Application of Decision Trees for Optimal Allocation of Harmonic Filters in Medium-Voltage Networks
Round 1
Reviewer 1 Report
This paper uses decision tree to optimize the filter that applied to the medium voltage networks. This is an interesting topic, because there usually were many factors that may affect the power quality in such network.
Comments to the authors:
1. Decision tree is a commonly used method. It would be better if the authors would compare several common methods in this field, for example, random forest and gradient boosted trees. Here are some related references
- Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q. and Liu, T.Y., 2017. Lightgbm: A highly efficient gradient boosting decision tree. Advances in neural information processing systems, 30, pp.3146-3154.
- Couronné, R., Probst, P. and Boulesteix, A.L., 2018. Random forest versus logistic regression: a large-scale benchmark experiment. BMC bioinformatics, 19(1), pp.1-14.
- Tian, W., Lei, C. and Tian, M., 2018, December. Dynamic Prediction of Building HVAC Energy Consumption by Ensemble Learning Approach. In 2018 International Conference on Computational Science and Computational Intelligence (CSCI) (pp. 254-257). IEEE.
2. It could be clearer if the authors could explain where the distortions in Figure 7 comes from.
3. The work will be more convenient if the authors can show improved current/voltage waveform after the decision tree optimization, and compare with several bench marks in THD.
Author Response
The authors would like to thank the Reviewer for the effort in reviewing the paper. We would like to inform that the comments of the Reviewer have been taken into account and also the proofreading by a native speaker has been done and included in the new revision.
Comment 1:
- Decision tree is a commonly used method. It would be better if the authors would compare several common methods in this field, for example, random forest and gradient boosted trees. Here are some related references
Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q. and Liu, T.Y., 2017. Lightgbm: A highly efficient gradient boosting decision tree. Advances in neural information processing systems, 30, pp.3146-3154.
Couronné, R., Probst, P. and Boulesteix, A.L., 2018. Random forest versus logistic regression: a large-scale benchmark experiment. BMC bioinformatics, 19(1), pp.1-14.
Tian, W., Lei, C. and Tian, M., 2018, December. Dynamic Prediction of Building HVAC Energy Consumption by Ensemble Learning Approach. In 2018 International Conference on Computational Science and Computational Intelligence (CSCI) (pp. 254-257). IEEE.
Answer:
The proposed papers have been included in the references and referred to in the introduction.
Comment 2:
- It could be clearer if the authors could explain where the distortions in Figure 7 comes from.
Answer:
These distortions are voltage drops which are caused by the high steepness of the transformer output current. The oscillations around these drops result directly from the used model of nonlinear receivers and the simulation method in the frequency domain. The presented waveforms are a combination of a finite number of harmonics obtained as a result of simulations and not measured waveforms. A relevant explanation has been included to the paper. The cause of the oscillations resulting from the Gibbs effect can be clearly seen in the figure below showing the current waveform of one of the receivers.
Comment 3:
- The work will be more convenient if the authors can show improved current/voltage waveform after the decision tree optimization, and compare with several bench marks in THD.
Answer:
In the paper, the voltage and current waveforms (Fig. 18 - Fig. 21) have been added for the two selected cases after optimization (C and E routes). The waveforms can be compared with those obtained for the case without APFs (Fig. 6 and Fig. 7). Moreover, a new table (Table 5) summarizes values of THDV for all node voltages and THDI for all line currents for these cases. It can be compared with the data presented in Table 1.
Reviewer 2 Report
Please include a deeper review of the state of the art papers in chronological order in the introduction section.
The variable x and Rk in equation (2) have not been defined, nor an interpretation was given.
Please make a double-check of equations, especially (4).
Please, include a more robust and complete description of figures on the corresponding Captions.
Section 2 was named: FILTER OPTIMUM SIZING AND LOCATION, but the entire section only describes some optimization and classification methods, reorganized such a section.
Describe using more detail perturbations in waveforms in Figure 7. Check if the magnitude orders or the scales used in the vertical axis are correct.
Please include the Pareto frontier to support discussion for results in Figure 11.
Include a formal section to summarize all conclusions.
Some bibliographical references are outdated; please use novelty research in journals from the last five years to support this interesting work.
Author Response
The authors would like to thank the Reviewer for the effort in reviewing the paper. We would like to inform that the comments of the Reviewer have been taken into account and also the proofreading by a native speaker has been done and included in the new revision.
