SEDEA-based lorry driver safe driving evaluation method

Abstract

The invention discloses a truck driver safe driving evaluation method based on SEDEA, and the method comprises the steps: regarding the vehicle characteristics, behavior characteristics, road characteristics and environment characteristics of a vehicle in a driving process as external resources of a driver safe driving task, and regarding the characteristics of overspeed driving, fatigue driving, distraction driving and the like as risk results after the resources are utilized; the method comprises the following steps: taking resource utilization efficiency evaluation as a core means to measure the safe driving ability of a truck driver, and constructing a weight system containing multiple inputs and multiple outputs and a resource utilization efficiency evaluation model to calculate a safe driving score of the driver; and various safety indexes and quantitative classification management measures of the driver are given based on the model. The model is in close contact with the safe driving management situation of the truck driver, not only considers the resource input and the efficiency output of the safe travel task of the truck driver, but also sets the optimization direction between different inputs and outputs through a weight system.

Description

TECHNICAL FIELD

[0001] The present application relates to transportation, in particular to a truck driver safe driving evaluation method based on SE_DEA. BACKGROUND

[0002] The safety risk of freight vehicles has the following characteristics: first, due to the large mass and large inertia of the truck, the braking distance is significantly longer than that of ordinary vehicles, and it is more likely to cause serious accidents in emergency situations; second, the long-distance transportation and time pressure in the freight industry lead to frequent violations such as fatigue driving and overspeeding; third, the driving behavior of truck drivers is affected by factors such as load state and cargo type, making the risk characteristics more complex. Domestic road traffic accident statistics show that although the number of casualties has fluctuated in the past five years, the loss caused by truck accidents remains high.

[0003] The causes of truck road traffic accidents involve many factors such as people, vehicles and roads, and human factors account for more than 90% of accident causes. Studies have shown that driving behavior (such as overspeeding, fatigue driving, sudden acceleration, sudden braking, etc.) is significantly related to traffic safety and is the core cause of truck accidents. However, existing research on truck driving safety has obvious shortcomings. On the one hand, these studies are often limited to a single driving scenario and small-scale data, and lack of systematic discussion on the deeper safety factors such as vehicle characteristics, driving environment and transportation tasks; on the other hand, most of the research is from the perspective of car drivers, focusing on the relationship between visual features, overspeeding, distracted driving, fatigue driving and driving risk, while the research on the relationship between truck driving behavior and risk is relatively scarce. In addition, previous studies have mainly focused on high-frequency GPS data with a frequency of no less than 1Hz. However, in actual operation, more than 60% of GPS trajectory data is low-frequency sampling, which has the characteristics of large data volume, high authenticity and strong reliability. Although high-frequency data has high accuracy, its acquisition and storage cost is also relatively large. Heavy freight vehicles in China have been equipped with satellite positioning devices, which can transmit low-frequency trajectory data to public management platforms, laying a solid foundation for the research work of this paper. In summary, the current research has shortcomings in the exploration of the relationship between truck driving characteristics and traffic safety, especially the wide-scale analysis based on low-frequency GPS data still needs to be further explored. SUMMARY

[0004] In view of the above shortcomings in the prior art, the present application provides a truck driver safe driving evaluation method based on SE_DEA.

[0005] In order to achieve the above-mentioned application purposes, the technical scheme adopted by the present application is as follows: A truck driver safe driving evaluation method based on SE_DEA, comprising the following steps: S1. Construct a multi-source traffic risk database, integrating vehicle characteristics, driving behavior characteristics, road characteristics, and environmental characteristics data; S2. Based on the multi-source traffic risk database, the SE-DEA model is applied to calculate the driver's comprehensive safety index; S3. Based on the comprehensive safety index of the drivers, classify and manage the drivers, and establish a hierarchical monitoring mechanism and targeted management strategies; Furthermore, S1 includes: S11. Collect multi-source data, including GPS data of commercial trucks, vehicle information data, weather data, and street view data; S12. Preprocess the data, including deduplication, deletion of abnormal data, and trajectory division; S13. Extract feature indicators, including vehicle features, driving behavior features, road features, and environmental features, to form a risk database.

