Dynamic Risk Assessment and Control Framework for Work Zone and Its First Implementation under Simulation Environment

,


Introduction
Maintenance is an essential guarantee for maintaining roads in good condition.As China's road network enters the maintenance era, road maintenance operations are becoming increasingly important.However, the presence of maintenance work zones can interfere with the normal operation of trafc and pose signifcant safety hazards.Data from the United States shows that in the past 15 years, road maintenance work zones have caused many trafc accidents, with an average of 734 deaths, 348947 injuries, and nearly 1430000 property losses per year [1].Although there is no such detailed accident data in China, considering the large population base, signifcant vehicle ownership, and more complex transportation environment, it is conservatively estimated that the trafc situation in China's work zones is more severe [2].
Te safety issues in the work zones have attracted widespread attention from researchers domestically and internationally.Researchers have conducted qualitative and quantitative analyses of various factors in the work zone based on trafc accident databases, attempting to reveal the mechanism of accidents in the work zone and proposing a series of trafc control methods.Many of these experiences and achievements have been included in the maintenance and construction technical guidelines.For example, China released the frst version of the "Safety Work Rules for Highway Maintenance" in 2004 [3] and revised it in 2015 [4], playing an important role in standardizing maintenance construction and risk control.However, these rules have not been able to prevent accidents in the work zone.Te safety issues in the work zone remain serious.One possible reason is that these rules can only provide general guidance and not targeted dynamic control for each work zone scenario [5,6].Te trafc environment on the road is dynamically changing.It is unknown whether the proposed control methods meet the requirements of the current work zone scenario, whether there are risks, and whether adjustments are needed [7][8][9][10].
In response to the above issues, this research proposes a dynamic risk assessment and control framework applicable to the work zones.Based on various advanced data collection methods, the risk assessment of the work zone can be completed in a short period (only 20 minutes at a moderate trafc level), and targeted control suggestions can be proposed when there are safety hazards in the work zone.Iterative/closed-loop control optimizes the work zone until it meets safety requirements.As the frst implementation of the framework, this study uses simulation methods for data collection.Te assessment and risk control model have been implemented, laying a solid foundation for subsequent case validation based on measured data.Te contribution of this study is twofold: (1) A complete dynamic risk assessment and control framework for the work zone has been proposed, which can timely diagnose and control the risks in the work zone; (2) a detection and assessment method has been proposed to cover single vehicle conficts, two vehicle conficts, and multiple vehicle conficts at one time.
Te arrangement of other sections is as follows: Section 2 reviews the development of risk assessment methods for work zones.Section 3 presents the risk assessment and control framework.Section 4 shows the frst implementation of this framework in a simulation environment.Section 5 summarizes the entire paper and provides prospects for future research.

