Comprehensive quantitative evaluation method for operation risk of automatic driving commercial vehicle

Abstract

The invention relates to the technical field of intelligent automobiles, and particularly discloses an automatic driving commercial vehicle operation risk comprehensive quantitative evaluation method comprising the following steps: obtaining evaluation information of an automatic driving commercial vehicle; calculating an evaluation score of the autonomous operating vehicle based on the evaluation information; acquiring a risk level corresponding to the assessment score, taking assessment information corresponding to the assessment score of the same risk level as target information, and generating a target vector based on the target information of the single automatic driving commercial vehicle; and clustering the target vectors to obtain clusters, obtaining representative vectors in a single cluster, generating a disposal strategy based on the representative vectors, and executing the disposal strategy on the automatic driving commercial vehicles corresponding to all the target vectors in the cluster. According to the invention, the accuracy and comprehensiveness of risk assessment and the refinement degree of handling after risk assessment are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent vehicle technology, and in particular to a comprehensive quantitative assessment method for the operational risks of autonomous driving commercial vehicles. Background Art

[0002] Driven by both technological progress and commercial value, autonomous vehicles are gradually moving from testing and demonstration to commercial applications, and the commercialization process of autonomous vehicles is gradually accelerating.

[0003] With the continuous iteration of autonomous driving technology and the continuous expansion of application scenarios, the operational risks of autonomous driving vehicles have shown new characteristics such as complexity and dynamism, which have put forward higher requirements for the accuracy of risk assessment and the targeted handling.

[0004] In existing technologies, most of the processing is based on the scores obtained after parameter quantification. For example, if a certain score corresponds to a low-risk situation, routine monitoring may be carried out. However, this approach ignores the differences in risk composition. Although the scores are the same, the specific causes of the risks may be significantly different. Treating them uniformly will lead to a lack of accuracy in risk management and make it difficult to reduce risks at the root. Second, it does not consider transportation risks, but focuses more on driving risks. It lacks operational characteristics and makes it difficult to accurately assess the risks of operating vehicles. Third, it does not fully utilize accident information to correct risk situations. Summary of the Invention

[0005] The purpose of the present invention is to provide a comprehensive quantitative assessment method for the operating risks of autonomous driving commercial vehicles to solve the above technical problems.

[0006] The purpose of the present invention can be achieved through the following technical solutions: The comprehensive quantitative assessment method for the operational risk of autonomous driving commercial vehicles includes the following steps: Obtain assessment information on autonomous driving operating vehicles, including risk information and accident information. Risk information includes risk type and corresponding number of occurrences. Risk types include transportation risk and driving risk. Accident information includes the number of accidents and the accident score of each accident. The accident score is determined based on the accident level. A first score is obtained by weighting and summing the driving risk and the corresponding number of occurrences, and a second score is obtained by weighting and summing the transportation risk and the corresponding number of occurrences. The first score is modified using accident information to obtain a third score, and the second and third scores are summed to obtain an assessment score. Obtaining the risk level corresponding to the assessment score, using the assessment information corresponding to the assessment score of the same risk level as the target information, and generating a target vector based on the target information of a single autonomous driving operating vehicle; The target vectors are clustered to obtain clusters, a representative vector in a single cluster is obtained, a disposal strategy is generated based on the representative vector, and the disposal strategy is executed on the autonomous driving operating vehicles corresponding to all the target vectors in the cluster.

[0007] As a further solution of the present invention, obtaining the risk level corresponding to the assessment score includes: Set evaluation score thresholds P1 and P2, with P1 < P2; When P<P1, it is judged as a low risk level; When P1≤P<P2, it is determined to be a medium risk level; When P2≤P, it is judged as a high risk level.

[0008] As a further solution of the present invention, obtaining the target vector includes: Obtain the proportion of responsibility allocated to the autonomous vehicle operating at the time of the accident, and multiply the proportion by the accident score to obtain the impact score of the accident; Then the target vector XL=(A1,A2,…,A4,B1,B2,…,B5,C1,C2,…,C n ); Among them, A1 represents the number of occurrences of the first type of transportation risk, B1 represents the number of occurrences of the first type of driving risk, C n Represents the impact score when the nth accident occurs.