[0006] Furthermore, in S13, vehicle characteristics include vehicle tonnage and manufacturing date; driving behavior characteristics include average daily driving time, speeding frequency, and fatigue driving time; road characteristics include the number of lanes and the proportion of lane types; and environmental characteristics include average temperature and weather conditions.

[0007] Furthermore, S2 includes: S21. Set the input variable as the feature index in the risk database, and the output variable as the reciprocal of the dangerous driving behavior; S22. Construct an SE-DEA linear programming model to output comprehensive efficiency, technical efficiency, and scale efficiency indicators, and at the same time output the reference objects of ineffective drivers and the slack variables of the corresponding indicators. S23. Based on the comprehensive safety index, identify drivers who are highly efficient and safe, those with improved technology, and those with resource deficiencies.

[0008] Furthermore, the SE-DEA model in S22 is represented as follows:

[0009]

[0010] In the formula, Indicates driver j The superefficiency value; j o Indicates the driver currently being evaluated; Represents non-Archimedean infinitesimals; The output dataset represents the drivers. j The reciprocal of the abnormal driving behavior parameters; This indicates the input dataset, for drivers. jfour characteristic indexes of the four characteristics; 、 relaxation variable of input and relaxation variable of output; represent the input and output shares of the driver being evaluated.

[0011] Further, the S3 specifically includes the following steps: S31, judging the effectiveness of the decision-making unit DMU based on the super-efficiency value of the driver and determining the value of the corresponding index relaxation variable; S32, for a driver whose comprehensive safety index is less than 1, finding a linear combination of a group of drivers whose comprehensive safety index is greater than or equal to 1 to generate the same output as the non-effective decision-making unit as a reference object, and assigning weights to each reference object driver; S33, decomposing the comprehensive technical efficiency TE into pure technical efficiency PTE and scale efficiency SE, where the comprehensive technical efficiency TE is the ratio of the distance from the origin to the current driver being evaluated ( ) to the distance from the origin to the projection point on the frontier, i.e., the efficiency reference object ( ). Further, the S3 for classifying and managing drivers specifically includes: A1, definition: the comprehensive safety index, technical safety index, and resource allocation index correspond to the comprehensive technical efficiency, pure technical efficiency, and scale efficiency calculated in S33, respectively, where the comprehensive technical efficiency represents the driver's comprehensive safety index, the technical efficiency represents the driver's technical safety index, and the scale efficiency represents the driver's resource allocation index, and the drivers are sorted according to their comprehensive safety index efficiency values.

[0012] A2, define classification criteria: High-efficiency and safe type: comprehensive safety index ≥ 1 and relaxation variable = 0, which is the "peer benchmark" driver; Technical improvement type: comprehensive safety index < 1 and technical safety index < 1; Resource deficiency type: comprehensive safety index < 1 and resource allocation index < 1; A3, based on the classification criteria, classify the drivers according to their efficiency reference objects ( ) and relaxation variable values to set the quantitative basis for safety training.

[0013] Further, the specific way of establishing a hierarchical monitoring mechanism and targeted management strategy in the S3 is: B1, design a hierarchical monitoring mechanism: implement basic monitoring for high-efficiency and safe drivers; implement strict level monitoring for technical improvement drivers; implement standard level monitoring for resource deficiency drivers; B2, formulate targeted management strategy: high efficiency and safety type mainly with benchmarking and points reward; technical improvement type mainly with behavior correction and compulsory training; resource deficiency type mainly with resource allocation optimization and intelligent scheduling.

[0014] The present application has the following beneficial effects: First, as a non-parametric analysis method, SE-DEA does not require prior assumptions about the specific form of the production function, nor does it require artificial setting of the weight distribution of input and output variables. It skillfully uses the principles of operations research and linear programming to comprehensively consider the input and output variables, and automatically generates the weight of the variables through mathematical programming. This greatly reduces the errors that may be introduced in the model setting process.

