Potential of Model-Free Control for Demand-Side Management Considering Real-Time Pricing
Round 1
Reviewer 1 Report
.
Author Response
At very first we would like to appreciate the valuable comments from reviewers. We have addressed all the comments and suggestions in this revised version and provided point-to-point response. We believe that these changes have greatly improved the paper. All the changes in the revised version of the manuscript are written in blue for easy identification.
We hope that this revised version of the manuscript will be acceptable for publication.
We sincerely thank you for your help with this paper.
Author Response File: 
Author Response.docx
Reviewer 2 Report
In the paper a DPC method applied to energy management system is presented. Two model-free algorithms of ANN and RF are selected to make control strategy prediction on system operation. The results are presented for four-storey building with electric heaters and basic user loads.
The methods used by the authors are appropriate but there are some points that should be addressed before publication.
Introduction
A review of the literature has been included, although it is rather modest.
System overview and case study description
What do the basic user loads mean? This issue needs clarification. Has any room usage (internal heat gains) schedule been adopted. Was it the same on all floors?
What about indoor temperature? It is stated that the used control scheme ensures users’ thermal comfort. Consequently, what the thermal comfort model was used? Because calculated indoor temperature varies within relatively large limits (Figures 9, 10, 13 and 14).
Model predictive control description
How the solar radiation and the variability of internal heat gains were taken into account. According to the presented equations, only the variability of the external temperature is taken into account.
What about variability of ventilation airflow (infiltration)?
Figure 6
Explain the negative value of electricity price in Figure 6.
Table 1
It should be the same accuracy for the same parameters.
Author Response
At very first we would like to appreciate the valuable comments from reviewers. We have addressed all the comments and suggestions in this revised version and provided point-to-point response. We believe that these changes have greatly improved the paper. All the changes in the revised version of the manuscript are written in blue for easy identification.
We hope that this revised version of the manuscript will be acceptable for publication.
We sincerely thank you for your help with this paper.
Author Response File: Author Response.docx
Reviewer 3 Report
Create a nomenclature table and place it at the beginning of the manuscript
Improve significantly English language use, at some points reading gets quite complex.
More details on the building plant description as well as the sizing of the energy loads is needed. For example, what kind of electric heater is needed to heat 400m2 of space? What kind of insulation and heat exchange dynamics is considered in the building plant? What is the exact location of the building (i.e. what kind of environmental seasonal conditions are considered? Winter? Heavy winter? Autumn? Dry Autumn? Spring? What is the size of the windows? How many windows are there? What is the orientation of the building?)
The authors try to explain the operational setup of the simulative tests. Please use a table to indicate what kind of data are used for the MPC, what kind of data for the DPC, what kind of data are shared among these two techniques and how. An operational workflow diagram would be also quite helpful.
Improve and enrich the list of references, 19 items is a quite small number weakening the novelty of the proposed methodology.
The novelty of the proposed methodology is not adequately justified. The authors mention [15] and [16] as similar methodologies in similar applications. Please elaborate more on this matter.
Comments for author File: 
Comments.pdf
Author Response
At very first we would like to appreciate the valuable comments from reviewers. We have addressed all the comments and suggestions in this revised version and provided point-to-point response. We believe that these changes have greatly improved the paper. All the changes in the revised version of the manuscript are written in blue for easy identification.
We hope that this revised version of the manuscript will be acceptable for publication.
We sincerely thank you for your help with this paper.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Problems of solar radiation and ventilation have not been clarified. If these factors and internal heat gains have been ignored, the presented model is very simple and will not translate into real buildings. Please, clarify this problem.
Author Response
Thank you for pointing it out. The marginal impact of solar radiation in the pre-set environmental seasonal condition of winter is not considered in the paper at this stage. This is because of the main point of this paper is to evaluate if the model-free approach can simulate the control strategies according to the control actions generated by the Model-Predictive control. The ventilation consists of natural ventilation (cross ventilation, architectural layout, etc.), mechanical ventilation (exhaust system, air supply system, etc.) and compound ventilation according to the design code for civil buildings [27]. The complex influence of ventilation and meagre internal heat gains (caused by loads and human activities) on user’s comfort is neglected in the proposed simple testbed because we focus more on the assessment of the learnability of control strategy using model-free method. Our future work is concentrated on revealing the detailed and precise impact of these factors on the much bigger and realistic building models.
Author Response File: Author Response.docx
Reviewer 3 Report
Improve English grammar. Not adequately tackled in the first revision phase. Justify text in section 5. Provide insights and replication hints for future follow up research works (e.g. how many layers and how many neurons per layer are there in the ANN?).
Author Response
We are so sorry for the English grammar. We have made a revision to the whole paper. The text in section 5 is justified. The insights and replication hints for future follow up research works are provided (The proposed model g is a three-layer feed backward neural network with one hidden layer which contains 50 neurons. The input and output layers have 72 and 24 neurons respectively. The activation function of Rectified Linear Units is applied in the network. The constant parameter of tolerance for convergence is set to 0.0001. Adam algorithm which performs well in bias-correction across different hyperparameter settings is selected as the stochastic optimizer [27]. The number of estimators (trees) in the proposed forest is set to 100.).