Length Requirements for Urban Expressway Work Zones’ Warning and Transition Areas Based on Driving Safety and Comfort

Abstract

As aging urban expressways become more pronounced, maintenance and construction work on these roadways is increasingly necessary. Some lanes may need to be closed during maintenance and construction, decreasing driving safety and comfort in the work zone. This situation often leads to traffic congestion and a higher risk of traffic accidents. Notably, 80% of work zone traffic accidents occur in the warning and upstream transition areas (or simply warning and transition areas). Therefore, it is crucial to appropriately determine the lengths of these areas to enhance both safety and comfort for drivers. In this study, we examined three different warning lengths (1800 m, 2000 m, and 2200 m) and three transition lengths (120 m, 140 m, and 160 m) using the entropy weighting method to create nine simulation scenarios on a two-way, six-lane urban expressway. We selected various metrics for driving safety and comfort, including drivers’ eye movement, electroencephalogram, and driving behavior indicators. A total of 45 participants (mean age = 23.9 years, standard deviation = 1.8) were recruited for the driving simulation experiment, and each participant completed all 9 simulation scenarios. After eliminating 5 invalid datasets, we obtained valid data from 40 participants. We employed a combination of the analytic network process and entropy weighting method to calculate the comprehensive weights of the eight evaluation indicators. Additionally, we introduced the fuzzy theory, utilizing a trapezoidal membership function to evaluate the membership matrix values of the indicators and the comprehensive evaluation grade eigenvalues. The ranking of the experimental scenarios was determined using these eigenvalues. The results indicated that more extended warning lengths correlated with increased safety and comfort. Specifically, the best driver safety and comfort levels were observed in Scenario I, which featured a 2200 m warning length × 160 m transition length. However, the difference in safety and comfort across different transition lengths diminished as the warning length increased. Therefore, when road space is limited, a thoughtful combination of reasonable lengths can still provide high driving safety and comfort.

1. Introduction

Urban expressways are integral to urban roads. They are provided with a central separation belt and have four or more lanes, which are either fully or partially controlled by grade separation, allowing vehicles to drive at a higher speed. They serve short commutes within the city, enhancing urban traffic efficiency and shortening travel time. Their flow is concentrated during peak hours. After decades of construction, China’s expressway system has undergone gradual improvement, and urban expressway maintenance and construction work will become the norm. The driving environment becomes more complicated, and the road traffic capacity is reduced because some lanes need to be closed during maintenance and construction of urban expressways, which decreases driving safety and comfort in work zones. The work zone becomes the bottleneck section on urban expressways and the accident-prone point of urban roads.

According to the Standardization Administration of the People’s Republic of China [1], the work zone is divided into six elements: advance warning area, upstream transition area, buffer area, work area, downstream transition area, and termination area, as shown in Figure 1 [1]. For simplicity, the first two areas are referred to here as warning and transition areas. The warning area allows drivers to adjust their speeds and change lanes by looking for gaps, making its length crucial for traffic diversion. The transition area helps prevent sudden changes in traffic flow during lane changes, facilitating a smooth transition of vehicles; therefore, its length is significant. Relevant research shows that traffic accidents in the warning and transition areas account for 80% of the total traffic accidents in the work zone [2,3]. Therefore, reasonably setting the warning and transition lengths is essential for improving driving safety and comfort in the work zone.

1.1. Studies on Warning and Transition Areas

There are few studies available in the literature on urban expressway work zones. However, studies on highway work zones have achieved specific results that can serve as a reference for urban expressways. Li et al. [4] recommended considering the total length required for lane changing without deliberate deceleration as the length of the warning area based on the vehicle lane change model. Ge et al. [5] proposed a method for determining the length of the highway warning area using the VISSIM simulation model and the surrogate safety assessment model for agents (SSAM). Liu et al. [6] established a model for setting the length of the transition area under various construction schemes based on the theory of acceptable gap. Based on the lane change conflict model, Liu [7] evaluated the length of warning and transition areas from dynamic and driver psychological perspectives. Wang et al. [8] selected three spatial safety evaluation indicators to analyze the spatial safety of the transition area under different operating speeds and mixed traffic flow conditions. Zhang et al. [9] proposed optimal work zone lengths for different traffic volumes using optimization. Singh et al. [10] set up a work area model using a cellular automaton model and improved traffic congestion by optimizing the work area length. Han et al. [11] analyzed the variation in road network evaluation parameters for the work zone length under different traffic volumes and organization modes. They proposed reasonable recommended values for the work zone length.

There are quite a few norms and standards for work zones in China, such as the Road Traffic Signs and Markings Part 4: Work Zones [1], and Safe Operation Regulations published by the Ministry of Transport [12], as well as the management measures for construction sections formulated by the various provinces and the subordinate local departments. Although the standards stipulate the layout of the work zone, only the minimum value is determined for the length of various construction sections, which makes it difficult for highway management and construction units to accurately determine the length during implementation. At the same time, a series of safety accidents have exposed problems: the existing norms have low standards, broader provisions, and a mismatch with the current traffic situation.

In summary, existing research primarily focuses on highway work zones, and the length of the warning and transition areas is based solely on road traffic characteristics, without considering drivers’ driving factors or the influence of different combinations of section lengths. Although urban expressways are similar to highways, they differ in traffic functions and characteristics, making it challenging to apply existing results directly. Therefore, analyzing the length of warning and transition areas on urban expressways is necessary, considering traffic and driving characteristics.

1.2. Studies on Driving Safety and Comfort

There are few comprehensive evaluations of driving safety and comfort because most studies analyze an indicator or specific indicators regarding the evaluation mode of construction areas. Omkar et al. [13] analyzed traffic safety in terms of the conflict probability at selected roads with and without work zone sections along the same road. Tang et al. [14] evaluated vehicle safety in highway work zones by analyzing collision risk using vehicle trajectory data collected from work zones. Garber et al. [15] analyzed traffic accident data in work zones and summarized the leading causes and characteristics of accidents. Zhang et al. [16,17] used various analytical models to evaluate the impact of different traffic flow conditions on risk in highway work zones. Ishak et al. [18], Umar et al. [19], and Li et al. [20,21] analyzed various construction sites. They found that the frequency of traffic accidents was higher in the warning and transition areas and the work area. Li et al. [22] examined the effect of exceptional weather, different driving experiences, and gender on traffic safety in work zones. Heaslip et al. [23] used algorithms to assess the impact of drivers’ familiarity with the road and driving behavior on traffic safety. Several studies have found that drivers’ behavior and cognitive abilities also impact road capacity [24,25]. Khan [26] divided subjective comfort into 5 levels and studied the subjective comfort of 108 drivers when the length of the transition area was decreased by 10% to 40% (10% interval). The study did not consider the safety indicators of vehicle operation in the work zone.

With the advancement of science and technology, measurement methods for assessing people’s perceptions and emotions are becoming increasingly accurate, providing strong support for establishing the relationship between the characteristics of the road complex environment and the mental and physiological indicators of the driver [27,28]. Based on visual indicators, Belyusar et al. [29] found that dynamic billboards significantly affect drivers’ driving, increasing driving risk. Costa et al. [30] found a linear relationship between visual recognition distance and duration, which can affect the acquisition of key information on the sign. Li et al. [31] concluded that tunnel brightness has a significant impact on mental and physiological indicators of the driver, including gaze duration, saccade amplitude, and blink frequency, etc. Zhou et al. [32] conducted vehicle tests in four kinds of tunnels, using EEG power spectral density, pupil diameter, and other indicators to investigate the driver’s visual load and tension. Wang [33] investigated the relationship between traffic safety and biomass index at the entrance of highway tunnels based on pupil diameter and EEG power ratio (α + θ)/β, as well as other indicators.

