Traffic flow operation safety risk evaluation method

Abstract

The invention discloses a traffic flow operation safety risk evaluation method. The method comprises the steps that the traffic density and the traffic speed of a traffic flow are acquired; obtaining the vehicle distance and relative change information of different vehicles in the traffic flow, and obtaining a traffic safety operation index according to the vehicle distance and relative change information; judging the traffic safety operation index, and calculating based on a judgment result to obtain a safety risk value; and fitting the relationship between the traffic density and the traffic speed and the safety risk value to obtain a traffic flow risk prediction model, and identifying the traffic density and the traffic speed obtained in real time through the traffic flow risk prediction model to obtain a traffic flow risk value monitored in real time.

Description

Technical Field

[0001] The present invention belongs to the technical field of traffic safety risk assessment, and particularly relates to a method for assessing the safety risk of traffic flow operation. Background Art

[0002] Analyzing accident or conflict data and discovering and rectifying safety defects in road infrastructure are the main measures for traffic accident prevention. With the shift of accident prevention towards "before-the-fact", real-time assessment and reduction of collision risks in traffic flow have become crucial. The traditional mode relying on accident data analysis has problems such as data lag and lack of details, making it difficult to meet the requirements of proactive prevention. Existing research mostly focuses on "event-based" methods, ignoring a large amount of "safety interaction" information and making it difficult to comprehensively evaluate the safety state of traffic flow. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a method for assessing the safety risk of traffic flow operation to solve the problems existing in the above prior art.

[0004] To achieve the above object, the present invention provides a method for assessing the safety risk of traffic flow operation, including:

[0005] Obtaining the traffic density and traffic speed of the traffic flow;

[0006] Obtaining and based on the vehicle distances and relative change information of different vehicles in the traffic flow, obtaining a traffic safety operation index;

[0007] Judging the traffic safety operation index and calculating based on the judgment result to obtain a safety risk value;

[0008] Fitting the relationship between the traffic density and traffic speed and the safety risk value to obtain a traffic flow risk prediction model, and identifying the real-time traffic density and traffic speed obtained in real time through the traffic flow risk prediction model to obtain the final real-time monitored traffic flow risk value.

[0009] Optionally, the obtaining process of the traffic safety operation index includes:

[0010] Obtaining distance interaction information based on the vehicle distance; obtaining relative interaction information based on the relative change information; calculating the traffic safety operation index based on the distance interaction information and the relative interaction information.

[0011] Optionally, the obtaining process of the distance interaction information CRF includes:

[0012]

[0013] Where n represents the number of vehicles in the influence area, and are respectively the target vehicle and the The distances of the vehicle in the transverse and longitudinal directions and are the parking distances in the transverse and longitudinal directions , where v represents the speed, g represents the acceleration due to gravity, and f represents the ground friction

[0014] Optionally, the process of obtaining the relative interaction information DBF(t) includes:

[0015]

[0016] where is the relative vehicle distance change between the target vehicle and the th vehicle at time , is the relative speed change between the target vehicle and the th vehicle at time , is the current vehicle distance between the target vehicle and the th vehicle; is the speed of the th vehicle

[0017] Optionally, the traffic safety operation index is obtained by calculating the product of the distance interaction information and the relative interaction information

[0018] Optionally, the process of judging the traffic safety operation index includes:

[0019] Judging the traffic safety operation index according to the upper threshold of the maximum safe operation domain

[0020] The upper threshold of the maximum safe operation domain is:

[0021]

[0022] where is the maximum safe operation area distance is the tolerance probability

[0023] Optionally, the process of obtaining the safety risk value TSC(x,t) includes:

[0024]

[0025] where is the time step, representing the resolution of the trajectory data. nT i is the total number of time steps traveled by the i-th vehicle within the spatio-temporal window, Conf ij is the conflict discrimination index, used to judge whether the i-th vehicle is in a conflict state at time j;

[0026] Among them, .

[0027] Optionally, the traffic flow risk prediction model adopts a deep learning model, where: the traffic flow risk prediction model is used to characterize the fitting relationship between traffic density, traffic speed and safety risk value:

[0028]

[0029] Among them, TSC represents the safety risk value, represents the traffic flow risk prediction model, represents the traffic density, and V represents the traffic speed.

[0030] Optionally, the traffic flow risk prediction model adopts a support vector machine, a random forest regression model or a gradient boosting model.

[0031] Optionally, the process of obtaining the traffic density ρ and traffic speed V of the traffic flow includes:

[0032]

[0033]

[0034] Among them, and represent the spatial and temporal sizes of the window, n is the number of vehicles within the window, tt i and Dt i are respectively the driving time and driving distance of vehicle i.

[0035] Compared with the prior art, the present invention has the following advantages and technical effects:

[0036] The present invention constructs an evaluation framework based on information theory and control theory, and proposes a safety limit hypothesis for traffic flow. Further, a distributed stochastic predictive control model under single-lane non-free flow conditions is constructed, and the mapping relationship between the macroscopic characteristics of traffic flow and safety risk is deduced. To quantify the microscopic interaction risk, the "Safe Operating Domain" (SOD) index is proposed, the safety risk value is constructed as a function of density and speed, and a macroscopic traffic flow risk prediction model based on machine learning is established, which can effectively conduct risk analysis in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0038] Figure 1Schematic diagram of the influence area of the target vehicle and the vehicles within the influence area according to the embodiments of the present invention;

[0039] Figure 2 Schematic diagram of the target vehicle and the vehicles within its area according to the embodiments of the present invention;

[0040] Figure 3 Schematic diagram of the verification result of SOD according to the embodiments of the present invention;

[0041] Figure 4 Schematic diagram of the trend line of the number of accidents varying with speed when the traffic flow density is 30 veh / h according to the embodiments of the present invention. Detailed implementation manners

[0042] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0043] It should be noted that the steps shown in the flowchart of the drawings may be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than here.

[0044] Aiming at the problem in the prior art that it is difficult to comprehensively evaluate the traffic flow safety state for the "safety interaction" information, the present invention proposes a traffic safety operation domain (SOD) index to quantify the microscopic interaction risk, and based on this index, it identifies whether a traffic conflict occurs. When a traffic conflict occurs, the conflict time is used as an index to measure the microscopic traffic safety risk, constructs the safety risk value as a function of density and speed, establishes a macroscopic traffic flow risk prediction model based on machine learning to effectively cope with the prediction scenario of traffic flow risk, and learns the effectiveness of the SOD index and the accuracy of the prediction model through relevant cases.

