A multi-dimensional safety evaluation method and system based on driving behavior of freight vehicles

Abstract

The present application relates to the technical field of image recognition, and more particularly to a multi-dimensional safety evaluation method and system based on freight vehicle driving behavior, which comprises the following steps: obtaining a risk event sequence containing event types and severity; calculating time sequence coupling degree in combination with the time interval and logical correlation of adjacent risk events; calculating physical continuity based on the vehicle motion data within the time interval and the safety limit corresponding to the current vehicle speed; determining connection confidence by fusing the time sequence coupling degree and the physical continuity, and constructing a risk chain; determining risk gain according to the severity upgrading trend in the risk chain, and calculating an evolution index; and generating a driving safety score based on the evolution index by using a nonlinear function. The present application introduces a physical state verification and multi-dimensional linking mechanism, eliminates coincidental events, accurately evaluates the chain risk, and improves the accuracy of freight safety evaluation.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a multi-dimensional safety evaluation method and system based on the driving behavior of freight vehicles. Background Technology

[0002] With the rapid development of the logistics industry, the driving safety of freight vehicles has received increasing attention. Currently, mainstream commercial vehicle safety management systems mainly rely on driver monitoring systems and advanced driver assistance systems to identify abnormal events during driving, such as fatigued driving, distracted driving, lane departure, or following too closely.

[0003] To assess the driving safety of freight vehicles, existing evaluation methods typically employ statistically based linear summation algorithms or simple rule-matching algorithms. These algorithms usually pre-define basic penalty weights for various risk events. During driving, they perform simple linear weighted summation based on the frequency and severity of detected risk events, or simply correlate adjacent events based on a fixed time window threshold, thereby calculating the driver's safety score.

[0004] The aforementioned evaluation algorithms based on discrete event accumulation or fixed time thresholds have significant limitations in practical applications. First, linear accumulation algorithms ignore the inherent logical causality and evolutionary trends between risk events, isolating the continuous risk escalation process—such as abnormal driver physiological state leading to vehicle motion control instability—as isolated events, resulting in insufficient assessment of the severity of cascading risks. Second, relying solely on time intervals to determine event correlation is too one-sided, easily misjudging coincidental events with close temporal proximity but no actual causal relationship as strongly correlated, and failing to verify the authenticity of causal relationships through the vehicle's physical motion state. Consequently, safety scores cannot accurately reflect the actual nonlinear impact of driving behavior on road safety. Summary of the Invention

[0005] To address the technical problems of the aforementioned evaluation algorithms, which neglect the logical causal relationship and evolutionary trends between risk events and lack correlation verification based on vehicle physical motion data, thus making it difficult to accurately assess severe cascading risks and reflect the nonlinear impact of driving behavior on driving safety, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a multi-dimensional safety evaluation method based on the driving behavior of freight vehicles, the method comprising the steps of:

[0007] Several risk events during vehicle driving are acquired and sorted according to their occurrence time to construct a risk event sequence. Each risk event includes an event type and severity. The temporal coupling degree of adjacent risk events is determined based on the time interval between them and the logical correlation between event types. Vehicle motion data within the time interval of the adjacent risk events is acquired, and motion disturbance values ​​are calculated. The motion disturbance values ​​are processed using a safety limit determined by the vehicle's current speed to obtain the physical continuity of adjacent risk events. The connection confidence of adjacent risk events is calculated based on the temporal coupling degree and physical continuity. Adjacent risk events whose connection confidence meets a preset threshold are connected to obtain several risk chains. The sum of the severity of each risk event in the risk chain is recorded as the risk accumulation. The risk gain is determined based on the severity escalation trend and connection confidence of adjacent risk events in the risk chain. The evolution index of the risk chain is calculated based on the risk accumulation and risk gain. The driving safety score of the vehicle is calculated based on the evolution index of each risk chain.

