https://orcid.org/0000-0003-2193-2133
Figures
Abstract
Identifying traffic congestion accurately is crucial for improving the expressway service level. Because the distributions of microscopic traffic quantities are highly sensitive to slight changes, the traffic congestion measurement is affected by many factors. As an essential part of the expressway, service areas should be considered when measuring the traffic state. Although existing studies pay increasing attention to service areas, the impact caused by service areas is hard to measure for evaluating traffic congestion events. By merging ETC transaction datasets and service area entrance data, this work proposes a traffic congestion measurement with the influence of expressway service areas. In this model, the traffic congestion with the influence of service areas is corrected by three modules: 1) the pause rate prediction module; 2) the fitting module for the relationship between effect and pause rate; 3) the measurement module with correction terms. Extensive experiments were conducted on the real dataset of the Fujian Expressway, and the results show that the proposed method can be applied to measure the effect caused by service areas in the absence of service area entry data. The model can also provide references for other traffic indicator measurements under the effect of the service area.
Citation: Liao L, Li Z, Lai S, Jiang W, Zou F, Yu X, et al. (2023) An expressway traffic congestion measurement under the influence of service areas. PLoS ONE 18(1): e0279966. https://doi.org/10.1371/journal.pone.0279966
Editor: Zhihong (Arry) Yao, Southwest Jiaotong University, CHINA
Received: October 23, 2022; Accepted: December 19, 2022; Published: January 6, 2023
Copyright: © 2023 Liao et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All data files are available on GitHub (https://github.com/lee-zr/service_area_traffic_congestion/blob/main/data.csv).
Funding: 1. Innovative Research Group Project of the National Natural Science Foundation of China. (41971340) Lyuchao Liao 2. Innovative Research Group Project of the National Natural Science Foundation of China. (41471333) Lyuchao Liao 3. Innovative Research Group Project of the National Natural Science Foundation of China. (61304199) Lyuchao Liao 4. Projects of Fujian Provincial Department of Science and Technology. (2021Y4019) Lyuchao Liao 5. Projects of Fujian Provincial Department of Science and Technology. (2020D002) Lyuchao Liao 6. Projects of Fujian Provincial Department of Science and Technology. (2020L3014) Lyuchao Liao 7. Project of Fujian Provincial Universities Engineering Research Center for Intelligent Driving Technology (Fujian University of Technology). (KFJ21012) Lyuchao Liao The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Traffic congestion has become a growing global problem due to its impacts on economics, safety, and the environment [1]. Hence, identifying traffic congestion is critical to enhancing their operational efficiency and improving the level of service. Expressways have gotten increasing attention from academic and industrial circles as an essential element of traffic networks [2–5]. Specifically, service areas, as a critical part of the expressway, provide rest and vehicle parking services for travellers [6]. When users get into service areas, the travel time for users will be longer than before. This impact will decrease the accuracy of traffic congestion measurement. In addition, this impact negatively impacts estimating driving speed. Therefore, it is necessary to conduct a comprehensive analysis to quantify the impacts of service areas on expressway operations, verify their roles in measuring traffic congestion and propose efficient measures to erase their effects when measuring traffic congestion.
To measure traffic congestion more accurately, existing studies proposed different measurements based on the causes of congestion [7, 8] and the influences of congestion on operations [9] under different scenarios. These measurements can be divided into the following categories: 1) speed-based methods, such as speed reduction index [10, 11], speed performance index [12, 13]; 2) travel time-based method, for example, travel time reliability [13]; 3) delay-based methods, including delay rate [14], delay ratio [14]; 4) level of service, such as V/C [15].
Due to that the distributions of microscopic traffic quantities are highly sensitive to slight changes [16] in the real traffic system, the measurement of these traffic quantities faces abound of effects caused by different factors, such as work zone [17, 18], traffic incident [19–22], inclement weather [23–25], demand/capacity [26] and the environment around. Hence, the measurement of traffic quantities accurately arouses increasing interests from the academic and industrial circles. Existing studies could be generally classified into two categories: 1) analyzing the impact of dynamic traffic environment; 2) providing the adaptability of measurements. For example, Kwon, J summarized the impact of various traffic factors on the measurements’ effectiveness, including traffic-influencing events (traffic incidents, work zone activity, and weather), traffic demand (fluctuations in demand, special events), and physical features (traffic-control devices, inadequate base capacity) [27].