A driver’s vehicle handling behavior can also reflect driving safety and comfort to some extent. This behavior includes controlling vehicle speed, controlling the driving route, and handling emergencies. It indicates that the driver’s driving safety and comfort are poor if they frequently change lanes, speed, or brake suddenly [34,35]. Therefore, vehicle handling behavior can also characterize driving safety and comfort during the driving process in the work zone.

1.3. Study Objectives

This paper focuses on enhancing driving safety and comfort in work zones along urban expressways. The study creates a simulated environment of a work zone on a two-way, six-lane urban expressway to carry out indoor driving simulation experiments. The objectives of the study are threefold: (1) to evaluate the performance of various combinations of warning and transition area lengths, (2) to implement comprehensive performance indicators, including visual, cognitive, and driving behavior effects, on driver’s safety and comfort, including drivers’ eye movements and electroencephalogram (EEG), and (3) to develop a systematic evaluation methodology that combines the analytic hierarchy process (AHP) and the entropy weighting method (EWM) to address the limitations of both subjective and objective weighting approaches. The results of this research can serve as a reference for determining the appropriate lengths of work zones on urban expressways in the future. This paper breaks through the limitations of traditional single evaluation indices and develops a multidimensional evaluation system encompassing physiology, psychology, and behavior that addresses the comprehensive evaluation of driver safety and comfort. This study focuses on the impact of the combination of the warning area and transition area lengths, rather than the length of a single section. It examines the interaction of different combinations on driving safety and comfort, providing a basis for refining the design of the work area.

2. Experimental Design

2.1. Experimental Scenarios

Since there is no stipulation for the length of each section of the work zone on the urban expressway in the existing norms, this paper selected the length of each section according to the standards of the first-class highway [12]. The length of the warning area was gradually increased by 200 m based on meeting the minimum length of 1600 m in the norms, and a total of six lengths was established. The length of the transition area gradually increased by 20 m, meeting the minimum length of 100 m in the norms, and a total of 6 lengths was set up. Additionally, an 80 m buffer area, a 1000 m work area, a 30 m downstream transition area, and a 30 m termination area were established according to the regulations [12].

Peng et al. [36], Ren [37], and Yang [38] used VISSIM simulation. They selected the delay rate, conflict rate, and average travel time as evaluation indicators to determine the optimal ranges of the lengths of warning and transition areas for different types of work zones. Therefore, in this study, six VISSIM simulation models of the warning length and six VISSIM simulation models of the transition length were constructed using the control variable method. The VISSIM simulation experiment output the results of five evaluation indicators: average queue length, number of traffic conflicts, average travel delay, average travel time, and parking times. According to the results of the five evaluation indicators, the lengths of the warning and transition areas were comprehensively evaluated, respectively, based on the entropy weight method [36,37,38,39,40]. The results showed that the warning area’s lengths were ranked from best to worst, as follows: 1800 m > 2000 m > 2200 m > 2400 m > 1600 m > 2600 m. Therefore, the warning lengths of 1800 m, 2000 m, and 2200 m were selected as the control variables for driving simulation experiments. The transition area’s lengths were ranked from best to worst, as follows: 140 m > 120 m > 160 m > 100 m > 180 m > 200 m. Therefore, the transition lengths of 120 m, 140 m, and 160 m were selected as the control variables for driving simulation experiments.

Apart from the warning and transition lengths, all other elements in the experimental scenarios remained consistent. The experimental road was a two-way, six-lane urban expressway with a total length of approximately 8 km and a single-lane width of 3.5 m. The design speed in the basic section was 120 km/h, and the speed limit was 60 km/h. The work zone was arranged following SAC [1], and the work conditions were set for closed outer lane operation (Figure 2).

Considering the influence of traffic flow, a steady flow of 2000 pcu/h was selected as the traffic volume loaded in this experiment. Therefore, this study used the warning and transition lengths as the control variables. The experimental design involved three warning lengths (1800 m, 2000 m, and 2200 m) and three transition lengths (120 m, 140 m, and 160 m). The differences in the nine experimental scenarios are shown in

Table 1. The experimental scenarios.

Scenario	Warning Length	Transition Length
A	1800 m	120 m
B	1800 m	140 m
C	1800 m	160 m
D	2000 m	120 m
E	2000 m	140 m
F	2000 m	160 m
J	2200 m	120 m
H	2200 m	140 m
I	2200 m	160 m

.

2.2. Participants

The minimum sample size required was determined based on the expected variance, target confidence, and error magnitude, as follows:

$$N={\displaystyle \frac{Z^2{\sigma }^2}{E^2}},$$

(1)

where N = experimental sample size, Z = standard normal distribution statistic, σ = standard deviation (SD; 0.25 to 0.5), and E is the maximum error. This experiment used the following variables: Z = 1.96 (95% confidence level), σ = 0.3, and E = 10% [41]. The minimum sample was calculated to be 35.

In this experiment, 45 participants (25 males and 20 females) with valid driving licenses were recruited to ensure a sufficient sample size after excluding invalid data. Their ages ranged from 20 to 28 years (M = 23.9, SD = 1.8). The driving experience of participants ranged from 1 to 6 years (M = 1.6, SD = 1.9). The participants were self-reported to be in good physical condition with normal visual function and normal or corrected visual acuity of 1.0 or above. All participants had driving experience on urban expressways and were unfamiliar with the driving simulation scenarios of this experiment. All participants were aware of this experiment’s procedures and had signed an informed consent form. All participants were guaranteed to have a regular diet and rest for 24 h before the experiment, without drinking alcohol or taking drugs or strenuous exercise.

2.3. Performance Indicators

Relevant studies have shown that the driver’s mental and physiological characteristics and driving behavior during the driving process are the most critical factors affecting driving safety. The data on the driver’s eye movement, EEG, and driving behavior could reflect the driver’s safety and comfort levels during the driving process. This experiment collected three types of indicators: visual effect (eye movement) indicators (U₁), cognitive effect (EEG (U₂)) indicators, and driving effect (manipulation behavior) indicators (U₃).

2.3.1. Visual Effect Indicators

Pupil area (U₁₁, pixel): The pupil area reflected the driver’s comfort level with vision. The pupils will dilate when the visual effect is poor, and the pupils will constrict when the visual effect is good. Meanwhile, the pupil area will also change during the operation, and the magnitude of the change is closely related to the difficulty of the operation [42].

Saccade frequency (U₁₂, times/s): Saccade is a quick movement of the eyes from one point to another with a viewpoint dwell time of less than 120 ms. Saccade frequency is calculated as the number of saccades occurring per unit time. The saccade frequency is lower when the drivers are familiar with the driving environment, or the road environment is simple. The saccade frequency is higher when the driver is unfamiliar with the driving environment or the road environment is complex, which indicates the challenge of searching for information [43].

2.3.2. Cognitive Effect Indicators

Absolute power of α wave (U₂₁, μV²): This wave will increase when the drivers are relaxed and their attention is focused. At this time, the driver’s cognitive load is moderate, allowing them to deal with road emergencies more calmly, which is more conducive to driving safety [44].

Absolute power of β wave (U₂₂, μV²): The beta waves generally appear when people are stimulated by external events, with intense alertness and high cognitive load. The continuous increase in the absolute power of β waves often indicates that people gradually change from alertness to tension, causing drivers’ fatigue and reducing driving safety and comfort [45].