Literature Review
In recent decades, research has mainly been based on accident databases [11].Te accident rate is the most direct refection of the safety level in the work zone.Using accident rate as an evaluation indicator, researchers have identifed many risk factors, including the length of the work zone, duration, trafc volume, speed limits, and trafc composition.Tey have qualitatively and quantitatively studied the impact of these factors on accident rates and established a relationship model between multiple factors and accident rates.Tis method, based on accident databases and using accident rates as an evaluation indicator, has signifcantly contributed to improving the safety of work zones.However, this method has apparent limitations [12,13], mainly manifested in its high dependence on accident databases.If the database is incomplete, it is easy to draw one-sided or even incorrect conclusions.
Recognizing the limitations of the above methods, researchers have proposed surrogate indicators for accidents.Tese indicators are relatively easy to observe and have a strong correlation with accident indicators, which means that accident rates can be estimated through surrogate indicators.Te proposal of surrogate indicators has extensively promoted the development of safety-related research in work zones [6].Among numerous surrogate indicators, trafc confict-based indicators have been widely recognized, mainly due to their similarity to the causes and processes of trafc accidents and the good correlation between particularly severe conficts and accidents [13].Tis technology, which uses trafc confict as a surrogate evaluation indicator for safety assessment, is also known as the trafc confict technique (TCT).
Te well-known defnition of trafc confict is proximity-based: Trafc confict is an observable situation in which two or more road users approach each other in space and time to such an extent that there is a risk of collision if their movements remain unchanged [14][15][16].Based on this defnition, indicators such as time to collision (TTC), proportion of stopping distance (PSD), and so on were proposed [17].In promoting TCT, the Federal Highway Administration (FHWA) has made signifcant contributions.Its software, Surrogate Safety Assessment Model (SSAM), can import TRJ fles generated by simulation software such as Vissim, detect two or multiple vehicle conficts promptly, and output various indicators.Te TRJ fle is specially designed for SSAM; therefore, by organizing the data into this format (no matter whether the data is obtained through simulation or measured methods) and importing it into SSAM, the trafc confict can be analyzed efectively.
However, as mentioned in its defnition, it is targeted at two-or multivehicle conficts without considering singlevehicle conficts.Data from the United States indicates that in addition to two or multi-vehicle accidents, the situation of single-vehicle accidents is also severe in work zones [1]: from 2007 to 2021, a total of 9888 fatal accidents occurred, wherein 49.%, i.e., 4854 accidents were single-vehicle ones.In addition, nighttime construction is being accepted by many maintenance departments to avoid serious congestion caused by maintenance.Nighttime trafc volume is lower, and it is difcult to observe conficts between two or multivehicles, while single-vehicle conficts are relatively more severe [18].Te limitations of existing TCT being unable to detect single-vehicle conficts are even more evident in this circumstance [17,19].
Our previous research [16] proposed a single-vehicle confict detection method based on another classic evasive behavior-based trafc confict defnition.Te defnition suggests that a trafc confict is a phenomenon in which a road user must take evasive behavior (e.g., lane changing, braking) to avoid collision [15,16,20].By executing automatic segmenting on the microscopic vehicle behavior data (MVBD) and setting thresholds, evasive behaviors can be detected, which means that the single vehicle conficts are detected.Furthermore, through integrating the SSAM method, a method for all types of trafc confict detection was proposed, which can efectively meet the needs of risk assessment in work zones.
Detecting trafc conficts of all types is based on evasive behavior and proximity indicators, which can only characterize the possibility of accidents and cannot represent the severity of accidents.For example, TTC is used to detect two 2 Journal of Advanced Transportation or multiple vehicle conficts.When TTC is smaller than 1.5 s, a trafc confict can be considered to occur.Assuming two conficts are detected, with TTC less than 1.5 s, one confict has a speed of 100 km/h for both vehicles, and the other has a speed of 5 km/h.Obviously, the former is more severe than the latter at the time of the accident, but the above severity judgment cannot be obtained solely from TTC indicators.
For another example, deceleration is used to detect singlevehicle conficts.When the deceleration is smaller than −3.92 m/s 2 , a trafc confict can be considered to occur.Assuming two conficts are detected, with deceleration less than −3.92 m/s 2 , one has a vehicle speed of 100 km/h, and the other has a speed of 5 km/h.Obviously, the former is more severe than the latter at the time of the accident, but the above severity judgment cannot be obtained solely from deceleration indicators.In fact, the assessment of conficts should be conducted from two dimensions: the possibility of causing accidents and the severity of the accidents.Te aforementioned detection of all types of trafc conficts has only completed the frst aspect of work, and a further improvement in its severity assessment is needed, which is a critical issue that needs to be addressed in this study.We presented a preliminary result of safety assessment and risk control for work zones in 2019 [2], proposing a fast safety assessment and correction framework.However, that work did not involve detecting all types of trafc conficts and did not consider the severity of establishing safety assessment methods and standards.Tis paper has signifcantly improved [2], with complete theory and excellent practicality.

Framework
Te dynamic risk assessment and control framework is shown in Figure 1.Te framework contains mainly fve parts, i.e., (1) collect MVBD on-site, (2) execute risk assessment based on the collected data, (3) make the judgment whether the work zone is safe or not, (4) make risk control decisions when the work zone is at risk, and (5) execute the control measure.It can be seen that this is a dynamic closedloop control process.After the execution of risk control measures, data will be continuously collected.Ten, the next round of assessment will be conducted to evaluate the effectiveness of control measures until the work zone is safe.In addition, the safety status of the work zone will also change due to dynamic changes in road trafc fow.Terefore, this framework continues conducting dynamic data collection and assessment to ensure the work zone remains safe.Te details of this framework are introduced as follows.
3.1.Data Collection.Te frst part is to collect the MVBD at the site of the work zone.Te MVBD refers explicitly to the vehicle's trajectory, speed, and accelerations in this framework.Te behavior data can be obtained using advanced data acquisition technologies such as video detection, naturalistic driving, and simulation-based data collection methods [16].Te video detection-based method uses cameras to record the trafc and machine vision to extract MVBD from the record.Especially at this stage, the maturity of multicamera technology makes it more convenient to obtain large-scale MVBD in the work zone [21].Te naturalistic driving-based method installs data acquisition instruments on the vehicle and collects MVBD in a naturalistic state.With the popularization of smartphones and smart cars, obtaining MVBD has become increasingly easy.With the help of crowdsourcing mode, a massive amount from the MVBD of vehicles passing through the work zone can be obtained [7].Te simulation-based method is to collect data directly from the simulation software.Trough high-precision scene reconstruction and calibration, simulation software can obtain high-fdelity data and output rich indicators, making simulation methods popular among researchers [22].

