Safety evaluation method for intersection of suburban arterial road in cold region and improvement scheme evaluation method

Abstract

The cold region suburb trunk road intersection safety evaluation method and improvement scheme evaluation method belong to the technical field of intersection safety evaluation, and solve the technical problem that the existing technology cannot accurately judge the safety risk when facing complex environmental conditions such as cold regions. A plurality of detection points are arranged in a preset area of a target plane intersection, and a plurality of physiological indexes of a plurality of drivers passing through each detection point are obtained respectively; a comprehensive physiological load evaluation model is constructed to calculate the safety score of the target plane intersection; the safety score is used for risk classification; a safety improvement scheme is proposed; a simulation experiment is constructed, and the corrected data is used as the input of the comprehensive physiological load evaluation model to calculate the safety score; the safety score is used for risk classification to complete the safety evaluation of the safety improvement scheme. The present application is used for efficient and accurate safety evaluation of cold region suburb trunk road intersections.

Description

Technical Field

[0001] This invention relates to the field of intersection safety evaluation technology, specifically to safety evaluation methods and improvement scheme evaluation methods for intersections of urban and suburban trunk roads in cold regions. Background Technology

[0002] Suburban arterial roads are vital transportation corridors connecting cities and suburbs, serving both regional traffic flow and access to land along their routes. With the continuous advancement of urbanization in my country, traffic flow on suburban arterial roads is constantly increasing. As key nodes in the road network, the safety of intersections is becoming increasingly prominent, especially in cold regions with low temperatures and snowfall, which further exacerbates driving risks at intersections of suburban arterial roads in cold areas.

[0003] Traditional methods for assessing intersection safety primarily include accident statistics and traffic conflict analysis, both of which rely on historical accident data or observations of potential conflicts. However, for intersections on suburban arterial roads, historical accident data is often insufficient; and traffic conflict observations reflect the consequences of driving behavior rather than its causes. Furthermore, accident statistics are affected by reporting rates, and the identification and severity assessment of traffic conflicts depend on the observer's subjective experience and standards, which may lead to significant biases.

[0004] In the prior art, Chinese patent document CN118629195A discloses "a traffic risk analysis method for unsignalized intersections." This method constructs a static evaluation index system based on accident causation data and calculates the static safety level of the intersection using a fuzzy evaluation method. It also classifies flow conflict types based on the flow conflict characteristics of unsignalized intersections and designs corresponding SSMs to construct a dynamic evaluation index system. The method obtains the conflict probability level through dynamic evaluation indicators and then calculates the dynamic safety level of the intersection by combining the conflict severity level. Finally, it constructs a comprehensive risk assessment model based on clustering methods to determine the comprehensive traffic risk level of the intersection. However, this technical solution only evaluates from the perspective of road facilities and vehicle conflicts, without considering the impact of weather and environmental factors on the driver's physiological and psychological state. When the road surface adhesion coefficient decreases significantly due to ice and snow, and visibility is significantly reduced due to snowfall, the driver's tension and physiological load increase, but a detectable immediate conflict may not necessarily occur between vehicles. In this case, the dynamic index system of this solution is unlikely to trigger a high-risk warning, resulting in significant deficiencies in its applicability and sensitivity in complex road sections with cold environments.

[0005] In summary, existing technologies have the technical problem of being unable to accurately assess safety risks when facing complex road sections with cold environments. Summary of the Invention

[0006] This invention solves the technical problem that existing technologies cannot accurately assess safety risks when facing complex road sections with cold environments.

[0007] The safety evaluation method for intersections of urban and suburban trunk roads in cold regions described in this invention includes the following steps: Step 1: Set up multiple detection points in a preset area of ​​the target intersection and obtain multiple physiological indicators of multiple drivers when they pass through each detection point. Step 2: Based on the multiple physiological indicators obtained in Step 1, construct a comprehensive physiological load evaluation model and calculate the safety score of the target plane intersection; Step 3: Based on the safety score, classify the risks and complete the safety assessment of the intersection.

[0008] Furthermore, in one embodiment of the present invention, the multiple physiological indicators in step 1 include heart rate growth rate, root mean square value of electromyography, and rate of change of skin resistance increment.

[0009] Furthermore, in one embodiment of the present invention, step 2, which involves constructing a comprehensive physiological load evaluation model and calculating the safety score of the target intersection, specifically includes: ; in, This is the heart rate correction factor for cold, low-temperature environments. For the correction factor of skin electrical stress in suburban environments, This is a correction factor for muscle tension in traffic flow density at intersections. For the first Heart rate growth rate per sample For the first Increment of skin resistance in each sample For the first Root mean square value of electromyography for each sample The time it takes for a driver to pass through an intersection. The average travel time for vehicles at intersections of urban and suburban trunk roads. For the angle of conflicting traffic flow at the intersection, This is the attenuation term for travel time. For conflict risk correction items, and These represent the maximum and minimum values ​​of the driver's overall physiological load at multiple testing points. and These are the maximum and minimum values ​​of the growth rate of the center rate of all samples, respectively. and These represent the maximum and minimum values ​​of skin resistance increments across all samples. and These represent the maximum and minimum root mean square values ​​of electromyography (EMG) across all samples.

[0010] Furthermore, in one embodiment of the present invention, the risk classification in step 3 specifically includes: when Time-sharing is considered high-risk; when The risk level is medium when the time frame is determined; The time frame is considered low-risk.

[0011] The method for evaluating safety improvement schemes at intersections of urban and suburban trunk roads in cold regions, as described in this invention, is based on any of the aforementioned safety evaluation methods and includes the following steps: Step A: Propose safety improvement plans based on the safety assessment results; Step B: Based on the safety improvement plan, construct a simulation experiment to obtain the simulation data to be corrected; Step C: Correct the simulation data to obtain the corrected data; Step D: Use the corrected data as input to the comprehensive physiological load assessment model to calculate the safety score; Step E: Based on the safety score, classify the risks and complete the safety evaluation of the safety improvement plan.

[0012] Furthermore, in one embodiment of the present invention, the correction of the simulation data in step C specifically includes: A data correction model is constructed, using physiological index data obtained from simulation experiments as input and corresponding physiological index data from real vehicle experiments as output. The data correction model is trained by inputting the simulation data to be corrected into the trained data correction model to obtain the corrected data.

[0013] Furthermore, in one embodiment of the present invention, the data correction model uses multiple simulated physiological indicators as input features to predict the corresponding measured physiological indicators.

[0014] Furthermore, in one embodiment of the present invention, the data correction model employs the random forest algorithm.