[0008] This invention first utilizes temporal coupling to preliminarily screen potentially related events from both temporal and logical dimensions. Then, by calculating the physical continuity within time intervals and using vehicle motion data and safety limits based on current vehicle speed, it verifies the continuity between events from a physical state perspective, thereby accurately connecting adjacent risk events with causal relationships into a risk chain. Furthermore, when calculating the evolution index, this invention not only considers the severity of the risk event itself but also assesses the impact of risk escalation trends on safety through risk gain, ultimately calculating a driving safety score based on the evolution index. This can more realistically reflect the potential danger level of freight vehicles during dynamic driving, improving the accuracy of safety assessments and the ability to identify accident precursors.

[0009] Preferably, determining the temporal coupling degree of adjacent risk events based on the time interval between adjacent risk events and the logical correlation between event types includes: obtaining the logical correlation value of adjacent risk events from a preset logical correlation table; calculating the decay value of the time interval between adjacent risk events based on a Gaussian kernel function; and multiplying the logical correlation value by the decay value to obtain the temporal coupling degree of adjacent risk events.

[0010] This invention employs a combination of a pre-defined logical association table and a Gaussian kernel function to determine temporal coupling. The logical association table allows calculations to be performed only on specific event pairs with potential causal relationships based on prior knowledge, reducing computational redundancy and excluding logically impossible events. The Gaussian kernel function processes time intervals, utilizing its smooth, non-linear decay characteristics to simulate the natural law that the impact of risk gradually weakens over time. This allows for the assessment of the likelihood of association at different time intervals, ensuring that logically related events occurring consecutively within a short period are assigned higher coupling weights.

[0011] Preferably, the temporal coupling degree of the adjacent risk events Satisfying the relation:

[0012] ;

[0013] in, , They are the first The, the The moment a risk event occurs; It is the time decay value; It is the natural exponential function; It is the first The, the The logical connection between the risk events.

[0014] Preferably, the preset logical association table includes causal relationships between different event types; the event types include at least: prolonged visual deviation, closing eyes, yawning, and lane departure.

[0015] Preferably, the vehicle motion data includes steering wheel angular velocity and longitudinal jerk; the process of processing the motion disturbance value using the safety limit determined by the vehicle's current speed includes: constructing a lateral safety limit function inversely proportional to the vehicle speed; obtaining a preset longitudinal safety limit threshold; normalizing the steering wheel angular velocity using the lateral safety limit function; and normalizing the longitudinal jerk using the longitudinal safety limit threshold.

[0016] Preferably, the physical continuity of the adjacent risk events Satisfying the relation:

[0017] ;

[0018] in, , They are the first The, the The moment a risk event occurs; yes The steering wheel angular velocity at any given moment; yes The longitudinal jerk at any given moment; yes Vehicle speed at any given moment; yes The lateral safety limit function at time t; It is the longitudinal safety limit threshold; These are the horizontal feature weights; It is the absolute value symbol.

[0019] This invention utilizes an integral form to calculate the physical continuity within the interval between adjacent risk events, capturing the cumulative changes in vehicle motion state over the event interval, rather than just instantaneous peak values. By introducing lateral feature weights to weighted sum the normalized values ​​of steering wheel angular velocity and longitudinal jerk, the degree of attention to lateral instability can be flexibly adjusted according to the characteristics of freight vehicles. This can assess whether the vehicle remains in a non-stationary state between two adjacent risk events, thus corroborating the existence of a physical causal continuity between the two adjacent risk events.

[0020] Preferably, the connection confidence of the adjacent risk events Satisfying the relation:

[0021] ;

[0022] in, It is the first The, the The temporal coupling degree of risk events; It is the first The, the The physical continuity of risk events.

[0023] This invention calculates connectivity confidence by combining temporal coupling and physical continuity. The introduction of physical continuity modulates the temporal coupling, meaning that even if two events are highly correlated in time and logic, connectivity confidence will be suppressed if the vehicle's physical state remains extremely stable during their interval. This effectively avoids misjudging independent events that are merely temporal coincidences but have safe intermediate processes as continuous risk chains, reducing false alarm rates and improving the credibility of risk chain construction.