Risk Assessment.
Based on the massive amount of MVBD obtained through collection methods, a risk assessment of the work zone will be carried out, which will be introduced in three steps.

Detection of All Types of Trafc Conficts
. Te frst step is to perform all types of trafc confict detection based on the MVBD collected in Section 3.1.Considering that the detection method was detailed in our previous work, this section will only provide a brief introduction.For more detailed information, please refer to [16].
Te method can be summarized in Figure 2. Conduct detections for single-vehicle confict and two-or multivehicle confict based on MVBD, respectively.Wherein single-vehicle confict detection utilizes vehicle behavior analysis methods, including automatic segmentation of vehicle behavior and behavior recognition (i.e., extracting fragments of risk avoidance behavior and determining their types), the confict detection between two or multiple vehicles adopts an SSAM-based analysis method.It should be noted that the MVBD should be organized into TRJ fle format and imported into SSAM software for automatic analysis.Trough the above analysis, two detection results can be obtained.In fact, there is some overlap between these two sets of results.Some two-or multivehicle conficts may also appear in the single-vehicle confict detection results, and they should be removed.We adopted a straightforward but efective method: iterating the fragments of singlevehicle confict and determining whether they are in the SSAM analysis results (based on spatial and temporal distance).If so, it indicates a confict of two or multiple vehicles; if not, it is a confict of a single vehicle.

Comprehensive Assessment Indicator Calculation.
Based on our previous work, this paper proposes an indicator that can comprehensively evaluate the possibility and severity of accidents caused by conficts, namely, the UTECN (unit total equivalent confict number).Te calculation process is shown in Figure 3. First, detect conficts of single-vehicle, two-vehicle, or multivehicle separately.Ten, calculate the severity and possibility of accidents caused by conficts, and then calculate the risk values.
Finally, the equivalent number of conficts is calculated by introducing the standard confict risk value.
(1) Trafc confict severity evaluation indicator establishment based on vehicle collision energy theory.

(i) Two or multiple vehicle conficts
Tere is a loss of mechanical energy during the collision process, which can be used as an indicator to evaluate the severity of trafc conficts.Te following basic assumptions are made [23,24]: ①Considering only the most unfavorable scenario of accidents, a vehicle collision is defned as a completely inelastic collision, where the driver does not take any avoidance measures throughout the entire process, i.e., maintaining the current speed and trajectory state unchanged.②Vehicle collision is an ideal energy and momentum conversion process without considering the infuence of external forces.③Te vehicle is a rigid body whose attribute parameters do not change before and after the collision, and vehicles become a whole after the collision.
Tere are only rear-end collisions and lanechanging collisions in work zones.Te diference lies in whether the collision angle α is 0 or not.When α equals 0, it is a rear-end collision; otherwise, it is a lane-changing collision.Te vehicle collision process is shown in Figure 4, and the calculation of energy loss of collision can be uniformly derived as follows.
According to the above assumptions, all mechanical energy loss during a vehicle collision is converted into plastic deformation energy, which satisfes the conservation of energy and momentum theorems.Let E i,i−1 be the energy loss during the collision.Based on the conservation of energy before and after the collision, the following relationship can be obtained: Before and after the collision, the conservation of momentum theorem is satisfed in both the x and y directions. ( By combining (1) and ( 2), the energy loss in collision E i,i−1 can be calculated as follows: wherein m i and m i−1 is the mass of vehicle i and i − 1 (kg); v i and v i−1 is the speed of vehicle i and i − 1 instantaneously before the collision happens (m/s); v x ′ and v y ′ is the speed component of v ′ in the x and y directions, respectively (m/s);