[0045] Regarding the relationship between traffic safety risk and interaction. From a physical perspective, traffic safety risk can be microscopically defined as the product of the probability and consequence of spatio-temporal overlap occurring in an interaction. Without interaction, there will be no traffic accidents. Therefore, for interactions involving different environments, different vehicles, and different people, there are significant differences in traffic safety risk, that is, different interaction types. Ideally, if every interaction occurring in the traffic flow can be captured, then by integrating all interaction risks, the number and consequences of traffic accidents occurring on a second-scale can be predicted. However, current technical means are difficult to obtain the massive heterogeneous interactions occurring in a vast spatio-temporal range, and it is impossible to perform safety risk analysis, evaluation, and prediction through the ideal method.

[0046] Considering that interaction is a source of traffic accidents, if traffic accident data is used as the basis for traffic safety evaluation, then for the evaluation of road sections, the number of traffic accidents per unit annual mileage can be used, which is called the annual traffic accident rate; for the evaluation of intersections, the number of traffic accidents per unit annual traffic volume in the weaving direction can be used, which is called the annual weaving accident rate. From the physical process of accident occurrence, micro risks come from the uncertainty of people's perception of surrounding information and driving decisions, which is positively correlated with driving speed and negatively correlated with people's environmental information acquisition and decision-making control capabilities; the risk of a certain traffic flow comes from the instability of longitudinal and lateral interactions, which is negatively correlated with vehicle spacing and the level of environmental information supply and positively correlated with traffic flow speed.

[0047] The present invention provides a traffic flow safety structured analysis framework for analysis, providing a theoretical analysis basis for subsequent safety risk prediction.

[0048] "Interaction" is the core of traffic safety analysis. In terms of form, interaction can be divided into following interaction, lane-changing interaction, and intersection interaction. Among them, the physical process of following interaction is the simplest and easy to quantify and model. This invention mainly takes following as the research object, and there is a similar logic for other types of interaction. Considering the randomness of the driving controller model parameters, for the internal dynamic interaction within a platoon formed under the condition of single-lane non-free flow, each vehicle optimizes the local objective function only according to its own situation and the state of the vehicle in front, and establishes a distributed stochastic model predictive control (DSMPC) model of traffic flow. Further analysis is carried out from the aspects of stability, randomness, and safety, demonstrating that under a specific traffic flow density, there is a safety upper limit for the average speed of traffic flow, and below this upper limit, the accident rate is relatively low.

[0049] This embodiment discloses a traffic flow operation safety risk evaluation method, including:

[0050] Obtain the traffic density and traffic speed of the traffic flow;

[0051] Obtain and obtain the traffic safety operation index according to the vehicle distance and relative change information of different vehicles in the traffic flow;

[0052] Judge the traffic safety operation index, and calculate based on the judgment result to obtain the safety risk value;

[0053] Fit the relationship between the traffic density and traffic speed and the safety risk value to obtain a traffic flow risk prediction model, and identify the traffic density and traffic speed obtained in real time through the traffic flow risk prediction model to obtain the traffic flow risk value monitored in real time.

[0054] As an implementation manner of this embodiment, the process of obtaining the traffic safety operation index includes:

[0055] Obtain distance interaction information according to the vehicle distance; obtain relative interaction information according to the relative change information; calculate a traffic safety operation index according to the distance interaction information and the relative interaction information.

[0056] The theoretical analysis of this embodiment specifically includes:

[0057] First, model a distributed stochastic predictive control model, and the process is as follows:

[0058] Establish a distributed stochastic predictive control model that can simulate human driving characteristics and their randomness, and describe the vehicle kinematics, driver perception and reaction, predictive control strategy and their randomness and uncertainty in the vehicle platoon. Specifically as follows:

[0059] (1) Vehicle kinematics modeling:

[0060] (1)

[0061] Among them, represents the position of vehicle i at time k, represents the speed of vehicle i at time k, represents the acceleration of vehicle i at time k, represents the sampling time interval.

[0062] (2) Driver perception and reaction modeling:

[0063] Considering the reaction time of human drivers, introduce a time delay , and at this time the acceleration is expressed as:

[0064] (2)

[0065] Among them, represents the distance between vehicle i and the vehicle in front , f( ) represents the decision function of the driver, represents the reaction time of driver i, and it is assumed to follow a normal distribution .

[0066] (3) Predictive control strategy:

[0067] Assume that each driver is constantly predicting the behavior of the vehicle in front and adjusting the target vehicle acceleration based on these predictions. The predictive control strategy can be expressed as:

[0068] (3)

[0069] Among them, N represents the prediction horizon, represents the desired vehicle distance, represents the desired speed, represent the target weight coefficients of the vehicle distance, speed, and acceleration, , represents the acceleration constraint, represents the speed constraint, represents the vehicle distance constraint.

[0070] (4) Randomness and uncertainty:

[0071] To simulate the randomness of human driving behavior, multiple random variables are introduced, including the reaction time, state estimation error , prediction error and control error . In addition, a random parameter is introduced to represent the personality characteristics of the driver (such as aggressive or conservative), and this parameter can affect the selection of the weight coefficient , where can be assumed to follow the standard normal distribution N(0, 1).

[0072] Safety analysis based on the distributed stochastic predictive control model:

[0073] Based on the established distributed stochastic predictive control model, stability, randomness, and safety analyses are carried out to further demonstrate the existence of the upper limit of the safe operating average speed of traffic flow.

[0074] (1) Stability analysis:

[0075] Analyze the stability of the system. In the ideal case (without randomness and uncertainty), the stability of the system mainly depends on the parameter selection of the predictive control strategy, and the stability condition is determined by analyzing the characteristic equation of the system. Consider the simplified linear model:

[0076] (4)

[0077] where and are the proportional gain and speed difference gain, respectively. Then the characteristic equation of the system is:

[0078] (5)

[0079] According to the Routh-Hurwitz stability judgment condition: , then the sampling time interval is less than a given value, and the system can converge to a stable state, and this value is related to the speed difference gain. It shows that when the sampling time of the human (driver) is large (such as distracted driving), the system stability deteriorates, the traffic flow tends to become unstable or a collision accident occurs, and the higher the speed, the higher the sampling frequency requirement for the human, and the smaller the tolerable driving error.