θ/β (U₂₃): Since a single EEG energy value can only represent the drivers’ physiological changes in a single frequency band, scholars often combine the EEG energy ratios (α/β and θ/β, etc.) to objectively and comprehensively analyze the driver’s driving status [44,45,46]. To avoid repeated characterization of EEG signals, we selected low-frequency θ waves and high-frequency β waves with low correlation for the ratio calculation. When the theta wave increases, the brain is relaxed, indicating that a person’s mental stress has been relieved. Therefore, the θ/β value can characterize the change in cognitive level. The larger the θ/β value is, the more relaxed the driver is. At this time, the driver’s cognitive load is moderate, allowing them to handle road emergencies more calmly, which is more conducive to driving safety [44,47,48].

2.3.3. Driving Effect Indicators

Longitudinal acceleration (U₃₁, m/s²): It refers to the change in speed per unit time during the braking process of the vehicle, which can intuitively reflect the operational stability of the drivers during speed changes. Rapid acceleration or deceleration behavior is considered relatively unsafe driving behavior [49]. The greater the fluctuation in acceleration is, the greater the risk of driving.

Instantaneous speed of entering the work area (U₃₂, km/h): This refers to the vehicle speed in the normal direction when it enters the work zone. To some extent, it reflects the rationality of the warning and transition areas’ lengths and the speed control effect of the signs and markings [50].

Lane change duration (U₃₃, s): It refers to the difference between the time the vehicle travels to the centerline of the target lane and the time when the wheels first press the lane line. The duration of a lane change can reflect its difficulty to some extent [51].

2.4. Equipment

This study collected data on eye movement, EEG, and driving behavior during simulation driving, using corresponding data acquisition systems. The experiment was conducted indoors, controlling the effects of climate, light, and noise. Three types of equipment were used in this study.

2.4.1. Driving Simulator

The driving simulation experiment was conducted using the DSR-1000TS2.0 automobile driving simulation system (Kunming University of Science and Technology Science & Technology Industrial Management Company, Kunming, China), which included the cockpit, 120° ring-view display system, and the independent console (Figure 3). The cockpit included individual car seats and vehicle-mounted devices (steering wheel, wing mirrors, and pedals). The 120° ring-view display system could display driving scenarios for the participants. The independent console could record and export the experiment’s driving behavior and vehicle dynamics parameters, such as speed, longitudinal acceleration, and lane change duration. The system featured an auxiliary sound system that realizes the sound of the experimental environment and the vehicle.

2.4.2. Eye Tracker and D-Lab Software

The Dikablis eye-tracking system (Ergoneers GmbH, Geretsried, Germany) was used to track and measure the drivers’ eye movement features during the simulation experiment, working with the D-Lab analysis software (version 3.55, Ergoneers GmbH, Geretsried, Germany). The device could track the drivers’ sights, and the opening and closing states of their eyes in real time, recording the changes in eye movement parameters, such as pupil area, gaze duration, and saccade frequency. The eye-tracking device collected samples at a frequency of 60 Hz with an accuracy range of 0.1° to 0.3°.

The D-Lab software uses the Dikablis eye tracker to track and measure the driver’s eye movement characteristics. The D-Lab platform creates a time segment based on the scenario video playback that needs to be analyzed, called a “task segment”, and selects the eye movement data items required in the “task segment”.

2.4.3. 32-Channel NE Wireless EEG System

The 32-channel NE wireless EEG (Neuroscan, Charlotte, NC, USA) collected the drivers’ EEG data during the simulation experiment. The equipment had 32 experimental channels, each with a bandwidth of 0 to 250 Hz. The sampling rate was 500 SPS, the resolution was 24 bits, and the noise was less than 1 uVRMS (0~250 Hz). It achieved EEG data transmission through wireless technology and obtained EEG signals, such as α, β, and θ waves.

2.5. Procedure

The participants were first asked to sign an experimental consent form and complete a personal information questionnaire, which included name, gender, age, number of years of driving, and mental state on that day. Before the formal experiment, the participants read the experimental instructions and understood the experimental tasks. Then, the participants travelled for a few minutes on a non-experimental scenario to familiarize themselves with the operation of the driving simulation cabin. After the participant was familiar with the operation, the experimenter assisted them in wearing EEG caps and Dikablis Glasses and calibrated them. If the participants felt uncomfortable, they could leave the experiment at any time.

The experimenter created the experiment number information in the D-Lab and started the formal experiment. To avoid learning effects, the experimenter randomly presented the experimental scenarios. According to the experimental requirements, the participants entered the section at a speed of 100 km/h. The participants changed their driving behavior and controlled their speed according to the sign prompts. The experimenters needed to confirm whether the participants’ EEG signals and eye movement images were normal during the experiment. After each scenario, the participants had a 5 min break before completing the nine-scenario experiments.

The above experimental process was repeated until all 45 participants had completed the experiment. At the end of the experiment, each participant was paid 50 yen as a reward. If a participant had an accident in the experimental scenario, such as crashing into another vehicle or going off the road, they were asked to stop and rerun the experiment.

2.6. Data Pre-Processing

The raw eye movement data were imported into D-Lab software. Then, the eye movement indicator data, such as pupil area and saccade frequency, were extracted based on the determined area of interest. Finally, the driving behavior data were output from the vehicle dynamics module, and the collected EEG data were exported to MATLAB software (https://ww2.mathworks.cn/products/matlab.html) for pre-analysis.

3. Methodology

The fuzzy method is a highly effective multi-factor tool for comprehensively evaluating things affected by multiple factors. It quantifies some factors that are not easily measurable and systematically reflects the strengths and weaknesses of the evaluation object from various perspectives and levels, transforming the problems into a quantitative evaluation. Calculating the weights is the key step of the fuzzy comprehensive evaluation. A combination of the analytic network process and entropy weighting method (ANP-EWM) was used to calculate comprehensive weights to overcome the shortcomings of subjective and objective weighting methods. This method not only refers to experience for weighting but also avoids excessive interference from subjective factors and overcomes the lack of objectivity in the weight determination of traditional fuzzy evaluation methods.

3.1. Combined Subjective and Objective Weights

The ANP and EWM were combined to determine the combined subjective and objective weights. The procedure involved the following steps:

Step 1: Determine the subjective weights using ANP

The ANP is a commonly used subjective weighting method in which experts measure the relative importance between various evaluation indicators and score them on a 1–9 scale. Then, the experimenter obtained judgment matrices and conducted consistency tests. Finally, the experimenter obtained the subjective weights ($w_{cj}$) of the indicators and the subjective weight vector Wc = [$w_{c1}$, $w_{c2}$, …, $w_{cn}$] [52].

Step 2: Determine the objective weights using EWM [53,54]

Using the entropy weight method to determine the driver’s response evaluation index system’s weight can objectively and realistically reflect the evaluation effect. It is necessary to normalize the original data to eliminate inconsistencies in units and dimensions between indicator data. Therefore, the min–max normalization method was used to normalize the data and obtain the normalized decision matrix B, as follows:

$$B_{m\times n}=\left(\begin{array}{cccc}{b}_{11}& b_{12}& \dots & b_{1n}\\ b_{21}& b_{22}& \dots & b_{2n}\\ \dots & \dots & \dots & \dots \\ b_{m1}& b_{m2}& \dots & b_{mn}\end{array}\right)=b_{ij}\left(0\le i\le m,0\le j\le n\right),$$

(2)

where m = the number of evaluation objects, n = the number of evaluation indicators, and $b_{ij}$ = the normalized value of the j-th evaluation indicator for the i-th evaluation object.