Furthermore, Kamga, C explored the temporal patterns of average travel time, standard deviation, and coefficient variation in urban areas under different weather conditions [28]. They found that inclement weather indeed increases average travel time. Caceres, H further estimated route travel time distribution by the probabilistic model with considering the time of day, inclement weather, and traffic incidents; then, he found that route travel time variability increases when the mean travel time increases [24]. Some studies show that when the length of a segment is long enough and the impact of traffic accidents is small, the value of Travel Time Reliability (TTR) will be normal [29].
However, as a part of expressway systems, service areas are always ignored when analyzing the impacts of factors on traffic congestion measurement performance. Most of existing studies focused on the pause rate [30–33], travel time estimation [34–36], and the service area itself [37–39]. Specifically, they often notice the impact of service areas while estimating travel time. For example, Y. Ohba noticed the unusual travel time and removed them [34]; unfortunately, he failed to distinguish between the data with long travel time caused by congestion and the data caused by service areas. Interestingly, YangYu employed large-scale streaming ETC transaction data to estimate fine-grained edge travel time, but he regarded these data of long travel time as the outliers and ignored them [36]. Additionally, the pause rate is proposed to evaluate as the rate of the number of vehicles entering the service area to the total vehicles during a specific time. Some works predict pause rate by judging whether the average speed meets the extremely low speed caused by service areas [32]. In fact, the influence of service areas is obvious and we can use the pause rate to figure out these impacts; however, the data of entering service areas are generally absent, making the pause rate hard to be obtained. Therefore, the accurate analysis of impacts caused by service areas is still a challenge for identifying traffic congestion.
In this work, we propose a traffic congestion measurement under the influence of expressway service areas with the ETC transaction data, named speed performance index correction method (SPIco). Unlike the PeMS dataset, the ETC transaction data show a finer granularity and provide a shorter length of segments. Firstly, we explore the travel time difference between the users who get into the service area and those who don’t. Then, the traffic congestion with the influence of service areas is evaluated by three modules: 1) the pause rate prediction module; 2) the fitting module for the relationship between effect and pause rate; 3) the measurement module with correction terms.
The contributions of this work could be summarized as follows.
- We firstly employ the Kolmogorov–Smirnov (KS) test to examine the travel time difference between users who get into the service area and others.
- We establish a linear model to examine the relationship between pause rates and measurement bias caused by service areas.
- We propose a framework SPIco to enhance congestion measurements for segments involving service areas in expressways under the absence of data in the service area.
The remainder of this work is organized as follows. Section 2 describes the research problem; Section 3 introduces the details of the SPIco; Section 4 describes the experiments on real ETC dataset, and the results are discussed in section 5. Finally, the conclusion is summarized in section 6.
Problem definition
To analyze the impact of service areas, we focus on an expressway segment with a service area in this work. As shown in Fig 1, detector A and B are road detectors (ETC gantries), and the service area are between two detectors. When drivers enter the service areas, their travel time will be longer, influencing the accuracy of traffic congestion measurement. Hence, we explore the relationship between pause rate and the influence to get the real travel time and erase the impact of service area.
When the ith driver gets cross detectors, we can get the time tAi and tBi, then we define the ith driver’s travel time tABi, and it can be calculated as Eq (1):
(1)
For the segment like Fig 1, tABi can be divided into two parts tdi and trei. td i is the time of the ith driver getting through this segment directly, and trei is the time of the ith driver staying in the service area. Therefore, tABi could also be represented as Eq (2):
(2)
The average travel time TAB of this segment can be defined as Eq (3):
(3)
where m is the number of drivers. As shown in Eq (3), TAB contains all drivers’ travel time. In addition, when ith driver didn’t enter service areas, the value of trei will be 0 and tABi will be the same as tdi, which is defined as the situation
. We could then get the average travel time of drivers who didn’t enter service areas by Eq (4):
(4)
Where s is the number of drivers who don’t enter service areas.
