Discovering Electric Vehicle Charging Locations Based on Clustering Techniques Applied to Vehicular Mobility Datasets

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript employs various existing clustering techniques to determine EV charging stations. In the Abstract, there is no description about what and how techniques are utilized to achieve the objective of the manuscript. In the Material and Method, the readability is reduced mainly because only descriptions about clustering procedures are given without functional graphs for illustration. Moreover, the procedure of the proposed method is not summarized in a graphical form.  In the Results, no descriptions are given about how the dataset preprocessing was carried out. Based on the simulations, it is unclear about the accuracy of identifying an available EV charging station nearby using the proposed method.     

Author Response

Thank you Reviewer 1 for allowing us to resubmit our work and at the same time address your comments/suggestions/questions.  Below are our answers to your review.

1. The manuscript employs various existing clustering techniques to determine EV charging stations. In the Abstract, there is no description about what and how techniques are utilized to achieve the objective of the manuscript.

Thank you for this comment.  We have revised our abstract to show how we employed clustering techniques to reveal possible locations of the EV chargers in an urban setup.  The changes to the abstract are written below.

In this study, we determine the possible EV charging station locations based on an urban city’s vehicular capacity distribution obtained from taxi and ride-hailing mobility GPS traces.  To achieve this, we first transform the dynamic vehicular environment based on vehicular capacity into its equivalent urban single snapshot.    We then obtain the various traffic zone distributions by utilizing initially k-means clustering to allow flexibility in the total number of wanted traffic zones in each dataset.  In each traffic zone, iterative clustering techniques employing Density-based Spatial Clustering of Applications with Noise (DBSCAN) or Clustering by fast search and find of density peaks (CFS) reveals various area separation where EV chargers are needed.  Finally, to find the exact location of the EV charging station, we last run k-means for centroid location, depending on the constraint on how many EV chargers are needed.

 

Our study shows how CFS + k-means clustering techniques have been able to pinpoint EV charger locations.  However, when utilizing DBSCAN initially, the results did not present any notable outcome.

 

2. In the Material and Method, the readability is reduced mainly because only descriptions about clustering procedures are given without functional graphs for illustration. Moreover, the procedure of the proposed method is not summarized in a graphical form.

 

Thank you for this comment.  We first separated the dataset and methodology into two sections.  In section 4, we added figures to aid readers on how the various clustering techniques achieve the objective of the study.

 

We revised Section 4.2 to incorporate illustrations on how the k-means, DBSCAN, and CFS were implemented using the enumerated procedure.  Please check 4.2 of the revised paper for these changes, especially the figures involved.

 

3. In the Results, no descriptions are given about how the dataset preprocessing was carried out. Based on the simulations, it is unclear about the accuracy of identifying an available EV charging station nearby using the proposed method.

 

Thank you for this comment.  We apologize for the confusion brought about by the “data preprocessing” we mentioned in Section 3.  We added a statement in Section 5. Results what we have done in preprocessing.  It is written as follows:

 

In pre-processing the datasets of BJG, JKT, and SIN, we only chose the features common to all datasets, i.e., Trajectory ID, timestamp, latitude, and longitude coordinates.  The other features found in the JKT and SIN datasets were simply disregarded.

 

We apologize for confusing our reviewer about identifying the EV charging location.  We revised Figure 14 to highlight the location of EV chargers.  These locations were derived by employing k-means clustering to determine the area’s centroid.  We also added a finding that states:

 

One important observation revealed by the lower right section of Figure 14 is the possibility of having EV chargers inaccessible to roads or can be situated in private spaces.  Such scenario will require more than engineering design and need government intervention.

Reviewer 2 Report

Comments and Suggestions for Authors

This research is well written and a very interesting it but our comments is as bellows;

Comment : In the discussion,  please state the limitations on the research, and in the conclusion, provide future recommendations in detail.

Comment : The equations need better formatting as some do not fully appear in the PDF version. Especially, Figure 2  requires revision for clarity. 