(ii) Single-vehicle confict
Similarly, the severity of a single-vehicle collision can also be measured by the energy loss of the collision.Te same assumptions as two or multivehicle conficts are made.Te single-vehicle collision process is shown in Figure 5.According to the above assumptions, all mechanical energy loss during a vehicle collision is converted into plastic deformation energy, which satisfes the conservation of energy and momentum theorems.Let E be the energy loss during the collision.Based on the conservation of energy before and after the collision, the following relationship can be obtained: Before and after the collision, the conservation of momentum theorem is satisfed in the x direction: By combining ( 4) and ( 5), the energy loss in collision E can be calculated as follows: Specifcally, when the collided fxed object is a facility with infnite mass (such as walls and trees), i.e., m 2 is infnite, and all kinetic energy of the vehicle is lost.
(2) Trafc confict possibility evaluation indicator establishment based on the probability theory.Journal of Advanced Transportation (i) Two or multiple vehicle conficts TTC is essentially the time it takes for two or more vehicles to maintain their current physical state until a collision occurs.If the driver fails to take efective avoidance actions (braking and deceleration) to resolve trafc conficts during this period, the confict will become an accident.
On the contrary, it can be considered a safe scenario if the driver makes timely avoidance actions within the TTC time range.In fact, in a safe scenario, TTC can be further divided into four parts, including driver response time T r , driving operation coordination time T 0 (can take a value of 0.3 s according to [23]), shortest avoidance behavior time T s , and safety margin time ts (under the most unfavorable conditions, the value is 0), i.e., TTC � T r + T 0 + T s + t s .Te shortest avoidance behavior time T s is defned as the shortest time required for the driver to slow down to the preceding vehicle's speed.According to the relationship between the front and rear vehicles, calculations can be divided into three scenarios, as shown in the following equation:

Vehicle i-1
Vehicle i  6 Journal of Advanced Transportation Wherein v i−1 and v i are the speed of the front and rear vehicle, respectively (m/s); a max is the maximum deceleration of the rear vehicle (m/s 2 ) and can be taken as 4.51 m/s 2 according to [23,25]; and α is the lane change angle ( °).Te driver's reaction time T r is infuenced by individual diferences, and adverse road conditions and fatigue can also signifcantly increase reaction time.Its probability density function is shown in the following equation ( 9) [26]: Terefore, in the most unfavorable scenario, when TTC < Tr + T0 + Ts, i.e., the driver has insufcient time to resolve trafc conficts, resulting in a trafc accident, the probability is P (TTC < T r + T 0 + T s ).It is equivalent to calculate the probability P(TTC − T 0 − T s < T r ), i.e., (ii) Single-vehicle confict Te evasive behavior-based trafc confict defnition suggests that if the deceleration exceeds the threshold, it is considered a confict, but will this confict lead to an accident?Tis deceleration is the actual deceleration of the vehicle; is this deceleration sufcient?If it's not big enough, for example, if the actual deceleration is smaller than the needed deceleration, it can collide.Terefore, the probability of a single-vehicle accident occurring is P single � P(a actual < a need ).
Assuming the most unfavorable situation, the vehicle brakes sharply and stops just in front of the fxed object; that is, the distance between the vehicle and the fxed object is 0, the speed is 0, and the acceleration required for this process is a need .Assuming that the vehicle adopts a uniform deceleration motion, a need is a constant value, and a need � v actual /t need , then, the probability of a single-vehicle accident can be calculated as follows:

Journal of Advanced Transportation
Wherein T 0 can be taken as 0.3 s; Ts can be calculated as v actual /a max .
According to the basic theory of risk management, the risk is equal to the probability of an accident occurring multiplied by the accident loss.According to the previous deductions, the probability of an accident occurring is P, and the loss of the accident is E. Te accident risk value can be obtained from the following equation: (4) Number of equivalent conficts.Since most of the existing models for accident analysis are count models, they can be fully utilized by converting the above risk indicator into the number of trafc conficts.For example, for a specifc confict, if the risk value is calculated to be 1000 and the risk value of a standard confict is assumed to be 100, then this confict is relatively severe, equivalent to 10 standard conficts.Terefore, the key point of the equivalence method lies in selecting standard conficts and determining their risk values.
According to statistical theory, when the risk value of trafc confict reaches the 85% percentile, it is very close to an accident [27].Terefore, the standard confict risk indicator value can be obtained by drawing a cumulative percentage frequency chart of risk indicator values and selecting the 85% percentile as the risk indicator value.Tat is to say, when the risk indicator value of a confict exceeds the standard indicator value, the confict is already dangerous and is highly likely to evolve into a trafc accident and cause signifcant harm.Terefore, the conversion formula for Equivalent Confict Number (ECN) can be defned as follows: Wherein RI is the risk indicator value of the confict, and RI b is the standard confict risk indicator value.Ten, the calculation method for the trafc confict evaluation index UTECN of a specifc work zone within a certain evaluation period is as follows: Wherein n is the total number of trafc conficts; Length is the length of the assessed road section (km).[28].