[0080] (2)Randomness impact analysis

[0081] Considering randomness and uncertainty, analyze the impact of these factors on system performance. Use stochastic stability theory to analyze the mean-square stability of the system:

[0082] Define the state vector , the system can be expressed as:

[0083] (6)

[0084] where A(k) and B(k) are random matrices, is the control input, is the random perturbation. The mean-square stability condition of the system can be expressed as:

[0085] (7)

[0086] where E[] represents the expectation and M is a bounded constant. Through the analysis of the Lyapunov equation, the mean-square stability expression of the system is:

[0087] (8)

[0088] where P is a positive definite matrix and Q is a semi-positive definite matrix, and the condition for the mean-square stability of the system is obtained at this time. This shows that in traffic flow, if the speed perturbation or position control error decreases with time, the system tends to a stable state, that is, the error caused by randomness is one of the main factors determining the upper limit of system stability.

[0089] (3)Safety analysis

[0090] Define a prediction-based safety index, and the expression is:

[0091] (9)

[0092] where N is the prediction horizon, is the minimum safety distance. This safety index considers the difference between the minimum predicted inter-vehicle distance and the safety distance within the next N time steps. When , it means that an unsafe situation may occur within the prediction range. Therefore, it is necessary to define the safety probability, and its expression is:

[0093] (10)

[0094] Under the DSMPC framework, this probability is affected by many factors, such as and the distribution characteristics of; the control strategy parameters The choice, the driver's reaction time and personality parameters as well as the perception control error. Therefore, the optimization objective of DSMPC is expressed as:

[0095] (11)

[0096] where E[] represents the expectation, is a weight coefficient. By directly considering safety in the control strategy in this way, the controller will weigh the two goals of pursuing the desired state and ensuring safety when making decisions. Take as a relatively large value. If the above objective function has a solution, it indicates that there exists a driver predictive control strategy such that tends to 0. Whether there is a solution depends on the distribution characteristics of the aforementioned random parameters and the desired headway and desired vehicle speed. Therefore, the driver predictive control strategy can be regarded as a generalized traffic control strategy. The traffic management department can make the predictive control strategy tend to the optimal strategy through control strategies such as speed limits and headway warnings, and then the accident rate can tend to 0. At this time, there is a steady relationship between the desired headway and vehicle speed.

[0097] (4) Analysis of the relationship between average speed and safety based on DSMPC

[0098] Under the DSMPC framework, the relationship between average speed and safety becomes more complex because they are both functions of the control strategy, prediction model, and random factors. Define an average speed based on DSMPC, and the expression is:

[0099] (12)

[0100] where M is the number of vehicles in the platoon. At this time, the upper speed limit for safe operation should satisfy:

[0101] (13)

[0102] where, is a small tolerance probability and can tend to 0.

[0103] All randomness and uncertainties of the DSMPC system need to be considered, such as the prediction error distribution and the state estimation error distribution and the control error distribution the driver characteristics distribution and the control parameter distribution 、 and Further analysis of the relationship with various DSMPC parameters. First, the influence of the prediction horizon N. When a larger N generally improves safety but also increases the computational complexity (attention consumption). Second, the influence of the control parameter . When different parameter combinations will lead to different . Further, the driver characteristic parameters and affect the upper limit of the safe speed. Finally, the influence of the error distribution parameter (such as variance). When there is uncertainty in the system, the upper limit of the safe speed is also affected. Therefore, a more comprehensive expression is obtained:

[0104] (14)

[0105] It can be proved through the above analysis that at a specific traffic flow density, there is indeed a safe upper limit for the average speed of vehicles. Below this upper limit, the accident rate of vehicles is relatively low because SMPC (the driver) can adjust the acceleration of the vehicle by itself, making the local platoon segment stable and coping with random disturbances. However, when the random disturbance exceeds a certain threshold, collisions are inevitable.

[0106] In response to the above theoretical analysis, a subsequent detailed elaboration is carried out on the traffic flow operation safety evaluation method based on the traffic safety operation domain provided by the present invention:

[0107] The traffic safety analysis concept model provides the goal and path for the quantitative analysis of traffic safety. In response to the above analysis results, the present invention provides a traffic flow operation safety evaluation method based on the traffic safety operation domain. On the basis of the above traffic flow safety mechanism analysis, a quantitative analysis method is tried to construct a relationship model of traffic flow speed, density and safety. First, an alternative metric for the microscopic interaction risk of traffic flow is proposed to realize the quantitative calculation of the risk value; then, based on the key traffic flow parameters, a machine learning prediction model for the traffic flow operation safety risk is established. Furthermore, the traffic management department can use the traffic flow high-risk warning value to actively adjust the traffic flow state through means such as variable speed limit, lane control, vehicle type control, and induction prompt, improve the high-risk state, and realize the dynamic traffic safety risk control.

[0108] Quantification of traffic flow operation safety risk based on microscopic interaction:

[0109] In view of the effectiveness and limitations of different conflict indicators in evaluating traffic safety, and based on the macroscopic traffic flow state, this invention proposes a safety evaluation indicator, the Safety Operating Domain (SOD). By quantifying the safe interaction area between vehicles based on macroscopic traffic flow characteristics such as traffic density, and introducing features such as lane changes and traffic flow instability, it realizes the measurement of microscopic interaction safety risks in traffic flow. Specifically as follows: Based on macroscopic traffic flow trajectory data, the macroscopic traffic state variables traffic density ρ, traffic flow Q, and speed V are calculated respectively as shown in the following formulas:

[0110] (15)

[0111] Among them, and represent the spatial and temporal sizes of the window, n is the number of vehicles within the window, tt i and Dt i are the travel time and travel distance of vehicle i respectively. The definitions of the macroscopic state variables of Edie are consistent and automatically satisfy the hydrodynamic relationship:

[0112] (16)

