Safety Assessment of Auxiliary Lanes in Freeway Interchange Weaving Areas based on Traffic Conflict Technique

Auxiliary lanes can provide an improved weaving environment rather than a forced or direct merge or diverse for vehicles entering and departing freeway weaving segments. Considering China’s traffic composition and operational status, this paper proposed safety assessment of auxiliary lanes in freeway interchange weaving areas based on traffic conflict technique in order to estimate its safety impacts and related influential factors considered in design. Based on freeway interchange weaving areas data and actual traffic characteristics, simulation models were built with VISSIM. This paper used Surrogate Safety Assessment Model (SSAM) to calculate conflicts by inputting vehicle trajectories from VISSIM, analysed the influence of ramp spacing, lane number of mainline, traffic volume, weaving ratio and percentage of heavy vehicles on safety performance of auxiliary lanes in freeway weaving areas. Finally, Traffic Conflict Modification Factor was presented to evaluate effectiveness of implementing auxiliary lanes. The paper found the safety impacts and related influential factors of auxiliary lanes. T he findings can provide support to effective decision making with regards to constructing future auxiliary lanes in freeway interchange weaving areas.


Introduction
As an important part of freeway, interchange and its operation performance is of great importance to the entire road network.Traffic flow features in weaving areas are much more co mplicated than other function areas, therefore, weav ing areas have a great influence on interchange safety.Auxiliary lane (A L) is a common facility in interchange weaving areas and is often used to improve weav ing environment by adding lanes between entrance-and exit-ramp pairs when the ramps are close.There are many standards specifications about AL setting and designing methods: A Policy on Geo met ric Design of Highway and Streets (AASHTO Green Book, 2004) [1], Manual on Uniform Traffic Control Devices (MUTCD, 2009) [2], Japan Highway Design Essentials (1991) [3], Design Specificat ion for Highway Align ment (2006) [4] and Gu idelines for Design of Highway Grade-separated Intersections [5].However, the setting methods and geometric design guidelines listed in these specifications are mainly based on capacity analysing and lack detail, which could lead to uncertainty to traffic engineers when considering designing ALs.Besides, a large proportion of AL-related researches focus on the operational impacts of ALs, very few on safety impacts.
As for the researches on safety impacts of ALs, Beverly Kuhn.et al analysed roads setting conditions and traffic crash rates in ramp influential areas and found that implementing ALs in merge and weaving areas can effectively reduce traffic crash and conflicts [6].According to principles of lane balance, Hongjun Tian took Wang Guantun interchange on Ji Nan-He Ze highway as an example to evaluate impacts of ALs and found that ALs can reduce the interruptions of main line vehicles, make traffic running smoother and safer [7].Worku Mergia identified crash inju ry severity influential factors were diver-related, traffic, environ mental and geometric design factors.A generalized ordinal logit model and part ial proportional odds model were applied to identify the factors that increase the likelihood of one of five levels of injury severity: no in juries, possible/invisible in juries, non-incapacitating injuries, incapacitating injuries, or fatal injuries.It was found that the use of continuous auxiliary lanes between an entrance ramp and an exit ramp tends to increase the likelihood of severe injuries near the diverge areas [8].Sarhan.etal studied 34 weaving segments in 26 interchanges to investigate the effects of ramp terminal spacing and traffic volu mes on safety performance.So me findings were found included: 1) the historical crash records show that the implement of A Ls did not improve the safety performances significantly at these sites.2) Weaving Type A was associated with relatively lower collision frequencies when compared with weaving Type B. 3) the number of collisions will decrease with increasing length of speed-change lane [9].Richard Glad analysed datasets and found that two-thirds of the rear-ends and one-third of the sideswipe occurred in ALs in weaving areas, although the crashes were mainly attributed to weaving traffic instead of the presence of ALs [10].Yi Qi.et al presented Traffic Conflict Modificat ion Factor (TCM F) as an indicator to analyse the changes in conflict ICTTE 2017 frequency before and after imp lementing A Ls under different conditions by simulation software.It was found that the reduction of traffic conflicts could be especially significant when high weaving volu mes were present [11].
Fro m the literature review, we can conclude that the safety impacts of ALs and related influential factors haven't been studied clearly.Besides, the analysis on safety impacts of ALs are mainly based on historical crash records.Crash records have issues of small simp le size, difficulty of obtaining, low data reliability and large time span.What is more, the method of analysing crash records is not available for facilities or managements to be or are being planned.It has been proved that traffic conflict is in correlat ion with traffic crash and is an effective surrogate measure of evaluating t raffic safety.Traffic Conflict Technique (TCT) has the advantages of short periods, large simp le size and high data reliability, etc. Safety performance can be evaluated under different conditions by combining TCT with simulation.
Therefore, TCT is applied in this paper to assess safety impacts of ALs in freeway interchange weaving areas.Take the basic setting form of A Ls as an examp le, different influential factors were studied in order to provide detailed guidelines on A Ls planning and designing.This paper also provide support to effective decision making with regards to weaving areas geometric design, management and prevention of accidents.