[0024] Preferably, the step of determining the risk gain based on the severity escalation trend and connectivity confidence of adjacent risk events in the risk chain, and calculating the evolution index of the risk chain based on the cumulative risk and the risk gain, includes: calculating the severity increment of adjacent risk events within the risk chain, and taking increments greater than zero as positive differences; accumulating the product of the connectivity confidence of each adjacent risk event within the risk chain and the positive differences, and multiplying the accumulation result by a preset risk constraint factor to obtain the risk gain; and recording the sum of the cumulative risk and the risk gain as the evolution index of the risk chain.

[0025] This invention introduces a risk gain based on the severity escalation trend when calculating the evolution index. By identifying the severity increments of adjacent events in a risk chain, it captures the process of risk deterioration and amplifies this trend by combining connectivity confidence. This approach results in a higher evolution index for risk chains that progressively escalate in severity. This non-linear assessment of risk evolution highlights the high sensitivity of early warning signals for accident precursor sequences.

[0026] Preferably, the step of calculating the vehicle's driving safety score based on the evolution index of each risk chain includes: dividing the sum of the evolution indices of all risk chains by a preset risk tolerance value to obtain a comprehensive risk coefficient; and multiplying the negative natural exponential function value of the obtained comprehensive risk coefficient by a preset full score value to obtain the driving safety score.

[0027] In a second aspect, the present invention provides a multi-dimensional safety evaluation system based on the driving behavior of freight vehicles. The multi-dimensional safety evaluation system based on the driving behavior of freight vehicles includes a memory and a processor. The memory stores computer program instructions. When the computer program instructions are executed by the processor, the multi-dimensional safety evaluation method based on the driving behavior of freight vehicles according to the first aspect of the present invention is implemented.

[0028] By adopting the above technical solution, a computer program is generated from the multi-dimensional safety evaluation method based on the driving behavior of freight vehicles according to the first aspect of the present invention, and stored in a memory so that it can be loaded and executed by a processor. A terminal device is then made based on the memory and the processor for convenient use.

[0029] The beneficial effects of this invention are as follows: First, by acquiring a sequence of risk events during vehicle driving, this invention calculates the connectivity confidence using both temporal coupling degree and physical continuity indicators, thereby constructing discrete risk events into a risk chain with causal relationships. By introducing a speed-based safety limit to normalize vehicle motion data, a unified standard for assessing vehicle physical stability under different driving conditions is ensured. Furthermore, this invention calculates an evolution index by analyzing the escalation trend of severity within the risk chain and uses a nonlinear function to generate a safety score, achieving an accurate assessment of the risk accumulation and deterioration process. Attached Figure Description

[0030] Figure 1 A flowchart illustrating a multi-dimensional safety evaluation method based on the driving behavior of freight vehicles, provided in an embodiment of the present invention;

[0031] Figure 2 This is a schematic representation of the logical associations provided in the embodiments of the present invention;

[0032] Figure 3 This is a structural block diagram of a multi-dimensional safety evaluation system based on the driving behavior of freight vehicles, provided as an embodiment of the present invention. Detailed Implementation

[0033] The first aspect of this invention provides a multi-dimensional safety evaluation method based on the driving behavior of freight vehicles, such as... Figure 1 As shown, the method includes steps S100-S500:

[0034] Step S100: Obtain several risk events during vehicle driving, sort them according to the time of occurrence of the risk events to construct a risk event sequence, and each risk event includes event type and severity.

[0035] It should be noted that, in order to accurately assess driving risks, the raw data collected by the vehicle-mounted terminal must first be transformed into risk units with unified definitions and dimensions. This step utilizes existing technology to uniformly define various abnormal states detected during driving as risk event sequences, providing a data foundation for subsequent calculations of the coupling relationships between events.