[0113] According to formula (16), the macroscopic state variables essentially contain all the traffic flow efficiency characteristics reflected by the underlying car-following interaction behaviors. These interactions between vehicles will trigger traffic conflicts, and these conflicts usually have microscopic characteristics and essentially depend on the behavior patterns of drivers and pedestrians. However, in traffic safety evaluation, it is not feasible to obtain the microscopic behaviors of vehicles or pedestrians in real time (such as steering wheel operations, temporarily changing walking directions, etc.). Therefore, safety evaluation can be carried out through the macroscopic traffic flow states (such as traffic density, traffic flow, and speed, etc.) reflected by these microscopic behaviors. To quantify the traffic flow safety risk from a macroscopic aggregated perspective, the "Conflict Time" (TSC) can be used as an indicator to measure the microscopic traffic safety risk. TSC refers to the total time experienced by all vehicles when a conflict occurs, and this time is calculated based on a specific conflict metric. For a specific traffic flow condition, a higher TSC value indicates that vehicles stay in the conflict situation for a longer time under this condition, and the risk of collision is higher. Considering the vehicle interactions within the spatio-temporal window, TSC can be expressed by the following formula:

[0114] (17)

[0115] Among them, is the time step (seconds), representing the resolution of the trajectory data. nT i is the total number of time steps that the i-th vehicle travels within the spatio-temporal window, Conf ijIt is a conflict discrimination index, used to determine whether the i-th vehicle is in a conflict state at the j-th moment, where x represents the label of the interacting vehicle and t represents the corresponding time step.

[0116] When calculating the TSC, the present invention is mainly based on the following two considerations: First, the TSC calculated based on the conflict index should maintain a significant correlation with the macroscopic traffic index; Second, the TSC calculated using different conflict indices should have clear physical meanings and interpretability. Among them, the two indices of Time to Collision (TTC) and Deceleration Rate to Avoid Crash (DRAC) are based on a premise assumption: traffic conflicts are limited to the interaction between the front and rear vehicles in the same lane. This assumption condition limits its applicability to traffic safety assessment at the macroscopic level to a certain extent, especially unable to effectively reflect the potential traffic safety hazards caused by the interaction between vehicles in adjacent lanes. Specifically, when using TTC or DRAC as the conflict index, it only considers the situation where the speed of the following vehicle is greater than that of the leading vehicle (i.e., the rear-end collision scenario), so it can only reflect the interaction information on the collision path and cannot fully capture other potential dangerous interaction behaviors that may exist between vehicles under high-speed driving conditions. In contrast, the TSC index based on Potential Stopping Distance (PSD) can more effectively establish the correlation between macroscopic traffic flow variables and traffic safety levels by measuring the time when the stopping distance is less than the relative distance between vehicles.

[0117] However, the medium congestion state is usually accompanied by a series of complex traffic phenomena, such as increased speed differences, traffic instability, and hysteresis effects, etc. These phenomena are of great significance in traffic safety assessment. Therefore, in order to deeply reveal the internal relationship between traffic safety and traffic flow, the TSC calculated based on any conflict index should be able to effectively explain this phenomenon. For example, if the Time to Collision (TTC) is used as the traffic conflict index, the congestion state will be regarded as the most critical influencing factor. Under congestion conditions, due to the significant reduction in vehicle spacing, time-based conflict indexes (such as TTC, TIT, TET) usually identify congested traffic as the most important safety risk factor. However, as a distance-based conflict index, the Potential Stopping Distance (PSD) considers the importance of the congested traffic state to be relatively low because under low-speed driving conditions, the stopping distance of vehicles is longer. The medium congestion state or the transition state from non-congestion to congestion is the most critical stage for safety. Under the medium congestion state, due to the small vehicle spacing and the more frequent dynamic behaviors of traffic participants, the potential risk of conflicts will indeed increase significantly. In order to improve the operation stability and safety of dynamic traffic flow, a new method for evaluating operation safety risks is urgently needed, which can comprehensively consider characteristics such as traffic density, speed, lane change, and traffic flow instability to achieve a more comprehensive description of the relationship between macroscopic traffic flow and safety.

[0118] (1) Analysis of interactive safety between vehicles

[0119] Under traffic conditions with high homogeneity and lane compliance, the interaction between drivers is mainly reflected in the longitudinal dimension, such as adjusting the speed to follow the vehicle in front, while the interaction in the lateral dimension is relatively less. The lateral interaction between vehicles usually occurs during lane change or overtaking. This means that the reaction of the target vehicle is mainly affected by the immediately preceding vehicle. However, under traffic conditions with heterogeneity and no lane discipline, drivers not only interact in the longitudinal dimension but may also interact in the lateral dimension. In this case, the reaction of the target vehicle is affected not only by the leading vehicle but also by other surrounding vehicles, such as multiple leading vehicles, following vehicles, and laterally adjacent vehicles, and even by the action of other vehicles within the influence area of the target vehicle. The "influence area" of the target vehicle refers to a hypothetical area within which the surrounding traffic environment affects the reaction of the target vehicle. To deeply study this phenomenon, different sizes and shapes of influence areas have been explored and field investigated to obtain reasonable behavior predictions. The present invention adopts a fan-shaped influence area because during driving, the driver's field of vision is a fan-shaped area and the rearview mirror area, and the road boundary is easily approximated as the two sides of the fan-shaped influence area size.

[0120] Figure 1Shows the hypothetical influence area and the vehicles within it at a specific time step. With the target vehicle at the center position, the influence area of the target vehicle is divided into five regions: the left adjacent region, the right adjacent region, the front region, the left rear guiding region, and the right rear guiding region. Among them, the front region includes the leading region, the left front leading region, and the right front leading region. It should be noted that if the target vehicle moves left or right, the boundaries of the influence area will be correspondingly shortened. In addition, the influence area and the vehicles within it will be evaluated at each time step, so it is limited by the traffic environment at each time step. During the movement of the target vehicle at each time step, the combination of vehicles within the influence area will change accordingly.

[0121] (2)Construction of traffic safety operation domain index based on conflict time

[0122] Considering that the target vehicle interacts with other vehicles within the influence area simultaneously in the lateral and longitudinal dimensions, the target vehicle may conflict with surrounding vehicles in the lateral or longitudinal direction, or simultaneously in both directions. In this chapter, the conflict time is modeled and calculated through the newly constructed SOD index.