Standard
LOSS is proposed with reference to the concept of road level of service.It refers to the level of safety service that road facilities can provide to all trafc participants or the feeling of trafc participants towards the level of safety service that road facilities can provide.In the LOSS analysis, indicators such as accident frequency, accident rate, number of deaths and injuries, and trafc volume can defne the safety service level of road sections.
Generally, the LOSS is divided into four levels, and the corresponding safety service levels from high to low are LOSS-1, LOSS-2, LOSS-3, and LOSS-4.Te corresponding conditions for each LOSS are as follows.
(i) LOSS-1: Te safety status of facilities is excellent, and the safety service quality that trafc participants can feel is high.Road safety is at a high level, and there is almost no possibility of further improvement and the department only needs to maintain the current situation.(ii) LOSS-2: Te safety status of facilities is good, and the safety service quality that trafc participants can feel is higher than expected.Certain measures can be taken to improve the safety situation further and achieve the optimal state.(iii) LOSS-3: Te safety status of facilities is fair, and the safety service quality that trafc participants can feel is lower than expected.Tere is still signifcant room for improvement in the current safety situation, and corresponding measures must be taken to improve it.(iv) LOSS-4: Te safety status of facilities is poor, and the safety service quality that trafc participants can feel is far lower than expected.Generally, it can be considered that sections with LOSS-4 are accident-prone sections, which pose signifcant safety hazards.Risk control measure is required to execute imperatively; otherwise, more trafc accidents will occur.
Te establishment of LOSS standards can be carried out as follows: (1) Draw a scatter plot of trafc volume and assessment indicator (for this study, it is the ECN).(2) Divide the trafc volume into several equal and continuous intervals.(3) Calculate the mean E i and standard deviation δ i of the assessment indicator of each trafc volume interval according to equation: Wherein, E i is the mean of the assessment indicators of all the sections within the i volume interval; δ i is 8 Journal of Advanced Transportation the standard deviation of the assessment indicators of all the sections within the i volume interval; A k is the assessment indicator value of the k road section; j is the number of road sections within the I volume interval.(4) Calculate the E i + 1.5δ i , E i , E i − 1.5δ i within each trafc volume interval.Ten three sets of data can be obtained, namely, (E i + 1.5δ i , volume i ), (E i , volume i ), and (E i − 1.5δ i , volume i ). ( 5) By conducting regression analysis on the three sets of data, the following models can be obtained: Wherein y 1 is the upper bound of the LOSS gradation; y 2 is the mean of the LOSS gradation; y 3 is the lower bound of the LOSS gradation; and volume is the trafc volume, vehicles/hour.( 6) Te three curves y1, y2, and y3 divide the scatter plot into four regions, from bottom to top, which are LOSS-1, LOSS-2, LOSS-3, and LOSS-4.Te schematic diagram is shown in Figure 6, where y1, y2, and y3 are the dividing lines of LOSS.
With LOSS, evaluating whether the work zone is safe is very simple; simply compare it with the LOSS of normal road sections.For example, suppose the LOSS of the section where the work zone is located is LOSS-1 under normal conditions.In that case, the ideal state for the operation period is maintaining the safety service level at LOSS-1 through control measures.However, in reality, it is not easy to maintain a level of safety service consistent with normal road sections due to work zones.In other words, achieving this goal requires a very high cost (at the extreme, by detouring all the upstream vehicles, the safety service level of this work zone section must be LOSS-1).Tus, the department can establish an acceptable LOSS reduction after comprehensively considering various factors.For example, the department may accept that during the construction period, the safety service level of the road section decreases by one level.Ten, when the actual LOSS in the work zone is downgraded by more than two levels, control measures must be taken.

Decision-Making for Risk Control.
Using the abovementioned method, a dynamic risk assessment can be conducted for the work zone.When it is found that the risk is unacceptable, i.e., the safety service level is lower than the acceptable level, control measures should be taken.Ten, what control measures should be taken to quickly eliminate risks and adjust the level of safety service to an acceptable range?Tis can be answered by constructing a relationship between multiple factors and UTECN based on a large amount of historical data.
Existing research has explored the impact of multiple factors such as working area length, trafc volume, and the proportion of large vehicles on accidents in work zones [11].For example, research has shown that the proportion of large vehicles signifcantly impacts accidents in the work zone.Conversely, when it is found through measured data that the large vehicle proportion is high and the risk value is high, the risk can be reduced by reducing the proportion of large vehicles.Terefore, selecting factors and modeling the relationship between factors and accidents are keys to this framework.Reviewing existing literature [11], it was found that factors afecting the safety of work zones include the proportion of large vehicles, trafc volume, speed limit, merging type, road grades, lane closures, and adverse weather conditions.Considering that the purpose of modeling in this section is to serve risk control decisionmaking, it is recommended to choose factors controllable by the department as the key consideration.As for modeling the relationship between multiple factors and accidents in work zones, it includes Poisson and negative binomial (NB), zero-infated NB and Poisson, truncated regression, generalized additive model, Conway Maxwell Poisson model, and negative multinomial model, etc [11,29].After establishing the model, analyzing the marginal efects of each factor can guide control decisions.

