Efficient Traffic Flow Guidance Feedback Strategy Considering Drivers’ Disobedience in ITS

Round 1

Reviewer 1 Report (Previous Reviewer 1)

--

Author Response

Reviewer 1 comments: there was no remark.

Response: We would like to thank the reviewer.

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 2)

I think the authors have carefully revised the manuscript according to the comments. Now, the manuscript can be acccepted due to that the manuscript quality have significantly improved.

Author Response

Reviewer 2 comments: I think the authors have carefully revised the manuscript according to the comments. Now, the manuscript can be accepted due to that the manuscript quality have significantly improved.

Response: We would like to thank the reviewer for his valuable remarks that allowed us to improve greatly the quality of the paper.

Author Response File: Author Response.docx

Reviewer 3 Report (New Reviewer)

In this paper, the deep learning methods are used to mine the relationship among the multiple traffic parameters including traffic flow, average travel speed, and traffic density to better predict large-scale traffic flow. Some problems should be explained:

(1) The normalized method should be given in Fig. 2.

(2) In 4.2, “LSTM block is used to extract the spatial features of the traffic data”. The spatial features should be temporal features?

(3) In table 1, the “SVR” should be “SVM”?

Author Response

Reviewer 3 comments

(1) The normalized method should be given in Fig. 2. (2)

Response: We would like to thank the reviewer for the valulable remarks.

We would like to note that there is no normalized method used in Fig.2. This figure shows our first experiments when no instructions given by system. The goal is to see what happen on the road during a congestion if no interference is made by the ITS system. At the beginning, the traffic lights created a congestion on the road. The waiting queue become longer as vehicles kept entering the route. Although congestion was reduced a little bit when the traffic lights turned to green, but not all the vehicles in the waiting queue were able to exit the route before traffic lights turned to red. The queue kept growing until the route was full of vehicles and the dynamic vehicle density reached its peak.

Figure 2a shows the mean velocity of the vehicles on the route. From the graphs of simulation, we can see that if no interference was taken, the congestion will not be eliminated by itself. The mean velocity reaches the peak at first, but when congestion happens, it goes down, and because of zero interference, the congestion keeps getting worse and just a few vehicles can pass through the route when the traffic light is green. The original route was full of vehicles when the simulation ends.

 

(2) In 4.2, “LSTM block is used to extract the spatial features of the traffic data”. The spatial features should be temporal features? (3) In table 1, the “SVR” should be “SVM”?.

Response: We thank the reviewer, we corrected the related issue in the new version of the manuscript.

Author Response File: Author Response.docx

Reviewer 4 Report (New Reviewer)

The paper is well presented. 

Author Response

Reviewer 4 comments: The paper is well presented. 

Response: We would like to thank the reviewer for the valuable remarks.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Authors present an approach to regulating the traffic flow dynamics considering drivers' obedience rate to instructions provided by an intelligent transportation system.

First two sections are devoted to the literature review. Authors clearly separate their aim, but the paper feels a little bit lengthy. Having a goal oriented introduction where only the main contributions are cited would increase the paper's impact. In case of being needed, at the beginning of the current section 3, a summary of the main assumptions required to develop the model can be incorporated. A flow chart of the methodology adopted or designed in this research highlighting the main contribution w.r.t. the previous publications would help as well in section 3.

One main concern regarding the paper is that is fully based on well established techniques to analyze the produced dataset. So, no new developments are presented in terms of methods. This drawback could be overcome by introducing relatively complex case studies. However, the case study presented in this paper is a very simple one. In order to consider the paper for publication, it would be necessary the inclusion of at least one more realistic scenario where the full potential of the proposed method can be better evaluated.

Authors made several mistakes with the conjugation of singular and plural verbs.

The quality of the figures should be increased. The labels-text ratio should be improved and captions should be more descriptive. Figure 5 is not a figure but a table and it is adding nothing to the analysis. It is made up of the same value 5 times. 

Conclusions are written in a non-standard way. First the limitations of the proposed method are listed, then what was done. Maybe reorganizing conclusions would help the reader.

Author Response

We would like to thank the anonymous referees and the Editor for their reviews, comments and remarks. We have produced a revised version of our article, based on these comments, and have performed a few extra changes to improve the quality of the manuscript. In the sequel, we provide a point-by-point answer to the reviewer’s comments

Response to the first reviewer

1. Does x in the k-means algorithm refer to the input vector? If so, how are these vectors quantified.

Yes, the x in the algorithm refers to the input vector; they were generated randomly but they were divided on two sets: training dataset and test dataset.