[0123] As Figure 2 shown, assuming that the target vehicle interacts with its left front leading vehicle, based on the longitudinal and lateral speeds of the target vehicle and the relative longitudinal and lateral spacings between the target vehicle and different vehicles within the influence area, the SOD at a certain time step (𝜏) will show different values in the lateral and longitudinal directions, and its mathematical representation is:

[0124] (18)

[0125] Where = , and are the distances between the target vehicle and the th vehicle in the lateral and longitudinal directions respectively, = , and represent the lateral position and longitudinal position of the vehicle. The subscript i represents corresponding to vehicle i, and the subscript target represents corresponding to the target vehicle. and are the lateral and longitudinal stopping distances, which generally refer to the distance required for the measured vehicle to completely stop to avoid collision, that is ,v represents speed, g represents gravitational acceleration, and f represents ground friction; where It is 3.92 m / s² in the longitudinal dimension and 1.47 m / s² in the lateral dimension. These values represent the deceleration values that the vehicle under test needs to achieve to avoid collisions. n is the number of vehicles in the influence area. In a congested state, due to the small vehicle spacing, the potential risk of conflicts will indeed increase significantly. The volatility and instability of the local traffic flow are crucial for the risk of traffic conflicts. Therefore, optimize SOD from the characteristics of traffic flow density.

[0126] When the traffic flow density suddenly increases, the volatility and instability of the local traffic flow, such as the target vehicle suddenly accelerating, etc., will bring a higher risk of traffic conflicts. The conflict risk caused by the temporary sudden behavior of the target vehicle is higher when the traffic density is high. Considering that it is impossible to obtain the behavior such as the speed change of the target vehicle in real time, but it can be represented by the relative speed change and relative spacing change, so the present invention measures this instability by calculating the relative speed change and relative spacing change between the target vehicle and the surrounding vehicles, as shown below:

[0127] (19)

[0128] Wherein, is the relative vehicle distance change between the target vehicle and the th vehicle at time , is the relative speed change between the target vehicle and the th vehicle at time . is the current vehicle distance between the target vehicle and the th vehicle. is the speed of the th vehicle.

[0129] In summary, in order to more comprehensively reflect the impact of the dynamic behavior of the target vehicle on the conflict risk in a high-density environment, introduce the instability characteristics brought by the sudden microscopic behavior of the target vehicle, and SOD is expressed as:

[0130] (20)

[0131] When the SOD value is high, it means that in the current traffic environment, the driving safety of the target vehicle is low, there is a greater conflict risk, which may be due to the increased instability caused by the excessive traffic density, or the increased uncertainty due to the failure to maintain a safe braking distance. When the SOD value is low, it means that the driving environment of the target vehicle is relatively stable, the surrounding traffic environment has little impact on safety, and the risk of conflicts is low. By designing and introducing the evaluation index of the safe operating domain of traffic (SOD) in this paper, traffic safety can be more comprehensively evaluated, especially in a complex traffic environment, and the accuracy and effectiveness of traffic safety management can be improved.

[0132] According to the theoretical analysis of the previous traffic flow safety structured analysis framework, there is an upper limit value of the safety threshold. When the SOD value exceeds this safety upper threshold, it means that the driving environment of the target vehicle is very close to or has already occurred a conflict state. The SOD upper limit is closely related to the stability and safety of the system. Considering random perturbations and uncertainties comprehensively, the upper limit condition of SOD can be obtained as and , which represent the expected safety and expected stability of the system respectively. Combining the relationship between speed and safety in the DSMPC framework, the upper limit of the safe operating domain is defined as:

[0133] (21)

[0134] where is the maximum safe operating area distance, and is a small tolerance probability, which can approach 0, representing a very low conflict probability.

[0135] Use to identify whether a vehicle conflict occurs. If in the interaction between the target vehicle and its surrounding vehicles, SOD is greater than or equal to , then the interacting vehicle and the target vehicle are marked as in conflict. The conflict calculation is:

[0136] (22)

[0137] Among all vehicles in the traffic flow, traffic conflicts are identified at each time step. Once a traffic conflict in the traffic flow is identified, the TSC value is calculated using formula (17).

[0138] Operating safety risk prediction model based on macroscopic traffic flow state:

[0139] According to the established traffic flow operation safety risk quantification calculation model, the safety risk value of a certain traffic flow can be calculated in real time. However, in traffic management practice, it is difficult to obtain the full amount of vehicle movement trajectory data and it cannot be used for real traffic flow safety control. According to the theoretical analysis of the traffic flow safety structured analysis framework, traffic flow density and speed are the key factors affecting the traffic flow safety state, and they are also common traffic flow parameters that are relatively easy to obtain in traffic management scenarios. Therefore, the present invention attempts to establish a prediction model for the traffic flow safety risk value based on macroscopic traffic flow parameters. Using a machine learning model to directly predict TSC from macroscopic traffic flow state parameters, its expression is as follows:

[0140] (23)

[0141] Machine learning models not only need to have high prediction accuracy, but also good generalization performance. Therefore, the selected machine learning models are evaluated from the following aspects:

[0142] a. Goodness of Fit (GoF): The mapping degree between the model prediction value and the actual value. For this criterion, R-squared and the scatter plot between TSC and density are used.

[0143] b. Measure of Performance (MoP): The accuracy of the model prediction is mainly evaluated using different errors, such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Square Error (MSE).

[0144] c. Model Reliability: The reliability of the model is mainly evaluated through the error distribution of the model.

[0145] d. Meaningful Predictability: The meaning of the model prediction results is mainly from the non-negative property of TSC. If the predicted SOD value is negative, the model suitability is considered poor.

[0146] Based on the preliminary analysis of the problem, the present invention selects three machine learning algorithms that have been widely verified in the field of traffic safety research, including Support Vector Machine (SVM), Random Forest Regression (RF), and Gradient Boosting (GB). By optimizing the hyperparameters of different machine learning models, the best hyperparameter values of different machine learning models are obtained using K-fold cross-validation and grid search techniques. At the same time, five-fold cross-validation and grid search method are used for hyperparameter tuning.

[0147] (1) Support Vector Machine (SVM) Algorithm

[0148] The basic idea of SVM is to convert the regression problem into a classification problem that can be handled by the support vector machine method. The core goal is to find a hyperplane to maximize the geometric margin between two types of data points. Therefore, the optimization objective is transformed into minimizing the norm of the weight vector:

[0149] (24)

[0150] The constraint condition is: . Where is the class label, and define the hyperplane. For non-linearly separable or noisy data, SVM introduces slack variables and regularization parameter C. The data expression of the optimization objective is:

[0151] (25)

[0152] The constraint condition is: . Therefore, the corresponding loss function is:

[0153] (26)

[0154] The solution to this optimization problem can be expressed by the method of Lagrange multipliers as follows:

[0155] (27)

[0156] Where, is the Lagrange multiplier. The present invention adopts a radial basis kernel function (RBF), and its expression is as follows:

[0157] (28)

[0158] Where, is a hyperparameter, and this hyperparameter determines the influence degree of a single sample on the training process.