First Implementation
To verify the feasibility of this framework, this section implements it for the frst time in the form of a case study.Tis maintenance operation was carried out on the Shanghai Waihuan Expressway S20 (a two-way, eight-lane expressway with a speed limit of 80 km/h) to repair the pothole in lanes 1 and 2. Only lanes 1 and 2 are closed to reduce construction's impact on trafc (see Figure 7).Te lengths of the warning area and work area are 500 m and 170 m, respectively.Te trafc volume is 3500 vehicles/hour.Te proportion of large vehicles is 0.22.Tis case uses simulation methods to collect MVBD; the simulation software is Vissim.Accurate restoration of the work zone site was conducted in Vissim, and the model was calibrated using measured data.

Data Collection under Simulation Environment.
Vissim can output detailed data.Tis case sets the simulation resolution to 20-time steps per simulation second.Te output fle selects the vehicle number, acceleration, world coordinate front x, world coordinate front y, world coordinate rear x, world coordinate rear y, speed, simulation time, vehicle type, and weight.Vissim can also directly output the TRJ fle data required by SSAM.[16] has established the trafc confict detection method and implemented it through coding in MATLAB.After importing the MVBD exported from Vissim into the single-vehicle confict detection program, a total of 30 single-vehicle conficts were identifed.When the TRJ fle generated from Vissim was imported into SSAM, a total of 188 two or multivehicle conficts were detected.Te number of single-vehicle conficts is 25, and the number of two or multivehicle conficts is 188.Part of the conficts and their key parameters are shown in Tables 1 and 2.

Calculation of Assessment
Indicator.Based on the data in 4.3, the confict severity and possibility calculation method in 3.2.2 were implemented in MATLAB.Te results were obtained, as shown in Tables 3 and 4, respectively.Due to the uncertainty of the standard confict risk value, it is not yet possible to calculate the equivalent number of trafc conficts, which will be determined by simulating a large scale of the S20 and conducting a statistical analysis, as shown in Section 4.4.To establish a risk assessment standard for the S20 Expressway, this study simulated a large scale of the S20 for 75 kilometers in Vissim.Using the same confict analysis method as for the above work zone, the conficts, severity, possibility, and risk index corresponding to each confict were obtained for this model.Draw the cumulative frequency distribution curves of the risk indices of single vehicle confict and two or multiple vehicle confict in S20, respectively; take the 85% percentile risk value as the standard risk value for each type of confict.For single-vehicle conficts, the standard confict risk value is 58000 J; for two or multivehicle conficts, the standard confict risk value is 490000 J. Te equivalent number of conficts for each confict in Section 4.3 can be calculated using the standard risk value.For example, for the frst single vehicle confict in Table 3, with a risk value of 1846203 J, the equivalent number of trafc conficts is 1846203/58000 � 31.8.For the frst two or multivehicle conficts in Table 4, with a risk value of 1928.3J, the equivalent number of trafc conficts is 1928.3/490000� 0.004.Te total number of equivalent conficts in the work zone is obtained by adding up all the equivalent numbers of conficts.
And the UTECN can be obtained by dividing the length of the work zone, i.e., 1 km in this case.For single-vehicle conficts, the UTECN value is 173 per km; for two or multivehicle conficts, the UTECN value is 170 per km.
To determine the risk level of the work zone, it is necessary to establish risk assessment standards.According to the steps introduced in Section 3.2.3,divide S20 into units with a length of 1 km and calculate the UTECNs and trafc volume for each section.Tese sections are grouped according to their trafc fow with a width of 1000 vehicles per hour, and the mean, standard deviation, mean −1.5•standard deviation, the mean-+ 1.5•standard deviation of UTECNs in each group are calculated to obtain the distribution shown in Figure 8.
It can be seen that the ftting line of the mean and the one of the mean + 1.5 standard deviation divide the region into three regions (since the mean −1.5 standard deviation is less than 0, the (0, mean −1.5 standard deviation) and (mean −1.5 standard deviation, mean) regions can be merged).
Area A indicates a low risk and belongs to LOSS 1 and 2. Area B has a high risk and belongs to LOSS 3. Area C has a very high risk and belongs to LOSS 4.
Te trafc volume in this work zone is 3500 vehicles/hour.Te UTECN for single-vehicle confict is 173.It is located in Area C in Figure 8, of which the LOSS is 4. Terefore, it can be determined that the risk value of single-vehicle conficts is very high.Te UTECN of two or multivehicle conficts is 170.It is located in Area C in Figure 8, of which the LOSS is 4. Terefore, it can be determined that the risk value of conficts between two or more vehicles is also very high.In fact, through S20 simulation, it was found that the LOSS of the section where this work zone is located is 1 or 2 in the normal state (i.e., when no work zone exists).It can be seen that the presence of the work zone seriously afects the LOSS, and specifc measures need to be taken.If conditions permit, optimizing and elevating the LOSS to LOSS 1 and LOSS 2 is recommended.If conditions do not permit, the LOSS should also be elevated to at least LOSS 3; otherwise, it is easy to cause accidents.So, what specifc measures should be taken?Te answer will be obtained in 4.5.