Data collection
The traffic survey was conducted at a semi-cloverleaf interchange with a typical weaving areas, which is composed of a one-lane mainline and a one-lane ramp (one-lane entrance and one-lane exit with one A L between).(Fig. 1) The width of every lane is 3.75m.The ramp spacing is about 165m and the rad ius of the ramps is 60m.Note that the mainline is in fact a collectordistributor road, so the weaving volume was high in this area, satisfying the requirement of sample size.In order to observe and analyse the traffic conflicts clearly, un manned aerial vehicle (UA V) photography was allo wed to shoot video in good conditions.High definit ion of the whole view of the interchange weaving areas can be obtained from the video.To get enough traffic volu me and traffic conflicts and eliminate random errors,four t imes shooting separately in the morn ing and afternoon.Besides, because of the limitation of battery lasting time, every one time lasts about 15 minutes.Finally t wo hours high-precision video was obtained.The aerial photo was below.(Fig. 2)

Model Assumption
Specific to the characteristics of the weaving area and the function of the simu lation software, so me as sumptions are present as below.
• According to Chapter 12 of HCM 2010 [12], there is at least one AL in freeway weav ing area.It also states that a freeway segment between an on-ramp and an offramp but without an A L should be analysed separately as isolated merge and diverge areas.However, for continuity and simp licity in analysis for the rest of this paper, we assume that a weav ing area can be designed with or without ALs.
• Assume that the main line and the ramps are straight, the influence of horizontal curve is neglected.Just consider the geometrical parameters such as length, width of the lane in the simulation experiment.
• The influence of gradient is considered to be 0.
• The angle between the ramp and mainline is 25° and the typical au xiliary lanes analysed in this paper is parallel type introduced in HCM 2010.
• The influences of weather conditions and lateral clear width on the traffic flow are neglected.
• Only small cars (Car) and heavy vehicles (HGV) are considered in the discussion.
• Assume that the design speed of the mainline is 80km/h and the ramp is 40 km/h.

Parameter Calibration
Calibrat ion in simulat ion mainly includes the representative vehicle parameters and three types of driver behaviours models: the car-following model, the lane-changing model and the transvers e behaviour model [13].Refer to the previous related research and take this paper's research object into consideration, parameters are set as follows: Representative Vehicle Parameters Calibration Cars and HGVs are considered in this paper and the calibrat ion of vehicle characteristic parameters include the length, maximu m/ minimu m and expectation of acceleration/deceleration. Specific settings are shown in Table 1.
The Car-following model The Wiedemann99 Car-following model was adopted and the specific parameters are shown in Table 2.
The lane-changing model

ICTTE 2017
Line-changing is the most significant behavior in weaving areas.Frequent lane changes may lead to disruption and congestion in traffic flow.The lanechanging model parameters include M in.Head way, Lane change Behavior and to slower lane if collision time above.Specific parameters setting are shown in Table3.
• The transverse behavior model Desired Positon at Free Flow, M in.Lateral Distance and Overtake on same lane are the main parameters in the transverse behaviour model.
• Simulation time period and number of times Considering the field investigation data, sample size requirement and in order to eliminate the influence of randomness of both traffic conflicts and simulation, every simu lation under each different comb ination of conditions conduct 10 times, every time lasts 4500s and the average is adopted as the final result.

Simulation Scheme
This paper's purpose is to evaluate safety impacts of auxiliary lanes in freeway interchange weaving areas.Safety influential factors can be main ly divided into geographic conditions (ramp spacing, lane nu mbers of ramp, lane numbers of main line, lane width, etc.), t raffic conditions (through volume, weav ing ratio, ramp -to-ramp volume, percentage of HGV, etc.) and other management conditions.
To avoid complication of processing data and to get representative results at the same t ime, kinds of combinations of conditions are established in the simu lation, the specific conditions of simulation schemes are shown in TA BLE5.We can get a total of 72 simu lation scenarios to compare safety conditions with an AL or not.

Traffic Conflicts Model and Asse ssment Index
SSAM is used to process data describing the trajectories of vehicles provided in simulation software, VISSIM in this paper.SSAM calculates surrogate measures of safety among every two vehicles and discern conflicts if the interaction satisfies the criteria for conflicts.[14] In this paper, TTC is chosen as the criteria for conflicts and it is defined as the minimu m t ime to collision value observed during the interaction of two vehicles on co llision course.
In this paper, we use 4s as the threshold and if at any time step the TTC is below this value, the interaction is tagged as a conflict [15].Traffic Conflict Rate is used as assessment index, and we use TCM Fs (Traffic Conflict Modification Factor) to analyze the safety impacts of A Ls. Similar to the CMFs presented in the AASHTO HSM , the TCMFs were provided for estimating the expected changes of traffic conflict rate at a location after imp lementing specific geometric treat ments associated with an A L. The TCMFs were calculated as follows [11]: If a TCMF value is less than 1.0, it means the treatment may potentially reduce the traffic conflicts and improve the safety performance.

