A Dynamic Risk Assessment Method for Vehicle Transportation Based on Closed-Loop Coupling of FTA and Index Method

Abstract

This invention provides a dynamic risk assessment method for vehicle transportation based on a closed-loop coupling of FTA and the exponential method. It employs a multi-subjective and multi-objective weighting method to eliminate reliance on subjective experience and utilizes an exponential decay model to dynamically adjust static weights, thereby driving the construction of a dynamic risk quantification system. Simultaneously, it overcomes the static limitations of traditional fault trees by adding real-time data-driven dynamic risk perception capabilities through the exponential method. Furthermore, it combines the time-varying analysis capabilities of the Cox proportional hazards model with the logical structure of fault trees to construct a novel dynamic failure rate update framework. Finally, it introduces an adaptive threshold function to dynamically classify warning levels, analyze the contribution of each event in real time, accurately locate key risk transmission paths, and ultimately achieve a complete closed loop from real-time monitoring and accurate assessment to rapid tracing and handling.

Description

Technical Field

[0001] This invention belongs to the field of dynamic risk assessment technology for vehicle transportation, and particularly relates to a dynamic risk assessment method for vehicle transportation based on the closed-loop coupling of FTA and index method. Background Technology

[0002] Road transport safety is a core issue in the fields of traffic engineering and public governance, rooted in the fundamental role of the road system in national operations. Road transport accidents are characterized by both high frequency of exposure and high destructiveness, easily leading to significant casualties, property damage, and disruption of the logistics system, among other socio-economic risks. In the process of transport management, in addition to formulating systematic laws and regulations, it is also necessary to continuously monitor and scientifically assess the safety status of vehicle operations to understand risk trends and formulate corresponding accident prevention strategies, thereby achieving the safety governance goal of prevention first.

[0003] Traditional safety assessments, relying excessively on historical accident data and employing a single methodological approach, struggle to comprehensively capture dynamic risks, resulting in significant limitations in accuracy and foresight. While index methods and fault tree analysis (FTA) hold significant value in risk identification and quantification, they exhibit limitations when facing the increasing complexity of road transport and dynamic environmental changes. Index methods can quickly quantify risks and are highly operational, but they are subjective; FTA can rapidly locate the root cause of an accident and calculate the probability of the top event through logic gates, but it cannot dynamically respond to real-time changes, limiting probabilistic accuracy, and its high construction cost makes it difficult to meet the needs of precise assessment. The singular use of these two methods leads to risk assessment analysis easily remaining at the representational level, failing to reach the deeper causes of risks, resulting in a significant imbalance between analytical depth and coverage. The mechanical overlay of fault tree and index methods severely restricts the accuracy, evolution, and decision-making effectiveness of risk assessments.

[0004] Therefore, there is an urgent need for a vehicle dynamic risk assessment method to overcome the limitations of using a single analytical method, as well as to improve the dynamism, real-time performance, objectivity, and accuracy of the risk assessment system. Summary of the Invention

[0005] To address the shortcomings and deficiencies of existing technologies, a dynamic risk assessment method for vehicle transportation based on the closed-loop coupling of FTA and the index method is provided. This method can solve the limitations of using existing analysis methods in isolation and the problems of insufficient dynamism, real-time performance, objectivity, and accuracy of risk assessment systems.

[0006] To achieve the objectives of this invention, a dynamic risk assessment method for vehicle transportation based on closed-loop coupling of FTA and index method is provided, comprising the following steps:

[0007] S1, Data Preprocessing: Multi-source data such as vehicle driving, environment, vehicle freight, and historical accidents are standardized using a unified interface and converted into a unified internal format. Then, the unified format data is subjected to anomaly detection, missing data handling, and time-series alignment. Finally, a structured and standardized risk indicator feature set is generated through feature fusion.

[0008] S2, Risk indicator system determination: The standardized risk indicator feature set is used to form a risk indicator system through multiple rounds of structured consultation;

[0009] S3, Dynamic Weight Determination: A multi-subject / objective weighting mechanism is used to assign weights to the risk indicator system, and an exponential decay model is used for dynamic adjustment to obtain the dynamic weights. ;

[0010] S4, Dynamic Fault Tree Risk Assessment: Construct a three-layer fault tree topology consisting of top event, intermediate events, and bottom event, and apply dynamic weights. Probability of event occurrence The consequence value C is calculated, and the real-time data is transformed into the dynamic failure probability of the bottom event using the Cox proportional hazards model. Then, the probability P of the top event is calculated through the fault tree structure. Finally, the comprehensive risk value R is obtained through the composite model R=C×P.