[0159] (2)Random Forest (RF) algorithm

[0160] Random Forest (RF) belongs to the tree-based ensemble learning method, and improves the final prediction accuracy by combining multiple weak learners. In the Random Forest (RF) method, each tree is independently constructed by the bootstrap sampling method, which is called the Bagging method. In this process, the probability that each data point is selected in different random subsets is equal. Compared with a single decision tree, Random Forest has stronger prediction ability, mainly due to its improved node splitting strategy. In a standard decision tree, all features are considered when splitting nodes, while in Random Forest, each node is split only based on a randomly selected subset of features. There are many hyperparameters in the RF algorithm. Usually, the key hyperparameters affecting the model performance include the number of trees ( ), the minimum number of samples required for each leaf node and the maximum depth of the tree ( ).

[0161] (3)Gradient Boosting (GB) algorithm

[0162] The Gradient Boosting (GB) algorithm also belongs to the tree-based ensemble learning method. It constructs decision trees in a step-by-step iterative manner, and the results of each tree are used to improve the prediction performance of subsequent trees. The optimization objective of the GB algorithm is to minimize the prediction error. Its objective function is expressed as follows:

[0163] (29)

[0164] Where, represents the th decision tree, is the splitting variable, is the weight of the tree, which determines the influence degree of the tree on the final prediction result. The core of GB is to optimize to minimize the loss function between the predicted value and the true value. Its training process can be summarized into three steps: First, randomly initialize ; Then, construct each decision tree in turn, calculate the residuals of each sample, calculate the gradient based on the loss function, and use the residuals to fit a new decision tree estimate , and update the model predicted value ; Finally, achieve the optimal model performance. In the GB algorithm, the key hyperparameters affecting the model performance include the number of trees and the learning rate.

[0165] For the above technical solutions, relevant data examples of the present invention are used to conduct traffic flow operation safety simulation experiments and result analysis:

[0166] Dataset:

[0167] (1) Traffic condition: The road scene is set as an urban expressway with two-way four lanes, each lane is 3.5 meters wide, and the road length is 5 km; At the same time, the expected speed conforming to the Gaussian distribution is set; Limited by the vehicle dynamics engine version, the main vehicle types put into use are cars.

[0168] (2) Experimental data collection: Under different working conditions, as shown in Table 1, after setting different working condition parameters and scenarios based on the microscopic traffic safety simulation platform, Monte Carlo simulation experiments are carried out to generate traffic flow and vehicle movement trajectory data under each working condition.

[0169] Table 1

[0170]

[0171] For the technical solution provided by the present invention, traffic safety risks are evaluated simultaneously from both microscopic and macroscopic perspectives based on the simulated vehicle movement trajectory data. Macroscopic traffic flow characteristic parameters, namely traffic flow, speed and density, are extracted at a 5-minute aggregation interval. The vehicle movement trajectories are extracted. The vehicle trajectories are extracted at a resolution of 0.5 seconds, covering a 200-meter section of the road, and the total duration is 60 minutes.

[0172] Trajectory data preprocessing:

[0173] In the processing of vehicle motion trajectory data, due to possible local errors in the calculation process, there may be drastic fluctuations in instantaneous acceleration, deceleration, or speed values, which in turn affect the accuracy of the data. Therefore, the present invention needs to smooth the trajectory data to reduce noise. To reduce the noise in the data set, a seven-point moving average method is used to smooth the vehicle motion trajectory extracted by simulation, so as to reduce the fluctuations caused by errors in the calculation process. This method smooths the data by taking the average value of each point in the trajectory data and its six adjacent points, thereby effectively reducing the instantaneous fluctuations and improving the stability and reliability of the trajectory data. Specifically, for a given trajectory point , its smoothed value is calculated by the following formula:

[0174] (30)

[0175] where represents the coordinates of the (i - k)-th point in the trajectory. Through this smoothing process, high-frequency noise in the data can be effectively removed, the overall trend of the vehicle trajectory can be maintained, and the accuracy and reliability of subsequent analysis can be improved.

[0176] Analysis of traffic flow characteristics based on the traffic safety operation domain:

[0177] According to the established traffic flow operation safety risk quantification method, through the simulation experiment data, a brief analysis of the traffic flow safety characteristics is carried out. First, determine the limit value of the traffic safety operation domain. Calculate the traffic conflict time TSC based on the traffic safety operation domain, and the results are as Figure 3 shown. When CRF and DBF are less than 1.09, TSC is relatively gentle. When CRF and DBF continue to increase, TSC rises sharply. Therefore the upper limit value is set to 1.188. When SOD is higher than 1.188, it is considered that the risk of an accident rises sharply.

[0178] By calculating and statistically analyzing the simulation trajectory data, it can be seen that the safety risks are mainly concentrated in the medium-density area; as the density increases, the speed gradually decreases, and the low speed in the high-density area is accompanied by a decrease in risk. It should be noted that high-risk points are mostly distributed in the intersection area of medium density and medium-high speed, which may reflect that the interaction behaviors such as overtaking and lane-changing between vehicles are frequent in this state, resulting in an increase in the accident risk.

[0179] In addition, taking the traffic flow density of 30veh / h as an example, intercept the data related to the number of collision accidents and speed. The change trend line of the number of collision accidents with speed is as Figure 4As shown in the figure. When the average speed of the traffic flow is between 18 km / h and 30 km / h, the number of accidents shows a slight increase. After analyzing the vehicle trajectory data, it is found that when the speed of the vehicle in front is slow, a small number of vehicles choose to overtake and change lanes, resulting in collisions with low-speed vehicles and an increase in the number of accidents.

[0180] Therefore, it can be inferred that under specific traffic flow density conditions, there is a safe range of traffic flow speed with a relatively low accident rate, within which efficiency and safety are balanced.

[0181] Comparative analysis of the effectiveness of the traffic safety operation domain:

[0182] This part explores the relationship between the TSC results based on different conflict measurement indicators such as TET, TIT, and SOD and the macroscopic traffic flow state parameters, compares the relationships between different TSC values and macroscopic state parameters in the dataset, and different TSCs show different relationship patterns for macroscopic state parameters.