[0009] As a further solution of the present invention: obtaining the representative vector includes: In a single cluster, the cosine value of the angle between any two target vectors is greater than the preset value; In a single cluster, the sum of the cosine values ​​of the angles between the target vector a and the remaining target vectors is obtained as the screening value of the target vector a; The target vector corresponding to the maximum screening value is used as the representative vector.

[0010] As a further solution of the present invention: generating a disposal strategy includes: The first four dimensions in the representative vector are grouped as one group, the five dimensions between the 5th and 9th dimensions in the representative vector are grouped as one group, and the remaining dimensions in the representative vector are grouped as one group; Taking the ratio of the value of each dimension in the representative vector to the modulus of the representative vector as the target value, and obtaining the average target value of each group respectively; Sort the average target values ​​in descending order, and mark the groups as primary risk group, secondary risk group, and supplementary risk group according to the sorting order; Sort the target values ​​of each dimension in the primary risk group by size to obtain the primary risk sequence, and obtain the secondary risk sequence and supplementary risk sequence.

[0011] As a further solution of the present invention, generating a disposal strategy further includes: Obtain the operation and maintenance records of autonomous vehicles when risks occur in the same dimension, use text clustering methods to extract the central description and establish a dimension-action comparison table; Read the corresponding actions one by one in the order of primary risk sequence, secondary risk sequence, and supplementary risk sequence, remove duplicate content and maintain the original sequence, and splice the obtained text line by line to form a maintenance strategy draft; A semantic consistency check is performed on the maintenance policy draft, and after the check, the maintenance policy draft is converted into a maintenance policy body, which is a disposal policy.

[0012] As a further solution of the present invention, generating a disposal strategy further includes: If there is a target value that is less than the preset target value threshold, the action on the corresponding dimension will not be read.

[0013] As a further solution of the present invention, executing a disposal strategy for the autonomous driving operating vehicles corresponding to all target vectors in the cluster includes: If the risk level is low, the first 1 / 3 of the actions in the disposal strategy are executed; If the risk level is medium, the first two-thirds of the actions in the disposal strategy are executed; If the risk level is high, all actions in the disposal strategy will be executed.

[0014] The beneficial effects of the present invention are as follows: 1) This invention effectively identifies transportation and driving risks for autonomous vehicles, fully coupling the risk characteristics of autonomous vehicles with those of commercial vehicles, and enables risk identification of the operational safety level and transportation safety level of autonomous vehicles. 2) This invention integrates transportation process risks, driving behavior risks (and their weights), and comprehensively incorporates multi-dimensional factors such as accident frequency, accident severity, and responsibility attribution; on this basis, by constructing a dynamic coupling mechanism between driving risks and accident data, it significantly enhances the assessment accuracy and characterization capabilities of the risk exposure level of autonomous vehicles in actual operating scenarios. 3) This invention performs cosine clustering on the target vectors of vehicles with the same risk level. After selecting representative vectors, it calculates target values ​​by dimension grouping and generates primary, secondary, and supplementary risk sequences. This is then combined with a dimension-action comparison table to automatically assemble and maintain drafts, forming a differentiated disposal strategy for the entire cluster of vehicles, significantly improving the targeted risk disposal and operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below with reference to the accompanying drawings.

[0016] Figure 1 It is a flow chart of the comprehensive quantitative assessment method for the operation risk of autonomous driving commercial vehicles of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] See also Figure 1 As shown, the present invention is a comprehensive quantitative assessment method for the operation risk of an autonomous driving commercial vehicle, comprising the following steps: Step 1: Obtain assessment information of autonomous driving operational vehicles; Assessment information includes risk information and accident information. For risk information, specifically: 1. Transportation risk - passing through the wrong route (including lanes) or exceeding the fence: The management platform can set the operating area or route of the autonomous vehicle. The on-board terminal detects the position and trajectory of the autonomous vehicle through various means such as vehicle positioning and roadside detection. If the autonomous vehicle is detected passing through the wrong route (including lanes) or exceeding the fence, the positioning information is recorded and an alarm is sent to the management platform.