According to the normalized decision matrix, the entropy value of the j-th indicator in the evaluation index system is given by:

$$e_j=-{\displaystyle \frac{1}{\ln \left(m\right)}}{\displaystyle \sum _{i=1}^m{f}_{ij}\ln \left(f_{ij}\right)},$$

(3)

$$f_{ij}={\displaystyle \frac{A+b_{ij}}{{\displaystyle \sum _{i=1}^m\left(A+b_{ij}\right)}}},$$

(4)

where A is the translation amount (0.00001). It is necessary to add a translation amount to the value of 0 after standardization to obtain a small positive value, ensuring the completeness and validity of the data, and making the logarithm in Equation (3) meaningful.

The objective weights for indicators are given by:

$$w_{pj}={\displaystyle \frac{1-e_j}{n-{\displaystyle \sum _{j=1}^n{e}_j}}}.$$

(5)

Step 3: Combination weights method [55]

The subjective and objective weights were combined and assigned as follows:

$$w_j={\displaystyle \frac{w_{cj}·w_{pj}}{{\displaystyle \sum _{j=1}^n(}{w}_{cj}·w_{pj})}},$$

(6)

where $w_j$ = the combined weight for the j-th indicator, $w_{cj}$ = the subjective weight for the j-th indicator, and $w_{pj}$ = the objective weight for the j-th indicator.

3.2. Ranking Using Fuzzy Evaluation Method [56,57]

The specific steps of the improved fuzzy comprehensive evaluation method based on the combined weights were as follows:

(a) Determine the factor set U and evaluation set V: The factor set is a set of elements composed of various factors affecting the evaluation object, represented by $U=(u_1,u_2,\dots ,u_m$). The evaluation set is a set composed of various results that the evaluator may make on the evaluation object, represented by $V=(v_1,v_2,\dots ,v_n$). Generally, the evaluation set was divided into five levels, and each was regarded as a fuzzy subset.

(b) Determine the membership degree: The membership degree of each factor was determined using a trapezoidal membership degree function, divided into three segments: partial-small, intermediate, and partial-large (Figure 4). For the partial-small segment (a to b), the membership degree increased from 0 at a to 1 at b. For the intermediate segment (b to c), the membership degree was 1. For the partial-large segment (c to d), the membership degree decreased from 1 at c to 0 at d. These segments represent the range where the indicator value was considered small, intermediate, and large.

(c) Construct the membership matrix R: The evaluation indicators were processed and normalized to the [0, 1] interval using the min–max normalization method to eliminate the inconsistency of units and dimensions between indicators. Then, the membership degrees of the factor set were determined at each evaluation level, and the membership degree matrices, R_mxn, were constructed as follows:

$$R_{m\times n}=\left(\begin{array}{ccc}{r}_{11}& \dots & r_{1n}\\ ⋮& \ddots & ⋮\\ r_{m1}& \dots & r_{mn}\end{array}\right)=r_{ij}(0\le i\le m,0\le j\le n),$$

(7)

where $r_{ij}$ is the membership degree of the i-th element in the factor set U at the j-th level in the evaluation set V. The ANP-EWM method was applied to determine the weight vector $W=(w_1,w_2,\dots ,w_m$).

(d) Obtain the fuzzy comprehensive evaluation result vector S: The weight vector W and the membership matrix R were synthesized to obtain each evaluation object’s fuzzy comprehensive evaluation vector S, as follows:

$$S=W\circ R_{m\times n}=\left(w_1,w_2,\dots ,w_m\right)\left(\begin{array}{ccc}{r}_{11}& \dots & r_{1n}\\ ⋮& \ddots & ⋮\\ r_{m1}& \dots & r_{mn}\end{array}\right)=\left(s_1,s_2,\dots ,s_n\right),$$

(8)

where $s_i$ is the membership degree of the evaluation object at the j-th level in the evaluation set V.

(e) Analyze the evaluation results: The weighted average method was used to process the membership degree to calculate the comprehensive evaluation grade eigenvalue G, as follows [56]:

$$G={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^{m}j\times s_i^k}}{{\displaystyle {\sum }_{i=1}^m{s}_i^k}}}$$

(9)

where k is the undetermined coefficient, taken equal to 2 in this study [57].

Unlike the ineffective maximum membership degree of the traditional fuzzy comprehensive evaluation, the presented (improved) comprehensive evaluation index is more efficient and more intuitively reflects the specific results of the evaluation object.

4. Analysis and Results

4.1. Statistical Analysis

The participants’ eye movement, EEG, and driving behavior in each scenario were obtained through driving simulation experiments. After excluding invalid data (such as Dikablis Glasses not calibrated, missing data, and incomplete driving tasks), the data of 40 subjects were retained. A two-way repeated measures analysis of variance (ANOVA) was used to analyze the influence of three warning lengths (1800 m, 2000 m, and 2200 m) and three transition lengths (120 m, 140 m, and 160 m) and their interactions on the participants’ visual effects, cognitive effects, and driving effects at a significance level (α) of 0.05. Before the ANOVA was performed, the normality of the data was checked using the Shapiro–Wilk method, which indicated that the data in each group passed the normal distribution test and had no outliers [53,54,55,56,57,58].

In this experiment, the IBM SPSS 27.0 Statistics software was used for the two-way repeated measures ANOVA. If the data did not meet Mauchly’s sphericity test (p < 0.05), they were corrected using the Greenhouse–Geisser method. The average and test results of participants’ visual, cognitive, and driving behavior effects are shown in

Table 2. Results of two-way repeated measures ANOVA of participants’ visual effects.

Evaluation Indicator	Work Zone Element	Length	Mean	SE	95% CI of the Difference	F	Sig.	Partial η2
Pupil area	Warning length	1800 m	733.363	28.089	(676.547, 790.178)	39.050	<0.001	0.500
		2000 m	590.179	22.716	(544.233, 636.126)
		2200 m	540.127	21.933	(495.764, 584.490)
	Transition length	120 m	679.476	27.022	(624.820, 734.133)	13.601	<0.001	0.259
		140 m	621.293	24.094	(572.558, 670.028)
		160 m	562.900	21.443	(519.527, 606.273)
	Warning length*Transition length					1.850	0.136	0.045
Saccade frequency	Warning length	1800 m	1.029	0.052	(0.923, 1.135)	23.722	<0.001	0.378
		2000 m	0.763	0.036	(0.690, 0.836)
		2200 m	0.708	0.037	(0.633, 0.783)
	Transition length	120 m	1.091	0.047	(0.996, 1.186)	40.813	<0.001	0.511
		140 m	0.784	0.039	(0.705, 0.864)
		160 m	0.625	0.044	(0.536, 0.715)
	Warning length*Transition length					0.928	0.429	0.023

,

Table 3. Results of two-way repeated measures ANOVA of participants’ cognitive effects.

