholds) in various fields. Adding a reference here also helps.

Line 217. Please explain what the confusion matrix is or give a reference.

Line 285. “…its graphical…” not “…it’s graphical…”

Line 291. Please explain in the Discussion why certain classes are often confused as shown in Figures 10, 11, 12. Also, please mention whether the same classes are also confused using the other two models: CNN1 and CNN3. Does that depend on the number of convolutional layers in CNN?

A Conclusions section needs to be added that briefly summarizes what you proposed in this research, advantages/disadvantages, and a sentence for Future outlook.

Regards,

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors have presented an ML-based algorithm to identify the smart grid demand response flexibility. Given below are my comments to improvise upon the quality of the manuscript:

1. A clear distinction on the novelty of the work is missing in the introduction. The authors must clarify the research gap and how their work is beneficial for advancing knowledge in this particular field.
2. The authors have considered 1-year dataset from May 2018-April 2019. Can this data be generalized for any year? In the last two decades the use of electrical equipment have considerably improved, and the coming two decades it will further improve. In such a situation will your model holds valid even after 25 years from now? Kindly explain.
3. The authors have used a k-means algorithm for clustering. Can you indicate, what was the motivation behind using such a fundamental algorithm for clustering?
4. Line 196-197: Why the qualitative evaluation was not verified? What have you done in order to make your model robust?
5. Table 5: Why such a high deviation is observed in some of the substations located in Swansea, Campsie and Darlinghurst?
6. Can you indicate the limitations of your study in the discussion section?

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper presents the theoretical framework for quantifying network flexibility potential by introducing a machine learning-based node characterization. This Reviewer has the following comments:
1) Rewrite the Introduction section by mentioning the main contributions of this work.

2) Simulation Results should be enriched. More detailed scenarios should be added.

3) A separate section should be added for discussion of obtained results and main achievements.

4) Results and discussion are not adequate.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Reviewer’s Comments to the Authors

I would like to thank the authors for addressing my comments. The paper looks much better in my view. I have a few more comments, though:

Please remove the following two sentences in lines 276-278 where it says:

“Time-series data evaluated in this paper are equally spaced and consisting only of 96 data-points. Therefore, it may be beneficial to classify raw data without unnecessary pre-processing that introduces more bias.”

And replace them by

"The spectral and wavelet analyses are very useful for estimating trends and seasonal components of any time series and identifying their patterns in the time-frequency domain [here you may also cite your work: PODROUŽEK, J.; BUCHER, C.; DEODATIS, G. Identification of critical samples of stochastic processes towards feasible structural reliability applications. Structural Safety, 2015/47]. Herein, we directly classify the time series data and shall leave the use of wavelets to future research."

Please note that wavelet analysis does not have any limitations related to the size and type of time series. The way that you worded these two sentences could be misleading in that wavelets are not useful and introduce biases. So, please follow my recommendation as mentioned above.

Line 338 in Conclusions. Please merge the last sentence to the previous sentence because it is unusual for a paragraph to contain only one sentence.

Regards,

Author Response

Please see the attachment.

Author Response File: Author Response.pdf