[0019] 2. Transportation risk - exceeding the rated passenger capacity. The vehicle terminal can detect the number of people in the vehicle in real time through visual and liveness detection, mainly for taxis. When the vehicle is overcrowded, the vehicle terminal records image and video information to provide in-vehicle prompts and alert the management platform.

[0020] 3. Transportation Risk - Speeding. The onboard terminal can recognize speed limit signs, and the management platform can issue speed limits for specific road sections and areas. The management platform can also issue temporary speed limits to the onboard terminal based on events such as traffic accidents and pedestrian crossings. When the autonomous vehicle exceeds the lowest of the three speed limits, the onboard terminal records the speeding information and alerts the management platform.

[0021] 4. Transportation Risk - Cargo Spillage. The onboard terminal can identify the cargo box of an autonomous truck, capture images of the box from multiple angles at a specific frequency, and compare them with the moment loading is completed. If the comparison results are inconsistent, it is detected that the autonomous truck has spilled cargo. The terminal can record image and video information and send an alarm to the management platform.

[0022] 5. Driving Risk - Lane Change Collision Risk. After the autonomous vehicle activates its turn signal and its wheels cross the lane, the onboard terminal can detect the distance between the autonomous vehicle's target lane and the vehicles in front and behind. If a collision risk is determined, an alarm will be sent to the management platform.

[0023] 6. Driving risk - unstable speed control. The vehicle terminal monitors average deceleration, average deceleration change rate, average acceleration, and other data in real time. If it detects that the threshold is exceeded, it will alert the management platform.

[0024] 7. Driving Risk - Unstable Lateral Control. The vehicle terminal collects the standard deviation of lateral displacement within the lane within 2 seconds. If the standard deviation exceeds the threshold, an alarm is sent to the management platform.

[0025] 8. Driving Risk - Following too closely. When an autonomous vehicle's speed exceeds 30 km / h, the onboard terminal monitors the following distance and TTC to the vehicle ahead in real time. If it detects that the threshold is exceeded, an alarm is sent to the management platform.

[0026] 9. Driving risk - exceeding operating conditions. The vehicle terminal can match the input vehicle operating conditions, including light, weather, etc. When it detects that the road environment exceeds the vehicle operating conditions, the vehicle terminal will send an alarm to the management platform.

[0027] Specific alarm parameter indicators are shown in Table 1 below; Table 1:

[0028] Among them, V 自 、V 后 They represent the speeds of the rear vehicle and the current vehicle in the target lane during lane change, N represents the total number of sampling times, and Di represents the lateral displacement at the time of the i-th sampling; For accident information, specifically: The accident score is determined based on the accident level, and the values ​​are shown in Table 2; Table 2:

[0029] Step 2: Calculate the evaluation score based on the evaluation information; The calculation model of the evaluation score is: ; Among them, R is the evaluation score, A j represents the number of occurrences of the jth type of transportation risk, w k represents the weight of the j-th transportation risk, B k represents the number of occurrences of the kth type of driving risk, m k represents the weight of the k-th driving risk, C xThe impact score of the x-th accident is the product of the accident score of the x-th accident and the responsibility allocation ratio. For example, if the autonomous vehicle bears 30% of the responsibility in an accident, the responsibility allocation ratio is 30%. X represents the total number of accidents. Step 3: Obtain the risk level corresponding to the assessment score; In a preferred embodiment of the present invention, obtaining the risk level corresponding to the assessment score includes: Set evaluation score thresholds P1 and P2, with P1 < P2; When P<P1, it is judged as a low risk level; When P1≤P<P2, it is determined to be a medium risk level; When P2≤P, it is judged as a high risk level.