Evaluation Indicator	Work Zone Element	Length	Mean	SE	95% CI of the Difference	F	Sig.	Partial η2
Absolute power of α	Warning length	1800 m	718.732	45.998	(625.693, 811.771)	83.756	<0.001	0.682
		2000 m	514.580	30.440	(453.008, 576.152)
		2200 m	385.020	24.433	(335.599, 434.441)
	Transition length	120 m	586.494	37.016	(511.622, 661.366)	13.034	<0.001	0.250
		140 m	544.993	33.278	(477.683, 612.303)
		160 m	486.846	29.503	(427.171, 546.520)
	Warning length*Transition length					15.863	<0.001	0.289
Absolute power of β	Warning length	1800 m	362.532	12.936	(336.366, 388.697)	49.972	<0.001	0.562
		2000 m	457.846	19.254	(418.901, 496.791)
		2200 m	606.741	21.826	(562.594, 650.888)
	UT length	120 m	495.299	19.626	(455.601, 534.997)	2.149	0.123	0.052
		140 m	447.988	16.894	(413.816, 482.160)
		160 m	483.831	17.448	(448.539, 519.124)
	Warning length*Transition length					6.104	<0.001	0.135
θ/β	Warning length	1800 m	1.216	0.017	(1.181, 1.251)	281.991	<0.001	0.879
		2000 m	0.856	0.013	(0.830, 0.882)
		2200 m	0.748	0.010	(0.727, 0.769)
	Transition length	120 m	1.053	0.013	(1.027, 1.079)	99.684	<0.001	0.719
		140 m	0.923	0.011	(0.900, 0.946)
		160 m	0.844	0.008	(0.828, 0.861)
	Warning length*Transition length					33.057	<0.001	0.459

and

Table 4. Results of two-way repeated measures ANOVA of participants’ driving effects.

Evaluation Indicators	Work Zone Element	Length	Mean	SE	95% CI of the Difference	F	Sig.	Partial η2
Longitudinal acceleration	Warning length	1800 m	−2.124	0.021	(−2.167, −2.080)	1162.637	<0.001	0.968
		2000 m	−1.381	0.019	(−1.420, −1.342)
		2200 m	−0.833	0.017	(−0.867, −0.799)
	Transition length	120 m	−1.760	0.020	(−1.801, −1.719)	302.734	<0.001	0.886
		140 m	−1.471	0.019	(−1.510, −1.433)
		160 m	−1.106	0.018	(−1.144, −1.069)
	Warning length*Transition length					27.145	<0.001	0.410
Instantaneous speed of entering the work area	Warning length	1800 m	47.025	0.346	(46.326, 47.725)	110.153	<0.001	0.739
		2000 m	43.620	0.325	(42.963, 44.277)
		2200 m	40.806	0.233	(40.334, 41.278)
	Transition length	120 m	46.189	0.326	(45.529, 46.850)	63.379	<0.001	0.619
		140 m	44.050	0.320	(43.402, 44.697)
		160 m	41.212	0.302	(40.601, 41.823)
	Warning length*Transition length					2.673	0.034	0.064
Lane change duration	Warning length	1800 m	9.587	0.199	(9.184, 9.990)	20.195	<0.001	0.341
		2000 m	8.653	0.149	(8.353, 8.954)
		2200 m	8.013	0.153	(7.703, 8.323)
	Transition length	120 m	9.547	0.192	(9.158, 9.936)	20.086	<0.001	0.340
		140 m	8.827	0.153	(8.518, 9.135)
		160 m	7.880	0.181	(7.515, 8.245)
	Warning length*Transition length					0.056	0.994	0.001

. When the main effect existed, post hoc multiple comparisons were performed. When the interactive effect existed, simple effects analyses were performed.

4.2. Visual Effect Results

4.2.1. Pupil Area (U₁₁, Pixel)

The pupil areas of participants were collected. The distribution of participants’ pupil areas in different experimental scenarios is shown in Figure 5a. As the warning length increased, the participants’ pupil areas gradually decreased and showed a convergence trend. As the transition length increased, the participants’ pupil areas gradually decreased and showed a convergence trend. This showed that adequate warning and transition lengths can improve drivers’ visual comfort.

Table 2 shows that the main effects of warning and transition lengths on pupil areas were statistically significant. Still, the interaction between warning and transition lengths in pupil areas was not statistically significant. Therefore, post hoc multiple comparisons were performed for within-group effects where significant differences existed, as shown in

Table 5. Post hoc multiple comparisons tests of participants’ pupil areas (pixel).

Warning or Transition Length	Warning or Transition Length	Mean Difference (I–J)	SE	Sig.	95% CI of the Difference
Warning length (I)	Warning length (J)
1800 m	2000 m	143.184 *	23.453	<0.001	(84.511, 201.856)
	2200 m	193.236 *	26.709	<0.001	(126.420, 260.052)
2000 m	2200 m	50.052 *	16.791	0.015	(8.047, 92.058)
Transition length (I)	Transition length (J)
120 m	140 m	58.183 *	15.337	0.002	(19.816, 96.550)
	160 m	116.577 *	26.286	<0.001	(50.818, 182.335)
140 m	160 m	58.393	23.931	0.058	(−1.473, 118.259)

* The significance level of the mean value difference was 0.05.

.

Table 5 shows that the pupil areas between different warning lengths all showed significant differences, and the pupil area was most prominent when the warning length was 1800 m, indicating that the driver’s visual comfort was the worst. When the transition length was 120 m, the pupil areas differed significantly from the other transition lengths. The pupil areas showed no significant difference between the 140 m and 160 m transition lengths, indicating that participants’ visual comfort was equivalent in these two transition lengths.

4.2.2. Saccade Frequency Results

The saccade frequencies of participants were collected. The distribution of participants’ saccade frequencies in different experimental scenarios is shown in Figure 5b. As the warning length increased, the participants’ saccade frequency decreased, and the changing trend of different transition lengths was similar. As the transition length increased, the participants’ saccade frequency decreased, and the changing trend of different warning lengths was similar. When the transition length was 120 m, the participants showed frequent saccade behavior, and the saccade frequency was highest when the warning length was 1800 m.

As shown in Table 2, the main effects of both warning and transition lengths on saccade frequency were statistically significant. Still, the interaction between warning and transition lengths on saccade frequency was not statistically significant. Therefore, post hoc multiple comparisons were performed for within-group effects where significant differences existed, as shown in

Table 6. Post hoc multiple comparisons tests of participants’ saccade frequency (times/s).

Warning or Transition Length	Warning or Transition Length	Mean Difference (I–J)	SE	Sig.	95% CI for Difference
Warning length (I)	Warning length (J)
1800 m	2000 m	0.266 *	0.057	<0.001	(0.122, 0.410)
	2200 m	0.321 *	0.052	<0.001	(0.190, 0.452)
2000 m	2200 m	0.055	0.038	0.459	(−0.040, 0.150)
Transition length (I)	Transition length (J)
120 m	140 m	0.306 *	0.049	<0.001	(0.183, 0.429)
	160 m	0.465 *	0.053	<0.001	(0.332, 0.598)
140 m	160 m	0.159 *	0.055	0.018	(0.023, 0.295)

* The significance level of the mean value difference was 0.05.

.

Table 6 shows that the saccade frequency was the highest when the warning length was 1800 m, indicating that drivers had a high search demand for the surrounding road environment and poorer visual stability. The saccade frequency showed no significant difference between 2000 m and 2200 m warning lengths. The saccade frequencies between different transition lengths all showed significant differences, and the saccade frequency was most prominent when the transition length was 120 m, indicating that the driver’s visual stability was the worst.

4.3. Cognitive Effect Results

4.3.1. Absolute Power of α Wave

The absolute power of the α wave of participants was measured. The distribution of this indicator across different experimental scenarios is shown in Figure 6a. As the warning length increased, this indicator decreased gradually. This indicated that the drivers’ cognitive load and driving pressure increased gradually as the warning length increased, which was not conducive to driving safety. However, the difference in the absolute power of the α wave under different transition lengths gradually decreased as the warning length increased.

As shown in Table 3, the main effects of the warning and transition lengths and the interaction effect on the absolute power of the α wave were statistically significant. Therefore, simple effects comparisons were performed for the warning and transition lengths separately.