[0015] Second, the SE-DEA model shows high flexibility in setting input and output variables, and can comprehensively include various factors that affect the risk level of drivers, including but not limited to the characteristics of the vehicle itself, the driving behavior characteristics of the driver, the actual conditions of the road, and external environmental factors. This comprehensive approach makes the safety evaluation results more realistic and objective.

[0016] Third, there is no need to set subjective benchmark combinations as evaluation standards. On the contrary, the model automatically generates the production frontier by analyzing the data of the decision-making unit (DMU), which serves as an objective basis for judging the effectiveness of each DMU. This approach makes the evaluation results more objective and independent, reducing the bias caused by human factors.

[0017] Fourth, compared with traditional DEA models, the SE-DEA model has significantly improved processing capacity. It not only can accurately further rank the drivers with a comprehensive safety index of 1, but also can support more in-depth difference analysis, sensitivity analysis and efficiency analysis. This makes the SE-DEA model not only help drivers understand their risk ranking when evaluating their safety level, but also provide direction for driver training and improvement for enterprise managers. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 The present application is based on the SE_DEA truck driver safety driving evaluation method flowchart.

[0019] Figure 2 The present application is based on the SE_DEA truck driver safety driving evaluation method flowchart.

[0020] Figure 3 The present application is based on the SE_DEA truck driver safety driving evaluation method flowchart.

[0021] Figure 4 The present application is based on the SE_DEA truck driver safety driving evaluation method flowchart.

[0022] Figure 5 Driving safety evaluation framework for the embodiment of the present application. DETAILED DESCRIPTION

[0023] The specific embodiments of the present application are described below to facilitate the understanding of the present application for those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the inventions utilizing the concept of the present application are within the scope of protection.

[0024] A truck driver safe driving evaluation method based on SE_DEA, as shown in Figure 1 The method comprises the following steps: S1, constructing a multi-source traffic risk database, integrating vehicle characteristics, driving behavior characteristics, road characteristics and environmental characteristic data; The driving risk elements in this embodiment are divided into four traffic dimensions and ten risk factors, and the specific data information is shown in Table 1. In order to construct a comprehensive truck driver risk element database, the above-mentioned various data needs to be summarized and integrated. In this process, the license plate number of the driver will be used as the key identification to realize accurate matching and deep fusion of information from different sources. The following are the specific operation steps for constructing the database: (1) According to the license plate number of the truck driver, match the vehicle information; (2) On the basis of GPS trajectory data, refer to the definition of driving behavior indicators to determine the value of each dimension of each truck driver; (3) Use the longitude and latitude in the GPS data of the truck driver to match the population density data, road data, street view data, and weather data; The final truck driver risk database table formed is shown in Table 2.

[0025] Table 1 Data Information Table

[0026] Table 2 Risk Database

[0027] Since the total mileage of the drivers involved in the experiment is large enough, and the influence of the vehicle's own characteristics, personal driving habits, road conditions, and weather and other external factors is fully considered, the evaluation results of the data envelopment analysis can accurately reflect the driving safety level of different drivers. Based on the truck driving data, the vehicle characteristics (such as tonnage, factory time), driving behavior characteristics (such as daily driving behavior), road characteristics (such as street population density, road street view, etc.), and environmental characteristics (such as weather, driving period, route familiarity, etc.) are used as the model input of each driver, that is, indicators X1-X8 and X14-X42, and the dangerous driving behavior characteristics are used as the model output, that is, indicators X9-X13, but considering the dangerous driving behavior, the smaller the better in actual driving. To meet the selection criteria of the input and output indicators of the DEA, the dangerous driving behavior related characteristics are respectively taken as the reciprocal processing, and the increase of 1 / speeding frequency (Y1), 1 / speeding driving time (Y2), 1 / fatigue frequency (Y3), 1 / fatigue driving time (Y4) and 1 / distraction frequency (Y5) is used to represent the reduction of actual dangerous driving behavior, that is, "behavior safety", which is used as the output, so as to construct the safety evaluation index system of truck drivers.