Without the data of , most existing works employ TAB to measure traffic congestion, but
is better to reflect the real state of the segment. Unfortunately, it’s hard to get
directly. In this work, we aim to estimate the real state by analyzing the relationship between TAB and
. At first, we define this subtractive value between them as ΔTAB, as shown in Eq (5):
(5)
In this work, we explore the impact of service areas to ΔTAB by mapping ΔTAB to the bias of SPI and exploring the relationship between bias and pause rate. The pause rate can be calculated as Eq (6) [30]:
(6)
where m represents the number of all users in the statistical time and s represents the number of users who don’t enter service areas in the statistics time.
Methods
In this section, after judging whether the service area affects the measurement of traffic state, we propose a speed performance index correction method, named SPIco. At first, we employ ETC transaction data and entrance data of service area to extract and predict pause rate by XGBoost, and then we explore the relationship between bias and pause rate. Finally, we generate a correction term SPIco for estimating and correcting the impact from service areas.
As shown in Fig 2, SPIco contains three modules: 1) Pause rate prediction; 2) Regression of bias; 3) Correction term. Firstly, we extract the training dataset and the bias from ETC transaction data and service area data. Then in the module of Pause rate prediction, the training dataset is put into the XGBoost model to predict the pause rate. After predicting the pause rate, we explore the regression relationship between bias and pause rate. Then we obtain the corresponding bias with the predicted pause rate and put them into the correction term module. At last, the real traffic state is put out with the correct term.
To make sure SPIco still work under absence of data for service areas, we use ETC transaction data and entrance data of service area to extract pause rate and the correspond variables as input of XGBoost. Then we can predict the pause rate under absence of data for service areas. To fit the relationship between bias and pause rate, we calculate bias and use linear regression to fit the curve. At last, we add the correction term into SPIco. Fig 2 presents the proposed research framework.
Significance test of difference
The significance test of difference (STD) is used to examine whether there is a difference between two samples and whether the difference is significant in scientific experiments. To figure out whether there is a difference in the distribution of travel time between users who get into the service area and users who don’t, the steps of the difference test are as follows:
- Proposing a hypothesis H0. In this work, H0 is that there is no difference in the distribution of travel time between different user groups.
- Using Kolmogorov–Smirnov (KS) test to examine the truth of the null hypothesis.
- Compare it to a reference value to establish significance, the P-value. Based on that, either reject or not reject the null hypothesis (H0).
KS-test is a nonparametric test of the equality of continuous (or discontinuous), one-dimensional probability distributions that can be used to compare two samples [40]. It quantifies the distance between the empirical distribution functions of two samples.
Speed performance index
Speed performance index (SPI) is a measurement of traffic congestion, and it can be defined by the ratio between vehicle speed and the maximum permissible speed, as shown in Eq (7) [12]:
(7)
where vavg is the average speed of the vehicles within the segment or road; vmax is the maximum permissible speed of the road segment. Previous studies define the value of maximum permissible speed differently, including road-designed speed and the average vehicle speed during midnight40 (24:00–1:00) [41]. In this work, we use road-designed speed to simulate the actual traffic condition more accurately.
From Eq (7), it is evident that the value of SPI ranges from 0 to 100. Previous studies present that the value can be classified with three quantiles, and the classification criterion of the road traffic state is shown in Table 1.
SPIco to correct traffic congestion measurement with expressway service areas
Pause rate prediction.
In this work, we classified features into three parts: temporal features, traffic flow features, and traffic composition features.
- Temporal features
The temporal-based feature contains the periods of a day (15-minutes for each time interval) and weekday indicator (if the day is a workday, the value of this feature will be 1, otherwise 0). The temporal-based features are as follows:(8) where γ1 is the period, and its value ranges from 0 to 95; γ2 is a binary whose value is set to 0 for a workday and 1 for a non-workday.
- Traffic flow feature
More users are willing to get into service areas when the traffic flow is large and the traffic condition is heavy. Therefore, the traffic flow has also been considered, and the feature is shown as follows:(9) Where vi represents ith traffic flow.