[0036] Specifically, firstly, in-vehicle sensing devices synchronously collect video streams from the in-vehicle DMS camera, the ADAS forward-facing camera, and the vehicle's CAN bus. Secondly, existing image processing, computer vision, and data processing algorithms are used to analyze the collected multi-source data and detect various types of abnormal events. For example, existing image processing algorithms are used to analyze the DMS video stream to detect abnormal events related to the driver's physiological state, such as prolonged gaze deviation, eye closure, and yawning; existing computer vision algorithms are used to analyze the ADAS video stream to detect abnormal events related to the vehicle's external environment, such as lane departure and following too closely; and existing data processing algorithms are used to analyze the CAN bus data to detect abnormal events related to vehicle motion, such as rapid acceleration, rapid deceleration, and sharp turns.

[0037] Finally, each detected anomaly is defined as a risk event. For each type of risk event, its severity is normalized based on its physical hazard and the time of occurrence to obtain a severity score. Taking a risk event as an example, it should contain at least three attributes: ,in, For the first The moment when a risk event occurs For the first The event type of each risk event, For the first The severity of each risk event.

[0038] At this point, the risk event sequence has been obtained.

[0039] Step S200: Determine the temporal coupling degree of adjacent risk events based on the time interval between adjacent risk events and the logical correlation between event types.

[0040] It should be noted that, unlike methods that simply accumulate events based on frequency or severity, this embodiment fully considers the logical relationship between risk events over time. In the driving scenario of freight vehicles, for example, there is a causal relationship between prolonged line-of-sight deviation and subsequent lane departure, which carries a higher risk than two isolated events. To assess whether such an inherent logical relationship exists between adjacent risk events, this step introduces temporal coupling degree. The calculation preferably uses a Gaussian kernel function to describe the nonlinear decay of causal relationship over time. This is because the Gaussian kernel function has a smooth nonlinear decay characteristic, avoiding the abrupt changes and inaccuracies caused by traditional fixed time thresholds. This ensures that the shorter the time interval, the higher the temporal coupling degree, thus more reasonably assessing the possibility of causal relationship.

[0041] First, the logical correlation between adjacent risk events is obtained. It should be noted that this logical correlation can be determined by whether the adjacent risk events have a causal relationship in driving behavior. If there is a causal relationship, the logical correlation is 1; if there is no causal relationship, the logical correlation is 0.

[0042] Taking prolonged visual deviation and lane departure as examples, there is a causal relationship between the two; prolonged visual deviation can lead to lane departure. Therefore, the logical connection between the two is as follows: In one feasible implementation, logical relationships can be obtained by querying a logical relationship table, such as... Figure 2 The logical relationship table shown in the diagram is only an example of some event types. In practice, the logical relationships between risk events need to be determined based on daily scenarios and implementation requirements.

[0043] Next, the temporal coupling degree of adjacent risk events is obtained. It should be noted that, to ensure that only logically related risk event pairs have their temporal proximity contributing to the temporal coupling degree, this invention preferably uses logical correlation to logically discriminate the results of the Gaussian kernel function. When the logical correlation between adjacent risk events is 0, regardless of the shortness of the time interval, the temporal coupling degree is cleared to zero. This effectively excludes logically unrelated event pairs, thereby ensuring that the final temporal coupling degree is an effective value based on causal relationship filtering.

[0044] Based on the above logic, adjacent risk events and Temporal coupling Satisfying the relation:

[0045] ;

[0046] in, , They are the first The, the The moment a risk event occurs; It is the time decay value; It is the natural exponential function; It is the first The, the The logical connection between the risk events.

[0047] In this relation, Used to determine whether there is a causal relationship between adjacent risk events and driving behavior; By time interval Nonlinear processing is used to evaluate the temporal closeness; this term ensures that the smaller the time interval, the closer the term is to the expected value. This indicates a higher degree of temporal coupling; conversely, the larger the time interval, the closer this term is to... This indicates that the two risk events are separated in time and have a weak causal relationship.

