A Hybrid Forecast Model for Household Electric Power by Fusing Landmark-Based Spectral Clustering and Deep Learning

Round 1

Reviewer 1 Report

The paper deals with the problem of forecasting household power load which is important from the point of view of operating and planning of power grids. Authors used clusters basing on historical data whichg were developed by means of landmark based spectral clustering and deep learning model. To enlarge the dataset it was expanded by means of bootstrapping. Finally, neural networks and long short-term memory were used to predict power consumption.

The paper consists of five chapters. The Introduction part shows both practical and mathematical background of the paper. A short review of other similar works is done showing the possibilities of applying various methods.

Second part is dedicated to materials and methods. Authors used an Individual Household Electric Power Consumption dataset which consists a data from Paris, France from 2006 till 2010. They showed haw the constructed and completed matrix, normalized data nad generated samples. Finally, the hybrid model was constructed what allowed to introduce algorithm which seems to be correctly designed. In third part all steps were presented, including landmark based spectral clustering, bootstrap aggregating, neural network application, long short-term memory and finally integration of all mentioned stages.

In fourth part the results of presented methodology were provided. The task of the model was to predict the global active power of household electricity in the next hour basing on the previous 24 hours. This chapter shows the results on all stages of methodology, including results comparison with and without clustering. The Authors proposal gave a very good results.

Final chapter shows the main conclusions.

Generally, the topic is very important and in the time when economy of power becomes more and more important such methodology can be treated as a very good way to predict households power loads. Therefore, I find the paper as a well prepared, scientifically sound and worthy.

Author Response

We thank you for your careful review and evaluation of the manuscript. We are greatly touched by your patience and rigorousness. Your review has brought us great help.

Reviewer 2 Report

Dear Authors, 

"A hybrid forecast model for household electric power by fusing landmark based spectral clustering and deep learning" is an interesting paper that combines CNN and LSTM as LSC-CNN-LSTM. The paper is getting benefit from CNN feture extraction and LSTM to train. I have following concerns which will need to be addressed. 

1. I know that Deep learning help us find features from data set to estimate the power consumption. However, I just wonder if the Data set belonging to 2006-2010  will really help us estimate todays power consumption since the as technology increases the power consumption behavior complexity also increases.

2. I also see that dataset have missing points about 7.1%. I can see that you are using matrix completion technique to fill the gaps. Dont you think that this will couse overfitting for your estimation ? 

3.  The algorithm 1 kind of confusing. Could you please rearrange all the steps. Please check out mdpi guidelines for that.

4.  I can also see that you have used bootstrap aggregating to increase the number of samples. As I said increasing the data variability with artificial technique may cause overfitting. Which means that you may have over fitting for the future estimations. 

5. The cost of the functions represented in Figure 10 are so smooth and without any variations. I have some concerns about the health of the data. please check again.

6. For some reason,  k-means-CNN-LSTM have almost same accuracy with the proposed method. In some cluster it has even better result. 

          k-means-CNN-LSTM--> 0.0888 0.0995 0.0806 0.0896

          LSC-CNN-LSTM --> 0.0392 0.1072 0.1040 0.0835

 

 

Author Response

SEE WORD

Author Response File: Author Response.pdf

Reviewer 3 Report

To address the prediction issue of household power consumption in power grid, the authors choose the time series of historical power consumption as the feature variables. Furthermore, the authors use the landmark based spectral clustering (LSC) and the deep learning model to cluster and predict the power consumption dataset respectively. The research topic is topical and interesting. The work is organically structured and coherent in all its parts. The description of the methodology is also precise and easy to read.

However, the article has two limitations: 1. In my opinion, the correct prediction of household power consumption in power grids has important implications for energy transition and the pursuit of sustainable development goals. This is an aspect that is completely ignored by the authors. Furthermore, highlighting this aspect can add value to the article. In this regard, it is appropriate to review the relevant literature. I suggest reading and considering for references: AlKhars, M.; Miah, F.; Qudrat-Ullah, H.; Kayal, A. A Systematic Review of the Relationship Between Energy Consumption and Economic Growth in GCC Countries. Sustainability 2020, 12, 3845. https://doi.org/10.3390/su12093845. Nesticò, A. and Maselli, G. Declining discount rate estimate in the long-term economic evaluation of environmental projects. Journal of Environmental Accounting and Management 2020, Vol. 8, Issue 1, pp. 93-110. https://doi.org/10.5890/JEAM.2020.03.007. 2. What are the limitations of the applied analysis models? What are the research perspectives, both in strictly theoretical terms and in terms of practical applications? it is important to answer these questions at length.

Author Response

SEE WORD

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have addressed my concern.