A Novel Load Scheduling Mechanism Using Artificial Neural Network Based Customer Profiles in Smart Grid

Round 1

Reviewer 1 Report

In my opinion it is a good paper, well organized and with enough references. English language is also fine.

The authors propose a novel approach to optimize load demand and storage management in response to dynamic pricing using machine learning and optimization algorithms. The results and analysis made in the paper try to validate the method developed by the authors, but in my opinion some clarification must be done.

As the authors refers there are many other optimization methods to do the same (loads scheduling, peak shaving, low tariffs, power constraints...), and I did not completely understand the main value of the work, authors should emphasize the differences and explained what distinguished their approach and their results from the others. Also, ESS is many times used to peak shaving and to provide users flexibility, why is your work different?

The quality of the paper is interesting from the point of view of the content, and it is also a topic of interest for the researchers in the related areas.

Thus, in my opinion the manuscript needs some clarifications and changes before acceptance for publication.

Comments

Abbreviations should be must be placed in alphabetical order. Figure 4, “original price” and “price for user 1” have the same symbol, it is very hard to perceive when it one or the other. Please change the symbol of one of them. The same problem in figure 5, please change the color of one of the blues. Line 363, how did authors get to (20-38-1)? Also, authors state that “Results show the better performance as compared to other algorithms used in literature.” Please quantify how better it is with example. Power SI unit is “W” and authors may write “kW”, not “Kw”. Please correct it in the pictures. Why Figures 2 and 5 have a maximum of 5 kW and the other examples Figures 7, 8, 10 and 11 have 70 kW? Just because they are two users? Where is the users description? Line 390, authors state that “It is also assumed that, each user has a battery backup with some initial level in such that its capacity must not be below the half of rated capacity.”, but when is considered “initial” time? For instance, in figure 10c the SOC is many times below half value, even in the final of the day. Please clarify. Line 403, please verify if reference to Fig.10c is correct. In figure 10 authors are only analyzing one single home using several solvers, but why two user’s in Fig.10d?

Author Response

                                      Reviewer-1                        

Paper title:  A Novel Load Scheduling Mechanism using Artificial Neural Network Based Customer Profiles in Smart Grid

Date: 12-02-2020

 

Dear Reviewer, thank you very much for your time and efforts to deeply review the paper. Your comments really guided us to improve the quality of the paper up to the level of the journal. Please note that responses to your comments are given green color.

__________________________________________________________________

 

Authors refers there are many other optimization methods to do the same (loads scheduling, peak shaving, low tariffs, power constraints...), and I did not completely understand the main value of the work, authors should emphasize the differences and explained what distinguished their approach and their results from the others.

Ans: Dear Reviewer, thank you very much for your comment. In literature, authors have discussed various techniques to deal energy management problems with diverse constraints. Most of the techniques perform energy management on the basis of a day-ahead pricing signal provided by the utility. Whereas, we have proposed a slightly different pricing scheme which is based on combination of RTP and IBR signals. Initially, the low tariff area based on historical load demand profile is obtained. Then using artificial neural network (ANN), the same low tariff area is forecasted due to the fact that load consumption trends are dynamic in nature. Finally, based on low tariff area, the individualized price profiles for all users are obtained. These load profiles are designed in such a way that electricity tariff of each user would be modified based on load demand and market consumption trends. So that, rebound peaks are minimized providing incentives to users in terms of minimum discomfort. For simplicity, the objectives of the proposed work are explicitly discussed in [Section 1, Page 4, Line 93-108].

 

Abbreviations should be must be placed in alphabetical order. Power SI unit is “W” and authors may write “kW”, not “Kw”. Please correct it in the pictures.

Ans: Dear Reviewer, thank you very much for your comments. We have incorporated your       comments as per your suggestion. Abbreviations have been written in alphabetical order                (Page 2 (abbreviations)). Moreover, the figures have also been modified, accordingly.

Line 363, how did authors get to (20-38-1)? Please quantify how better it is.

Ans: Dear Reviewer, (20-38-1) gives total number of neuron in each hidden layer. Our model architecture gives the best performance in terms of forecasting error at (20-38-1). It is obtained by using coarse-to-fine approach by randomly changing the number of neurons in each hidden layer of ANN architecture.