[0183] The TSC value obtained based on the TTC - type indicators has a significant relationship with density, and the range of the TSC size is different, mainly significant in the case of high density. This experimental conclusion can be explained by two mechanisms: First, in the case of high density, the vehicle spacing decreases, so vehicles are more likely to have TTC - based conflicts with a small speed difference; second, as the density increases, the total driving time of all vehicles in each spatio - temporal window also increases, which leads to the accumulation of more such conflict instances, ultimately resulting in an increase in the TTC - based TSC value. At the same time, in a highly congested state, the TSC value based on the TTC - type indicators may be unstable because the vehicle speed is low and the SOD of the vehicle is higher than the critical threshold, meaning that the vehicle is more likely to stop safely before a collision. In a highly congested state, the TSC value based on SOD is low, and even in this state, due to the increase in density, the total driving time of all vehicles in each spatio - temporal window will also increase.

[0184] As a conflict indicator based on events, the TTC - type indicators have certain limitations in their application scope, which makes the TSC based on the TTC - type indicators show obvious deficiencies in the following two aspects: First, the incidence rate, frequency, and severity of conflict events are greatly affected by the driver's following behavior, and the driving behavior characteristics often have significant regional differences; second, the TSC assessment based on TTC only targets a single following process in a specific following scenario and cannot comprehensively reflect the safety conditions during other driving periods in each spatio - temporal window. Therefore, although the TTC - type indicators perform well in evaluating the single - vehicle collision risk at the micro level, they have limitations in the evaluation of traffic flow operation safety risks.

[0185] When analyzing the relationship between SOD-based TSC and macroscopic state variables, the present invention finds that, compared with the correlation between other event-based TSCs and macroscopic state variables, the correlation of SOD-TSC is more significant and substantial. The experimental results show that the moderate congestion state is identified as one of the critical high-risk states due to its highest TSC value. This phenomenon is closely related to various complex traffic phenomena, such as increased speed fluctuations, traffic flow instability, and the propagation of stop-and-go traffic waves, which are widely regarded as the main factors leading to rear-end collisions. In addition, multiple studies further confirm that the moderate congestion state is one of the traffic states with the most prominent rear-end collision risks. Therefore, TSC calculated based on SOD has higher effectiveness and authenticity compared with traditional TTC-based alternative indicators when reflecting the real moderate congestion traffic state. SOD is essentially a non-event-driven metric, and its input parameters (including vehicle speed, vehicle distance, etc.) are significantly correlated with macroscopic traffic flow parameters (such as average speed and density). This characteristic enables the relationship pattern between the SOD-based TSC value and macroscopic traffic flow parameters to maintain good generalization performance in different traffic scenarios with similar flow-density curve characteristics.

[0186] In summary, from the perspective of quantitative evaluation and practical application, compared with TTC-based alternative evaluation indicators, the SOD-based traffic safety state identification and classification method has outstanding advantages: First, its quantitative characterization results can accurately reflect the essential characteristics of traffic flow; second, this indicator has better applicability and effectiveness in actual traffic control scenarios.

[0187] Performance analysis of traffic flow operation safety risk prediction model:

[0188] The present invention establishes a data-driven machine learning model and evaluates its prediction ability. As mentioned above, three machine learning algorithms are used to model TSC as a function of macroscopic traffic flow parameters (density, speed), namely random forest (RF), support vector machine (SVM), and gradient boosting (GB). These machine learning algorithms are implemented in the Python language, and the hyperparameters of each machine learning model are adjusted or optimized through k-fold cross-validation and grid search techniques. The optimal hyperparameter values are shown in Table 2.

[0189] Table 2

[0190]

[0191] Analysis of the error distribution of the prediction results shows that the error distribution of the RF model approximately follows a normal distribution, which mainly benefits from its unique feature randomness mechanism. During the model construction process, each decision tree only uses some features for training, thereby enhancing the diversity of the model and improving the robustness to noise data and redundant features.

[0192] Table 3 shows the results of the machine learning models on the training and test datasets. The experimental results in Table 3 indicate that the performance of all machine learning models is consistent on the training set and the test set. Through comparative analysis, it can be seen that among the three machine learning models of random forest (RF), support vector machine (SVM), and gradient boosting (GB), the RF model achieves the minimum values in evaluation metrics such as root mean square error, mean absolute error, and mean square error, indicating that its prediction performance is slightly better than other models.

[0193] Table 3

[0194]

[0195] Among them, RMSE represents the root mean square error; MAE represents the mean absolute error; MSE represents the mean square error.

[0196] To deeply evaluate the performance differences of different machine learning models, the present invention is verified by comparing and analyzing the correlation between the TSC values predicted by the models and the actual observed values. Specifically, based on the distribution characteristics of the TSC - density relationship and the TSC - speed relationship, a visual analysis of the relationship between the observed values and the predicted values is carried out. The results show that under the given traffic flow density and speed conditions, the TSC values predicted by the models are highly consistent with the actual observed values. It should be noted that although all the machine learning models constructed in the present invention show good consistency, the SVM and GB models predict negative TSC values in some cases, which neither conforms to the physical meaning nor has practical application value. In contrast, the RF model does not show negative values in all predictions, and this result further confirms the effectiveness and reliability of the RF model in prediction performance.

[0197] Empirical analysis of traffic flow operation safety risk assessment and intervention:

[0198] Empirical analysis of traffic safety operation domain:

[0199] Based on the MFTD dataset, the macroscopic traffic flow evaluation indicators SOD and TSC are calculated. When dividing the spatio - temporal window, an appropriate time dimension can capture the dynamic instability in the traffic flow, and a suitable space dimension can ensure that the window is large enough when calculating traffic variables to avoid noise. Therefore, the spatio - temporal plane is discretized into multiple spatio - temporal windows with sizes of and In the traffic scenario of Huaxiang Bridge on the South Fourth Ring Road in Beijing, according to the method proposed in this chapter, the macroscopic traffic flow safety risk assessment indicator TSC of the spatio - temporal window is calculated.