[0028] In this embodiment, the truck trajectory data used is obtained from a certain operating vehicle supervision platform in Chengdu. The platform manages enterprise safety education and training information, and its main business includes operating vehicle dynamic supervision, enterprise high-risk processing feedback, enterprise hidden danger self-check, enterprise information reporting, enterprise information display, etc., providing information management means such as source management for transportation industry supervision departments, and effectively regulating enterprise behavior in safety management through data. For GPS preprocessing of operating trucks, mainly through data deduplication, deletion of abnormal data, etc. Subsequently, according to the actual data of Chengdu operating trucks, trajectory data preprocessing analysis is carried out, in order to effectively prevent and significantly reduce the incidence of heavy traffic accidents, Chengdu is currently working hard to scientifically and systematically manage and evaluate the driving risk of operating trucks. Under the strict supervision and management of the government departments concerned, Chengdu operating trucks have all been installed with GPS vehicle positioning system, and the enterprise needs to transmit the company's operating truck data to the relevant supervision platform in real time to collect vehicle time location data and driving data, so as to obtain a large amount of GPS data. A total of 2 data sets related to driving data are obtained, which are GPS data and vehicle information data set.

[0029] The GPS data is usually composed of 8 fields such as GPS upload time point, license plate number, longitude, latitude, altitude, speed, direction and mileage. The data collection time is from June 1 to June 30, 2024, and the GPS data collection frequency is between 1s-30s.

[0030] Table 3 Chengdu operating truck GPS data example

[0031] The operating range is Chengdu urban area. The data size is 868 MB, there are 6 transport companies, 91 vehicles, and the total operating GPS data is 6418138. Table 4 is a part of the example of Chengdu operating truck GPS data.

[0032] The fields contained in the operating truck GPS data include deviceTime, plateNumber, latitude, longitude, altitude, speed, direction, and mileage. The specific description of the fields is as follows: deviceTime: data collection time, with seconds as the minimum unit.

[0033] plateNumber: vehicle number. At present, Chengdu has basically realized the binding of people and vehicles on the road, so a license plate number represents a driver.

[0034] Lat, Lon: GPS point latitude and longitude coordinates, which belong to spatial data, used to determine the position of the operating vehicle.

[0035] altitude: altitude, indicating the altitude of the vehicle at this position point, in meters.

[0036] speed: instantaneous speed of the operating vehicle, in kilometers per hour (km / h).

[0037] Direction: direction angle, usually refers to the angle between the vehicle and the north direction, between 0-360 degrees.

[0038] Extract the corresponding enterprise, license plate number, tonnage, and factory date fields. Table 3 is a part of the example of the extracted data set.

[0039] Table 3 Operating Truck Information Data

[0040] Identify and process the repeated items in the data set. Use the precise matching algorithm based on hash value to jointly verify the license plate number, timestamp, and latitude and longitude coordinates. If they are exactly the same, it is determined as redundant data and deleted. The basis for deletion: redundant data will distort the trajectory continuity analysis results, leading to deviation in mileage statistics and driving behavior evaluation.

[0041] Identify and process abnormal data. ①Speed anomaly: according to the maximum speed limit of urban road 120km / h, if the vehicle speed is not in the normal driving speed range of the city, that is, the vehicle speed is less than 0km / h or greater than 120km / h, delete the data; ②Direction angle anomaly: based on the principle of geometry (angle closure domain 0°~360°), the vehicle direction angle is less than 0 degree or greater than 360 degrees, then the data is considered invalid data, and is deleted. The information comparison before and after the GPS data set preprocessing is shown in the chart 4: Table 4 Comparison of data before and after data preprocessing

[0042] The preprocessed data is sorted according to the license plate number and time, and the time interval of adjacent data of the same driver is counted, as shown in Figure 2 The data with a collection time interval of 30s is the most, accounting for more than 35%. Road driving experiments and operation data of transportation enterprises mainly focus on high-frequency trajectory data with a collection frequency of 1Hz or more. From the perspective of data collection frequency, the data set of the embodiment is a low-frequency collection data set in analyzing driving behavior.