 

Why Figures 2 and 5 have a maximum of 5 kW and the other examples Figures 7, 8, 10 and 11 have 70 kW?

Ans: Dear Reviewer, in figures 2 and 5, we have considered 5kW as a test data to clarify the concept of user flexibility. It reveals that end users would be able to decide the value of Q, as given in Eq. 8, and section 3.2.5. Regarding implementation [figures 7,8,10 and 11], we have considered 70kW in order to ensure that any capacity can be incorporated by using proposed algorithm.

 

Line 403, please verify if reference to Fig.10c is correct.

Ans: Dear Reviewer, we apologize. It should be referred as Fig 9. [Section 5.1, Page 17, Line 430]

                                                                                               

In figure 10 authors are only analyzing one single home using several solvers, but why two user’s in Fig.10d?

Ans: Dear Reviewer, thank you very much for your comments. To quantify the proposed approach, two types of users are considered in our scenario. User_2 is considered without physical flexibility mode, to make the difference in cost and PAR minimization as already explained in Fig 9.  Also, please refer to [Section 5.1, Page 16]

 

Line 390, authors state that “It is also assumed that, each user has a battery backup with some initial level in such that its capacity must not be below the half of rated capacity.”, but when is considered “initial” time? For instance, in figure 10c the SOC is many times below half value, even in the final of the day. Please clarify.

Ans: Dear Reviewer, we have rechecked the working of proposed algorithm and respective constrains and identified that there was a problem with control variable.  The correction has been made in the revised version of manuscript [Section 5.1, Page 18, Fig. 10c]

Author Response File: Author Response.docx

Reviewer 2 Report

The paper copes an extensive dissertation related to the user of ANN in Load Scheduling. I believe the work is fairly good enough to be published in this journal, but the presentation and writing needs a big improvement. It seems to be a very interesting work but poorly redacted, explained and written.

On the first hand is very strange for me to have a paper devoted to the ANN use on DR with 4 pages of introduction and background without any mention to similar works based on ANN. Working with ANN in power system has been long used and many of them has been used on similar environments proposed by the authors, even on the same journal. Some works of Matallanas et. or Mancedo et al should be checked. Some relevant works could be:

- Haque MT, Kashtiban AM (2007) Application of neural networks in power systems: a review. Int J Energy Power Eng 1(6):897–901
- Kurbatsky V, Sidorov D, Tomin N, Spiryaev V (2014) Optimal training of artificial neural networks to forecast power system state variables. Int J Energy Optim Eng 3(1):65–82
- Eduardo Matallanas, Manuel Castillo-Cagigal, Estefanía Caamaño-Martín and Álvaro Gutiérrez (2019). Neural Controller for the Smoothness of Continuous Signals: an Electrical Grid Example. Neural Computing and Applications, 32(6):1—16
- Eduardo Matallanas, Manuel Castillo-Cagigal, Álvaro Gutiérrez, Félix Monasterio-Huelin, Estefanía Caamaño-Martín, Daniel Masa and Javier Jiménez-Leube. Neural Network Controller for Active Demand Side Management with PV Energy in the Residential Sector. Applied Energy
- Macedo, M.N.Q. & Galo, Joaquim & Luz de Almeida, Luiz & Lima, Antonio. (2015). Demand side management using artificial neural networks in a smart grid environment. Renewable and Sustainable Energy Reviews. 41. 128–133. 10.1016/j.rser.2014.08.035.
- M.N.Q.MacedoJ.J.M.GaloL.A.L.de AlmeidaA.C.de C. Lima, Demand side management using artificial neural networks in a smart grid environment, Renewable and Sustainable Energy Reviews Volume 41, January 2015, Pages 128-133
- RenzhiLu Seung HoHong, Incentive-based demand response for smart grid with reinforcement learning and deep neural network, Applied Energy Volume 236, 15 February 2019, Pages 937-949.

Which are the main differences with those works??

I would even suggest to make a review of Demand Response and Demand-Side management systems and algorithms published since 2018 in Energies journal.

Moreover, linear programming and mixed integer linear programming are mentioned in the text as if it is common knowledge for the readers of the Energies journal. I will say that it is not a common concept for Energies readers, so I will suggest to devote some time to explain the algorithmic implication of the solution proposed.