[0011] S5, Early Warning Feedback: A fixed baseline threshold is set based on the comprehensive risk value R, and an adaptive threshold function T(t) is introduced to generate a dynamic early warning level with a real-time threshold setting; when an early warning response is triggered, the contribution of each event in the dynamic fault tree is analyzed in real time. Quickly locate the key risk events that trigger the early warning; then form the risk path contribution value of the minimum cut set from the key risk events. Identify the main risk paths and analyze the system's weaknesses; then, use Bayesian methods combined with fusion weights, real-time probabilities, and logical relationships to calculate the posterior probability of events and screen for core root causes; finally, conduct a comprehensive score for risk events to determine the final main risk paths and root causes.

[0012] The beneficial effects of this invention are:

[0013] Compared with existing technologies, this invention provides a dynamic risk assessment method for vehicle transportation based on a closed-loop coupling of FTA and exponential methods. It employs a multi-subjective and multi-objective weighting method to eliminate reliance on subjective experience and utilizes an exponential decay model to dynamically adjust static weights, thereby driving the construction of a dynamic risk quantification system. Simultaneously, it overcomes the static limitations of traditional fault trees by adding real-time data-driven dynamic risk perception capabilities through the exponential method. Furthermore, it combines the time-varying analysis capabilities of the Cox proportional hazards model with the logical structure of fault trees to construct a novel dynamic failure rate update framework. Finally, it introduces an adaptive threshold function to dynamically classify warning levels, analyze the contribution of each event in real time, accurately locate key risk transmission paths, and ultimately achieve a complete closed loop from real-time monitoring and accurate assessment to rapid tracing and handling. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the process of the present invention;

[0015] Figure 2 This is a schematic diagram of the data preprocessing process in this invention;

[0016] Figure 3 This is a schematic diagram of the topology of the fault tree in this invention;

[0017] Figure 4 This is a schematic diagram of the early warning feedback process in this invention. Detailed Implementation

[0018] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings:

[0019] according to Figure 1 As shown, this invention provides a dynamic risk assessment method for vehicle transportation based on closed-loop coupling of FTA and exponential method, including the following steps:

[0020] S1, Data Preprocessing: Multi-source data on vehicle driving, environment, vehicle freight, and historical accidents are standardized using a unified interface, converted to a unified internal system format, and then subjected to anomaly detection, missing data handling, and time-series alignment. Finally, feature fusion is used to generate a structured, standardized risk indicator feature set, specifically as follows... Figure 2 As shown.

[0021] The unified interface is a standard data access layer that converts different formats of multi-source data into a unified internal format (including timestamps, transmission protocols, units, etc.). The unified multi-source data then enters the parallel processing stage: based on... The system performs anomaly detection based on principles, IQR, and business logic verification. It also fills in missing values, aligns multi-source time-series data, and generates a structured, standardized risk indicator feature set through feature fusion.

[0022] S2, Risk Indicator System Determination: The standardized risk indicator feature set is transformed into a risk indicator system through multiple rounds of structured consultation. Specifically: In the first round, an open-ended consultation generates an initial risk indicator set; in the second round, the initial risk indicator set is summarized and classified, and multiple rounds of quantitative scoring and iterative feedback are conducted based on criteria such as importance, measurability, independence, and data availability; finally, structured risk indicators are obtained through statistical convergence.

[0023] S3, Dynamic Weight Determination: A multi-subject / objective weighting mechanism is used to assign weights to the risk indicator system, and an exponential decay model is used for dynamic adjustment to obtain the dynamic weights. The multi-subjective / objective weighting mechanism for risk indicators refers to the following: the Delphi method and the AHP method constitute a multi-subjective weighting method, the entropy weight method and the CRITIC method constitute a multi-objective weighting method, and the subjective and objective weights are combined into the final weights using the multiplicative combination method to obtain the static weights.

[0024] The specific content of the multi-subjective weighting method, which consists of the Delphi method and the AHP method, is as follows:

[0025] ① Delphi Law Empowerment

[0026] To assign weights to transportation risk indicators using the Delphi method, an expert panel first independently scores the questionnaire for the designed risk indicator system. The standard deviation, interquartile range, median, and mean of the statistical indicators are anonymously fed back to the expert panel for iterative revision until a stable consensus is reached or a convergence criterion is met. The convergence criterion refers to a significant decrease in the standard deviation or interquartile range of each indicator, and a stabilization of the median or mean of each indicator. Both conditions must be met simultaneously to achieve the convergence criterion.