Author Response

Thank you Reviewer 2 for allowing us to resubmit our work and at the same time address your comments/suggestions/questions.  Below are our answers to your review.

1. This research is well written and a very interesting it but our comments is as bellows:

Thank you for this comment.  We appreciate that the reviewer sees this as an interesting research topic.

2. In the discussion, please state the limitations on the research, and in the conclusion, provide future recommendations in detail.

Thank you for this comment.  We included in our paper the major limitation in the implementation of this study.  It is written below and included in the major contributions section.

Our work is limited to studying urban and rural places which can be characterized by their rich empirical vehicular mobility datasets.  Places with sparse vehicular representation are excluded in this study.  As the features of the available datasets are varied, we only chose the common characteristics for comparison and contrast provided by our methodologies utilizing various clustering techniques.

At the end of our discussion section, we revised our summary of findings and challenges.  It is written as follows:

In summary, our methods provided possible EV charging locations for future EV users based on unsupervised learning clustering techniques, specifically, the CFS + k-means combination. The number of installed EV chargers is easily changed by varying the k value when k-means is re-run and evaluated by the validation indices.  Ideally, more chargers are needed to remove driving anxiety when compared to internal combustion engine vehicles. This work has also presented the importance of removing mobility traces outliers, as shown in the BJG dataset.  The road networks of BJG have been clearly visualized and interpreted because of the initial method of determining the spatiotemporal stable network characteristics.  However, the silhouette index still reveals that there are still too many outliers that were not removed.

We revised our conclusion to provide future recommendations in detail especially when employing the DBSCAN technique.  Our new conclusion is stated below.

In our work, DBSCAN reveals the road networks of the mobility datasets, and we just differentiate highly traveled ones from roads less traveled through color map representation.  In the next phase of this research, we dwell more on how we can utilize the DBSCAN output better in our datasets, especially in creating traffic zones/areas.  We would like to devise a systematic approach on how to obtain the DBSCAN parameters from a given dataset and ensure that there are good separations between clusters having the same network characteristics.  We target these results especially those that include locations with abundant traces.

3. The equations need better formatting as some do not fully appear in the PDF version. Especially, Figure 2 requires revision for clarity. 

Thank you for this comment.  We have already improved the quality of Figure 2, now labeled as Figure 3.  For the equations, we are using Latex to write all our equations and ensured that conversion to pdf allows correct display of all text, equations, figures, and tables.

Reviewer 3 Report

Comments and Suggestions for Authors

The paper, via various clustering methods, examines the formulation of strategic EV charging infrastructure plans, based on empirical vehicular mobility traces traversing an urban road network.

The topic is interesting and the authors have made an appreciable effort in their methodology. Please find below certain aspects that need to be addressed accordingly:

1.      What are the limitations of the study?

2. What other factors should be considered to ensure the study’s transferability?

3.      What constraints are anticipated when applying the methodology to urban and suburban motorways?

Comments for author File: Comments.pdf

Author Response

Thank you Reviewer 3 for allowing us to resubmit our work and at the same time address your comments/suggestions/questions.  Below are our answers to your review.

1. The paper, via various clustering methods, examines the formulation of strategic EV charging infrastructure plans, based on empirical vehicular mobility traces traversing an urban road network. The topic is interesting, and the authors have made an appreciable effort in their methodology.

 

Thank you for this comment.  We appreciate reviewer’s understanding of the effort the authors placed into the development of the manuscript before submission.

 

2. Please find below certain aspects that need to be addressed accordingly:
   1. What are the limitations of the study?

 

Thank you for this comment.  We included in our paper the major limitation in the implementation of this study.  It is written below and included in the major contributions section.

 

Our work is limited to studying urban and rural places which can be characterized by their rich empirical vehicular mobility datasets.  Places with sparse vehicular representation are excluded in this study.  As the features of the available datasets are varied, we only chose the common characteristics for comparison and contrast provided by our methodologies utilizing various clustering techniques.