About the maths in the manuscript:
- There are some variables that are difficult to follow while reading the text. It is true that most of them are defined in the nomenclature, but when first used in the text they should also be defined. For example, in Eq1. \bar{\tau_t} is not defined, the same with L_g in Eq . 5, it is previously mentioned but not defined. There are many of these typos.
- Furthermore, as an example, line 206 talks about \tau_{\star} as the average delay, but it is later noted as \tau_{avg}.
- Eq. 11 mentions 10 previous equations and Equation 15 mentions 14 previous equations. This does not look serious.
- All these problems should be corrected for making the manuscript readable.

Language: There are many writing problems in the manuscript. Some examples:
- Line 7, 8. The sentence is not understandable.
- Line 228 use in proposed  used in the proposed
- Line 249, thorugh battery storage system -> through a battery storage system
- Line 260, provided by utility  provided by the utility
- Linear programming is defined in lines 14, 140, 297
- Mixed inteter linear programming is defined in lines 128, 298
Finally, conclusions are very weak. One example:
“Also with the jelp of ANN, the effects on user comfort is measured in contrast to cost reduction”.
No information is provided for that claim. Results obtained from their works should provide objective results for every claim.

Author Response

                                      Reviewer-2                        

Paper title:  A Novel Load Scheduling Mechanism using Artificial Neural Network Based Customer Profiles in Smart Grid

Date: 12-02-2020

 

Dear Reviewer, thank you very much for your time and efforts to deeply review the paper. Your comments really guided us to improve the quality of the paper up to the level of the journal. Please note that responses to your comments are given blue color.

_________________________________________________________________

Which are the main differences with those works?

Ans: Dear Reviewer, thank you very much for your valuable comment. In literature, many works have been reviewed on advantages and drawbacks of using of ANNs applications in power system in contrast with the other conventional methods [1]. Major aspects of the ANN is of planning, expansion, development and load forecasting of the power system. ANNs are very fast and capable of direct coupling with electrical system to data acquisition without time constraints. Recent study shows that short-term load forecasting have greater percentage (62%) of work in research. [Section 1, Page 3, Line 71-77]

         In [3], authors used ANN for the smoothness of the aggregated load demand at grid level. Instead of controlling the large grid, the local load demand of grid elements on historical basis are managed in an automated manner having local generation and a battery storage system taking into account of aggregated consumption at the grid. The performance of the network is increased by adjusting the free parameters of neural structure using genetic algorithm. [Section 2, Page 4, Line 132-138]

The similar work is reported in [4], incorporates ANNs fixed by genetic algorithm to implement DSM controller for residential users. A distributed home automation system in which the controller comprises of scheduler and neural network system that coordinates with the utility and local PV generation having a storage system to maximize the local energy performance. In this system individual appliances equipped neural controller i.e., ANN and local generation in a distributed manner and are free to self-organize their output on the basis of preference and predicted generation. [Section 2, Page 4, Line 132-138]

The authors in [5] uses the ANNs to facilitate the implementation of DSM programs. The classification feature of the ANN is used in an intelligent environment to classify the load curves of each user from pool of load data generated by dynamic networks. With the prior greater knowledge of consumer habit, the optimization of the electrical system is carried out with the classified load data and by implementing DSM policies to each class to make it more sustainable and efficient. [Section 2, Page 4, Line 143-147]

ANN is used to assist service provider to discover the future rates to purchase energy from its customers to balance energy fluctuations in the power system. To coup up with the future uncertainties in power system due to its inherent nature, a supervised learning in deep neural networks (DNNs) is used to predict the real time unknown load demand and wholesale market prices instead of day-ahead to incentivize the active subscribed consumer [7]. [Section 2, Page 5, Line 132-138]

Dear Reviewer, Our work is contrary to all above referenced in the sense that we use ANN for the supervised learning to predict the low tariff load curve for a particular user depending on its flexibility i.e., either physical or virtual. After training and testing the network, average error is also calculated and we proposed that the error is basically the user compromised comfort in terms of reduction in electricity cost.

I will suggest to devote some time to explain the MILP algorithmic implication of the solution proposed.