 

2. In this study, the penalty factor is set to 0.1, but the process of parameter tuning is not shown.

Several combinations of penalty factor and hyper parameter ? were tested. There values were taken in [2−5,210]. The model was tuned by applying several tests in order to avoid overfitting or underfitting as much as possible and the selected values provided the best accuracy.

 

3. Why is there a difference between the original route and the spare route in the simulation scenario? Is there any basis for the difference of specific values?

This is due to the simulation settings of SUMO traffic generator; several traffic cases regarding the positons of vehicles were considered. When an entrance is congested, newly arriving cars will not enter the scenario; therefore, in order to avoid such situation, the possibility of congestions on the other route are limited at the lowest level. In the simulation, a three-route scenario was designed. The system will ask drivers to change to a spare route randomly.

 

4. In the three groups of simulation experiments part, only the linear changes of their results were analyzed respectively, but the advantages and disadvantages of the three groups of experiments were not comprehensively analyzed.

We thank the reviewer for the comment. In the new version of the manuscript, we set the following sub-section in the “Simulation result and discussion” Section to clarify the analysis of the three groups of simulations.

Analysis of different scenes

In this sub-section, 100 times of simulations are taken for each type of instructions (among the 3 types discussed above) in order to compare the mean duration of congestions under different scenes. To ensure the accuracy and credibility of the comparison, the 300 times of simulation all use the same set of driver data. The three types of instruction are: a) The instruction with pre-filtering using SVM, b) The instruction without pre-filtering using -means, and c) The instruction without pre-filtering using Fuzzy Logic.

We can observe in Figure 5 that the mean duration for instruction without pre-filtering using Fuzzy Logic varies considerably, although it has a better performance than SVM several time. The mean duration is too unstable, and the mean value for the 100 durations is much greater than that of SVM. The mean duration for instruction without pre-filtering using -means is more stable than that of Fuzzy Logic, but the mean duration value is still higher than with SVM. This result shows that although instructions without SVM can help to reduce the congestion, it is not as efficient as the instructions with SVM and -means or fuzzy logic. The last two methods have proven their effectiveness.

 

5. It seems that the manuscript version needs to be edited and revised more carefully. For example, two periods are used at the end of the sentence on line 190, and the same error occurs on line 326.

We thank the reviewer for the comment. In the new version of the manuscript, we tried to correct all typos and grammar errors.

 

6. The following studies were recommended to be properly cited: Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison, IEEE Sensors Journal, vol. 20, pp. 14317-14328, 2020. Two-Level Hierarchical Model-Based Predictive Control for Large-Scale Urban Traffic Networks, IEEE Transactions on Control Systems Technology, vol. 25, no. 2, pp. 496-508.

Thank you for the suggestion. Yes, we have added these 2 references in the new version of the manuscript:

• Chen et al., "Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison," in IEEE Sensors Journal, vol. 20, no. 23, pp. 14317-14328, 1 Dec.1, 2020, doi: 10.1109/JSEN.2020.3007809.
• Zhou, B. De Schutter, S. Lin and Y. Xi, "Two-Level Hierarchical Model-Based Predictive Control for Large-Scale Urban Traffic Networks," in IEEE Transactions on Control Systems Technology, vol. 25, no. 2, pp. 496-508, March 2017, doi: 10.1109/TCST.2016.2572169.

Author Response File: Author Response.docx

Reviewer 2 Report

（1）Does x in the k-means algorithm refer to the input vector? If so, how are these vectors quantified.

（2）In this study, the penalty factor is set to 0.1, but the process of parameter tuning is not shown.

（3）Why is there a difference between the original route and the spare route in the simulation scenario? Is there any basis for the difference of specific values?

（4）In the three groups of simulation experiments part, only the linear changes of their results were analyzed respectively, but the advantages and disadvantages of the three groups of experiments were not comprehensively analyzed.

（5）It seems that the manuscript version needs to be edited and revised more carefully. For example, two periods are used at the end of the sentence on line 190, and the same error occurs on line 326.