 

1. What other factors should be considered to ensure the study’s transferability?

 

Thank you for this comment.  In Section 3, we added a statement on recognizing the differences of the available datasets and presented the pre-processing procedure we have undertaken to ensure that there is feature uniformity when analyzing the data.  It is written as follows:

 

Given the differences in the dataset's features and characteristics, we anticipate that various mobility traces datasets will have their own collection and storage formats.  Ideally, the target mobility dataset should contain the timestamped trajectories of a unique vehicle identified through its identification number.  To ensure that this work is flexible to the other available datasets, a pre-processing procedure has been implemented.   The taxi traces are formatted to achieve feature uniformity.

 

1. What constraints are anticipated when applying the methodology to urban and suburban motorways?

Thank you for this comment.  We added a paragraph in Section 3 discussing the constraints when involving urban and suburban networks. It is written as follows:

One final note regarding our datasets is that suburban vehicular networks are sparse in nature because more vehicles tend to pick up and drop off passengers in places where commercial and industrial areas are found.  We will see later in Section \ref{Results:3.2} that the urban road networks are easily visualized at the urban map's center because of the area's vehicular density.  As vehicles move away from the center, roads are scattered and less formed.

 

Reviewer 4 Report

Comments and Suggestions for Authors

This study utilizes clustering techniques on vehicular mobility datasets to determine optimal locations for electric vehicle charging. However, it lacks a comprehensive analysis. I recommend revising the study before resubmission. Below are some comments that may enhance its quality:

(1)   In Section 1, the logic of the Introduction is unclear, making it challenging for readers to grasp the current challenges. I recommend that the authors clearly outline the limitations of current studies in a point-by-point format.

(2)   This manuscript should include a separate literature review section, which the authors need to add.

(3)   The authors should include a flowchart of the technical approach; otherwise, the research methods section may seem like a patchwork of existing work.

(4)   Are the authors improving upon existing clustering algorithms or applying them? If it's an improvement, have they compared their method with traditional approaches? If it's an application, the current experimental results lack deeper analysis.

(5)   The abstract and main text lack new findings or conclusions. The authors should succinctly summarize the results presented in the manuscript.

Author Response

Thank you Reviewer 4 for allowing us to resubmit our work and at the same time address your comments/suggestions/questions.  Below are our answers to your review.

1. This study utilizes clustering techniques on vehicular mobility datasets to determine optimal locations for electric vehicle charging. However, it lacks a comprehensive analysis. I recommend revising the study before resubmission. Below are some comments that may enhance its quality:

 

Thank you for this comment.  We revised, to the best of our abilities, the manuscript such that there is a comprehensive analysis of the work we want to present for possible publication.  We apologize for lacking a comprehensive analysis as this was an exploratory study dwelling first on finding EV charging locations and ruling out optimization steps in the process.  We intend to enhance the clustering techniques in our future studies and incorporate optimization guidelines.

 

2. In Section 1, the logic of the Introduction is unclear, making it challenging for readers to grasp the current challenges. I recommend that the authors clearly outline the limitations of current studies in a point-by-point format.

 

Thank you for this comment.  We revised our introduction to highlight the problem we want to solve and present to the readers our methodology.  We invite the reviewer to read the introduction in our revised manuscript for this revision as highlighted.

 

 

3. This manuscript should include a separate literature review section, which the authors need to add.

 

Thank you for this comment.  We added a literature review section and placed there relevant information and the difference of our work.  We invite the reviewer to read the introduction in our revised manuscript for this revision as highlighted.

 

4. The authors should include a flowchart of the technical approach; otherwise, the research methods section may seem like a patchwork of existing work.

 

Thank you for this comment.  We added a block diagram of our research and a paragraph discussing our approach in the introduction.  We acknowledge that the reviewer is correct that somehow our work just uses existing clustering techniques that were used iteratively to meet the objectives of our study.  We inform the reviewer that this is exploratory research using iterative unsupervised learning techniques and may look like a patchwork of existing work.