The pairwise comparison revealed that when the warning length was 2000 m, there was no significant difference in the absolute power of the α wave between the transition lengths of 120 m and 140 m (M(120–140) = −38.947, SE = 31.974, p = 0.692). In contrast, the pairwise comparisons showed significant differences between other transition lengths. When the transition length was 160 m, there was no significant difference in the absolute power of the α wave between the warning lengths of 2000 m and 2200 m (M(2000–2200) = 16.397, SE = 9.642, p = 0.291), while the pairwise comparisons showed significant differences between other warning lengths. When the warning length was 1800 m, the absolute power of the driver’s α wave was the highest, so the drivers’ cognitive load was the lowest.

4.3.2. Absolute Power of β Wave

The absolute power of the β wave of participants was collected. The distribution of participants’ absolute power of the β waves in different experimental scenarios is shown in Figure 6b. As the warning length increased, the participants’ absolute power of the β wave showed an increasing trend. This showed that the drivers’ cognitive load, driving tension, and pressure gradually increased as the warning length increased, which is not conducive to driving safety. When the warning length was 2200 m, the transition length was 120 m, and the participants’ absolute power value of the β wave was the highest.

As shown in Table 3, the main effects of the warning and transition lengths and the interaction effect on the absolute power of the β wave were statistically significant. Therefore, simple effects comparisons were performed for the warning and transition lengths separately.

The pairwise comparison revealed that when the warning lengths were 1800 m and 2200 m, there were significant differences in the absolute power of the β wave between the transition lengths of 120 m and 160 m (M(120–160) = −85.444, SE = 28.048, p = 0.012; M(120–160) = 174.706, SE = 59.965, p = 0.018). When the warning length was 2000 m, there were significant differences in the absolute power of the β wave between the transition lengths of 140 m and 160 m (M(140–160) = −114.461, SE = 41.052, p = 0.024). When the transition length was 140 m, there was no significant difference in the absolute power of the β wave between the warning lengths of 1800 m and 2000 m (M(1800–2000) = −37.154, SE = 25.897, p = 0.478). When the transition length was 160 m, there was no significant difference in the absolute power of the β wave between the lengths of 2000 m and 2200 m (M(2000–2200) = −17.738, SE = 50.765, p = 1.000), while the pairwise comparisons showed significant differences between other warning lengths.

4.3.3. θ/β Ratio

The θ/β values of participants were calculated. The distribution of the participants’ θ/β values across different experimental scenarios is shown in Figure 6c. As the warning length increased, the participants’ θ/β values gradually decreased. This showed that the drivers’ cognitive load and driving tension gradually increased as the warning length increased, which is not conducive to driving safety. However, the difference in the θ/β values under different transition lengths gradually decreased as the warning length increased. When the warning length was 2200 m, the θ/β values under different transition lengths were roughly the same, indicating that the drivers’ cognitive load and tension were comparable.

As shown in Table 3, the main effects of the warning and transition lengths and the interaction effect on the θ/β values were statistically significant. Therefore, simple effects comparisons were performed for the warning and transition lengths separately.

The pairwise comparison revealed that when the warning length was 2200 m, there were no significant differences in the θ/β values between the transition length of 120 m and the other transition lengths (M(120–140) = 0.035, SE = 0.018, p = 0.160; M(120–160) = −0.042, SE = 0.022, p = 0.186), while the pairwise comparisons showed significant differences between other transition lengths. When the transition length was 160 m, there was no significant difference in θ/β values between the warning lengths of 2000 m and 2200 m (M(2000–2200) = −0.034, SE = 0.029, p = 0.737), while the pairwise comparisons showed significant differences between other warning lengths. When the warning length was 1800 m, the drivers’ θ/β values were the highest, so the drivers’ cognitive load was the lowest.

4.4. Driving Effect Results

4.4.1. Longitudinal Acceleration

The longitudinal acceleration of participants was collected. The distribution of participants’ longitudinal acceleration in different experimental scenarios is shown in Figure 7a. As the warning length increased, the participants’ longitudinal acceleration showed an increasing trend and gradually converged. As the transition length increased, the participants’ longitudinal acceleration showed an increasing trend. When the warning length was 1800 m, the absolute value of the participants’ longitudinal acceleration was larger, indicating that the drivers had a large operating load and were eager to brake. It may be due to the shorter warning length, which shortened the deceleration distance.

As shown in Table 4, the main effects of the warning and transition lengths and the interaction effect on the longitudinal acceleration were statistically significant. Therefore, simple effects comparisons were performed for the warning and transition lengths separately.

The pairwise comparison revealed that when the warning length was 1800 m, there were no significant differences in the longitudinal acceleration between the transition lengths of 120 m and 140 m (M(120–140) = −0.072, SE = 0.05, p = 0.464). In contrast, the pairwise comparisons showed significant differences between other transition lengths. For any given transition length, the differences in the longitudinal acceleration among different warning lengths were significant. When the warning length was 2200 m, the absolute value of the drivers’ longitudinal acceleration was the smallest, indicating that the drivers slowed down gently.

4.4.2. Instantaneous Speed of Entering the Work Area

The participants’ instantaneous speed of entering the work area was collected. The distribution of participants’ instantaneous speed of entering the work area in different experimental scenarios is shown in Figure 7b. In the experiment, the speed limit of the work area was 60 km/h. Before entering the work area, all participants took deceleration measures, with an average speed between 40 km/h and 50 km/h, which met the speed limit requirements of the work zone. As the warning length increased, the participants’ instantaneous speed of entering the work area gradually decreased. Also, as the transition length increased, the participants’ instantaneous speed of entering the work area gradually decreased. It indicated that the longer the warning and transition areas, the longer the deceleration area could be provided for participants, which could reduce the deceleration pressure of the drivers.

As shown in Table 4, the main effects of the warning and transition lengths and the interaction effect on the instantaneous speed of entering the work area were statistically significant. Therefore, simple effects comparisons were performed for the warning and transition lengths one by one.

The pairwise comparison revealed that when the warning length was 1800 m, there were no significant differences in the instantaneous speed between the transition lengths of 120 m and 140 m (M(120–140) = 1.212, SE = 0.788, p = 0.396). When the warning length was 2200 m, there were no significant differences in the instantaneous speed between the transition lengths of 140 m and 160 m (M(120–140) = 1.872, SE = 0.762, p = 0.056). In contrast, the pairwise comparisons showed significant differences between other transition lengths. For any given transition length, the differences in the instantaneous speed among different warning lengths were significant.

4.4.3. Lane Change Duration

The lane change duration of participants was collected. The distribution of the participants’ lane change durations in different experimental scenarios is shown in Figure 7c. As the warning length increased, the participants’ lane change duration gradually decreased, and the change trend of different transition lengths was similar. As the transition length increased, the participants’ lane change duration steadily reduced, and the changing trend of different warning lengths was similar. The shorter warning or transition length may increase the cross-section traffic volume, and then the space for drivers to change lanes is compressed, making lane changing more difficult.

As shown in Table 4, the main effects of warning and transition lengths on lane change duration were statistically significant. Still, the interaction between the warning and transition lengths on lane change duration was not statistically significant. Therefore, post hoc multiple comparisons were performed for within-group effects where significant differences existed, as shown in

Table 7. Post hoc multiple comparisons tests of participants’ lane change duration (s).

Warning or Transition Length	Warning or Transition Length		Mean Difference (I–J)	SE	Sig.	95% CI for the Difference
Warning length (I)		Warning length (J)
1800 m		2000 m	0.933 *	0.234	<0.001	(0.349, 1.518)
		2200 m	1.573 *	0.264	<0.001	(0.913, 2.233)
2000 m		2200 m	0.640 *	0.249	0.042	(0.018, 1.262)
Transition length (I)		Transition length (J)
120 m		140 m	0.720 *	0.278	0.040	(0.026, 1.414)
		160 m	1.667 *	0.260	<0.001	(1.016, 2.318)
140 m		160 m	0.947 *	0.253	0.002	(0.314, 1.579)

* The significance level of the mean value difference was 0.05.