[0027] The median of all indicators after convergence is used as the final consensus importance score for that indicator. The following formula is used to normalize the consensus scores of all indicators and calculate the weight of each indicator. :

[0028] ,

[0029] in, Let be the consensus importance score of risk indicator j, and m be the m determined risk indicators. The subjective weights of risk index j are obtained using the Delphi method.

[0030] ②AHP method for weighting

[0031] Construct an AHP hierarchical structure: target layer, criterion layer, and indicator layer (weighting is only applied to risk indicators in the indicator layer; the criterion layer is not included). Compare each pair of risk indicators in the indicator layer and construct a judgment matrix as shown in the following formula:

[0032] ,

[0033] in , Each represents a different indicator. This indicates the importance of indicator i compared to indicator j. A value of 1 indicates that the two indicators are of equal importance. >0 and .

[0034] Using the geometric mean method (i.e., the root mean method) to calculate weights can effectively eliminate the influence of extreme values, making the calculation results more stable. Among these methods... Take the nth root of the product of all elements in each row. To obtain the subjective weight of risk index j using the AHP method.

[0035] ,

[0036] The largest eigenvalue of the judgment matrix is ​​then determined using the following formula. :

[0037] ,

[0038] Where: W is the weight vector of the judgment matrix. Let i be the i-th element of the product of matrix A and vector W. It is the i-th element of the weight vector.

[0039] The consistency index (CI) is obtained using the following formula.

[0040] ,

[0041] The smaller the CI value, the better the consistency of the judgment matrix. According to Table 1 below, the average random consistency index RI is found by the matrix order. RI reflects the level of inconsistency that judgment matrices of the same order may produce under random conditions.

[0042]

[0043] The consistency ratio CR is then calculated using the following formula.

[0044] ,

[0045] When CR < 0.10, the consistency of the judgment matrix is ​​considered acceptable; when CR ≥ 0.10, the consistency of the judgment matrix is ​​considered unacceptable.

[0046] After the consistency check passes, the subjective weights of the AHP method are output. .

[0047] ③ Combination of Delphi weighting and AHP weighting

[0048] First, the coefficient of variation (CV) after convergence of the Delphi method is used as a measure of its uncertainty. The smaller the CV value, the higher the certainty; the coefficient of variation (CV) = standard deviation / mean. The consistency ratio (CR) calculated by the AHP method is used as a measure of uncertainty. The smaller the CR value, the higher the certainty.

[0049] Secondly, the deterministic scores of the Delphi method and the AHP method are calculated using the following formula.

[0050] ;

[0051] Normalizing the deterministic scores, the combined weights of the Delphi method and the AHP method are as follows: and As shown below:

[0052] ;

[0053] Finally, the final subjective weights are synthesized using the following weighted formula:

[0054] .

[0055] The entropy weight method and the CRITIC method constitute the specific content of the multi-objective weighting method as follows:

[0056] ① Entropy weighting method

[0057] The identified m risk indicators are labeled as follows: ,in Because the units of measurement for each risk indicator are inconsistent, the risk indicators need to be standardized to eliminate dimensions, resulting in the following values: ; Determine positive and negative indicators (the larger the positive indicator value, the better; the smaller the negative indicator value, the better), among which It is the original value of the i-th sample on the j-th indicator. and These are the maximum and minimum values ​​of the j-th indicator, respectively.

[0058] ;

[0059] Calculate the proportion of each risk indicator to the total indicator value. The formula is as follows:

[0060] ;

[0061] Calculate the information entropy value of each risk indicator. The formula is as follows:

[0062] ;

[0063] Information redundancy Calculating the objective weights of the entropy weight method The formula is as follows:

[0064] .

[0065] ② CRITIC Law Empowerment

[0066] The CRITIC method for determining weights primarily uses two indicators: volatility and conflict. Volatility, or contrast strength, is represented by the standard deviation of the data; a larger standard deviation indicates greater volatility and a higher weight. Conflict, or correlation, is represented by the correlation coefficient between the data; a higher correlation coefficient indicates less conflict and a lower weight.