[0015] This invention solves the technical problem of existing technologies being unable to accurately assess safety risks when facing complex road sections in cold regions and other environmental conditions. Specific beneficial effects of this invention include: 1. This invention proposes a safety evaluation method and an improvement scheme evaluation method for intersections of main roads in cold suburban areas, such as... Figure 1 As shown, by collecting multimodal physiological signals such as electrocardiogram, electrodermal conductance, and electromyography of drivers in real time, the system can directly quantify the driver's tension, cognitive load, and operational pressure when passing through intersections. This fundamentally overcomes the technical problem that existing technologies rely solely on road infrastructure defects or observable conflicts between vehicles for safety evaluation, and cannot accurately determine safety risks when facing complex road sections with cold environments. 2. This invention proposes a safety evaluation method for intersections of urban and suburban trunk roads in cold regions. It uses the K-means clustering algorithm to classify the risk level of the safety score, and determines the classification boundary based on the inherent structure of the data itself, thus avoiding the classification bias caused by subjectively preset equal division thresholds in the existing technology. 3. This invention proposes an evaluation method for safety improvement schemes at intersections of urban and suburban trunk roads in cold regions. It constructs a data correction model based on multi-index joint regression, using three simulated physiological indicators as input features to predict the corresponding measured indicators. This fully utilizes the covariance characteristics between physiological indicators, enabling the effectiveness of safety improvement measures to be verified with high reliability through driving simulation when real vehicle experiments cannot be conducted. Attached Figure Description

[0016] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of the safety evaluation method for intersections of main roads in cold suburban areas as described in this invention; Figure 2 This is a conceptual diagram of the detection point setup described in Implementation Method 1; Figure 3 This is a schematic diagram of the convergence curve of the error as a function of the number of decision trees as described in Implementation Method 2; Figure 4 This is the heart rate growth rate correction comparison chart described in Implementation Method 2; Figure 5 This is a comparison chart of the skin resistance increment change rate correction described in Implementation Method 2; Figure 6 This is a root mean square comparison diagram of electromyography as described in Implementation Method 2. Detailed Implementation

[0017] Various embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. The embodiments described with reference to the drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0018] Implementation Method 1: Some existing suburban arterial roads suffer from inconsistencies between technical specifications and traffic flow characteristics, as well as inadequate traffic engineering facilities and management measures, resulting in significant driving safety hazards. Furthermore, regional climatic conditions further exacerbate these risks. Northeast my country is a typical cold region, where winter driving not only faces a significantly reduced coefficient of friction, but also suffers from ice and snow cover that severely impairs visibility. Coupled with the high proportion of large trucks on suburban arterial roads, driving difficulty is significantly increased. Under these multiple adverse conditions, drivers entering intersections must simultaneously deal with oncoming traffic and vehicles from side roads, leading to a concentrated surge in cognitive and decision-making pressure, easily exceeding physiological limits and significantly increasing safety risks.

[0019] Traditional methods for evaluating intersection safety often overlook the differences in the impact of seasonal conditions on drivers' physiological load, focusing only on external indicators such as road infrastructure defects or observable conflicts between vehicles. This results in significant limitations in applications to complex road sections with cold environments. When the road surface adhesion coefficient drops sharply due to ice and snow, and visibility is significantly reduced due to snowfall, drivers' stress and cognitive load increase dramatically. However, there may not be an immediate, detectable conflict between vehicles. In such cases, the existing evaluation system is unlikely to trigger a high-risk warning, leading to missed judgments.

[0020] To address the aforementioned technical problems, this implementation method proposes a safety evaluation method for intersections of urban and suburban trunk roads in cold regions, specifically including the following: Step 1: Set up multiple detection points in a preset area of ​​the target intersection and obtain multiple physiological indicators of multiple drivers when they pass through each detection point. Selecting driving time as a key indicator reveals the cumulative effect and fatigue evolution of a driver's physiological load over time. This has significant theoretical and engineering application value for understanding driving tolerance limits, developing reasonable work-rest schedules, and creating fatigue early warning systems. Heart rate load exhibits a non-linear increasing characteristic with a rapid initial increase followed by a slower increase after driving time. After two hours of continuous driving, the cardiovascular system may gradually enter a fatigue adaptation phase, but the load continues to accumulate. Furthermore, prolonged driving may have a cumulative activation effect on the sympathetic nervous system. The emotional stress and anxiety levels experienced by drivers after 3-4 hours of driving may further intensify, and fatigue and tension may have a cumulative effect. Simultaneously, the muscle exertion required to maintain vehicle control continues to increase without a clear trend of adaptation or relief. Therefore, heart rate shows a saturation-type increase with a rapid initial increase followed by a slow increase, skin electrical activity shows an acceleration-type increase with a slow initial increase followed by a rapid increase, while electromyography (EMG) accumulates at an approximately uniform rate.

[0021] Vehicle speed is the most direct quantitative reflection of the difficulty and demands of driving tasks. From a cognitive psychology perspective, changes in vehicle speed directly alter a driver's information processing rate and operational frequency. As vehicle speed increases, the amount of road environment information a driver needs to process per unit of time increases exponentially (i.e., the speed of visual scene changes accelerates), while corrective actions (such as fine-tuning the steering wheel) become more frequent. This high-intensity perception-decision-execution cycle directly leads to excessive consumption of cognitive resources in the central nervous system, resulting in physiological manifestations such as increased heart rate and elevated skin conductance. Conversely, at extremely low speeds, decreased alertness or "low-load stress" due to task monotony may also trigger specific physiological responses. Therefore, vehicle speed effectively distinguishes different levels of task demands and is a direct driver of changes in physiological load.

[0022] The increase in heart rate load per 5 km / h in the high-speed range (65-75 km / h) was significantly greater than that in the low-speed range (55-65 km / h). Furthermore, increased vehicle speed may further activate the sympathetic nervous system, and the driver's alertness and tension levels may increase non-linearly with speed. Simultaneously, the body increases the root mean square of electromyography (RMS) to cope with the dynamic load generated by higher speeds. Significance tests of influencing factors showed that the differences between driving speeds of 55 km / h and 70 km / h, and between 75 km / h, were significant across all three physiological indicators (p<0.01). However, the differences between adjacent speed ranges (e.g., 55 and 60 km / h) were mostly not significant (p>0.05), suggesting a threshold effect on physiological load, which is more pronounced in the high-speed range (≥70 km / h).