[0043] In order to fully show the spatio-temporal nature of trajectory data and its importance in urban traffic management, the embodiment analyzes the GPS data of the operating vehicles in Chengdu in depth and detail in the time and space dimensions. In the spatial dimension, the GPS trajectory data of the operating trucks is accurately projected onto the road network map of Chengdu. This step can intuitively observe the actual distribution and flow of the operating trucks on the complex road network. In order to ensure the representativeness and effectiveness of the analysis, the embodiment randomly selects 100,000 GPS trajectory data from the huge data set for projection analysis, as shown in Figure 3 The figure clearly depicts the density of operating trucks on the main roads and traffic hubs in Chengdu.

[0044] Obtain the instantaneous speed of the operating truck and draw a histogram, as shown in Figure 4 From the figure, it can be seen that the instantaneous speed is evenly distributed before 60km / h, of which the highest proportion is 49km / h, about 1.2%, and after 60km / h, it presents an inverted U distribution, and the highest proportion of instantaneous speed is 75km / h, about 4.1%. It can be seen that a part of the drivers in the data transport goods by means of highway, and the average speed of the truck in the highway is about 75km / h.

[0045] From Figure 5It can be seen that the expressway and the main road are more densely populated with GPS points, and the northern trajectory is significantly lower in density. The southern areas, including Wuhou, Jinjiang, and Shuangliu, have a significantly higher driving density. This indicates the distribution of freight transport enterprises and their work sites in Chengdu. The GPS trajectory density of secondary roads and branch roads is lower, and some branch roads have no GPS data. Trucks, due to their size and weight, generally do not take branch roads, so the actual operating truck situation is also more in line with reality S2、Based on the multi-source traffic risk database, apply the SE-DEA model to calculate the comprehensive safety index of the driver; Assuming that there are n drivers in the sample, each driver has p inputs and s outputs, and the SE-DEA model is a key extension based on the CCR model. To evaluate the driving efficiency of the driver j , , when evaluating the efficiency of a specific driver, the driver will be temporarily removed from the reference, and a linear programming model is constructed as shown in the following formula.

[0046]

[0047]

[0048] represents the super-efficiency value of the driver j , <1 indicates DEA non-efficiency; j o represents the driver being evaluated; represents a non-Archimedean infinitesimal; represents the output data set, which is the inverse of the abnormal driving behavior parameters of the driver j ; represents the input data set, which is the 4 major feature indicators of the driver j ; , are the slack variables of inputs and outputs; represents the input and output shares of the driver being evaluated.

[0049] Using the SE-DEA method, the driver's operation record data can be solved to obtain various parameters.

[0050] These parameters can be used to evaluate the safety level of truck drivers. Safety evaluation is divided into the following three dimensions: (1) Driving Effectiveness Judgment: Input and output indicators are selected and fed into the model. The model outputs three efficiency indicators through linear programming: comprehensive technical efficiency, technical efficiency, and scale efficiency. Comprehensive technical efficiency represents the driver's comprehensive safety index, technical efficiency represents the driver's technical safety index, and scale efficiency represents the driver's resource allocation index. Using this model, it is possible to determine whether the decision-making unit is strong DEA effective, weak DEA effective, or non-DEA effective. Based on the strength of effectiveness, the driving safety of the driver can be judged as high or low, i.e.: when =1 and hour, It is an effective unit, meaning there is no problem of excessive input or insufficient output. In this case:

[0051] when <1 hour, Invalid It can become efficient by improving input redundancy and increasing output redundancy, through the following formula (SBM projection):

[0052] based on definition The reference set is: The set of indicators that are greater than 0 is called The reference set.

[0053] In particular, when When >1, in the evaluation In the SE-DEA (VRS) model, any set of formulas that satisfy the given conditions... If at least one factor of production exists , making

[0054] Then it is called The set of SE-DEA (VRS) super-efficient decision units is denoted as SE.