[0030] It should be noted that the above calculation formula first multiplies the occurrence frequency of the four types of transportation risks by their respective weights and sums them to obtain the static risk at the transportation level. It then multiplies the occurrence frequency of the five types of driving risks by their corresponding weights and sums them. The formula is then modulated by a magnification factor of "one plus the sum of all accident impact scores". This ensures that the driving risk is magnified proportionally with the number of accidents and the proportion of responsibility, fully reflecting the coupling effect of accidents on the overall risk. This weighting and amplification mechanism can compress multi-source heterogeneous risks into a continuous score on the same scale, with larger scores representing higher risks. On this basis, two thresholds are set from small to large to form three intervals of low, medium and high. On the one hand, this ensures that the scores are strictly monotonic with the risk levels. On the other hand, a buffer zone is left between the thresholds to reduce misjudgments caused by measurement errors or occasional fluctuations.

[0031] After this processing, the comprehensive score not only retains the differential information of different risk components, but can also be directly used for subsequent clustering and strategy matching, achieving accurate classification and targeted treatment of running vehicles; Step 4: Use the assessment information corresponding to the assessment score of the same risk level as the target information, and generate a target vector based on the target information of a single autonomous driving operating vehicle; In another preferred embodiment of the present invention, obtaining the target vector includes: Obtain the proportion of responsibility allocated to the autonomous vehicle operating at the time of the accident, and multiply the proportion by the accident score to obtain the impact score of the accident; Then the target vector XL=(A1,A2,…,A4,B1,B2,…,B5,C1,C2,…,C n ); Among them, A1 represents the number of occurrences of the first type of transportation risk, B1 represents the number of occurrences of the first type of driving risk, C n represents the impact score when the nth accident occurs; Step 5: Cluster the target vector to obtain clusters and obtain the representative vector in a single cluster; In another preferred embodiment of the present invention, obtaining the representative vector includes: In a single cluster, the cosine value of the angle between any two target vectors is greater than the preset value; In a single cluster, the sum of the cosine values ​​of the angles between the target vector a and the remaining target vectors is obtained as the screening value of the target vector a; The target vector corresponding to the maximum screening value is used as the representative vector; Step 6: Generate a treatment strategy based on the representative vector, and execute the treatment strategy on all the autonomous driving vehicles corresponding to the target vectors in the cluster; In a preferred embodiment of the present invention, generating a disposal strategy includes: Based on the representative vectors in the cluster, the first four dimensions (i.e. A1 to A4) are classified as the transportation risk group, the fifth to ninth dimensions (i.e. B1 to B5) are classified as the driving risk group, and the remaining dimensions (i.e. C1 to C n ) are grouped as accident impact groups. The basis for grouping is that these dimensions correspond to the transportation link, driving link and accident consequences in business logic, and are homogeneous risk factors; Calculate the modulus of the vector and divide the value of each dimension by the modulus to get the target value. This ratio essentially measures the contribution of the risk of that dimension to the overall risk, allowing dimensions of different magnitudes to be compared on the same scale. Take the average of the three groups of internal target values. The average value represents the combined contribution of all risk factors in the group. The larger the average value, the more prominent the overall risk impact of the group. Then, sort the three groups according to the average value and mark them as the primary risk group, secondary risk group, and supplementary risk group, so as to determine the order of risk treatment; Within the primary risk group, the dimensions are arranged according to the target values ​​to obtain the primary risk sequence. Similarly, the secondary and supplementary risk sequences are obtained. This step further refines the risk factors within the group and provides a clear priority for action calls. Retrieve historical operation and maintenance records and extract the core description that best represents public experience from the text corresponding to each dimension. This allows for a comparison between the dimension and the operation and maintenance actions. This operation leverages the natural aggregation properties of similar text to summarize high-frequency and effective treatment measures from a large number of records. Actions are queried one by one in the order of primary, secondary, and supplementary risk sequences. Repeated actions between different sequences are removed to avoid