[0023] To investigate whether there is an interaction effect between driving duration and vehicle speed, a two-way repeated measures ANOVA was further employed. Two-way repeated measures ANOVA incorporates two within-group factors (driving duration and vehicle speed) on top of the single-factor analysis. The F-value of the interaction term is used to determine whether the combined effect of the two factors exceeds the sum of their independent effects. If the interaction effect is significant, it indicates that the effect of one factor is moderated by the level of the other factor; if it is not significant, the effects of the two factors on the dependent variable are independent. Prior to the analysis, the data were verified to satisfy the normality and sphericity assumptions. Under winter driving speed conditions of 55–75 km / h, driving duration is the dominant factor affecting the driver's physiological load, with an influence approximately 2–3 times that of vehicle speed. After two hours of continuous driving, all three physiological indicators showed a significant increase of 20%–26%, entering a stage of rapid fatigue accumulation. Although the effect of vehicle speed on physiological load is weaker than that of driving duration, it exhibits a characteristic of accelerated deterioration at high speeds, requiring extra attention in the range above 70 km / h. Therefore, this implementation method involves acquiring multiple physiological indicators when the driver's driving time is 3 to 4 hours. During this stage, the physiological load has reached its peak and the fatigue effect is fully manifested, which can realistically reflect the physiological stress characteristics under long-distance driving. Multiple physiological indicators are acquired at a driving speed of 70 to 75 km / h, which can sensitively capture the physiological responses caused by changes in vehicle speed. The combination of driving time and vehicle speed maximizes the representativeness of real-world driving scenarios and ensures the reliability of multiple physiological indicators.

[0024] After determining the experimental parameters, drivers sequentially passed through three types of intersections under specified conditions, and the BIOPAC physiological instrument recorded heart rate, skin conductance, and electromyography (EMG) in real time. All samples obtained were from drivers passing through different types of intersections at preset detection points in different seasons (summer and winter), and the different types of intersections included Y-shaped, cross-shaped, and T-shaped intersections. The detection point setup concept is as follows: Figure 2 As shown.

[0025] In this embodiment, multiple physiological indicators include heart rate growth rate, root mean square value of electromyography, and rate of change of skin resistance increment.

[0026] Compared to absolute heart rate, heart rate growth rate (HRG) exhibits stronger individual adaptability and task sensitivity. Absolute heart rate is continuously affected by background factors such as individual basal metabolic rate, emotional state, and ambient temperature, making it difficult to accurately reflect stress changes induced by specific driving tasks. HRG, using an individual's baseline as a reference, effectively filters out these slowly changing basal disturbances, more directly indicating the physiological stress response caused by the driving task itself, and offering stronger horizontal comparability. Therefore, this implementation method selects HRG as an electrocardiographic indicator for evaluating driver physiological workload.

[0027] Heart rate growth rate This is the percentage increase in heart rate relative to baseline; a positive value indicates an increase in heart rate, and a negative value indicates a decrease. The heart rate growth rate directly reflects the impact of changes in the intersection's driving environment on the driver's physiological and psychological state. As the first indicator for analyzing changes in a driver's physiological workload, it is shown in the following formula: ; in, The heart rate value measured at the current moment, in beats per minute (BPM). The baseline heart rate is measured in a quiet, relaxed state. It is usually the average of multiple measurements taken over a resting period and is used as a benchmark for comparison. The unit is beats / min.

[0028] The forearm flexor muscles are the core muscle group for drivers to control the steering wheel. Fine-tuning the steering, keeping the lane, and emergency avoidance all rely on the continuous or intermittent contraction of these muscles. It can stably and directly quantify muscle activity intensity, has good sensitivity to steering load and precision control during driving, and its calculation method is mature and noise-resistant, making it suitable for application in dynamic and complex real-world driving environments. Therefore, this embodiment selects the root mean square value of electromyography (RMS) of the forearm flexor muscle group as an electromyographic indicator for evaluating the physiological load of drivers.

[0029] The root mean square (RMS) value of electromyography (EMG) is a comprehensive index calculated by taking the root mean square of the amplitude of EMG signals over a period of time. It reflects the average activity intensity of the muscle during that period, and is expressed in units of 10. -2 mV, the calculation formula is as follows: ; in, For the first Electromyography signal amplitude at each sampling point (10 -2 mV). The total number of sampling points. The root mean square value of electromyography (10) -2 mV).

[0030] Changes in skin resistance are entirely governed by the sympathetic nervous system and are not subject to the driver's conscious control. This allows for the objective capture of unconscious emotional and physiological stress responses, avoiding biases from subjective reports. Compared to the absolute value of skin resistance, the rate of change in skin resistance increment uses an individual's baseline as a reference, eliminating the influence of differences in baseline skin conductivity between individuals and improving the effectiveness of cross-subject comparisons. Furthermore, the rate of change is more sensitive to rapid fluctuations caused by stress, enabling more precise location of moments of increased driver workload. Therefore, this implementation method selects the rate of change in skin resistance increment as an indicator for evaluating driver physiological workload in terms of skin electrical activity.

[0031] skin resistance increment rate This refers to the percentage change in skin resistance relative to baseline. Since skin resistance typically decreases upon stimulation, this indicator reflects the relative magnitude of the decrease in resistance; a larger change indicates a higher level of driver stress, as shown in the following formula: ; in, This is the baseline skin resistance value at rest, expressed in kΩ. The value of skin resistance measured during a stimulus or specific event is presented in kΩ.

[0032] In summary, the selected physiological load indicators are explained in Table 1.

[0033] Table 1 Explanation of Physiological Load Indicators

[0034] Step 2: Based on the multiple physiological indicators obtained in Step 1, construct a comprehensive physiological load evaluation model and calculate the safety score of the target plane intersection; Traditional methods are mostly used to classify driver status, driving style, or driver fatigue level. Their core object is the classification of human behavioral characteristics. They usually need to obtain a large number of behavioral parameters, such as the driver's speed, absolute value of acceleration, absolute value of impact, accelerator pedal opening, etc. PCA is used to compress high-dimensional data into a few principal components, which is used for subsequent clustering, noise reduction, and acceleration.

[0035] This implementation method does not classify individual driver states, but rather uses at-grade intersections on suburban trunk roads in cold regions as the evaluation object. It reflects the comprehensive impact of road environment, intersection geometry, and icy / snowy weather conditions on driving safety by analyzing changes in the driver's physiological load at different detection points. This implementation method only has three input indicators: heart rate growth rate, skin resistance increment rate, and root mean square value of electromyography. It has fewer dimensions and does not require high-dimensional compression. Directly applying conventional PCA methods in a low-dimensional context would face the following technical obstacles: The heart rate growth rate, skin resistance increment rate, and root mean square value of electromyography (EMG) are derived from the electrocardiogram (ECG), skin conductance (SDR), and electromyography (EMG) systems, respectively, and their dimensions, fluctuation ranges, and sensitivities differ significantly. Among these, SDR indices are extremely sensitive to changes in cold, snowy environments, while EMG indices are more easily affected by driving maneuvers. Direct cluster analysis could easily lead to a single index dominating the clustering results, distorting the risk classification.