.

As shown in Table 7, all lane change durations between different warning lengths exhibited significant differences, with the lane change duration being most pronounced when the warning length was 1800 m, indicating that lane changes were the most difficult. Driving safety was the worst for the drivers. Similarly, all lane change durations between different transition lengths showed significant differences, with the lane change duration being most pronounced when the transition length was 120 m, indicating that lane changes were the most difficult. Driving safety was also the worst for the drivers.

4.5. Results of Weights and Ranking Quality

4.5.1. Comprehensive Weights of Indicators

Seven experts in relevant fields were invited to assess the relative importance of various evaluation indicators on a scale of 1 to 9. The subjective weight of each evaluation indicator was calculated using the Super Decision software (version 8.5.6) using the ANP method. Using the entropy weight method, the objective weight was calculated according to Equations (2)–(5) based on the mean values of various evaluation indicators obtained from the driving simulation experiment. The comprehensive weights of the indicators were calculated from Equation (6), as shown in

Table 8. Comprehensive weights of safety and comfort evaluation indicator system (U).

Primary Indicators	Subjective Weights	Objective Weights	Comprehensive Weights	Secondary Indicators	Subjective Weights	Objective Weights	Comprehensive Weights
Visual effects (U1)	0.317	0.161	0.161	Pupil area (U11)	0.634	0.535	0.666
				Saccade frequency (U12)	0.366	0.465	0.334
Cognitive effects (U2)	0.203	0.489	0.312	Absolute power of α wave (U21)	0.304	0.348	0.318
				Absolute power of β wave (U22)	0.349	0.167	0.175
				θ/β value (U23)	0.347	0.484	0.506
Driving effects (U3)	0.480	0.350	0.527	Longitudinal acceleration (U31)	0.583	0.369	0.621
				Instantaneous speed of entering the work area (U32)	0.204	0.336	0.198
				Lane change duration (U33)	0.213	0.295	0.181

.

4.5.2. Grading Criteria for Indicators

To eliminate the inconsistency of units and dimensions between indicator data, the original data were forward normalized so that the data range fell within the interval of [0, 1]. Then, the K-means clustering algorithm was used to classify the evaluation indicators through SPSS software [58]. The grading criteria of the evaluation indicators obtained through the K-means clustering algorithm are shown in

Table 9. The grading criteria of evaluation indicators.

Evaluation Indicators	The Grading Criteria of Safety and Comfort
	Lowest	Lower	Medium	Higher	Highest
Pupil area (U11)	0–0.29	0.29–0.45	0.45–0.59	0.59–0.74	0.74–1
Saccade frequency (U12)	0–0.27	0.27–0.53	0.53–0.69	0.69–0.83	0.83–1
Absolute power of α wave (U21)	0–0.18	0.18–0.35	0.35–0.55	0.55–0.77	0.77–1
Absolute power of β wave (U22)	0–0.30	0.30–0.50	0.50–0.66	0.66–0.81	0.81–1
θ/β value (U23)	0–0.20	0.20–0.32	0.32–0.47	0.47–0.66	0.66–1
Longitudinal acceleration (U31)	0–0.23	0.23–0.39	0.39–0.56	0.56–0.71	0.71–1
Instantaneous speed of entering the work area (U32)	0–0.24	0.24–0.42	0.42–0.56	0.56–0.71	0.71–1
Lane change duration (U33)	0–0.26	0.26–0.44	0.44–0.59	0.59–0.74	0.74–1

.

4.5.3. Fuzzy Comprehensive Evaluation Results

Based on the experimental data and the established evaluation criteria, the membership degrees of the factor set were determined at each evaluation level, and the membership degree matrices were constructed. Combined with the comprehensive weight vector obtained by the ANP-EWM method, the fuzzy comprehensive evaluation result vector was calculated according to Equation (8). Then, the comprehensive evaluation grade eigenvalue G was calculated by Equation (9). Finally, the ranking of the experimental scenario was determined according to the eigenvalue G. The results are shown in

Table 10. Results of fuzzy comprehensive evaluation and ranking of experimental scenarios.

Scenario	Membership Degree of Safety and Comfort Level					Safety and Comfort Level	G	Comprehensive Sort
	Lowest	Lower	Medium	Higher	Highest
A	0.688	0.000	0.000	0.000	0.312	Lowest	1.682	8
B	0.643	0.045	0.021	0.267	0.024	Lowest	1.449	9
C	0.000	0.466	0.484	0.050	0.000	Medium	2.527	6
D	0.276	0.583	0.141	0.000	0.000	Lower	1.871	7
E	0.151	0.113	0.423	0.306	0.007	Medium	3.115	4
F	0.250	0.062	0.000	0.487	0.201	Higher	3.550	2
G	0.312	0.127	0.078	0.457	0.026	Higher	2.999	5
H	0.306	0.006	0.007	0.348	0.332	Higher	3.478	3
I	0.271	0.041	0.000	0.000	0.688	Highest	4.457	1

.

Table 10 presents the comprehensive ranking of safety and comfort for the nine experimental scenarios as follows: I > F > H > E > G > C > D > A > B. The higher the comprehensive evaluation grade eigenvalue G was, the higher the driver’s safety and comfort in this scenario was. The comprehensive evaluation grade eigenvalue of Scenario I (2200 m warning length × 160 m transition length) was the highest, indicating that the drivers’ safety and comfort were the best. The comprehensive evaluation grade eigenvalue of Scenario I (1800 m warning length × 140 m transition length) was the lowest, indicating that the drivers’ safety and comfort were the worst.

According to the evaluation results in Table 10, it is concluded that the comprehensive evaluation grade eigenvalue increased as the warning length increased when the transition length was fixed, indicating that the driver’s safety and comfort were improved. When the warning lengths were 2000 m and 2200 m, the comprehensive evaluation grade eigenvalue increased as the transition length increased, indicating improved driver safety and comfort. Therefore, in practical engineering applications, when the road space conditions permit, it is recommended to use a combination of 2200 m warning length × 160 m transition length.

The safety and comfort level of Scenario A (1800 m warning length × 120 m transition length) was higher than Scenario B (1800 m warning length × 140 m transition length), the safety and comfort level of Scenario C (1800 m warning length × 160 m transition length) was higher than Scenario D (2000 m warning length × 120 m transition length), the safety and comfort level of Scenario E (2000 m warning length × 140 m transition length) was higher than Scenario G (2200 m warning length × 120 m transition length), and the safety and comfort level of Scenario F (2000 m warning length × 160 m transition length) was higher than Scenario H (2200 m warning length × 140 m transition length). Therefore, a reasonable combination of lengths can also achieve high driving safety and comfort when the road space is limited.

5. Discussion

5.1. Results

Although the standards stipulate the work zone layout, only the minimum value is recommended for the length of various work zones [1]. Although the optimal length of work zones has been explored by some studies [59,60], few scholars have considered the influence of different length combinations of warning/transition zones. This study evaluated nine warning–transition length combinations on urban expressways using a comprehensive safety and comfort assessment system. The system evaluated the drivers’ visual, cognitive, and driving effects. This system is more comprehensive and systematic compared with the evaluation system that only uses road traffic characteristic indicators [36,61], eye movement indicators [62], or driving behavior indicators [34,35].