- Traffic composition features
Given that drivers who use different vehicle types have different travel behaviors. For example, if the vehicle is an electric vehicle, the possibility of the user getting into service areas is higher than that who use traditional vehicles. Also, coaches which carry more people are more likely to get service than private cars. Therefore, the traffic composition feature is as follows:(10) where θ1, θ2, θ3 and θ4 represent the percentage of vehicle groups with different vehicle types. θ1 represents the blue color which refers to fuel cars; θ2 represents the yellow color which refers to large vehicles such as trucks or buses with more than 2 tons (including trucks with ten seats or more), agricultural vehicles, tricycles, trailers, coach cars, and operational vehicles, such as taxis and buses; θ3 represents the green color which refers to new-energy vehicles, and θ4 represents the other color of the vehicle plate.
XGBoost is an integrated learning method based on the boosting algorithm [42], which uses multiple base learners to reduce the model error. Each base learner is relatively simple to avoid over-fitting. The schematic of the model is shown in Fig 3. In this model, the next learner’s generation depends on the previous learner’s results.
We extract the feature vector from raw ETC transaction data and the data collected by the entrance of the service area in the expressway. After labeling the data, we set the dataset as S = {(x, y)}, where x = (x1, x2, x3, …, xn)T, and xi = (γ1, γ2, vi, θ1, θ2, θ3, θ4)T (i = 1, 2, …, N) which represents the feature vector of the ith sample. Y = (y1, y2, y3, …, yn)T(i = 1, 2, …, N) represent the pause rate corresponding to xi. The XGBoost model has K decision trees, and the prediction result is shown in Eq (11) [42]:
(11)
where K represents the number of trees, fk(xi) represents the predicted value of the kth decision tree on sample xi, and F represents the integrated classifier composed of all decision trees.
XGBoost does not directly minimize the above loss function as the training target but adds the complexity of the tree to the above formula to avoid overfitting. Therefore, the objective function consists of the loss function and the regularization item, as shown in Eq (12):
(12)
where
is the loss function, and then we use Mean Square Error (MSE) for regression, as shown in Eq (13):
(13)Ω(ft) is L1 regularization, as shown in Eq (14):
(14)
where α represents the regularization penalty coefficient, which takes values in the range of [0, 1]. Tk presents the number of leaves of the kth tree and Wk represents the leaf weight of the kth tree.
Regression of the bias.
To obtain the bias, we divide the original data into two typical samples (pure dataset, original dataset). Pure dataset indicates that do not contain the data that vehicles entering service areas, and the original dataset is the whole sample. We then obtain the bias by calculating the difference of SPI under the two samples.
(15)
where SPIoriginal represents the value of SPI under effect by service areas, and SPIpure is the value of SPI without being affected.
In this work, we explore the relationship between bias and pause rate by linear regression, whose function is as follows:
(16)
where rate is the independent variable pause rate, ω and b are parameters of the linear model, and the
is the regression value calculated by a linear model.
Correction term.
After prediction and regression, we established the measurement named SPIco to measure the traffic congestion under the effect of the service area. When the road segment contains a service area, the value of SPI will be lower. Therefore, when we use SPIpure − SPIoriginal to get the bias, the value of the bias will be positive. Then, we add a correction term to make the value more accurate for obtaining the true value of SPI. The correction term is shown in Eq (17):
(17)
where the
is the regression value, and the
is the prediction value of the pause rate.
Finally, the novel measurement method, named SPIco, is shown in Eq (18):
(18)
Evaluation metrics
In this work, we employed RMSE and MAE to evaluate the measurement performance of SPIco [43].
(19)
(20)
Where h(xi) is the vector of observed values, yi is the vector of predicted values, and n is the number of samples.
Data
Firstly, a case study was conducted to verify the difference between the two scenes and the adaptability of the proposed measurement. Specifically, we select a road segment on Shenhai Expressway in the Quangang District, Quanzhou City, Fujian Province, China, to analyze the difference and measure the traffic state. As shown in Fig 4, this segment is between two ETC gantries (from Quangang to Yiban and from Xianyoufeng to Quangang, in up and down direction); the length of this segment is 15.695 km. The raw data is collected by the ETC system when the vehicle crosses the gantry, like passenger ticketing records in a metro system. The raw data contain the time collected, the time to enter the expressway, the entrance name, and the vehicle license plate. For the service area, the data collected by video data, we extract the main features containing the entrance of a service area, the time collected, and the vehicle license plate. All the data features are shown in Table 2.