Decision Making.
A regression relationship between multiple factors of the work zone and UTECN will be established in this section.Referring to previous research [11], 5 factors are selected for analysis, including the length of the warning area, the speed limit of the warning area, the length of the working area, trafc volume, and the proportion of large vehicles.Each factor considers four levels, as shown in Table 5. One-way analysis of variance (ANOVA) is adopted to explore the sensitivity of these factors to the impact of UTECN.20 sets of experiments were designed, each of which only changed the value of one factor based on the real scenario in Section 4.1.For example, when analyzing the sensitivity of trafc volume, only changing the trafc volume to 2000, 2500, and 3000 vehicles/hour and keeping other factors consistent with the real scenario.Perform two parallel experiments in each set.After the simulations are completed, the aforementioned method is used to calculate UTECNs, and one-way ANOVA is performed on SPSS.Te results are shown in Table 6.It can be seen that the proportion of large vehicle, trafc volume, and speed limit of the warning area have a signifcant impact on single-vehicle conficts, while the length of the warning area and the length of the working area do not show signifcant diferences (p > 0.05).Te proportion of large vehicles, trafc volume, speed limit of the warning area, and length of the warning area have a signifcant impact on multivehicle conficts, while the length of the working area does not show signifcant diferences (p > 0.05).
According to the ANOVA results, subsequent analysis will only consider the impact of the proportion of large vehicles, trafc volume, speed limit of the warning area, and length of the warning area on UTECN.Design 16 orthogonal experiments (as shown in Table 7).Set up two parallel experiments for each experiment.Conduct simulations separately, and then calculate the UTECNs in each work zone.Te average of parallel experiments was taken as the fnal result of each experiment set and summarized with the actual scenario to obtain the results shown in Table 7.
Based on previous research [11,30], a regression relationship between multiple factors and UTECNs was established using Poisson regression.Te results are shown in Table 8: Results show that, for single-vehicle conficts, the speed limit of the warning area, trafc volume, and large vehicle proportion signifcantly impact UTECN.Teir contribution ranking is trafc volume > proportion of large vehicles > speed limit in the warning area.McFadden R 2 is 0.652, which means that the three factors can explain the 65.2% change in the UTECN.Te regression model can be written as log (u) � −3.157 + 0.017 Speed limit of the warning area + 0.002, trafc volume + 2.995, and proportion of large vehicle (where u represents the expected mean).
For two-or multivehicle conficts, the speed limit of the warning area, trafc volume, and large vehicle proportion have a signifcant positive impact on UTECN, and the length of the warning area has a signifcant negative impact on UTECN.Teir contribution ranking is trafc volume > proportion of large vehicles > length of the warning area > speed limit in the warning area.McFadden R 2 is 0.789, which means that the fve factors can explain the 78.9% change in the UTECN.Te regression model can be written as log (u) � −0.832-0.001.Length of the warning area +0.009.Te speed limit of the warning area +0.001.Trafc volume + 3.871.Proportion of large vehicle (where u represents the expected mean).
It can be seen that for this work zone, reducing trafc volume or the large vehicle proportion can efectively reduce UTECN.Reduce the trafc volume to 2500 vehicles/hour through measures such as detours and rerun the simulation.Results show that the UTECN of two or multivehicle conficts is 9, the one for single-vehicle conficts is 43, and the new LOSS of the work zone is improved to LOSS 3.Alternatively, by controlling the proportion of large vehicles passing through upstream, such as by reducing the proportion of large vehicles to 0.1.Te results show that the UTECN of two or multivehicle conficts is 9, the one for single-vehicle conficts is 53, and the new LOSS of the work zone is improved to LOSS 3. Terefore, this risk control is efective and signifcant.
In fact, there are many other factors that can be considered, such as the number of closed lanes, lighting conditions, whether signal control is used, whether trafc police arrive at the scene, etc.By collecting a large amount of data and including the factors into the regression analysis model,  4.6.Discussion.As the frst implementation, this case adopts a simulation method that is easier to obtain data than actual measurement, which verifes the feasibility of the framework and methods to some extent.If measured data is used, such as data collected through a camera, the data extracted by the camera should be organized according to the requirements of the aforementioned program.For example, according to SSAM's requirements, trajectory data should be organized into TRJ format to be directly imported into SSAM for trafc confict analysis.For the MVBD acquisition of the entire road, collecting data on a large scale is not easy.It can be replaced by collecting data in several typical sections and calculating the total number of equivalent conficts, which can also achieve good results.Tese changes are only refected in data collection and do not require any changes to the framework, indicating that this framework can adapt well to the needs of on-site measurement.Tis case only studied the infuence of the length of the warning area, speed limit in the warning area, large vehicle proportion, trafc volume, and length of the working area on UTECN in the work zone.Many other factors have not been considered, such as weather conditions, lighting conditions, the length of the transition area, etc.It explains why the R 2 of the regression models is not particularly high.Further research will consider more factors for modeling, and these improvements do not require changes to this framework.
Finally, whether the LOSS of the work zone needs to be improved is actually a comprehensive optimization problem.Te presence of work zones will inevitably afect the LOSS, and improving LOSS comes at a cost, including owner cost (required to add/change facilities) and user costs (due to detours and congestions).Tis paper has not yet considered calculating costs and benefts and only proposes basic judgment principles.When it is found that the LOSS of the work zone has decreased, in principle, various measures should be taken to restore the level to the normal level.If conditions do not allow and the ideal state cannot be achieved, it should also be ensured that the service level in the work zone does not difer too much from the normal level.For example, descending by one level is acceptable, but measures should be taken when the descent reaches two levels.Comprehensive consideration of various costs and benefts to guide decision-making will be done as follow-up research.