[0036] To address the aforementioned issues, this implementation method first employs Z-score standardization (standard score) to unify different indicators into a standard space with a mean of 0 and a standard deviation of 1. Then, it uses PCA to extract the first principal component as a comprehensive load indicator, thereby effectively eliminating interference caused by dimensional differences and the coupling of multiple indicators.

[0037] To avoid bias in dimensionality reduction due to excessively large or small values ​​of multiple physiological indicators, Z-score standardization was applied to the three indicators (making the mean of each indicator 0 and the standard deviation 1) to eliminate the influence of dimensional differences on the dimensionality reduction results.

[0038] Prior to PCA analysis, suitability tests were performed on the three physiological indicators. The Shapiro-Wilk normality test showed that all indicators followed a normal distribution under all operating conditions (p>0.05). Pearson correlation analysis revealed significant positive correlations at the 0.01 level between heart rate increase rate and skin resistance increment rate (r = 0.751), skin resistance increment rate and root mean square value of electromyography (r = 0.733), and heart rate increase rate and root mean square value of electromyography (r = 0.586), indicating a moderate to strong correlation among the three indicators, providing a basis for principal component analysis. The KMO test statistic was 0.6911, greater than the acceptable threshold of 0.6; the Bartlett's test of sphericity showed a positive result. 2 =149.684, p<0.001, rejecting the null hypothesis that the correlation coefficient matrix is ​​an identity matrix, further confirming that the data are suitable for principal component analysis.

[0039] PCA (principal component analysis) was performed on the standardized data of multiple physiological indicators, and the eigenvalues ​​and variance contribution rates of the three principal components are shown in Table 2.

[0040] Table 2. Principal Component Eigenvalues ​​and Contribution Rates

[0041] As shown in Table 2, the eigenvalue of the first principal component F1 is 2.4075 (>1), with a variance contribution rate of 79.41%, explaining nearly 80% of the total variance. Based on the Kaiser criterion, only F1 is extracted as the comprehensive factor. The eigenvalues ​​of F2 and F3 are 0.4188 and 0.2053, respectively, both less than 1, and are therefore discarded. Therefore, in this implementation, the factor score of F1 is used as a proxy indicator of the driver's comprehensive physiological load.

[0042] Table 3 Factor Load Analysis Table

[0043] Table 3 shows that the loadings of the three physiological indicators on the first principal component F1 are: heart rate (HR) = 0.8784, skin conductance (SC) = 0.9381, and electromyography (EMG) = 0.8695. All three are above 0.86 and are positive, indicating that F1 comprehensively reflects the combined changes in the driver's sympathetic nerve activation level, emotional arousal level, and control muscle tension. Skin conductance (SC) has the largest loading on F1 (0.938), indicating that it contributes the most to the overall physiological load and is the most sensitive proxy indicator for overall load.

[0044] From the F1 score coefficients (factor weights): HR=0.5661, SC=0.6046, EMG=0.5604. The weights of the three indicators are almost evenly distributed (approximately 1 / 3), but the weight of skin conductance is slightly higher than the other two, which is consistent with the conclusion that skin conductance has the highest load on the F1. The overall load score can be expressed using the F1 score coefficients as follows: ; The F1 comprehensive load score is converted into a safety scoring standard formula of 0~100 points through a global linear mapping: ; in, To take into account physiological load, This is the standardized heart rate growth rate. This represents the standardized rate of change in skin resistance increment. The standardized root mean square value of electromyography (EMG) For the first The overall physiological load of each sample For safety rating, and These represent the maximum and minimum values ​​of the driver's overall physiological load at multiple testing points, ensuring global comparability of the scores. A lower safety score indicates a higher physiological load and greater driving risk at that testing point.

[0045] Furthermore, considering the specific effects of low temperatures in cold regions on physiological indicators such as heart rate, the impact of factors such as roadside disturbances and non-motorized vehicle traffic on emotional stress in suburban trunk roads, and how changes in traffic density at intersections alter the driver's control load, this implementation method, in constructing a comprehensive physiological load evaluation model, not only normalizes and weights the heart rate growth rate, skin resistance increment rate, and root mean square value of electromyography, but also introduces a cold-region low-temperature heart rate correction coefficient. To correct for the impact of temperature on heart rate, a skin conductance stress correction factor is introduced for suburban environments. To reflect the impact of complex suburban traffic environments on emotional load, a muscle tension correction coefficient based on intersection traffic density is introduced. This is to reflect the differences in handling load under different traffic flow conditions. Furthermore, considering that the longer a driver spends in the intersection area, the more significant the cumulative effect of physiological load and the lower the safety margin, a travel time attenuation term was added to the model. Considering that the smaller the angle of conflicting traffic flows at an intersection, the greater the driver's cognitive decision-making load and the higher the collision risk, a conflict risk correction term has been added. This reflects the geometric impact of the intersection angle on the safety score. Considering all the above factors, the final comprehensive physiological load assessment model is constructed as follows: ; in, A higher value indicates higher safety and lower physiological stress for the driver. The heart rate correction coefficient for cold-region low-temperature environments is determined by interpolation after establishing a heart rate baseline-temperature relationship curve by collecting the resting heart rate baseline of drivers at different ambient temperatures. The correction factor for skin conductance stress in suburban environments was determined through comparative experiments comparing the skin conductance stress levels of the same driver under the same driving task in suburban trunk roads and urban roads. The correction coefficient for muscle tension due to traffic flow density at intersections was determined through a comparative experiment of electromyographic load on drivers at the same intersection under different traffic flow density conditions. For the first Heart rate growth rate per sample For the first Increment of skin resistance in each sample For the first Root mean square value of electromyography for each sample The time it takes for a driver to pass through an intersection. The average travel time for vehicles at intersections of urban and suburban trunk roads. For the angle of conflicting traffic flow at the intersection, This is the attenuation term for travel time. For conflict risk correction items, and These represent the maximum and minimum values ​​of the driver's overall physiological load at multiple testing points. and These are the maximum and minimum values ​​of the growth rate of the center rate of all samples, respectively. and These represent the maximum and minimum values ​​of skin resistance increments across all samples. and These represent the maximum and minimum root mean square values ​​of electromyography (EMG) across all samples.