[0067] Calculate the standard deviation of each indicator The larger the standard deviation, the greater the variability of the indicator across samples. The formula is as follows:

[0068] ,

[0069] The correlation coefficients between the indicators were calculated using the Pearson correlation coefficient. , The correlation coefficient between indicator j and indicator k is given by the following formula:

[0070] ,

[0071] The conflict measure reflects the degree of conflict between indicator j and other indicators. The conflict measure of indicator j with other indicators is calculated. ,when When the value is smaller and closer to 0, it indicates that the two indicators have more overlapping information and less conflict between them; conversely, when... A larger value, closer to 1, indicates strong complementarity between the two indicators and a high degree of conflict between them. The calculation formula is as follows (m is the number of indicators):

[0072] ,

[0073] The information content of each indicator is obtained by combining the intensity of contrast and the degree of conflict. ,

[0074] ,

[0075] Information content The larger the value, the greater the weight of the indicator. The amount of information calculated is normalized to obtain the objective weights of the CRITIC method. :

[0076] .

[0077] ③ Combination of entropy weight method and CRITIC method weight

[0078] The two methods are combined to form the final objective weight based on the confidence level:

[0079] .

[0080] S2.3, Dynamic weights are obtained by dynamically adjusting using an exponential decay model. The specific content is as follows:

[0081] The obtained subjective and objective weights are combined using the geometric mean of the multiplicative combination method:

[0082] ,

[0083] Historical accident data from different time periods has a significant impact on the system. To increase the accuracy of system evaluation, an exponential decay model is adopted to dynamically adjust the weights.

[0084] .

[0085] S4, Dynamic Fault Tree Risk Assessment: Construct a three-layer fault tree topology consisting of top event, intermediate events, and bottom event, and apply dynamic weights. Probability of event occurrence The consequence value C is calculated, and the Cox proportional hazards model is used to transform the real-time data into the dynamic failure probability of the bottom event. Then, the probability P of the top event is calculated through the fault tree structure. Finally, the comprehensive risk value R is obtained through the composite model R=C×P.

[0086] S4.1, the specific content of constructing a three-layer fault tree topology consisting of top events, intermediate events, and bottom events is as follows: vehicle transportation accidents and property damage are set as top events; human factors, cargo factors, weather factors, and road factors are set as intermediate events; and operational errors, violations, cargo attributes, loading defects, extreme precipitation, low visibility, severe adverse weather, abnormal temperature, road conditions, temporary interference, and traffic restrictions are set as bottom events.

[0087] Physical characteristics, chemical characteristics, fatigued driving, illegal lane changes, speeding, overloading, concealment of dangerous goods, uneven loading, excessive stacking, insecure fixing, heavy rain, heavy snow, freezing rain, dense fog, sandstorm, smog, thunderstorm, hail, typhoon, high temperature, extreme cold, potholes, sharp bends, steep slopes, narrow roads, tunnels, landslides, road obstacles, construction encroachment, and height restriction poles are designated as base events; as detailed in Table 2.

[0088]

[0089] like Figure 3 As shown, the top event, intermediate events, and bottom events form a three-layer fault tree topology from top to bottom, with the basic event serving as a transition between the intermediate and bottom events;

[0090] Among them: operational errors and violations in basic events are attributed to human factors in intermediate events; cargo attributes and loading defects in basic events are attributed to cargo factors in intermediate events; extreme precipitation, low visibility, severe adverse weather, and abnormal temperature in basic events are attributed to weather factors in intermediate events; and road conditions, temporary interference, and traffic restrictions in basic events are attributed to road factors in intermediate events.

[0091] Basic operational errors include fatigued driving, illegal lane changes, and speeding; basic violations include overloading and concealment of dangerous goods; basic cargo properties include physical and chemical characteristics; basic loading defects include uneven loading, excessive stacking, and insecure securing; basic extreme precipitation includes heavy rain, blizzards, and freezing rain; basic low visibility includes dense fog, sandstorms, and smog; basic severe adverse weather includes thunderstorms, hail, and typhoons; basic temperature anomalies include high temperatures and extreme cold; basic road conditions include potholes, sharp bends, steep slopes, narrow roads, and tunnels; basic temporary disturbances include landslides, road obstacles, and road construction; and basic traffic restrictions include height restriction barriers.

[0092] The physical properties of bottom-level events include fragile items and liquids, while the chemical properties of bottom-level events include flammability and explosiveness.

[0093] S4.2, dynamic weights Probability of event occurrence The specific content of the consequence value C obtained from the calculation is as follows:

[0094] Dynamic weights As input to the fault tree logic gates, starting from the bottom event, the weighted logic gate formulas are used to calculate upwards level by level until the top event. The weighted logic gate formulas are as follows:

[0095] .

[0096] in, The dynamic relative importance weights of input events within the logic gate. Let be the probability of occurrence of the basic event j after the update.