Conclusions and Outlook
Tis paper constructs a dynamic risk assessment and control framework applicable to work zones, which has the following distinct characteristics: (1) collecting massive MVBD through multiple cameras or naturalistic driving or simulation.6) Adopting closed-loop control and dynamic evaluation for architecture.Te framework was frst implemented under a simulation environment using a work zone of the Shanghai Waihuan Expressway S20 as a case study.Te following conclusions can be obtained: (1) Tis framework adopts an all types of trafc confict analysis technology, which has the ability to quickly assess the safety condition of the work zone.At a moderate trafc level, it only takes 20 minutes to conduct a risk assessment for the work zone, which is very suitable for high-frequency and short-duration maintenance operations.(2) Tis framework has good practicality.Computer vision applications are very common, and large-scale multi-target tracking technology is relatively mature.Alternatively, naturalistic driving methods can obtain large-scale measured MVBD in the work zone.It can be input into various models for rapid analysis by organizing it in a predetermined format (for SSAM, it needs to be converted to TRJ format).
Next, we will conduct further research in the following areas: (1) designing a large-scale MVBD collection method and device based on multicamera and multitarget tracking technology suitable for the work zone; (2) conducting an (3) Establishing a road network LOSS database based on the total number of equivalent conficts to prepare conditions for risk assessment standards and guide the risk control decision-making of actual work zones.

Figure 1 :
Figure 1: Dynamic risk assessment and control framework for work zones.

Figure 2 :
Figure 2: Structure for detecting trafc conficts of all types.

Figure 3 :
Figure 3: Calculation method for the comprehensive assessment indicator.

Figure 4 :
Figure 4: Schematic diagram of physical state transition during the collision between two vehicles.

Figure 5 :
Figure 5: Schematic diagram of physical state transition during a single-vehicle collision.

Figure 7 :
Figure 7: Work zone model established in Vissim.

( 2 )
Detecting single-vehicle conficts and two or multivehicle conficts in the work zone based on risk avoidance behavior analysis and SSAM analysis.(3) Integrating the severity and possibility of accidents caused by conficts to obtain comprehensive risk indicators and the equivalent total number of trafc conficts.(4) Establishing a risk assessment standard based on the level of safety service.(5) Constructing a risk control decision-making method based on Poisson regression.(

Table 1 :
Single vehicle conficts and their key parameters (part).

Table 2 :
Two or multiple vehicle conficts and their key parameters (part).

Table 3 :
Risk value calculation for single vehicle conficts (part).

Table 4 :
Risk value calculation for two or multiple vehicle conficts (part).

Table 7 :
Experiment design and equivalent confict analysis results.

Table 8 :
14isson regression results.p<0.05 * * p < 0.01 z statistics in parentheses.14Journal of Advanced Transportation actual data collection for a certain work zone to further validate the feasibility, efectiveness, and superiority of the framework and method proposed in this paper in practice. *