[0048] It should be noted that the time decay characteristic constant The settings need to be combined with the physical characteristics of the target vehicle and the characteristics of the driving environment. For example, in scenarios such as highways, the driver's reaction time window is usually shorter than that in urban roads, so adjustments are necessary. The size. In one feasible implementation, for heavy semi-trailers in highway scenarios, considering the typical time window from driver reaction, vehicle movement to risk manifestation, Intervals longer than one second are generally considered irrelevant events, therefore, they are preferred. This makes in At second intervals, the timing coupling decreases significantly.

[0049] At this point, the temporal coupling degree of adjacent risk events has been obtained.

[0050] Step S300: Obtain vehicle motion data within the time interval of the adjacent risk events, and calculate motion disturbance value based on it; process the motion disturbance value using the safety limit determined by the current vehicle speed to obtain the physical continuity of adjacent risk events.

[0051] It should be noted that judging causality solely based on time intervals has limitations and may erroneously classify coincidental events that are close in time as having a strong correlation. In actual driving, if a vehicle is physically in a continuously unstable state during the interval between one risk event and the next—for example, continuous steering wheel corrections or slight body movement—this physical instability is crucial evidence connecting causal relationships and supporting the existence of a risk chain. Therefore, this step introduces physical continuity to assess the average physical instability of the vehicle during the interval between adjacent events. To avoid the limitations of artificially setting fixed thresholds, this invention uses dynamic physical limits based on real-time vehicle speed to normalize motion disturbances, ensuring that this physical continuity is comparable at different vehicle speeds.

[0052] First, vehicle motion data within the time interval between adjacent risk events is obtained. It's important to note that lateral instability affects a vehicle's ability to maintain its lane and is crucial evidence for determining the evolution of risks such as distraction and fatigue into lane departures; longitudinal instability affects the vehicle's longitudinal attitude stability and is the basis for determining behavioral chains such as rapid acceleration and deceleration. Therefore, this step requires extracting lateral and longitudinal instabilities.

[0053] Specifically, the steering wheel angular velocity at various times is obtained through CAN bus data. It is used to characterize lateral instability; the longitudinal jerk is obtained by taking the first derivative of the acceleration acquired from CAN bus data. , used to characterize longitudinal instability.

[0054] Next, a dynamic safety limit function is constructed. It should be noted that, to ensure the physical disturbance index has the same hazard dimension at different vehicle speeds, a dynamic safety limit function is needed to normalize the vehicle motion data. This limit function reflects the vehicle's dynamic characteristics: the faster the vehicle speed, the smaller the vehicle's tolerance limit to steering wheel and acceleration / deceleration operations.

[0055] Specifically, the lateral safety limit function based on vehicle speed. ;in, yes The vehicle's real-time speed is obtained via CAN bus data. ; The lateral control gain constant is the maximum permissible angular velocity of the vehicle at low speeds, and in a preferred embodiment, it can be set to 4000. For the longitudinal direction, considering that within the normal driving speed range, the physical limits of the vehicle's longitudinal abruptness are mainly limited by the powertrain response capability and occupant comfort requirements, the longitudinal safety limit threshold... This is set as an empirical threshold that is weakly correlated with or approximately constant with vehicle speed. For example, it is typically taken as a reference to human vibration comfort standards and real-vehicle test data. This value can be calibrated according to the braking / drive system performance of a specific vehicle model.

[0056] Finally, to calculate the physical continuity, it should be noted that, in order to obtain the average physical instability over the time interval, this invention uses a weighted average method to integrate the normalized lateral and longitudinal disturbances and divide by the length of the time interval.

[0057] Based on the above logic, adjacent risk events and physical continuity Satisfying the relation:

[0058] ;

[0059] in, , They are the first The, the The moment a risk event occurs; yes The steering wheel angular velocity at any given moment; yes The longitudinal jerk at any given moment; yes Vehicle speed at any given moment; yes The lateral safety limit function at time t; It is the longitudinal safety limit threshold; These are the horizontal feature weights; It is the absolute value symbol.