（6）The following studies were recommended to be properly cited: [1] Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison, IEEE Sensors Journal, vol. 20, pp. 14317-14328, 2020. [2] Two-Level Hierarchical Model-Based Predictive Control for Large-Scale Urban Traffic Networks, IEEE Transactions on Control Systems Technology, vol. 25, no. 2, pp. 496-508.

Author Response

Response to the reviewer 2

1. Does x in the k-means algorithm refer to the input vector? If so, how are these vectors quantified.

Yes, the x in the algorithm refers to the input vector; they were generated randomly but they were divided on two sets: training dataset and test dataset.

 

2. In this study, the penalty factor is set to 0.1, but the process of parameter tuning is not shown.

Several combinations of penalty factor and hyper parameter ? were tested. There values were taken in [2−5,210]. The model was tuned by applying several tests in order to avoid overfitting or underfitting as much as possible and the selected values provided the best accuracy.

 

3. Why is there a difference between the original route and the spare route in the simulation scenario? Is there any basis for the difference of specific values?

This is due to the simulation settings of SUMO traffic generator; several traffic cases regarding the positons of vehicles were considered. When an entrance is congested, newly arriving cars will not enter the scenario; therefore, in order to avoid such situation, the possibility of congestions on the other route are limited at the lowest level. In the simulation, a three-route scenario was designed. The system will ask drivers to change to a spare route randomly.

 

4. In the three groups of simulation experiments part, only the linear changes of their results were analyzed respectively, but the advantages and disadvantages of the three groups of experiments were not comprehensively analyzed.

We thank the reviewer for the comment. In the new version of the manuscript, we set the following sub-section in the “Simulation result and discussion” Section to clarify the analysis of the three groups of simulations.

Analysis of different scenes

In this sub-section, 100 times of simulations are taken for each type of instructions (among the 3 types discussed above) in order to compare the mean duration of congestions under different scenes. To ensure the accuracy and credibility of the comparison, the 300 times of simulation all use the same set of driver data. The three types of instruction are: a) The instruction with pre-filtering using SVM, b) The instruction without pre-filtering using -means, and c) The instruction without pre-filtering using Fuzzy Logic.

We can observe in Figure 5 that the mean duration for instruction without pre-filtering using Fuzzy Logic varies considerably, although it has a better performance than SVM several time. The mean duration is too unstable, and the mean value for the 100 durations is much greater than that of SVM. The mean duration for instruction without pre-filtering using -means is more stable than that of Fuzzy Logic, but the mean duration value is still higher than with SVM. This result shows that although instructions without SVM can help to reduce the congestion, it is not as efficient as the instructions with SVM and -means or fuzzy logic. The last two methods have proven their effectiveness.

 

5. It seems that the manuscript version needs to be edited and revised more carefully. For example, two periods are used at the end of the sentence on line 190, and the same error occurs on line 326.

We thank the reviewer for the comment. In the new version of the manuscript, we tried to correct all typos and grammar errors.

 

6. The following studies were recommended to be properly cited: Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison, IEEE Sensors Journal, vol. 20, pp. 14317-14328, 2020. Two-Level Hierarchical Model-Based Predictive Control for Large-Scale Urban Traffic Networks, IEEE Transactions on Control Systems Technology, vol. 25, no. 2, pp. 496-508.

Thank you for the suggestion. Yes, we have added these 2 references in the new version of the manuscript:

• Chen et al., "Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison," in IEEE Sensors Journal, vol. 20, no. 23, pp. 14317-14328, 1 Dec.1, 2020, doi: 10.1109/JSEN.2020.3007809.
• Zhou, B. De Schutter, S. Lin and Y. Xi, "Two-Level Hierarchical Model-Based Predictive Control for Large-Scale Urban Traffic Networks," in IEEE Transactions on Control Systems Technology, vol. 25, no. 2, pp. 496-508, March 2017, doi: 10.1109/TCST.2016.2572169.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

In this new version, authors are not following the journal's style and figures/labels are of poor quality. Captions are still not descriptive.

Author Response

We thank the Reviewer for their positive comment and careful review, which helped improve the manuscript.

We enhanced the resolution of the figures (however, we are not able to modify as the captions, as the original figures were modified to images).

Reviewer 2 Report

My comments have been addressed. 

Author Response

We thank the Reviewer for their positive comment and careful review, which helped improve the manuscript.

Round 3

Reviewer 1 Report

--