[0046] In this embodiment, PCA does not simply reduce dimensionality, but rather integrates three types of indicators with different physiological mechanisms—heart rate growth rate, skin resistance increment rate, and root mean square value of electromyography—into a unified comprehensive physiological load factor. This factor is used to eliminate dimensional differences and correlation interference between different physiological indicators, thereby improving the stability of safety evaluation results in complex traffic environments in cold regions.

[0047] Step 3: Based on the safety score, classify the risks and complete the safety assessment of the intersection.

[0048] Traditional methods often employ subjective, pre-defined equal-segmentation grading. However, these subjective grading boundaries are artificially set, lacking objective basis, ignoring the inherent distribution characteristics of the data, and easily leading to the separation of similar samples or the merging of dissimilar samples, resulting in biased grading results. Furthermore, due to the significant variations in the adhesion coefficient of icy and snowy road surfaces, different drivers exhibit large fluctuations in their physiological responses at intersections, leading to a significantly higher degree of dispersion between samples compared to ordinary road environments. Using traditional subjective threshold grading methods can easily result in overlapping boundaries between different risk levels.

[0049] To address this issue, this implementation introduces the K-means clustering algorithm, which uses the elbow rule and silhouette coefficient to determine the optimal number of clusters k=3, enabling the risk level boundary to be automatically generated based on the sample distribution, thereby improving the objectivity and stability of risk classification.

[0050] This implementation uses the K-means clustering algorithm to perform data-driven objective grading of the safety scores of all samples, ensuring that the grading boundaries are determined by the intrinsic structure of the data itself. The specific steps are as follows: (1) Determining the optimal number of clusters k: The elbow method was used to calculate the sum of squared errors within groups (SSE / Inertia) when k=2 to 7. The SSE decreased the most when k increased from 2 to 3. After k≥4, the curve tended to flatten out, forming a clear elbow. The optimal number of clusters k=3 was determined. At the same time, the silhouette score reached 0.5713 when k=3, which was higher than other k values, further verifying the rationality of the three-clustering scheme.

[0051] (2) K-means clustering execution: K-means clustering is performed on all one-dimensional data of safety scores with k=3 (the random seed is fixed at 42, and the initialization is repeated 10 times to obtain the optimal solution). After the algorithm converges, three cluster centers are obtained, which are arranged in ascending order as follows: 23.0 points, 48.1 points, and 77.0 points, corresponding to the typical centers of the three load levels of high risk, medium risk, and low risk.

[0052] (3) Calculation of hierarchical boundaries: Using the midpoint of adjacent cluster centers as the hierarchical threshold, two natural boundaries are obtained: Boundary 1 = (23.0+48.1) / 2 = 35.5 points; Boundary 2 = (48.1+77.0) / 2 = 62.5 points.

[0053] Therefore, the sample was divided into three risk levels: 0-35.5 was the high load / high risk zone, 35.5-62.5 was the medium load zone, and 62.5-100 was the low load / low risk zone.

[0054] In this implementation, K-means does not cluster driver behavior, but rather automatically generates road risk level boundaries based on comprehensive load scores from different intersections, seasons, and detection points. This allows the risk classification to reflect the true physiological load distribution characteristics of intersections in cold regions. Compared to traditional equal-interval grading methods, the cluster boundaries in this implementation are automatically generated based on the data's own distribution, thus more accurately identifying areas of abrupt physiological load changes in icy and snowy environments and improving the ability to identify high-risk intersections.

[0055] In this implementation, PCA and K-means are not used for driver feature identification, but rather for establishing a road risk assessment system. Driver physiological load is significantly higher in winter than in summer, especially with a more pronounced change in heart rate growth rate. Directly using single-season data to establish the evaluation boundary would lead to insufficient model generalization ability. Therefore, this implementation incorporates data from multiple intersection types, including winter and summer, Y-shaped, T-shaped, and cross-shaped intersections, during the clustering process. This allows the cluster centers to cover typical load characteristics under different environmental conditions, improving the model's applicability in complex traffic scenarios in cold regions. The physiological load varies significantly among different intersection types in winter, with Y-shaped intersections exhibiting the highest overall physiological load. Furthermore, the peak load location shows a stable clustering characteristic, indicating a clear correlation between driver physiological load and intersection geometry and the traffic environment.

[0056] Furthermore, in the PCA (Pressure Process Analysis) section, more physiological indicators of drivers, such as EEG, eye movement, and respiratory rate data, can be introduced to further enhance the comprehensive load model's ability to characterize driving cognitive states. In the clustering section, methods such as fuzzy C-means clustering (FCM) or Gaussian mixture model (GMM) can be combined to flexibly delineate the risk boundaries of intersections, thereby improving the identification accuracy of boundary transition areas. Additionally, real-time traffic flow parameters, road surface adhesion coefficients, and meteorological data can be combined to construct a dynamically updated intersection risk assessment model, enabling real-time safety warnings for roads in cold regions. Moreover, this implementation method currently employs offline data analysis, which can be further integrated with vehicle-to-everything (V2X) systems to achieve dynamic risk assessment and proactive safety intervention based on real-time physiological load. Therefore, this implementation method not only achieves an objective evaluation of safety risks at intersections on suburban trunk roads in cold regions but also provides a technical foundation for establishing a real-time, intelligent road safety management system in cold regions.

[0057] In summary, this implementation method employs a comprehensive physiological load scoring approach for drivers. By measuring physiological signals such as heart rate, skin conductance, and electromyography, combined with behavioral indicators like real-time vehicle speed data, it directly and objectively quantifies the psychological stress, tension, and workload experienced by drivers when navigating intersections. High load indicates unfriendly intersection design or traffic conditions, posing safety hazards. This approach can identify risk points before accidents actually occur, thus compensating for the shortcomings of traditional methods in predictability, objectivity, and root cause diagnosis.

[0058] Implementation Method 2: Evaluation Method for Safety Improvement Plans at Intersections of Suburban Trunk Roads in Cold Regions. This method is based on the safety evaluation method described in Implementation Method 1 and includes the following steps: Step A: Propose safety improvement plans based on the safety assessment results; Step B: Based on the safety improvement plan, construct a simulation experiment to obtain the simulation data to be corrected; Step C: Correct the simulation data to obtain the corrected data; Step D: Use the corrected data as input to the comprehensive physiological load assessment model to calculate the safety score; Step E: Based on the safety score, classify the risks and complete the safety evaluation of the safety improvement plan.

[0059] In order to provide safety recommendations for areas with high safety risks at intersections of urban and suburban trunk roads in cold regions, this implementation method is based on VTD (Vehicle Driving Simulation Platform) driving simulation. An optimized intersection driving scenario is built in a simulated driving environment, and the reliability of the safety recommendations is evaluated based on the driving experiment safety scores in the simulation experiment.