redundancy. The draft maintenance strategy is then assembled in the original order. After the draft is formed, a semantic consistency check is performed to ensure readability and enforceability by checking for contextual coherence and conflicting instructions. Drafts that pass the check are output as formal maintenance strategies. It's understandable that the core purpose of this design is to map complex dimensions to a clear priority system using a grouping-normalization-sorting approach. This allows dimensions and actions that contribute the most to risk to be extracted first and presented in the strategy, enabling rapid targeting of high-risk factors. Operational and maintenance actions are derived from massive amounts of real-world records and denoised through clustering, ensuring that measures are operational and reach industry consensus. Deduplication and consistency testing prevent redundant or contradictory strategies, improving execution efficiency and readability. It should be noted that if the target value is less than the preset target value threshold, the action on the corresponding dimension will not be read; It should be noted that the disposal strategies for the autonomous driving vehicles corresponding to all target vectors in the cluster include: Read the risk level label for each vehicle in the cluster and split the maintenance policy body shared by all vehicles in the cluster into a prioritized sequence of actions. For example, the sequence may include sensor self-check, route plan refresh, software patch update, hardware plug point tightening, redundant power supply detection, and driving strategy retraining. Extract subsequences of different lengths according to the risk level of the vehicle: When a vehicle is marked as low-risk, only the first third of actions are taken, such as performing sensor self-checks, route planning refreshes, and software patch updates. This allows the most common and easily accumulated hidden dangers to be addressed while minimizing the investment of maintenance resources. If the vehicle is at medium risk, the process is extended further, taking the first two-thirds of the steps and incorporating hardware connector tightening and redundant power supply testing to further cover weak links that could lead to escalating risks. When the vehicle is at high risk, all actions are fully executed, including time-consuming but critical in-depth measures such as driving strategy retraining. This layered and progressive approach ensures that higher-priority actions are triggered sooner, while lower-priority but still necessary measures are only implemented in high-risk situations. Through risk-driven, segmented O&M, limited resources are concentrated where they are most needed, achieving a balance between cost and safety. Low-risk vehicles are spared the downtime and financial burden of excessive maintenance, medium-risk vehicles can have potential escalation channels blocked promptly, and high-risk vehicles receive comprehensive intervention to quickly reduce the probability of accidents. Ultimately, this helps operators maintain a dynamically controllable O&M rhythm and risk level even as fleet size continues to expand. It should be noted that the execution of the disposal strategy for the autonomous driving vehicles corresponding to all target vectors in the cluster also includes: After completing the calculation of the comprehensive assessment score, first extract the weighted sum of transportation risk and the weighted sum of driving risk respectively, divide each of the two by the final assessment score to obtain the contribution ratio, and take the larger ratio as the current dominant risk type; if the result shows that transportation risk is the dominant risk type, first conduct a targeted review of the transportation safety system of the affiliated enterprise, such as improving the carrier qualification review, dynamic loading monitoring and shift scheduling process, and at the same time add or strengthen the model features related to cargo status identification, loading and unloading condition matching, and abnormal trip alarm on the vehicle algorithm side, so that the system and algorithm can work together to reduce the failure probability of the transportation link; if driving risk is the dominant risk type, then focus on reviewing the vehicle's driving scenario processing logic in the perception, decision-making and control chain, conduct data playback and model retraining for dimensions such as road obstacle recognition, lane keeping, speed curve planning and emergency braking strategy, and cooperate with road test verification to ensure that the risk factors of the driving link are directly weakened; By dynamically comparing the contribution of the two types of risks, resources can be focused on the most risky links, avoiding inefficient investment caused by average effort. Transportation risks are attributed to systems and operational processes, while driving risks are more derived from algorithms and control strategies. The improvement paths of the two are completely different. Classification and disposal can enable the plan to accurately identify the root cause, thereby quickly suppressing the source of high-incidence accidents and laying the foundation for the continuous improvement of the overall operational safety level.