[0060] In this relation, It is a lateral disturbance term, reflecting the safety margin consumed by the vehicle in lateral control, such as steering wheel operation; It is the longitudinal motion disturbance term, which reflects the longitudinal instability of the vehicle during rapid acceleration or deceleration. Used to balance the effects of lateral and longitudinal disturbances. This is used to calculate the average value of the integral result over a time interval, avoiding the incomparability of disturbance integral values ​​due to different durations of risk events, and ensuring that it can accurately reflect the average physical instability per unit time.

[0061] It should be noted that the horizontal feature weights The settings should be determined based on the type of risk event preference. Since freight vehicles travel at high speeds, lateral stability, such as lane drift and minor directional corrections, is often an early and critical indicator of loss of control due to fatigue or distraction; therefore, optimal settings are preferred. This assigns a higher weight to lateral perturbations.

[0062] At this point, the physical continuity of adjacent risk events has been obtained.

[0063] Step S400: Calculate the connection confidence of adjacent risk events based on the temporal coupling degree and physical continuity, and connect adjacent risk events whose connection confidence meets a preset threshold to obtain several risk chains.

[0064] It should be noted that the temporal coupling degree and physical continuity between adjacent risk events each reflect only one dimension of the causal relationship: temporal proximity or physical instability. To accurately determine whether a risk chain truly exists and effectively eliminate misjudgments caused by risk evolution trends that cannot be captured by traditional methods, this step uses physical continuity to modulate the temporal coupling degree. This effectively eliminates sporadic events that are close in time but have stable physical states, and reduces the weight of weakly correlated events caused by discontinuous physical processes. Ultimately, this provides a foundation of weights that have been verified by both spatiotemporal and physical factors for subsequent nonlinear risk accumulation calculations.

[0065] Based on the above logic, adjacent risk events and Connection confidence Satisfying the relation:

[0066] ;

[0067] in, It is the first The, the The temporal coupling degree of risk events; It is the first The, the The physical continuity of risk events.

[0068] In this relation, Evidence representing temporal relevance; Used to refine time-related evidence: when the physical instability of the vehicle is high during the interval, at this time... Approaching At that time, the term approaches This indicates enhanced temporal correlation; when the vehicle is physically stable, at this time... Approaching At that time, the term approaches This indicates that temporal correlation is suppressed, thus reducing the confidence of the connection.

[0069] The connection confidence of each adjacent risk event in the risk event sequence is obtained sequentially. A connection confidence threshold is set, and the risk event sequence is traversed: if the connection confidence of an adjacent risk event pair is greater than or equal to the connection confidence threshold, the two are considered to be continuous links in the same risk chain; if it is less than the connection confidence threshold, the chain is logically truncated at this point, thereby decomposing the entire risk event sequence into several independent risk chains with high confidence correlation.

[0070] At this point, several risk chains have been identified.

[0071] Step S500: The sum of the severity of each risk event in the risk chain is recorded as the risk accumulation. The risk gain is determined based on the severity escalation trend and connection confidence between adjacent risk events in the risk chain. The evolution index of the risk chain is calculated based on the risk accumulation and the risk gain. The driving safety score of the vehicle is calculated based on the evolution index of each risk chain.

[0072] It should be noted that the potential risks inherent in a complete risk chain, such as the progression from minor distraction to severe correction, are greater than the sum of their severity levels. Therefore, this step introduces a non-linear risk assessment mechanism. First, a risk chain evolution index is calculated, which can capture and amplify risk escalation patterns. Then, a non-linear mapping function is used to transform this index into a final safety score, ensuring that the final score accurately reflects the high-level risks corresponding to a complete risk chain.

[0073] First, the evolution index of each risk chain is calculated. It should be noted that, to ensure the risk chain index accurately reflects the risk escalation trend, this invention introduces a risk gradient gain term, incorporating the severity difference between adjacent risks into the calculation and multiplying it by the connectivity confidence level. This ensures that the risk chain evolution index only acquires a significant nonlinear gain when the causal relationship is clear and the risk escalates.