 

In this study, we formulate a deployment strategy based on traffic zones that are characterized by a vehicular attribute or their combination. Figure 1 illustrates the workflow of the study undertaken.  Given a vehicular mobility dataset and after preprocessing, we chose a network feature (e.g., vehicular capacity, speed, environment information) to characterize a given urban map.  Dataset preprocessing is done to ensure mobility feature uniformity.  Realizing that the vehicular traces vary both in time and space, we calculated its spatiotemporal stable network characteristic first.  This method provides a single snapshot of the urban map under study.  From this stable snapshot, we then explore various clustering techniques to reveal areas exhibiting same network behavior.

 Please see the revised manuscript for the block diagram of the study.

 

5. Are the authors improving upon existing clustering algorithms or applying them? If it's an improvement, have they compared their method with traditional approaches? If it's an application, the current experimental results lack deeper analysis.

 

Thank you for this comment.  Our study is exploratory research through the application of three clustering techniques.  Because of this, we have only focused on 1) obtaining the spatiotemporal stable network snapshot, 2) applying clustering to the urban snapshot, and 3) evaluating the formed clusters.  The first focus was a previous method developed by the authors and was tested on how it can be implemented directly with clustering techniques.  We really apologize for the lack of deeper analysis as this targets to be our work’s benchmark.

 

6. The abstract and main text lack new findings or conclusions. The authors should succinctly summarize the results presented in the manuscript.

 

Thank you for this comment.  We have revised our abstract to show how we employed clustering techniques to reveal possible locations of the EV chargers in an urban setup.  The changes to the abstract are written below.

 

In this study, we determine the possible EV charging station locations based on an urban city’s vehicular capacity distribution obtained from taxi and ride-hailing mobility GPS traces.  To achieve this, we first transform the dynamic vehicular environment based on vehicular capacity into its equivalent urban single snapshot.    We then obtain the various traffic zone distributions by utilizing initially k-means clustering to allow flexibility in the total number of wanted traffic zones in each dataset.  In each traffic zone, iterative clustering techniques employing Density-based Spatial Clustering of Applications with Noise (DBSCAN) or Clustering by fast search and find of density peaks (CFS) reveals various area separation where EV chargers are needed.  Finally, to find the exact location of the EV charging station, we last run k-means for centroid location, depending on the constraint on how many EV chargers are needed.

 

Our study shows how CFS + k-means clustering techniques have been able to pinpoint EV charger locations.  However, when utilizing DBSCAN initially, the results did not present any notable outcome.

At the end of our discussion section, we revised our summary of findings and challenges.  It is written as follows:

In summary, our methods provided possible EV charging locations for future EV users based on unsupervised learning clustering techniques, specifically, the CFS + k-means combination. The number of installed EV chargers is easily changed by varying the k value when k-means is re-run and evaluated by the validation indices.  Ideally, more chargers are needed to remove driving anxiety when compared to internal combustion engine vehicles. This work has also presented the importance of removing mobility traces outliers, as shown in the BJG dataset.  The road networks of BJG have been clearly visualized and interpreted because of the initial method of determining the spatiotemporal stable network characteristics.  However, the silhouette index still reveals that there are still too many outliers that were not removed.

We revised our conclusion to provide future recommendations in detail especially when employing the DBSCAN technique.  Our new conclusion is stated below.

In our work, DBSCAN reveals the road networks of the mobility datasets, and we just differentiate highly traveled ones from roads less traveled through color map representation.  In the next phase of this research, we dwell more on how we can utilize the DBSCAN output better in our datasets, especially in creating traffic zones/areas.  We would like to devise a systematic approach on how to obtain the DBSCAN parameters from a given dataset and ensure that there are good separations between clusters having the same network characteristics.  We target these results especially those that include locations with abundant traces.

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

I have no further comments