Results
Descriptive statistics analysis
After briefly describing the data, we tested the difference in travel time between users who get into the service area and those who don’t get into the service at first. Then, we plot the traffic flow change at different periods by continuous week data, as shown in Fig 5(A). To figure out the relevance of the pause rate, time period, and the feature workday, we plot the pause rate under the change of time period. As shown in Fig 5(B), the pause rate is positively correlated with traffic flow, and we can also find that the pause rate is less affected by the feature workday and is more volatile.
A: Time dependence of traffic flow. B: Time dependence of pause rate.
To explore the relationship between traffic composition and pause rate, we plot the change of traffic composition and the scatter of pause rate, shown in Fig 6. Fig 6(A) shows the change of traffic composition in the workday and non-workday. The x-axis is the time period, and the y-axis is the rate of the number of different vehicle types. The solid lines represent the rate of different vehicle types under non-workday, and the dashed lines represent the rate under workday. From 2:00 to 5:00, the number of yellow license plate vehicles is more than that of blue. In comparison, the number of blue license plate vehicles is more than the yellow from 6:00 to 24:00. This is because the yellow license plate vehicles are mainly trucks and coaches, the trunks tend to travel from 2:00 to 5:00, and the blue license plate vehicles are mainly the private fuel cars. In general, the rate of green license plate vehicles and other colors is low.
A: Time dependence of traffic composition. B: Correlation of traffic composition with pause rate.
As shown in Fig 6(B), the purple and yellow represent the green license plate vehicles and other color license plate vehicles; the blue and red represent the yellow license plate vehicles. The x-axis is the rate of traffic composition, and the y-axis is the pause rate. As shown in Fig 6, the blue license plate vehicles are positive to the pause rate; the yellow license plate vehicles are negative to the pause rate; the green license plate vehicles and the other color license plate vehicles have less correlation with the pause rate. The descriptive statistics of these variables are presented in Table 3.
Difference test
We select a segment and use a boxplot to plot the travel time from 13, May 2021 to 31, May 2021 to analyze the difference between the two samples. Fig 7 shows the boxplot of travel time for two scenes, the left item is the travel time for users who enter the service area during the travel, and the right item is the opposite. Because of the personalized service of service areas, such as lunch, taking a rest, and other activities, the travel time for users who get into the service area is much larger than those who go through the road directly in Fig 7. Hence, they need more time to finish the travel; it can be seen in Fig 7 that the upper boundary of the right boxplot lies in the lower 50% of the left boxplot. Generally, the spread and the range between the two scenes are very different, and the data for users who don’t get into the service area are more concentrated.
To explain the distribution of the two scenes more intuitively, we compared the travel time distributions in Fig 8. It is obvious that the distributions of the two scenes are different. The travel time distribution for users who get into service areas is more dispersed than others; this distribution is unimodal and skewed. But for travel time, when users do not get into the service area, the distribution is bimodal. Specifically, one of the modal centers is around 500s, and another is around 700s. We can transform time into speed, the first is about 113 km/h, and the second is about 80 km/h. Referring to Chen’s study, he compares the travel time under the free flow and peak period conditions, and the result shows that the distribution for these two conditions is different [41]. Therefore, in this case study, the first modal is the free flow travel condition, and the second is the peak period condition.
A: PDF of travel time. B: CDF curves of travel time.
After briefly viewing travel time distribution, we use the KS test to conduct a hypothesis test to verify the difference between these two conditions. In this study, the null hypothesis H0 is that there is no difference between the travel time under the two scenes, and the alternative hypothesis H1 is that there is a difference between the travel time under the two scenes. The result is rejecting the null hypothesis H0 and receiving the alternative hypothesis H1 that there is a difference between the travel time under the two scenes. The p-value is less than 0.001 under a 5% confidence level.
Measurement performance
Pause rate prediction.
We establish the dataset for XGBoost by using seven statistical features, as shown in Table 4. In the feature vector dataset, each vector contains seven dimensions of attributes, and the label y is the corresponding pause rate.
Before training the XGBoost model, we use Pearson correlation coefficient analysis to explore the correlation of these features and plot the heatmap, as shown in Fig 9. The color of the license plate feature θ, except θ4, has a positive and negative correlation with other features. In specific, the blue license plate rate θ1 has a positive correlation with time features and traffic flow, and the yellow license plate rate θ2 has a negative correlation with time features and traffic flow. In addition, the workday feature γ2 and the other color of the license plate rate θ4 have a very low correlation with other features.