[0074] Based on the above logic, the first Evolution index of a risk chain Satisfying the relation:

[0075] ;

[0076] in, It is the first The total number of risk events in a risk chain; , It is the first The first in the risk chain The, the The severity of each risk event; It is the first The first in the risk chain The first risk event to the first The connection confidence of the first risk event, which is the first... Local connectivity confidence of a risk chain; It is a risk constraint value; It is a maximum value function.

[0077] In this relation, A simple summation representing the severity of all risk events within the risk chain; Part Two It is the risk gradient gain term, where, Used to calculate risk from arrive The upgrade gradient, only when The gradient is positive only when the gradient is multiplied by the connection confidence, and then multiplied by the risk constraint value. Accumulation is performed to assess high-risk chains.

[0078] It should be noted that the risk constraint value The settings are used to control the degree to which risk escalation affects the final score. The setup should ensure that the increase in the index value from a complete risk chain is greater than the simple sum of two independent, low-severity events. This invention preferably... It should also be noted that the evolution index of a risk chain represents the cumulative risk energy of the entire risk chain, rather than the normalized probability of a single event. Therefore, its value is usually greater than 1. For example, a chain containing three medium-to-high-risk events may have a base cumulative value exceeding 2.0. The higher this value, the greater the impact of the risk chain on driving safety.

[0079] Then, a nonlinear decay function is used to map the evolution index of the risk chain to obtain a safety score. It should be noted that, considering the nonlinear and abrupt nature of driving risks, this step utilizes the decay characteristics of a natural exponential function to avoid insufficient punishment for high-risk events in the scoring model. This exponential decay model places the total risk index in the exponential term of the denominator, ensuring that even a small increase in total risk leads to an exponential decrease in the safety score. This makes the punishment for a single high-risk chain greater than the linear accumulation of multiple minor events.

[0080] Based on the above logic, the driver's safety score Satisfying the relation:

[0081] ;

[0082] in, It is the first An evolution index of a risk chain; This is the maximum score, usually set to 100, but implementers can set it according to their needs. It is the risk tolerance value; It is the natural exponential function; It represents the total number of risk chains.

[0083] This formula utilizes the decay characteristics of the natural exponential function to achieve a non-linear mapping between risk and score. In the low-risk range, the score decays relatively slowly; however, as the total risk accumulates, especially when approaching the set tolerance limit... At that point, the scoring curve will drop sharply. This exponential suppression effect ensures accurate assessment of high-risk chains.

[0084] It should be added that the risk tolerance value It is a configurable parameter for the safety score, and its setting is used to align the model output sensitivity with the risk tolerance of actual working conditions. It directly controls the rate of decay of the rating curve and the degree of response to risk accumulation: The smaller the value, the steeper the model's penalty curve, and the stronger the system's suppression effect on high risks; conversely, the larger the value, the flatter the curve. The determination of the threshold relies on a critical risk threshold pre-determined through statistical analysis, which is then mapped to a preset low-risk critical score line. In this embodiment, to achieve high suppression performance on high-risk chains, a value of 3.5 is preferred. This value is obtained through calibration using actual operating data, ensuring that when the evolution index of the risk chain reaches 1.8, the score can be effectively suppressed below the low-risk critical line. Implementers can set this value according to their needs.

[0085] The second aspect of this embodiment provides a multi-dimensional safety evaluation system based on the driving behavior of freight vehicles, such as... Figure 3 As shown, the multi-dimensional safety evaluation system based on the driving behavior of freight vehicles includes a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the first aspect of the present invention, a multi-dimensional safety evaluation method based on the driving behavior of freight vehicles, is implemented.