[0097] S4.3 uses the Cox proportional hazards model to transform real-time data into dynamic failure probabilities of bottom events, and then calculates the specific content of the top event probability P through a fault tree structure:

[0098] To improve the accuracy of risk prediction, the Cox model is used to update the bottom-event failure rate in real time, and the formula is as follows:

[0099] ,

[0100] in: It is a risk characteristic covariate vector. ; It is the vehicle at time t in the covariate vector The failure rate function is as follows; It is the baseline failure rate function; It is a vector of regression coefficients. This indicates the degree of influence of the covariate on the failure rate;

[0101] ① Risk characteristic covariate vector First, the underlying events are analyzed to obtain associated risk information and monitored in real time, collecting real-time data. Then, random forest importance screening is used to identify covariates that significantly influence risk prediction. Next, the data is preprocessed, handling missing and outlier values, and standardized or normalized to ensure data quality and comparability. Time series features are extracted from the variables to enhance their expressive power. Finally, the processed covariates are integrated into an ordered numerical vector, where each element represents the value of a specific covariate; this is the risk feature covariate vector, denoted as [missing information]. ;

[0102] ②Regression coefficient vector Because the model contains nonparametric parameters Then, the partial likelihood method is used to estimate the regression coefficient vector. Thus, the actual regression coefficient estimate is obtained. :

[0103] First, sort all risk events by their time points in ascending order, and denote them as follows: Then the partial likelihood function ,

[0104] ,

[0105] In the above formula, the numerator represents the risk score of an event, and the denominator represents the sum of the risk scores of the entire risk set; where... For the j-th event time point Risk set at any moment;

[0106] Then maximize this part of the likelihood function Obtain the regression coefficient estimate ,

[0107] ;

[0108] ③ Baseline Failure Rate Function Using regression coefficient estimators for For each covariate, a failure score is calculated, and the cumulative baseline risk function is derived using the Breslow estimator. :

[0109] ,

[0110] In the above formula, the denominator represents the sum of the fractional regression coefficient estimates for the entire risk set; where... For the j-th event time point Risk set at any moment

[0111] Then, a stepped baseline failure rate function is obtained by accumulating these steps. ;

[0112] ④ Cumulative Failure Rate Function The failure rate obtained from the Cox model is connected to the fault tree of the risk assessment model, thereby transforming the dynamic failure rate into a probability; through the failure rate function... Calculate the cumulative failure rate function ,

[0113] ,

[0114] ⑤ Top event probability P: First, the cumulative rate failure function Converted to failure probability ,

[0115] ,

[0116] Then the failure probability As input to the bottom event of the fault tree, the probability P of the top event is calculated from bottom to top.

[0117] .

[0118] S5, Early Warning Feedback: (e.g., ...) Figure 4As shown, a fixed baseline threshold is set based on the comprehensive risk value R, and an adaptive threshold function T(t) is introduced to generate a dynamic early warning level with a real-time threshold setting. When an early warning response is triggered, the contribution of each event in the dynamic fault tree is analyzed in real time. Quickly locate the key risk events that trigger the early warning; then form the risk path contribution value of the minimum cut set from the key risk events. The process involves identifying key risk paths and analyzing system weaknesses. Then, using Bayesian methods combined with fusion weights, real-time probabilities, and logical relationships, the posterior probabilities of events are calculated to screen for core root causes. Finally, risk events are comprehensively scored to determine the final key risk paths and root causes. Specifically:

[0119] S4.1, the main contents of dynamic early warning level setting are:

[0120] Four fixed threshold levels—blue, yellow, orange, and red—are preset as baseline levels. The adjustment coefficient output by the adaptive threshold function T(t) is combined with the set baseline thresholds to generate a dynamic threshold U that fluctuates with the risk situation. This allows the system to intelligently trigger corresponding level warnings based on real-time risk values. The adaptive threshold function T(t) is shown below:

[0121] ,

[0122] in, This is the base value for the adaptive threshold function. This is the sensitivity coefficient. E represents the weight of high-weight event i, and E represents the urgency of event i.

[0123] The dynamic threshold U is obtained by solving the following formula:

[0124] Dynamic threshold U = fixed threshold × T(t),

[0125] The response threshold for the early warning level is determined based on the dynamic threshold U. When the risk value R ≥ the dynamic threshold U, the response time is determined accordingly. The higher the early warning level, the shorter the response time. The early warning levels are shown in Table 3.