The XGBoost regression algorithm has numerous parameters: the number of decision trees(n_estimators), learning rate(learn_rate), maximum depth of the tree(max_depth), minimum weight in leaf nodes(min_child_weight), the parameter that controls the number of leaves(gamma), the proportion of sample sampling(subsample), the scale of feature sampling(colsample_bytree). To find the optimal parameters, we use the grid search method combined with the K-fold cross-validation method. In this study, we set K = 5 to tune the model parameters. The search range and the optimal value of parameters are shown in Table 5.
To evaluate the effectiveness of the XGBoost model, we use RMSE and MAE to measure the prediction. The RMSE for the XGBoost model is 0.033, and the MAE is 0.026. As shown in Fig 10, the x-axis is time (slice a day into 96 periods), and the y-axis is pause rate. The blue line is the real pause rate, and the red is the prediction pause rate. Generally, the model has learned the pause rate changing characteristics well.
The bias regression.
In this work, we extract the pause rate from ETC transaction data and service area data by counting the number of vehicles per 15 minutes and plotting the scatter of the data, as shown in Fig 11.
A: Scatterplot of bias versus pause rate. B: Regression line.
In Fig 11(A), the x-axis is the pause rate, and the y-axis is biased. It is obvious that there is a linear trend; the bias is affected by the pause rate, and when the pause rate gets higher, the bias also gets higher. We use linear regression to fit the curve. The regression line is shown in Fig 11(B). The equation of the straight line is y = −0.07264 + 28.26x; the estimate of intercept is -0.07264, and the estimate of the slope is 38.36. The evolution metrics for linear regression are shown in Table 6.
The performance of correction term.
After exploring the relevance between pause rate and bias and predicting pause rate, we add a correction term into SPI. To verify the new method, we choose data from day 13, May 2021, as shown in Fig 12, the red line is the raw traffic state, and in Fig 12(A), the blue line is the value of SPIco, in Fig 12(B), the blue line is the value of SPI. It is obvious that SPICo is more accurate under the situation of the service area segment.
A: Comparison before adding the correction term. B: Comparison after adding the correction term.
Hence, we use the RMSE and MAE to evaluate the performance of the correction term. As shown in Table 7. The performance is better than the SPI under specific scenarios.
Discussion and conclusion
In this work, we explore the influence of service areas on traffic congestion measurement. Then, we propose a comprehensive framework SPIco to improve the performance of traffic congestion measurement for the expressway. In the framework, we employ the XGBoost model to predict the pause rate; and then, we explore the relationship between bias and pause rate by linear regression. And at last, we propose a correction term into SPI. The main findings are as follows:
- The service area significantly influences the performance of traffic congestion measurement for expressways;
- The distribution of the travel time for users who don’t get into the service area is bimodal; the phenomenon is caused by a period of a day, like a free flow period and peak hour period;
- The proposed SPIco could be used to evaluate the expressway traffic states under the effect of service areas.
The proposed framework, SPIco, could be employed to measure expressway traffic congestion under the effect of service areas, which can also be applied into other traffic index measurement under the effect. However, there are still some limitations in this work: firstly, the value of R Square for linear regression is around 0.66, which is acceptable to some extent. However, the relationship between bias and pause rate may be more complex; secondly, we ignore the impact of weather and other sudden events. In the future work, we will do a more in-depth analysis to explore the relationship between bias and pause rate, making the pause rate prediction and traffic index measurement more accurately.
References
- 1.
Reed Trevor. Inrix global traffic scorecard. 2019.
- 2. Zeng J, Qian Y, Yin F, Zhu L, Xu D. A multi-value cellular automata model for multi-lane traffic flow under lagrange coordinate. Computational and Mathematical Organization Theory. 2022 Jun;28(2):178–92.
- 3. Zeng J, Qian Y, Lv Z, Yin F, Zhu L, Zhang Y, et al. Expressway traffic flow under the combined bottleneck of accident and on-ramp in framework of Kerner’s three-phase traffic theory. Physica A: Statistical Mechanics and its Applications. 2021 Jul 15;574:125918.