[0032] The above is a detailed description of an embodiment of the present invention. However, the content is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A comprehensive quantitative assessment method for the operational risk of autonomous driving vehicles, characterized by: The following steps are involved: Obtain assessment information on autonomous driving operating vehicles, including risk information and accident information. Risk information includes risk type and corresponding number of occurrences. Risk types include transportation risk and driving risk. Accident information includes the number of accidents and the accident score of each accident. The accident score is determined based on the accident level. A first score is obtained by weighting and summing the driving risk and the corresponding number of occurrences, and a second score is obtained by weighting and summing the transportation risk and the corresponding number of occurrences. The first score is modified using accident information to obtain a third score, and the second and third scores are summed to obtain an assessment score. Obtaining the risk level corresponding to the assessment score, using the assessment information corresponding to the assessment score of the same risk level as the target information, and generating a target vector based on the target information of a single autonomous driving operating vehicle; The target vectors are clustered to obtain clusters, a representative vector in a single cluster is obtained, a disposal strategy is generated based on the representative vector, and the disposal strategy is executed on the autonomous driving operating vehicles corresponding to all the target vectors in the cluster.

2. The comprehensive quantitative assessment method for the operational risk of an autonomous driving vehicle according to claim 1, characterized in that: The risk levels corresponding to the assessment scores include: Set evaluation score thresholds P1 and P2, with P1 < P2; When P<P1, it is judged as a low risk level; When P1≤P<P2, it is determined to be a medium risk level; When P2≤P, it is judged as a high risk level.

3. The comprehensive quantitative assessment method for the operational risk of an autonomous driving vehicle according to claim 1, characterized in that: Obtaining the target vector includes: Obtain the proportion of responsibility allocated to the autonomous vehicle operating at the time of the accident, and multiply the proportion by the accident score to obtain the impact score of the accident; Then the target vector XL=(A1,A2,…,A4,B1,B2,…,B5,C1,C2,…,C n ); Among them, A1 represents the number of occurrences of the first type of transportation risk, B1 represents the number of occurrences of the first type of driving risk, C n Represents the impact score when the nth accident occurs.

4. The comprehensive quantitative assessment method for operating risk of an autonomous driving vehicle according to claim 1, characterized in that: Obtaining the representative vector includes: In a single cluster, the cosine value of the angle between any two target vectors is greater than the preset value; In a single cluster, the sum of the cosine values ​​of the angles between the target vector a and the remaining target vectors is obtained as the screening value of the target vector a; The target vector corresponding to the maximum screening value is used as the representative vector.

5. The comprehensive quantitative assessment method for the operational risk of an autonomous driving vehicle according to claim 2, characterized in that: Generate disposal strategies include: The first four dimensions in the representative vector are grouped as one group, the five dimensions between the 5th and 9th dimensions in the representative vector are grouped as one group, and the remaining dimensions in the representative vector are grouped as one group; Taking the ratio of the value of each dimension in the representative vector to the modulus of the representative vector as the target value, and obtaining the average target value of each group respectively; Sort the average target values ​​in descending order, and mark the groups as primary risk group, secondary risk group, and supplementary risk group according to the sorting order; Sort the target values ​​of each dimension in the primary risk group by size to obtain the primary risk sequence, and obtain the secondary risk sequence and supplementary risk sequence.

6. The comprehensive quantitative assessment method for the operational risk of an autonomous driving vehicle according to claim 5, characterized in that: Generate a disposal strategy that also includes: Obtain the operation and maintenance records of autonomous vehicles when risks occur in the same dimension, use text clustering methods to extract the central description and establish a dimension-action comparison table; Read the corresponding actions one by one in the order of primary risk sequence, secondary risk sequence, and supplementary risk sequence, remove duplicate content and maintain the original sequence, and splice the obtained text line by line to form a maintenance strategy draft; A semantic consistency check is performed on the maintenance policy draft, and after the check, the maintenance policy draft is converted into a maintenance policy body, which is a disposal policy.

7. The comprehensive quantitative assessment method for operating risk of an autonomous driving vehicle according to claim 6, characterized in that: Generate a disposal strategy that also includes: If there is a target value that is less than the preset target value threshold, the action on the corresponding dimension will not be read.

8. The comprehensive quantitative assessment method for the operational risk of an autonomous driving vehicle according to claim 7, characterized in that: The disposal strategies for the autonomous driving vehicles corresponding to all target vectors in the cluster include: If the risk level is low, the first 1 / 3 of the actions in the disposal strategy are executed; If the risk level is medium, the first two-thirds of the actions in the disposal strategy are executed; If the risk level is high, all actions in the disposal strategy will be executed.

Images