[0200] In the trend of TSC varying with traffic flow density and speed, the higher the TSC value, the higher the risk. For traffic flow density less than 300 veh / km, the traffic flow is in a free flow or metastable state, with relatively high vehicle speeds and strong discreteness. TSC increases with the increase in density. This is because the driver has insufficient reaction time to sudden events (such as sudden braking of the vehicle in front). Although the absolute distance is relatively large, TSC still rises with the increase in density, reflecting speed-dominated risks. In the moderately congested traffic state (the transition from non-congested to congested), the speed difference increases significantly, triggering frequent acceleration and deceleration behaviors. Since the moderately congested traffic state is related to a more chaotic traffic state, the collision risk increases. When the traffic flow density is greater than 500 veh / km, the traffic flow enters a fully congested state, and the vehicles run at extremely low speeds. Instead, the TSC value decreases, reflecting the risk mitigation phenomenon under the density saturation effect, which is consistent with the relevant research consensus and indirectly verifies the effectiveness of the SOD index.

[0201] Feasibility analysis of risk intervention measures:

[0202] Traffic safety control is one of the important responsibilities of traffic management departments. Taking intervention measures based on dynamic risk assessment and early warning is a feasible means to prevent traffic accidents. For example, in the medium-density traffic state, the risk level classification of the traffic state can be released to the society in real time. By using three colors, red, yellow, and green, to represent high, medium, and low risks, drivers can be prompted to improve their attention and adjust their driving behaviors. When the traffic flow becomes unstable, a gradient speed limit strategy can also be dynamically implemented to improve the operation state of the traffic flow. For example, the speed limit can be gradually reduced, such as from 80 km / h to 60 km / h, and then to 40 km / h.

[0203] To evaluate the feasibility and effectiveness of implementing risk intervention measures, among them, the baseline scenario replicates the aforementioned real traffic conditions, calculates the baseline values of SOD and TSC. The proportion of high-risk areas is 12.3%, and the average TSC value in high-risk areas is 22.52. In the intervention scenario, the gradient speed limit strategy is simulated, and the proportion of high-risk areas and the average TSC value are compared, and the comparison before and after the intervention measures is calculated. The experimental results are shown in Table 4. After the intervention, the proportion of high-risk areas drops to 8.5%, and the average TSC value in high-risk areas drops to 20.41.

[0204] Table 4

[0205]

[0206] The above results show that risk intervention measures can effectively reduce the proportion of high-risk areas, improve the stability of the traffic flow, and reduce the risk of traffic accidents. The experimental results demonstrate that traffic safety control based on the results of traffic flow operation safety risk assessment has broad application prospects.

[0207] Based on this, the present invention proposes a traffic safety analysis method integrating macroscopic characteristics. First, based on the traffic loss stability mechanism, a safety evaluation theoretical framework integrating information theory and cybernetics is constructed, a traffic safety limit hypothesis is proposed, and a distributed stochastic predictive control model for single-lane non-free flow is established to reveal the internal relationship between macroscopic parameters of traffic flow and safety risks. An innovative quantitative index of "Safety Operating Domain" (SOD) is proposed to realize the quantitative characterization of microscopic interaction risks. A macroscopic risk prediction model is constructed in combination with machine learning, and the optimal prediction performance of the random forest model is verified by comparing SVM, RF, and GB algorithms. Relying on the microscopic traffic safety simulation platform, multi-density-speed combination experiments are carried out to verify the rationality and effectiveness of the SOD index and the prediction model. Empirical research shows that the proposed evaluation method can effectively evaluate the safety situation of traffic flow, and the application value in dynamic safety control of traffic flow is verified by comparing intervention measures, providing theoretical methods and technical tool support for the "pre-event" prevention of traffic accidents by traffic management departments.

[0208] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A traffic flow operation safety risk assessment method, characterized in that, Including: Obtain the traffic density and traffic speed of the traffic flow; Obtain and, based on the vehicle distances and relative change information of different vehicles in the traffic flow, obtain the traffic safety operation index; Judge the traffic safety operation index, and calculate based on the judgment result to obtain the safety risk value; Fit the relationship between the traffic density and traffic speed and the safety risk value to obtain a traffic flow risk prediction model, and identify the traffic density and traffic speed obtained in real time through the traffic flow risk prediction model to obtain the traffic flow risk value monitored in real time.

2. The method according to claim 1, wherein: The process of obtaining the traffic safety operation index includes: Obtain distance interaction information according to the vehicle distance; obtain relative interaction information according to the relative change information; calculate the traffic safety operation index according to the distance interaction information and the relative interaction information.

3. The method according to claim 2, wherein: The process of obtaining the distance interaction information CRF includes: where n represents the number of vehicles in the influence area, and are respectively the lateral and longitudinal distances between the target vehicle and the th vehicle, and are the lateral and longitudinal parking distances. The parking distance , where v represents the speed, g represents the acceleration due to gravity, and f represents the ground friction.

4. The method according to claim 2, wherein: The process of obtaining the relative interaction information DBF(t) includes: Among them, is the relative vehicle distance change between the target vehicle and the th vehicle at time . is the relative speed change between the target vehicle and the th vehicle at time . is the current vehicle distance between the target vehicle and the th vehicle; is the speed of the th vehicle.

5. The method according to claim 2, wherein: Obtain the traffic safety operation index by calculating the product of the distance interaction information and the relative interaction information.

6. The method according to claim 1, wherein: The process of judging the traffic safety operation index includes: Judge the traffic safety operation index according to the upper threshold of the maximum safe operation domain. The upper threshold of the maximum safe operating range is as follows: Among them, is the maximum safe operating area distance, is the tolerance probability.

7. The method according to claim 6, wherein: The process of obtaining the safety risk value TSC(x,t) includes: Among them, is the time step, representing the resolution of the trajectory data; nT i is the total number of time steps that the i-th vehicle travels within the spatio-temporal window, Conf ij is the conflict discrimination index, used to determine whether the i-th vehicle is in a conflict state at the j-th moment; Among them, .

8. The method according to claim 1, wherein: The traffic flow risk prediction model adopts a deep learning model, wherein: the traffic flow risk prediction model is used to represent the fitting relationship between the traffic density and traffic speed and the safety risk value: Among them, TSC represents the safety risk value, represents the traffic flow risk prediction model, represents the traffic density, and V represents the traffic speed.

9. The method according to claim 8, wherein: The traffic flow risk prediction model adopts a support vector machine, a random forest regression model or a gradient boosting model.

10. The method according to claim 1, wherein: The process of obtaining the traffic density ρ and traffic speed V of the traffic flow includes: Among them, and represent the spatial and temporal sizes of the window, n is the number of vehicles within the window, tt i and Dt i are the travel time and travel distance of vehicle i, respectively.

Images