- 4. Zeng J, Qian Y, Mi P, Zhang C, Yin F, Zhu L, et al. Freeway traffic flow cellular automata model based on mean velocity feedback. Physica A: Statistical Mechanics and its Applications. 2021 Jan 15;562:125387.
- 5. Qian Y, Zeng J, Wang N, Zhang J, Wang B. A traffic flow model considering influence of car-following and its echo characteristics. Nonlinear Dynamics. 2017 Jul;89(2):1099–109.
- 6.
Zhi P, Meng W, Wang JQ, Wu X, Zhou R, Zhou Q. Key Technology and Analysis of Expressway Intelligent Service Area. In2022 IEEE 25th International Conference on Computer Supported Cooperative Work in Design (CSCWD) 2022 May 4 (pp. 400–405). IEEE.
- 7.
Koźlak A, Wach D. Causes of traffic congestion in urban areas. Case of Poland. InSHS Web of Conferences 2018 (Vol. 57, p. 01019). EDP Sciences.
- 8. Pi M, Yeon H, Son H, Jang Y. Visual cause analytics for traffic congestion. IEEE transactions on visualization and computer graphics. 2019 Sep 10;27(3):2186–201.
- 9. Lu J, Li B, Li H, Al-Barakani A. Expansion of city scale, traffic modes, traffic congestion, and air pollution. Cities. 2021 Jan 1;108:102974.
- 10. Hau LV, Harris SE, Dutton Z, Behroozi CH. Light speed reduction to 17 metres per second in an ultracold atomic gas. Nature. 1999 Feb;397(6720):594–8.
- 11. Elkafas AG, Shouman MR. Assessment of energy efficiency and ship emissions from speed reduction measures on a medium sized container ship. International Journal of Maritime Engineering. 2021 Nov 16;163(A3).
- 12. He F, Yan X, Liu Y, Ma L. A traffic congestion assessment method for urban road networks based on speed performance index. Procedia engineering. 2016 Jan 1;137:425–33.
- 13.
Priambodo B, Ahmad A, Abdul Kadir R. Investigating Relationships Between Roads Based on Speed Performance Index of Road on Weekdays. InInternational Visual Informatics Conference 2019 Nov 19 (pp. 582–591). Springer, Cham.
- 14.
Huurne D, Andersen J. A quantitative measure of congestion in Stellenbosch using probe data. InProceedings of the 1st International Conference on the use of Mobile Informations and Communication Technology in Africa, Stellenbosch 2014 (pp. 62–67).
- 15. Roess RP, Vandehey MA, Kittelson W. Level of service: 2010 and beyond. Transportation Research Record. 2010 Jan;2173(1):20–7.
- 16. Krbálek M, Šeba F, Krbálková M. Super-random states in vehicular traffic–Detection & explanation. Physica A: Statistical Mechanics and its Applications. 2022 Jan 1;585:126418.
- 17.
Srivastava A, Chitturi MV, Ahn S, Rafferty P. Analytical methods for work zone travel time reliability. 2018 Jun.
- 18. Edwards MB, Fontaine MD. Investigation of travel time reliability in work zones with private-sector data. Transportation research record. 2012 Jan;2272(1):9–18.
- 19.
Tu H, van Lint H, van Zuylen H. The effects of traffic accidents on travel time reliability. In2008 11th International IEEE Conference on Intelligent Transportation Systems 2008 Oct 12 (pp. 79–84). IEEE.
- 20. Wright B, Zou Y, Wang Y. Impact of traffic incidents on reliability of freeway travel times. Transportation Research Record. 2015 Jan;2484(1):90–8.
- 21. Hojati AT, Ferreira L, Washington S, Charles P, Shobeirinejad A. Reprint of: Modelling the impact of traffic incidents on travel time reliability. Transportation research part C: emerging technologies. 2016 Sep 1;70:86–97.
- 22. Charlotte C, Sandra B. Empirical estimation of the variability of travel time. Transportation research procedia. 2017 Jan 1;25:2769–83.
- 23. Yazici MA, Kamga C, Singhal A. Weather’s impact on travel time and travel time variability in New York city. Transportation Research. 2013;40:41.