[0126]

[0127] S4.2, When the warning response is triggered,

[0128] First, based on the risk contribution of each event Identifying key risk events refers to the process of selecting key events that significantly contribute to the current risk level when a risk value R > dynamic threshold U triggers an early warning response. The event contribution and selection criteria are as follows:

[0129] ,

[0130] in, Contribution to event i in real time For the weight of event i, Let i be the probability of event i occurring at time t. The severity coefficient of the consequences of event i (1-10). The screening coefficient is (0.1-0.2).

[0131] Secondly, risk path contribution values ​​are formed by minimizing cut sets based on key events. Identifying the primary risk paths refers to: forming minimal cut sets based on identified key risk events, and contributing the identified minimal cut sets according to the risk paths. Sort the values ​​in descending order to identify the main risk paths. Calculate the contribution of each path containing the key risk event to the total risk using the following formula:

[0132] ,

[0133] in, The contribution of path j, Here, h is the time decay factor, h is the decay coefficient (0.05-0.1), and t is the current time. The first time the event was detected;

[0134] Then, Bayesian inference is used to obtain the root cause posterior probability of key risk events. This refers to:

[0135] Bayesian inference is performed on the events in the main risk paths to calculate the posterior probability of each risk event as a root cause. The formula used is as follows:

[0136] ,

[0137] in, Let i be the probability that event i is the root cause under the warning conditions. Let be the conditional probability that an early warning response is triggered when event i occurs. This is the weighting adjustment coefficient. Let i be the risk value of event i; and the prior probability. The real-time probability of event i occurring at time t. Probability of Evidence The overall risk value R(t) for triggering an early warning for the entire system;

[0138] Finally, based on the contribution of event risk and posterior probability of root cause Conducting a comprehensive assessment to determine the final risk path and root causes refers to:

[0139] Overall event rating: ,

[0140] Root cause determination criteria: ,

[0141] in, The contribution is weighted by the posterior probability (usually taken as 0.6), and max(D) is the maximum contribution of the event. This is the overall scoring threshold (usually set to 0.7). This is the posterior probability threshold (usually set to 0.8).

[0142] The above embodiments are not limited to the technical solutions of the embodiments themselves, and the embodiments can be combined with each other to form new embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the technical solutions of the present invention.

Claims

1. A dynamic risk assessment method for vehicle transportation based on closed-loop coupling of FTA and exponential method, characterized in that: Includes the following steps: S1, Data Preprocessing: Multi-source data such as vehicle driving, environment, vehicle freight, and historical accidents are standardized using a unified interface and converted into a unified internal format. Then, the unified format data is subjected to anomaly detection, missing data handling, and time-series alignment. Finally, a structured and standardized risk indicator feature set is generated through feature fusion. S2, Risk indicator system determination: The standardized risk indicator feature set is used to form a risk indicator system through multiple rounds of structured consultation; S3, Dynamic Weight Determination: A multi-subject / objective weighting mechanism is used to assign weights to the risk indicator system, and an exponential decay model is used for dynamic adjustment to obtain the dynamic weights. ; S4, Dynamic Fault Tree Risk Assessment: Construct a three-layer fault tree topology consisting of top event, intermediate events, and bottom event, and apply dynamic weights. Probability of event occurrence The consequence value C is calculated, and the real-time data is transformed into the dynamic failure probability of the bottom event using the Cox proportional hazards model. Then, the probability P of the top event is calculated through the fault tree structure. Finally, the comprehensive risk value R is obtained through the composite model R=C×P. S5, Early Warning Feedback: A fixed baseline threshold is set based on the comprehensive risk value R, and an adaptive threshold function T(t) is introduced to generate a dynamic early warning level with a real-time threshold setting; When an early warning response is triggered, the contribution of each event in the dynamic fault tree is analyzed in real time. Quickly locate key risk events that trigger early warnings; Then, the risk path contribution value is formed by the minimum cut set of key risk events. Identify the main risk paths and analyze the system's weaknesses; then, use Bayesian methods combined with fusion weights, real-time probabilities, and logical relationships to calculate the posterior probability of events and screen for core root causes; finally, conduct a comprehensive score for risk events to determine the final main risk paths and root causes.

2. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 1, characterized in that: The unified interface in S1 is a standard data access layer that converts different formats of multi-source data into a unified internal format. The unified multi-source data then enters the parallel processing stage. The system performs anomaly detection based on principles, IQR, and business logic verification. It also fills in missing values, aligns multi-source time-series data, and generates a structured, standardized risk indicator feature set through feature fusion.

3. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 2, characterized in that: The risk indicators formed by the multi-round structured consultation in S2 refer to: In the first round, an open consultation was conducted to generate an initial set of risk indicators; In the second round, the initial risk indicator set is summarized and classified, and multiple rounds of quantitative scoring and iterative feedback are conducted based on criteria such as importance, measurability, independence and data availability. Ultimately, a structured risk indicator is obtained through statistical convergence.

4. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 3, characterized in that: The multi-subjective / objective fusion weighting mechanism in S3 assigns weights to risk indicators as follows: the Delphi method and AHP method constitute a multi-subjective weighting method, the entropy weight method and CRITIC method constitute a multi-objective weighting method, and the subjective and objective weights are combined into the final weights by using the multiplicative combination method and then assigned to obtain static weights.

5. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 4, characterized in that: The dynamic adjustment of the exponential decay model in S3 refers to: using the exponential decay model to adjust the static weights in a timely manner to obtain dynamic weights.

6. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 5, characterized in that: The construction of a three-layer fault tree topology consisting of top events, intermediate events, and bottom events in S4 refers to: Vehicle transport accidents and property damage are designated as top events; Human factors, cargo factors, weather factors, and road factors are set as intermediate events; The basic events are set as operational errors, violations, cargo attributes, loading defects, extreme precipitation, low visibility, severe adverse weather, abnormal temperature, road conditions, temporary interference, and traffic restrictions. Physical properties, chemical properties, fatigued driving, illegal lane changes, speeding, overloading, concealment of dangerous goods, uneven loading, excessive stacking, insecure fixing, heavy rain, heavy snow, freezing rain, dense fog, sandstorm, smog, thunderstorm, hail, typhoon, high temperature, extreme cold, potholes, sharp bends, steep slopes, narrow roads, tunnels, landslides, road obstacles, construction encroachment, and height restriction poles are designated as base events. The top event, intermediate events, and bottom events are arranged in a three-layer fault tree topology from top to bottom, with the basic event serving as a transition between the intermediate and bottom events; Among them: operational errors and violations in basic events are attributed to human factors in intermediate events; cargo attributes and loading defects in basic events are attributed to cargo factors in intermediate events; extreme precipitation, low visibility, severe adverse weather, and abnormal temperature in basic events are attributed to weather factors in intermediate events; and road conditions, temporary interference, and traffic restrictions in basic events are attributed to road factors in intermediate events. Basic operational errors include fatigued driving, illegal lane changes, and speeding; basic violations include overloading and concealment of dangerous goods; basic cargo properties include physical and chemical characteristics; basic loading defects include uneven loading, excessive stacking, and insecure securing; basic extreme precipitation includes heavy rain, blizzards, and freezing rain; basic low visibility includes dense fog, sandstorms, and smog; basic severe adverse weather includes thunderstorms, hail, and typhoons; basic temperature anomalies include high temperatures and extreme cold; basic road conditions include potholes, sharp bends, steep slopes, narrow roads, and tunnels; basic temporary disturbances include landslides, road obstacles, and road construction; and basic traffic restrictions include height restriction barriers. The physical properties of bottom-level events include fragile items and liquids, while the chemical properties of bottom-level events include flammability and explosiveness.

7. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 6, characterized in that: The dynamic weights in S4 Probability of event occurrence The calculated consequence value C refers to: Dynamic weights As input to the fault tree logic gates, starting from the bottom event, the weighted logic gate formulas are used to calculate upwards level by level until the top event. The weighted logic gate formulas are as follows: ， in, The dynamic relative importance weights of input events within the logic gate. Let be the probability of occurrence of the basic event j after the update.

8. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 7, characterized in that: In step S4, the Cox proportional hazards model is used to transform real-time data into dynamic failure probabilities of bottom events, and then the probability P of the top event is calculated using a fault tree structure. To improve the accuracy of risk prediction, the Cox model is used to update the bottom-event failure rate in real time, and the formula is as follows: ， in: It is a risk characteristic covariate vector. ; It is the vehicle at time t in the covariate vector The failure rate function is as follows; It is the baseline failure rate function; It is a vector of regression coefficients. This indicates the degree of influence of the covariate on the failure rate; ① Risk characteristic covariate vector First, the underlying events are analyzed to obtain associated risk information and monitored in real time, collecting real-time data. Then, random forest importance screening is used to identify covariates that significantly influence risk prediction. Next, the data is preprocessed, handling missing and outlier values, and standardized or normalized to ensure data quality and comparability. Time series features are extracted from the variables to enhance their expressive power. Finally, the processed covariates are integrated into an ordered numerical vector, where each element represents the value of a specific covariate; this is the risk feature covariate vector, denoted as [missing information]. ; ②Regression coefficient vector Because the model contains nonparametric parameters. Then, the partial likelihood method is used to estimate the regression coefficient vector. Thus, the actual regression coefficient estimate is obtained. : First, sort all risk events by their time points in ascending order, and denote them as follows: Then the partial likelihood function , ， In the above formula, the numerator represents the risk score of an event, and the denominator represents the sum of the risk scores of the entire risk set; where... For the j-th event time point Risk set at any moment; Then maximize this part of the likelihood function Obtain the regression coefficient estimate , ； ③ Baseline Failure Rate Function Using regression coefficient estimators for For each covariate, a failure score is calculated, and the cumulative baseline risk function is derived using the Breslow estimator. : ， In the above formula, the denominator represents the sum of the fractional regression coefficient estimates for the entire risk set; where... For the j-th event time point Risk set at any moment Then, a stepped baseline failure rate function is obtained by accumulating these steps. ; ④ Cumulative Failure Rate Function The failure rate obtained from the Cox model is connected to the fault tree of the risk assessment model, thereby transforming the dynamic failure rate into a probability; through the failure rate function... Calculate the cumulative failure rate function , ， ⑤ Top event probability P: First, the cumulative rate failure function Converted to failure probability , ， Then the failure probability As input to the bottom event of the fault tree, the probability P of the top event is calculated from bottom to top. 。 9. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 8, characterized in that: The dynamic early warning level setting in S5 refers to: Four fixed threshold levels—blue, yellow, orange, and red—are preset as baseline levels. The adjustment coefficient output by the adaptive threshold function T(t) is combined with the set baseline thresholds to generate a dynamic threshold U that fluctuates with the risk situation. This allows the system to intelligently trigger corresponding level warnings based on real-time risk values. The adaptive threshold function T(t) is shown below: ， in, This is the base value for the adaptive threshold function. This is the sensitivity coefficient. E represents the weight of high-weight event i, and E represents the urgency of event i. The dynamic threshold U is obtained by solving the following formula: Dynamic threshold U = fixed threshold × T(t), The response threshold for the early warning level is determined based on the dynamic threshold U. When the risk value R ≥ the dynamic threshold U, the response time is shorter. The higher the early warning level, the shorter the response time.

10. The method for dynamic risk assessment of vehicle transportation based on closed-loop coupling of FTA and exponential method according to claim 9, characterized in that: When the warning response is triggered in S5, First, based on the risk contribution of each event Identifying key risk events refers to the process of selecting key events that significantly contribute to the current risk level when a risk value R > dynamic threshold U triggers an early warning response. The event contribution and selection criteria are as follows: ， in, Real-time contribution to event i For the weight of event i, Let i be the probability of event i occurring at time t. The severity coefficient of the consequences of event i (1-10). The selection coefficient is (0.1-0.2). Secondly, risk path contribution values ​​are formed by minimizing cut sets based on key events. Identifying the primary risk paths refers to: forming minimal cut sets based on identified key risk events, and contributing the identified minimal cut sets according to the risk paths. Sort the values ​​in descending order to identify the main risk paths; calculate the contribution of each path containing the key risk event to the total risk using the following formula: ， in, The contribution of path j, Here, h is the time decay factor, h is the decay coefficient (0.05-0.1), and t is the current time. The first time the event was detected; Then, Bayesian inference is used to obtain the root cause posterior probability of key risk events. This refers to: Bayesian inference is performed on the events in the main risk paths to calculate the posterior probability of each risk event as a root cause. The formula used is as follows: ， in, Let i be the probability that event i is the root cause under the warning conditions. Let be the conditional probability that an early warning response is triggered when event i occurs. This is the weighting adjustment coefficient. Let i be the risk value of event i; and the prior probability. The real-time probability of event i occurring at time t. Probability of Evidence The overall risk value R(t) for triggering an early warning for the entire system; Finally, based on the contribution of event risk Root cause posterior probability Conducting a comprehensive assessment to determine the final risk path and root causes refers to: Overall event rating: , Root cause determination criteria: , in, The contribution is weighted by the posterior probability (usually taken as 0.6), and max(D) is the maximum contribution of the event. This is the overall scoring threshold (usually set to 0.7). This is the posterior probability threshold (usually set to 0.8).

Images