Comment 1:
Please include a deeper review of the state of the art papers in chronological order in the introduction section.
Answer:
Due to the fact that this paper’s type is not a review but an article, we tried to present the effects of literature review rather substantively than chronologically. However, we have added 10 new references and positions within certain areas (e.g. connected with optimization algorithms applied into solve the problem under consideration) have been mentioned chronologically. The total number of references has decreased because due to the remark of another Reviewer we removed some older positions, included originally in the reference list, introducing more current ones.
Comment 2:
The variable x and Rk in equation (2) have not been defined, nor an interpretation was given.
Answer:
Interpretation of variable denoted by x has been added.
Comment 3:
Please make a double-check of equations, especially (4).
Answer:
The equation has been reformulated in order to use notation that occurs more commonly in the literature. The previous form however, was correct – it indicated the use of complex numbers which represented each harmonic current.
Comment 4:
Please, include a more robust and complete description of figures on the corresponding Captions.
Answer:
Figures captions have been revised and in several cases they have been changed in order to provide more reliable description of figure content.
Comment 5:
Section 2 was named: FILTER OPTIMUM SIZING AND LOCATION, but the entire section only describes some optimization and classification methods, reorganized such a section.
Answer:
Thank you for drawing our attention to the section title, it has been changed in order to reflect contents more precisely.
Comment 6:
Describe using more detail perturbations in waveforms in Figure 7. Check if the magnitude orders or the scales used in the vertical axis are correct.
Answer:
These distortions are voltage drops which are caused by the high steepness of the transformer output current. The oscillations around these drops result directly from the used model of nonlinear receivers and the simulation method in the frequency domain. The presented waveforms are a combination of a finite number of harmonics obtained as a result of simulations and not measured waveforms. A relevant explanation has been included to the paper. The cause of the oscillations resulting from the Gibbs effect can be clearly seen in the figure below showing the current waveform of one of the receivers.
Magnitude orders and the scales used in the vertical axis in Figure 7 are correct. Scale in figure is in kV. Line-to-line voltage of the analyzed system is 12.47 kV RMS, but Figure 7 shows phase voltage which is approximately 7.2 kV RMS and 10.2 kV in peak.
Comment 7:
Please include the Pareto frontier to support discussion for results in Figure 11.
Answer:
The Pareto frontier has been added in Figure 11.
Comment 8:
Include a formal section to summarize all conclusions.
Answer:
According to instruction for authors provided by Energies, a separate conclusions section is not mandatory. However, we have added a formal section for conclusions in accordance with the Reviewer’s suggestion.
Comment 9:
Some bibliographical references are outdated; please use novelty research in journals from the last five years to support this interesting work.
Answer:
Proposed literature has been revised, and some instances, especially the older ones, have been excluded from the paper. Additionally more up-to-date references were added (10 positions). However, in order to maintain continuity of the literature review some of the older articles remained cited due to their relevance.
Reviewer 3 Report
This paper proposes to use decision trees, a kind of machine learning algorithm, for the optimal active filter placement (AFP) for harmonic losses and cost minimisation in energy distribution systems. The proposed algorithm is compared with a simplistic Brute Force approach that checks all possible combinations. Both algorithms are applied to the same power system showing superiority of the proposed method. The experiments in Section 3 are discussed in great detail which is highly benefitial for the readers and those who would like to reproduce the results or apply the method to their own problems.
The paper is very well-written and clearly communicates authors’ ideas. The references are sufficient and the extensive discussion support the reported results. The reviewer believes that the paper can be published in its current form.
On a side note, the purpose of naming nodes in Fig. 4 and associated Table 1 was not clear to the reviewer.
Author Response
The authors would like to thank the Reviewer for the effort in reviewing the paper. We would like to inform that the proofreading by a native speaker has been done and included in the new revision.
Comment 1:
On a side note, the purpose of naming nodes in Fig. 4 and associated Table 1 was not clear to the reviewer.
Answer:
The node names are entered on the original test system proposed by W.M Gardy (“Understanding power system harmonics”) and used in other referenced papers. We have added additionally numbering of these nodes to make the line names shorter. The authors realize that this may not be entirely clear, but thanks to this, someone can compare the obtained results with the results from previous works.
Round 2
Reviewer 1 Report
The revised version has a significant improve compared with the original version. My previous comments have been addressed.
Reviewer 2 Report
The authors addressed all my concerns satisfactorily. The manuscript is technically sound and deserves to be considered for publishing.