[0060] Based on the safety evaluation results of Implementation Method 1, this implementation method proposes safety recommendations. Subsequently, a corresponding simulation scenario is constructed according to the safety recommendations, and simulation data (≥500 sets) is collected through large-scale repeated experiments. The effectiveness and reliability of the safety recommendations are quantitatively verified using a controlled experiment method.

[0061] Based on the results of real-vehicle tests, the following traffic design recommendations are proposed for at-grade intersections of main roads in cold suburban areas: (1) Suggestions for improving Y-shaped intersections. It is recommended to increase the intersection angle through geometric modification to improve sight distance; strengthen the setting of warning signs in winter and advance the speed limit sign setting distance by at least 50m from the current standard; add anti-skid treatment to the road surface and lay materials with a high coefficient of friction in the core area of ​​the intersection.

[0062] (2) Suggestions for improving cross-shaped intersections. It is recommended to set up a complete sight triangle at each of the four approach lanes, remove buildings and vegetation that obstruct the view, and add flashing warning signals in winter to remind drivers of the existence of the intersection.

[0063] (3) Suggestions for improving T-shaped intersections. T-shaped intersections have relatively low physiological load, but it is still recommended to strengthen guidance signs for drivers entering the main road from the side road in winter conditions, clarify the right-of-way rules, and reduce decision-making load.

[0064] Before the experiment, to simulate the physiological load caused by driving duration, drivers first needed to drive intermittently for three hours. After completing this, the drivers' resting physiological load index was tested indoors. After collecting the drivers' resting data, a recorder recorded the data, and the drivers, wearing physiological devices, conducted a simulated driving experiment in the experimental vehicle.

[0065] Due to the complexity of real-vehicle testing conditions and the limitations of the experimental environment, there is a systematic deviation between simulated data and actual collected data. Linear regression methods assume a fixed linear relationship between simulated and measured data, making it difficult to characterize the nonlinear fluctuations in driver physiological indicators under complex traffic conditions. While BP neural networks have some nonlinear fitting capabilities, they suffer from strong dependence on sample size, insufficient training stability, and a tendency to get trapped in local optima. Support vector machines have certain advantages under small sample conditions, but their parameter selection is sensitive to the complex mapping relationships between high-dimensional nonlinear features under multi-indicator coupling, limiting the model's generalization ability. Furthermore, most existing correction methods focus on independent correction of single indicators, lacking comprehensive utilization of the synergistic changes in multiple physiological indicators such as heart rate, skin conductance, and electromyography. Therefore, they are unable to effectively eliminate the systematic errors between simulated and measured data under complex traffic conditions in cold regions.

[0066] Especially in the scenario of intersections on main roads in cold suburban areas, the interaction of factors such as changes in the adhesion coefficient of icy and snowy road surfaces, frequent driving operations, and enhanced physiological stress on drivers causes physiological indicators to exhibit significant nonlinear, random fluctuations, and multivariate correlation characteristics. Traditional correction methods struggle to simultaneously ensure both fitting accuracy and model stability. Therefore, current technology lacks a data correction method that can adapt to the complex traffic environment in cold regions and simultaneously handle the nonlinear coupling relationships of multiple physiological indicators.

[0067] Therefore, to improve the reliability of simulated data, this implementation proposes a data correction method based on the Random Forest algorithm. A data correction model is constructed using the Random Forest algorithm to correct the simulated data. The Random Forest algorithm, through ensemble of multiple decision trees for nonlinear regression, can effectively handle complex mapping relationships between high-dimensional, multivariable, and nonlinear features, while possessing strong noise resistance and generalization ability. Compared to traditional linear models, it does not require pre-assumptions about functional relationships between variables; compared to single neural network models, it is less sensitive to sample size and parameters, and has stronger training stability; simultaneously, the multi-tree ensemble mechanism can reduce the risk of overfitting and improve the model's applicability in different driving scenarios. This implementation utilizes the Random Forest algorithm to simultaneously input multiple simulated physiological indicators such as heart rate growth rate, skin resistance increment rate, and root mean square value of electromyography, and jointly predicts the corresponding measured indicators, thereby fully utilizing the covariance characteristics between different physiological indicators to achieve high-precision correction of simulated physiological load data of drivers at cold-region intersections. The built-in OOB error estimation of the random forest algorithm provides a model evaluation method that does not require an additional validation set. The algorithm can also naturally handle multi-dimensional input features and reveal the degree of contribution of each input variable to the correction result through feature importance evaluation.

[0068] The steps of the random forest algorithm are as follows: Step 1: Bootstrap Sampling. Bootstrap sampling is the first mechanism for introducing diversity into random forests. Specifically, from the original training set containing N samples, N samples are randomly selected with replacement to form a bootstrap subset. Because sampling with replacement occurs, the same original sample may be selected repeatedly, while some samples will not be selected at all. Statistically, it can be shown that approximately 63.2% of the original samples will be selected at least once in each bootstrap sampling, leaving approximately 36.8% of the samples unused; these unused samples are called out-of-bag data (OOB).

[0069] For each Bootstrap subset, an independent CART (Classification and Regression Tree) decision tree is trained. Full growth means that the tree will continue to split until the number of samples in a leaf node is lower than the preset minimum number of samples in a leaf node (MinLeafSize), without pre-pruning.

[0070] Step 2: Decision Tree Construction and Feature Randomization The second mechanism by which random forests introduce diversity is feature randomization. In a regular decision tree, each time a node splits, the feature that best meets the splitting criterion is selected from all features. In a random forest, however, each time a node splits, only a subset (usually the square root or one-third of the total number of features) is randomly selected from all features, and the optimal split is then found from this subset. This constraint, while seemingly reducing the scope of individual trees, allows different trees to have different focuses on different features, thereby reducing the correlation between trees—the lower the correlation, the better the ensemble effect.

[0071] The node splitting criterion adopts the mean squared error (MSE) minimization principle: for samples within a node, a threshold for a certain feature is found such that after splitting into left and right child nodes according to this threshold, the sum of the MSEs of the samples within the two child nodes is minimized. MSE measures the squared mean of the difference between the predicted value and the true value; the smaller the value, the more similar the samples within the node are to each other.

[0072] Step 3: Integrated Prediction After training T decision trees, for any new input sample x, the prediction value of the random forest is the arithmetic mean of the tree prediction results: ; in, For the first Tree samples The predicted value.