- 24. Caceres H, Hwang H, He Q. Estimating freeway route travel time distributions with consideration to time‐of‐day, inclement weather, and traffic incidents. Journal of Advanced Transportation. 2016 Oct;50(6):967–87.
- 25. Zhang X, Chen M. Quantifying the impact of weather events on travel time and reliability. Journal of advanced transportation. 2019 Mar 3;2019.
- 26.
Dowling RG, Skabardonis A, Margiotta RA, Hallenbeck ME. Reliability breakpoints on freeways. 2009.
- 27. Kwon J, Barkley T, Hranac R, Petty K, Compin N. Decomposition of travel time reliability into various sources: incidents, weather, work zones, special events, and base capacity. Transportation research record. 2011 Jan;2229(1):28–33.
- 28. Kamga C, Yazıcı MA. Temporal and weather related variation patterns of urban travel time: Considerations and caveats for value of travel time, value of variability, and mode choice studies. Transportation Research Part C: Emerging Technologies. 2014 Aug 1;45:4–16.
- 29. Zhang X, Zhao M, Appiah J, Fontaine MD. Improving interstate freeway travel time reliability analysis by clustering travel time distributions. Transportation research record. 2021 Oct;2675(10):566–77.
- 30.
Chen Y, Wang J. A Study on Forecast Method of Pause Rate in Expressway Service Area. InCICTP 2012: Multimodal Transportation Systems–Convenient, Safe, Cost-Effective, Efficient 2012 (pp. 286–291).
- 31. Jian-wei WA, Yi TA. Transportation potential calculation model of pause rate in expressway service area. China Journal of Highway and Transport. 2008 Sep 1;21(5):109.
- 32. Cai Q, Yi D, Zou F, Zhou Z, Li N, Guo F. Recognition of Vehicles Entering Expressway Service Areas and Estimation of Dwell Time Using ETC Data. Entropy. 2022 Aug 29;24(9):1208. pmid:36141094
- 33. Zhao J, Yu Z, Yang X, Gao Z, Liu W. Short term traffic flow prediction of expressway service area based on STL-OMS. Physica A: Statistical Mechanics and its Applications. 2022 Jun 1;595:126937.
- 34.
Ohba Y, Ueno H, Kuwahara M. Travel time calculation method for expressway using toll collection system data. InProceedings 199 IEEE/IEEJ/JSAI International Conference on Intelligent Transportation Systems (Cat. No. 99TH8383) 1999 Oct 5 (pp. 471–475). IEEE.
- 35. Fagnant DJ, Kockelman KM. The travel and environmental implications of shared autonomous vehicles, using agent-based model scenarios. Transportation Research Part C: Emerging Technologies. 2014 Mar 1;40:1–3.
- 36.
Liu S, Yao S, Huang Y, Liu D, Shao H, Zhao Y, et al. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies.
- 37. Wu Y, Wu C, Zhou J, Zhang B, Xu C, Yan Y, et al. A DEMATEL-TODIM based decision framework for PV power generation project in expressway service area under an intuitionistic fuzzy environment. Journal of cleaner production. 2020 Feb 20;247:119099.
- 38. Gao J, Pan J, Hu N, Xie C. Hydrologic performance of bioretention in an expressway service area. Water Science and Technology. 2018 Apr 20;77(7):1829–37. pmid:29676740
- 39. Koo C, Hong T, Kim J. A decision support system for determining the optimal size of a new expressway service area: focused on the profitability. Decision Support Systems. 2014 Nov 1;67:9–20.
- 40. Eghbali HJ. KS test for detecting changes from Landsat imagery data. IEEE Transactions on Systems, Man, and Cybernetics. 1979 Jan;9(1):17–23.
- 41. Chen Z, Liu XC. Statistical Distance-Based Travel-Time Reliability Measurement for Freeway Bottleneck Identification and Ranking. Transportation Research Record. 2021 Nov;2675(11):424–38.
- 42. Chen T, He T, Benesty M, Khotilovich V, Tang Y, Cho H, et al. Xgboost: extreme gradient boosting. R package version 0.4–2. 2015 Aug 1;1(4):1–4.
- 43. Chai T, Draxler RR. Root mean square error (RMSE) or mean absolute error (MAE)?–Arguments against avoiding RMSE in the literature. Geoscientific model development. 2014 Jun 30;7(3):1247–50.