Ans:  Dear Reviewer, thank you very much for your comment. MILP is basically a LP-based solver with branch-and-bound algorithm. The optimal solution for the main problem is obtained by dividing the master problem into sub-problems and evaluated by using divide and conquer approach in the form of a subtree. Bounds on each node are evaluated linearly and selected that maximize/minimize the objective function. [Section 4, Page 10, Line 328-331]

 

In Eq1. \bar{\tau_t} is not defined, the same with L_g in Eq. 5, it is previously mentioned but not defined.

 

Ans: Dear Reviewer, thank you very much to identifying the mistake. [Eqn. 1] and [Eqn. 5] are now corrected and well described in the revised version. Please refer to the [Page 7, Line 233-234 & 224].

 

Line 206 talks about \tau_{\star} as the average delay, but it is later noted as \tau_{avg}.

Ans: Dear Reviewer, it was mistakenly written and we apologies on this. We have corrected the mistake. Please refer to [Page 7, Line 224]

 

11 mentions 10 previous equations and Equation 15 mentions 14 previous equations. This does not look serious.

Ans: Dear Reviewer, after the gradual modification in the objective function, the final form will          come in terms of all previously discussed equations and constraints. Our modified objective function is forced to follow all the constraints. Instead of writing all at same place          we have provided an ease to the reader to understand. [Section 3.2.4, Page 8, Line 252]

Line 7, 8. The sentence is not understandable.

Ans: Dear Reviewer, these lines referred to the inputs of the ANN model, e.g. historical, individual load demand and real time pricing signal data are made the inputs to our model for training and validating. We have rewritten these lines for better understanding [Page 1, Line 7-8]

Line 228 use in proposed  used in the proposed, and Line 249, through battery storage system -> through a battery storage system, also Line 260, provided by utility  provided by the utility.

Ans: Dear Reviewer, we apologies on this and manuscript is thoroughly revised and rectified all the errors and typos. Please see [Section 3.2.5, Page 9, Lines 269-270] and [Section         3.2.4, Page 8, Line 248-249]

Linear programming is defined in lines 14, 140, 297 and mixed integer linear programming is defined in lines 128, 298.

Ans: Dear Reviewer, we have incorporated the changes in the revised version.             Please refer to [Section 4, Page 10, Line 316-320] and [Section 2, Page 5, Line 147-149].

Finally, conclusions are very weak. One example: “Also with the help of ANN, the effects on user comfort is measured in contrast to cost reduction”.

Ans: Dear Reviewer, we have discussed the results in section 4.3 and fig 7 that explained the above claim. The actual power is the total amount of required power without any load shifting. While the suggested power defines the power profile obtained from historical power demand data, and the forecasted power profile is obtained from ANN. After testing the network, high error in hours is due to the low diversity in training data, and average error is also calculated which is 3.763 with 30% cost reduction. [Section 4.3, Page 13, Line 392-398]. To further strengthen our claim of proposed idea, we have modified conclusion section as well. Please refer to conclusion section.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors made a significant effort to respond to the reviewers suggestions. So, the quality of the paper increased substantially, and by this reason in my opinion the manuscript is now ready for publication.

Author Response

Dear Reviewer,

Thank you very much for reviewing our manuscript. 

Reviewer 2 Report

The authors have covered all the issues raised, although a greater description of demand response methods would have been desirable.

Moreover, a final english check is mandatory

Author Response

                                 Reviewer-2, Round-2                   

Paper title:  A Novel Load Scheduling Mechanism using Artificial Neural Network Based Customer Profiles in Smart Grid

Date: 21-02-2020

Dear Reviewer, thank you very much for your time and efforts to deeply review the paper. Your comments really guided us to improve the quality of the paper up to the level of the journal. Please, be noted that responses to your comments are given Blue color.

______________________________________________________________

 

 

1. The authors have covered all the issues raised, although a greater description of demand response methods would have been desirable.

Ans: Dear Reviewer, thank you very much for your concern. We have cited and discussed some relevant works in literature review section. [Section 2, Page 5, Line 159-177]

2. Moreover, a final English check is mandatory

Ans:  Dear Reviewer, we have gone through the entire manuscript very carefully and tried to address your concerns regarding English and spell check.