[0073] The advantage of averaging is that the random errors (with different directions of deviation) of each tree cancel each other out after averaging, while the true patterns captured by the trees are strengthened after averaging, making the ensemble model more accurate and stable than any single tree.

[0074] The mean data of the real vehicle data at 16 sampling points and the mean data of the simulation experiment were input into the data correction model. The comparison of the initial errors of the real vehicle experiment and the simulation data is shown in Table 4.

[0075] Table 4. Data before correction (taking a Y-shaped intersection as an example)

[0076] This implementation method employs a multi-indicator joint modeling scheme: three simulated indicators (simulated HR, simulated SC, and simulated EMG) are used as input feature vectors to predict each measured indicator. This scheme fully utilizes the covariant relationship between physiological indicators.

[0077] This implementation does not employ a one-to-one calibration method, meaning it uses a single simulated indicator to independently predict only its corresponding measured indicator. For example, it uses simulated heart rate (HR) to predict measured HR and simulated straight line (SC) to predict measured SC. While this method is simple in structure, it assumes that different physiological indicators are independent of each other, ignoring the significant coupling between driver physiological responses in real-world driving. In reality, when a driver passes through a cold-weather intersection, heart rate changes, skin conductance, and muscle tension usually change synchronously. When a driver is stimulated by factors such as icy roads, complex traffic conflicts, or limited visibility, the sympathetic nervous system is activated as a whole, and there is a clear synergistic relationship between ECG, skin conductance, and electromyography (EMG) signals. Therefore, if a one-to-one modeling approach is used, the model may only learn the local fluctuation characteristics of a single indicator and fail to recognize the joint changes between different physiological indicators, resulting in insufficient stability of the calibration results.

[0078] Especially in complex traffic environments in cold regions, single indicators are easily affected by random noise. For example, electromyography (EMG) signals may fluctuate momentarily due to short-term steering maneuvers by the driver, electrodermal signals may be affected by changes in ambient temperature, and heart rate indicators may be affected by individual differences in basal metabolic rate. If a one-to-one correction method is used, the model is prone to misinterpreting these random disturbances as actual load changes, thereby amplifying the prediction error.

[0079] Therefore, this implementation adopts a many-to-one joint input structure, that is, using three simulated indicators to jointly predict a single measured indicator, enabling the model to simultaneously learn the correlation and synergistic change characteristics between different physiological indicators. When a certain indicator shows local abnormal fluctuations, the model can use information from other indicators to compensate for and constrain it, thereby improving the robustness and stability of the correction results.

[0080] This implementation method compares two modeling approaches: one-to-one and many-to-one. Experimental results show that, under the same training sample conditions, the many-to-one joint modeling scheme has a higher average coefficient of determination (R²) for heart rate growth rate, skin resistance increment rate, and root mean square value of electromyography. 2The results showed that the multi-to-one approach was superior to the one-to-one approach, with a significant reduction in root mean square error (RMSE). Among these, the skin conductance index, due to its strong coupling relationship with emotional arousal and driving behavior, showed the most significant improvement in prediction accuracy after adopting a multi-indicator joint input. These results indicate that the multi-to-one structure is not an arbitrary choice of input format, but rather a targeted technical solution addressing the multi-indicator coupling characteristics of drivers' physiological load at cold-region intersections, effectively improving the accuracy of simulation data correction and the model's generalization ability.

[0081] Hyperparameter settings are crucial for machine learning models. Before training, a set of hyperparameters needs to be manually set; these are parameters that control the model's structure and training process, and differ from the parameters the model learns automatically from data. The values ​​of hyperparameters directly affect the model's performance, and a systematic approach is needed to select the optimal combination.

[0082] This implementation uses a grid search method: a candidate value set for each hyperparameter is pre-specified, and then all parameter combinations are traversed. The optimal parameter combination is selected based on minimizing the mean squared error of the out-of-bag (OOB) data. Three hyperparameters need to be searched; candidate values ​​are shown in Table 5.

[0083] Table 5 Hyperparameter Selection

[0084] To comprehensively measure the correction effect, this implementation method uses five complementary evaluation indicators: Coefficient of determination R 2 (R-squared): Measures the proportion of data variation explained by the model, ranging from 0 to 1. The closer to 1, the better the model captures the patterns of data variation. For example, R... 2 A value of 0.92 means that the model explains 92% of the data variation, while the remaining 8% is random error that the model fails to explain.

[0085] Root Mean Square Error (RMSE): The average magnitude of the prediction error (with the same dimensions as the original data), and it is more sensitive to larger errors. An RMSE of 0.40 beats / min for heart rate means that the average prediction error is approximately 0.40 beats / min, which is very small compared to the normal range of heart rate fluctuations (approximately 5–8 beats / min).

[0086] Mean Absolute Error (MAE): The average of the absolute values ​​of the error, which directly reflects the average deviation and is not overly affected by large errors.

[0087] Mean Absolute Percentage Error (MAPE): The percentage of error relative to the true value. It is dimensionless and can be used for cross-sectional comparisons between three indicators with different dimensions: heart rate, skin conductance, and electromyography. MAPE = 3% means that the predicted value deviates from the true value by an average of 3%.

[0088] OOB R 2 R calculated using out-of-bag samples 2 As an unbiased estimate of the model's generalization ability, it is used together with the test set R² to verify the model's reliability.

[0089] After optimizing the hyperparameters through grid search, the multi-index joint random forest model achieved good correction results on all three physiological indices. The specific evaluation indices are shown in Table 6.

[0090] Table 6 Summary of Evaluation Metrics for Joint Random Forest Model

[0091] As shown in Table 6, the random forest model performed well for all three indicators: heart rate on the test set R 2 The R² for OOB (Out-of-Body) was 0.9279, the R² for the test set was 0.9823, and the R² for EMG (Electromyography) was 0.8748. The differences between the OOB R² and the test set R² were all within a reasonable range, indicating that the model did not suffer from severe overfitting. Regarding MAPE (Magnetic Activity Perception), all three metrics were controlled within 6%.

[0092] Convergence analysis was performed on the model's OOB error to study the significance of the final model's decision tree settings. The convergence curve is shown below. Figure 3 As shown.

[0093] Depend on Figure 3 As can be seen from the convergence curve, when the number of decision trees reaches approximately 100, the OOB error has basically stabilized, and the improvement brought by further increasing the number of trees is extremely limited. Therefore, the final number of model trees is set between 100 and 200.

[0094] Input the simulation data to be corrected into the data correction model to generate a comparison chart of the corrected data and the actual data, such as... Figure 4 , Figure 5 and Figure 6 As shown, the trained random forest model was applied to the simulated data from the original 16 collection points. The corrected curve (green) closely matched the measured curve (red), especially in the high-load section (points 10-13), where the large nonlinear deviation was effectively corrected.