[0086] The multi-dimensional safety evaluation system based on the driving behavior of freight vehicles also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0087] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (DRAM), high-bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0088] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A multi-dimensional safety evaluation method based on the driving behavior of freight vehicles, characterized in that, include: Several risk events during vehicle driving are acquired, and a risk event sequence is constructed by sorting the risk events according to the time of their occurrence. Each risk event includes the event type and severity. Based on the time interval between adjacent risk events and the logical correlation between event types, the temporal coupling degree of adjacent risk events is determined, including: obtaining the logical correlation value of adjacent risk events from a preset logical correlation table; calculating the decay value of the time interval between adjacent risk events based on the Gaussian kernel function; and multiplying the logical correlation value and the decay value to obtain the temporal coupling degree of adjacent risk events. Acquire vehicle motion data within the time interval between adjacent risk events and calculate motion disturbance values ​​based on them; vehicle motion data includes steering wheel angular velocity and longitudinal jerk. The motion disturbance value is processed using the safety limits determined by the vehicle's current speed, including: Construct a lateral safety limit function inversely proportional to vehicle speed; obtain a preset longitudinal safety limit threshold; normalize the steering wheel angular velocity using the lateral safety limit function, and normalize the longitudinal jerk using the longitudinal safety limit threshold; obtain the physical continuity between adjacent risk events. The following relation is satisfied: ； in, , They are the first The, the The moment a risk event occurs; yes The steering wheel angular velocity at any given moment; yes The longitudinal jerk at any given moment; yes Vehicle speed at any given moment; yes The lateral safety limit function at time t; It is the longitudinal safety limit threshold; These are the horizontal feature weights; It is the absolute value symbol; Calculate the connectivity confidence of adjacent risk events based on temporal coupling degree and physical continuity. The following relation is satisfied: ； in, It is the first The, the The temporal coupling degree of risk events; It is the first The, the The physical continuity of risk events; Connect adjacent risk events whose connection confidence levels meet a preset threshold to obtain several risk chains; The sum of the severity of each risk event in the risk chain is recorded as the risk accumulation. The risk gain is determined based on the severity escalation trend and connection confidence between adjacent risk events in the risk chain. The evolution index of the risk chain is calculated based on the risk accumulation and risk gain. The driving safety score of the vehicle is calculated based on the evolution index of each risk chain.

2. The multi-dimensional safety evaluation method based on freight vehicle driving behavior according to claim 1, characterized in that, The temporal coupling degree of adjacent risk events Satisfying the relation: ； in, , They are the first The, the The moment a risk event occurs; It is the time decay value; It is a natural exponential function; It is the first The, the The logical connection between the risk events.

3. The multi-dimensional safety evaluation method based on freight vehicle driving behavior according to claim 1, characterized in that, The preset logical association table includes causal relationships between different event types; the event types include at least: prolonged gaze deviation, closing eyes, yawning, and lane departure.

4. The multi-dimensional safety evaluation method based on freight vehicle driving behavior according to claim 1, characterized in that, The method of determining risk gain based on the severity escalation trend and connectivity confidence among adjacent risk events in the risk chain, and calculating the evolution index of the risk chain based on the accumulated risk and risk gain, includes: Calculate the severity increment of adjacent risk events within the risk chain, and take increments greater than zero as positive difference values; The risk gain is obtained by summing the product of the connection confidence of each adjacent risk event in the risk chain and the positive difference, and then multiplying the summation result by a preset risk constraint factor. The sum of the accumulated risk and the risk gain is denoted as the evolution index of the risk chain.

5. The multi-dimensional safety evaluation method based on freight vehicle driving behavior according to claim 1, characterized in that, The calculation of the vehicle's driving safety score based on the evolution index of each risk chain includes: The sum of the evolution indices of all risk chains is divided by the preset risk tolerance value to obtain the comprehensive risk coefficient. The driving safety score is obtained by multiplying the negative natural exponential function value of the obtained comprehensive risk coefficient with the preset full score value.

6. A multi-dimensional safety evaluation system based on the driving behavior of freight vehicles, characterized in that, The multi-dimensional safety evaluation system based on freight vehicle driving behavior includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a multi-dimensional safety evaluation method based on freight vehicle driving behavior according to any one of claims 1-5.

Images