The results showed several key findings. First, the optimal combination: Scenario I (2200 m warning × 160 m transition) achieved the highest safety/comfort score (G = 4.457), with significantly improved visual stability (e.g., 61% lower saccade frequency vs. shortest lengths) and smoother driving behavior (longitudinal acceleration closest to zero). Second, shorter length combinations could also achieve higher driving safety and comfort (e.g., 1800 m × 120 m, ranked #8, but 1800 m × 140 m, ranked #9), proving that balanced combinations can mitigate spatial constraints. Finally, longer warnings increased the EEG-measured cognitive load (β wave power rose 67% from 1800 m to 2200 m), highlighting a trade-off between driver preparedness and stress.

Longer warnings consistently enhanced safety (e.g., 2000 m and 2200 m warnings reduced lane change duration by 9–16% vs. 1800 m). The results showed that the main effects of the warning/transition lengths on the participants’ eye movement and driving behavior indicators were statistically significant (Table 2 and Table 4). As the warning length increased, the participants’ visual effects and driving effects improved. Additionally, as the transition length increased, the visual effects and driving effects on the participants improved. This may be because longer work zones can provide drivers with a better visual search environment and longer deceleration and lane change distances [38,42]. The analysis results of EEG behavior data showed that the main effects of the warning and transition lengths, and the interaction effect on participants’ EEG indicators, were statistically significant (Table 3). As the warning length increased, the participants’ cognitive effects decreased. This may be because the longer warning area increased the drivers’ cognitive load, resulting in a sense of tension and pressure [44,45,46,47].

The proposed ANP-EWM method, which combines the weights of evaluation indicators, avoids the limitations of the single-weight method [57,63]. This method objectively reflects the important relationship between each evaluation indicator. It enhances the scientific validity of the results, providing a new approach for evaluating driving safety and comfort. The fuzzy theory was introduced using a combined weighting method, and the K-means clustering algorithm was employed to classify the evaluation indicators. The theory addresses the uncertainty of the evaluation system of different length combinations of warning and transition lengths. The trapezoidal membership degree function was selected to calculate the membership matrix values of evaluation indicators at all levels. The fuzzy comprehensive evaluation grade eigenvalues were calculated and combined with the comprehensive weight vector. The results showed that the drivers’ safety and comfort in Scenario I (2200 m warning length × 160 m transition length) was the best, consistent with the findings of others [38,64,65].

5.2. Practical Implications

The findings of this study offer actionable insights for optimizing work zone design on urban expressways. Below are key recommendations for practitioners:

(a)

Optimal length combination: Where the road space permits, adopt the 2200 m warning length × 160 m transition length (Scenario I) to maximize driver safety and comfort. This combination yielded the highest evaluation score and is recommended for new designs or major retrofits.

(b)

Space-constrained conditions: When right-of-way is limited, balanced combinations can mitigate spatial constraints. For example, use 1800 m warning × 120 m transition (Scenario A, ranked #8) instead of 1800 m × 140 m (Scenario B, worst-ranked). In addition, for the 2000 m warning, pair with the 160 m transition (Scenario F, ranked #2) rather than longer options (e.g., 2200 m × 140 m, ranked #3). This option would be vital in dense urban areas, leveraging dynamic message signs (DMSs) to create longer warning zones, thereby reducing physical space needs.

(c)

Interaction effects: The results indicated that there were interaction effects involving warning and transition zones (Table 2 and Table 4). Therefore, one should refrain from optimizing warning/transition lengths in isolation. For example, shorter length combinations (e.g., 2000 m × 140 m, ranked #4) became safer than longer combinations (e.g., 2200 m × 120 m, ranked #5), thereby reducing the need for excessive land acquisition.

(d)

Human-centered evaluation framework: The study’s physiology–psychology–behavior framework can be integrated into safety audits for high-risk work zones. In the future, designers can pilot this approach in intelligent work zones with IoT sensors to monitor real-time driver stress and adjust traffic controls dynamically.

(e)

Regulatory and design updates: The design standards, for example [1,12], should be updated to replace the minimum length requirements with context-dependent optimal ranges. Additionally, region-specific guidelines can be developed using the proposed framework to account for traffic volume, driver demographics, and alignment complexity.

6. Conclusions

This study examined the effects of various combinations of warning and transition lengths in urban expressway work zones on driving safety and comfort. Nine simulation scenarios were created using three warning lengths and three transition lengths. The research focused on the interaction between these lengths and their impact on driving performance by utilizing a multidimensional indicator system for evaluation. The analytic network process and entropy weighting method (ANP-EWM) were employed to determine indicator weights, while fuzzy comprehensive evaluation ranked each scenario. Key conclusions from the study include:

• Optimal design implementation should prioritize a combination of 2200 m warning length × 160 m transition length (Scenario I), which yielded the highest safety and comfort scores. This configuration significantly improved driver visual stability and smoother driving behavior. Where right-of-way is limited, practitioners should emphasize extending warning areas over transition zones, adopting balanced combinations like 2000 m × 160 m (Scenario F) or 1800 m × 120 m (Scenario A) to maintain safety without excessive land acquisition.
• Critical findings revealed that longer warning lengths (≥2000 m) consistently enhanced safety performance, reducing the lane change duration by 9–16% and minimizing abrupt deceleration compared to shorter warnings. Notably, EEG data confirmed a cognitive load trade-off: while longer warnings improved operational safety, they increased driver cognitive tension (67% rise in β wave power from 1800 m to 2200 m).
• This study demonstrated several notable contributions to traffic safety and work zone design. First, unlike traditional single evaluation index methods, this study developed a multidimensional evaluation system, incorporating eye movement data, EEG signals, and driving behavior indicators to assess driver safety and comfort. This holistic approach allowed for a more nuanced understanding of how combined warning and transition lengths impact drivers, capturing physiological and behavioral responses. By integrating these diverse data sources, the study provided a robust assessment of driver performance in work zones, which is more comprehensive than studies that rely solely on traffic characteristics or isolated driver behavior metrics.
• The methodology used in this study was rigorous. It employed advanced analytical techniques, including the ANP-EWM, to determine the weights of evaluation indicators. This combination of subjective and objective weighting methods reduced the influence of bias and enhanced the objectivity of the results. Furthermore, the introduction of fuzzy theory and the use of a trapezoidal membership function addressed the inherent uncertainty in evaluating safety and comfort levels, providing a more flexible and realistic assessment. Additionally, the study benefited from a relatively large sample size of 45 participants, which enhanced the reliability and generalizability of the findings. The study’s rigorous methodology and comprehensive evaluation framework make it a valuable contribution to the field, offering practical insights for improving work zone safety and comfort on urban expressways.
• For practical implementation, transportation agencies should revise standards, such as China’s [12], to recommend evidence-based length combinations. Dynamic message signs (DMSs) can simulate longer warning zones in space-constrained urban corridors. At the same time, the study’s multidimensional framework should be adopted for safety audits in high-risk work zones.
• This study has several limitations that present opportunities for future research. First, the study only examined the types of closed work zones on the outer lanes of urban expressways. Therefore, the results may not apply to other types of work zones. Although the nine length combinations of warning and transition areas were practical, this study did nt test longer combinations. Hence, Scenario I is optimal only within the tested range. Second, young drivers (aged 20–28 years) were selected as subjects due to their limited driving experience, which necessitates more operating space. Therefore, future work may focus on drivers of varying ages, including older drivers. Lastly, future research may address nighttime conditions, higher traffic volumes (>2000 pcu/h), and dynamic length adjustments employing IoT-enabled adaptive signage to maximize safety in variable traffic environments.