[0095] The physiological load data of the corrected intersection after improvement were used to make a safety score according to the comprehensive physiological load evaluation model in Implementation Method 1. The physiological load of the intersection after improvement was compared and analyzed. The scores before and after improvement are shown in Table 7.

[0096] Table 7. Comparison Scoring Table Before and After Improvement

[0097] To more clearly observe the distribution of each subject sample in the three risk zones, the sample size for each safety evaluation interval before and after improvement is shown in Table 8.

[0098] Table 8 Samples of each safety evaluation interval before and after improvement

[0099] As shown in Tables 7 and 8, the Y-shaped area showed the greatest improvement in winter, with an average increase of 7.39 points (+20.0%). The number of high-risk sampling points (score <35.5 points) decreased from 6 to 4 in winter. The average physiological indicators HR / SC / EMG in the core area (points 10-13) decreased by about 7%, indicating that signage warnings and road surface anti-skid measures have a significant effect on alleviating the acute stress of drivers in the core conflict area. However, the absolute score in the peak area is still at a moderately low level (13.2-24.6 points), indicating that the inherent cognitive load caused by the geometric complexity of the Y-shape is difficult to completely eliminate through passive signage facilities.

[0100] The average score for cross-shaped intersections improved by 5.4 points (+12.2%), and the number of high-risk points decreased from 6 to 4. The yield control markings and sight distance improvement measures have a relatively stable effect in medium-complexity intersections. The improvement in points 8 to 12 is relatively concentrated (4.7 to 7.2 points). Drivers' decision-making load in the core area has been reduced under the guidance of clear priority signs, but continued maintenance of the visibility of markings at night and in icy and snowy weather is still needed.

[0101] The average score for T-junctions improved by 3.76 points (+7.2%), the smallest and most even improvement. The score for merging sections (points 8-10) improved by about 3.5-5 points, which is consistent with the relatively limited intervention intensity, as the improvement measures mainly focused on optimizing sight distance and strengthening yield signs. The corresponding improvement in summer was about 1.67 points, indicating that the load of T-junctions under good weather conditions was already close to the low-risk zone, and there was limited room for improvement.

[0102] This implementation method addresses the safety risks identified in real-vehicle experiments by developing targeted improvement plans for three types of intersections. A data correction model constructed using the random forest algorithm effectively eliminates systematic deviations between simulation and actual measurements. The R² scores for all three indicators on the test set are above 0.87, and the MAPE is controlled within 6%. Comparative analysis before and after the improvements shows that the Y-shaped intersections saw the greatest improvement, with an average increase of 7.39 points (+20.0%), followed by the cross-shaped intersections with an increase of 5.40 points (+12.2%), and the T-shaped intersections with an increase of 3.75 points (+7.2%). The total number of high-risk sampling points in winter decreased from 16 to 8. The improvement is positively correlated with the geometric complexity of the intersections, validating the effectiveness of the proposed suggestions.

[0103] The above provides a detailed description of the safety evaluation method and improvement scheme evaluation method for intersections of urban and suburban trunk roads in cold regions proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A safety evaluation method for intersections of main roads in suburban areas of cold regions, characterized in that, Includes the following steps: Step 1: Set up multiple detection points in a preset area of ​​the target intersection and obtain multiple physiological indicators of multiple drivers when they pass through each detection point. Step 2: Based on the multiple physiological indicators obtained in Step 1, construct a comprehensive physiological load evaluation model and calculate the safety score of the target plane intersection; Step 3: Based on the safety score, classify the risks and complete the safety assessment of the intersection.

2. The safety evaluation method for intersections of urban and suburban trunk roads in cold regions according to claim 1, characterized in that, The physiological indicators in step 1 include heart rate growth rate, root mean square value of electromyography, and rate of change of skin resistance increment.

3. The safety evaluation method for intersections of main roads in cold suburban areas according to claim 1, characterized in that, In step 2, a comprehensive physiological load evaluation model is constructed to calculate the safety score of the target intersection, specifically as follows: ； in, This is the heart rate correction factor for cold, low-temperature environments. For the correction factor of skin electrical stress in suburban environments, This is a correction factor for muscle tension in traffic flow density at intersections. For the first Heart rate growth rate per sample For the first Increment of skin resistance in each sample For the first Root mean square value of electromyography for each sample The time it takes for a driver to pass through an intersection. The average travel time for vehicles at intersections of urban and suburban trunk roads. For the angle of conflicting traffic flow at the intersection, This is the attenuation term for travel time. For conflict risk correction items, and These represent the maximum and minimum values ​​of the driver's overall physiological load at multiple testing points. and These are the maximum and minimum values ​​of the growth rate of the center rate of all samples, respectively. and These represent the maximum and minimum values ​​of skin resistance increments across all samples. and These represent the maximum and minimum root mean square values ​​of electromyography (EMG) across all samples.

4. The safety evaluation method for intersections of urban and suburban trunk roads in cold regions according to claim 1, characterized in that, The risk classification in step 3 is specifically as follows: when Time-sharing is considered high-risk; when The risk level is medium when the time frame is determined; The time frame is considered low-risk.

5. An evaluation method for safety improvement schemes at intersections of main roads in cold suburban areas, wherein the method is implemented based on the safety evaluation method described in any one of claims 1 to 4, characterized in that... Includes the following steps: Step A: Propose safety improvement plans based on the safety assessment results; Step B: Based on the safety improvement plan, construct a simulation experiment to obtain the simulation data to be corrected; Step C: Correct the simulation data to obtain the corrected data; Step D: Use the corrected data as input to the comprehensive physiological load assessment model to calculate the safety score; Step E: Based on the safety score, classify the risks and complete the safety evaluation of the safety improvement plan.

6. The evaluation method for safety improvement schemes at intersections of main roads in cold suburban areas according to claim 5, characterized in that, Step C involves correcting the simulation data, specifically as follows: A data correction model is constructed, using physiological index data obtained from simulation experiments as input and corresponding physiological index data from real vehicle experiments as output. The data correction model is trained by inputting the simulation data to be corrected into the trained data correction model to obtain the corrected data.

7. The evaluation method for safety improvement schemes at intersections of main roads in cold suburban areas according to claim 6, characterized in that, The data correction model uses multiple simulated physiological indicators as input features to predict the corresponding measured physiological indicators.

8. The evaluation method for safety improvement schemes at intersections of urban and suburban trunk roads in cold regions according to claim 6 or 7, characterized in that, The data correction model uses the random forest algorithm.

Images