Results
Decreases in traffic conflict rate are found after implementing an AL by analyzing the statistical results.The safety performance of ALs in weaving areas is related to ramp spacing, lane nu mber of mainline, t raffic volume , weaving rat io and percentage of HGV, as shown in Table6~8.Overall, installing au xiliary lanes at weaving areas can significantly reduce traffic conflicts rate as opposed to the cases without auxiliary lanes.

Lane number of mainline
Traffic conflict rate and TCMFs when s L is 200m, w R is 0.2 and HGV R is 25% are shown in Fig. 3~5.TCM Fs values get lower when there are fewer lanes in mainline, in other words, the safety impact is mo re effective.Because when lanes in mainline are fewer, there is a more co mplicated situation that through and weaving vehicles mixing in the same line.For the lack o f space for changing speed and wait ing for a safety gap, more conflicts will occur.So when there are fewer lanes in main line of weaving areas, auxiliary lanes can be taken into consideration.

Ramp spacing
TCMFs values get lower when the ramp spacing are shorter (Fig. 6).Because when the length between onramp and off-ramp is short, there are not enough time and space for those weaving vehicles to perform a series of behaviors including acceleration/deceleration, lanechanging, etc.Large d ifferences of vehicle speeds can lead to d isturbance and crash risk, especially when the speed of vehicles on freeway are h igh.The merge and diverge influence areas are normally defined at a length of 450m in the HCM (2000, 2010), so when ramp spacing is shorter than this value, it's essential to design ALs.This is particularly true for weaving areas at cloverleaf type interchanges.ALs can be a cost effective method to increase capacity and safety particularly in cases where the facility experiences high entrance and exits fro m adjacent interchanges.The threshold value can be smaller when traffic volume and weaving volume are high.

Traffic volume
The reduction of traffic conflicts with A Ls could be especially significant when high traffic volu me, especially weaving volu mes were present.(Fig. 7) (e.g., with a through volume of 1400 veh/h/ln and weaving ratio of 0.4).Weaving is the most complicated behaviour in weaving areas, it consists of groups of vehicles performing speed-changing, lane-changing and other actions.Besides, driver factors is much more co mplicated and could also be important in this situation.ALs can provide sufficient space for those weaving vehicles and could also be taken as a buffer zone for those careless drivers.

Conclusion
This paper investigated the safety impacts of adding auxiliary lanes by analyzing the traffic conflicts derived fro m the traffic simu lation studies.The results showed that: • Adding auxiliary lanes can significantly reduce the ratio of traffic conflicts in freeway weaving areas.
• Safety influential factors considered in ALs design include geographic conditions, traffic conditions and other management conditions.In particular, this paper analyze the factors of ramp spacing, lane numbers of mainline, through volume, weaving rat io and percentage of HGV.The results show that these factors have great influence on safety impacts of ALs.
The findings can provide support to effective decision making with regards to constructing future auxiliary lanes in freeway interchange weaving areas.

Figure 1 .
Figure 1.Weaving area in survey

Figure 2 .
Figure 2. Weaving area in aerial photo

Figure 3 .
Figure 3. Traffic conflict rate before implementing ALs

Figure 6 .
Figure 6.Relation between ramp spacing and TCM F

Figure 7 .
Figure 7. Relation between traffic volume and TCM F

Figure 8 .
Figure 8. Relation between percentage of HGV and TCM F

Table 1
Vehicle Parameters in simulation software

Table 2
Parameters in Wiedemann99 Car-following model

Table 3 .
Parameters in lane-changing model me, percentage of HGV, etc. Mean travel t ime and traffic conflict counts are chosen as indicators of simu lation test, and mean absolute percentage error is used to calculate the errors.Mean travel time can be obtained by setting time samp ler on the roads and traffic conflict counts can be obtained by analyzing the vehicle trajectories through SSAM.The verification result is shown in TABLE4.It is shown that the errors of mean travel time and traffic conflict counts are 1.648% and 9.69% respectively.So the simulation model is valid.
2.2.3 Model VerificationWe carry out the simulation by setting up a simulat ion network of a freeway weaving area with an A L and take the practical survey data as input parameters, such as traffic volu

Table 4 .
M odel verification test results

Table 5 .
Conditions of traffic simulation.