[0055] In evaluation In the SE-DEA (VRS) model, if at least one output factor exists... Make Then it is called The set of SE-DEA (VRS) strong super-efficient decision units is denoted as SSE.

[0056] Therefore, when ≥1 and = When the overall safety index is 0, indicating a high overall safety index, the driver is considered a safe driver. The higher the value, the stronger the security.

[0057] when If the overall safety index is less than 1, the driver is considered to be driving unsafely. Optimal configuration and allocation for safe driving can be achieved by adjusting resource allocation or improving technical skills.

[0058] Meanwhile, the comprehensive security index = technical security index * resource allocation index, so the comprehensive security index can be further analyzed by breaking it down into two dimensions: technical security index and resource allocation index.

[0059] The comprehensive safety index reflects the input-output efficiency of drivers under optimal conditions, encompassing the combined performance of the technical safety index and the resource allocation index. The technical safety index represents the driver's technical performance; a low index indicates that the driver has not fully utilized resources to achieve desired outputs under current technological conditions. A technical safety index ≥1 indicates technical effectiveness, while a value <1 indicates technical management issues. The resource allocation index represents the driver's potential for resource efficiency; a low index suggests that the driver's failure to fully utilize resources may be due to unreasonable task allocation or unbalanced resource distribution. The comprehensive safety index is derived from the ratio of the comprehensive safety index to the technical safety index. A resource allocation index ≥1 indicates that the driver is in optimal resource allocation, while a value less than 1 indicates that there is a resource allocation problem.

[0060] (2) Basis for safety evaluation decision: The output results also include reference objects for DEA ineffective drivers and slack variables of indicators.

[0061] In order to evaluate drivers j To improve driving efficiency, a linear programming model is constructed and input into this model. Driver j efficiency value ( <1 indicates that DEA is invalid); Driver k The i Input quantity; Driver k The r Seed output; Driver k The weight represents its role in constituting a driver. j Contribution at the efficiency frontier; : The i-th input quantity of driver j; Let r be the r-th type of output for driver j. After solving, The driver k whose comprehensive safety index is greater than 0 is the reference object of the driver j. For the driver whose comprehensive safety index is less than 1, the model finds a group of drivers whose comprehensive safety index is greater than or equal to 1, and the linear combination of these drivers can generate the same output as the driver whose comprehensive safety index is greater than or equal to 1, but uses more inputs, that is, the reference object of the driver. The driving behavior of these reference object drivers constitutes the efficiency frontier of the driver whose comprehensive safety index is less than 1. The model assigns a weight to each reference object driver, indicating the importance of the reference object driver in constituting the efficiency frontier of the driver whose comprehensive safety index is less than 1. The greater the weight, the higher the reference value of the reference object driver.

[0062] According to the method and principle, a safety driving evaluation method of a data envelopment model SEA-DEA is provided. The dual model is used for "projection" analysis, so as to find the distance between the input or output of the non-DEA effective decision unit and the ideal value, and provide a decision basis for improving driving safety. The formula is as follows:

[0063] Among them, , , is the optimal solution of the corresponding linear programming of the driver j , , is the projection of the corresponding , of the decision unit on the data envelopment analysis frontier surface, which indicates the optimal value of the input and output of the evaluated land soil, and is considered as DEA effective. , The size directly reflects the contribution degree of each input variable to the overall output level: when it is greater than 0, it indicates that the driver has room for improvement in the safety field covered by the relevant index, which indicates the specific path to improve driving safety. On the contrary, if it is 0, it means that the influence of this input index on the output is relatively small, and the driver's performance in this aspect is already good, and there is no need for too much intervention, therefore, by adjusting this variable, the risk level can be effectively reduced.

[0064] In this embodiment, 66 drivers with relatively low risk are taken as reference objects of 25 drivers with relatively high risk, that is, for drivers with safety lower than the DEA level, their "peer benchmark" is exactly those drivers with safety higher than the DEA level. From the perspective of modeling, these DEA effective drivers set an upper limit or boundary of safety level for the whole driver group. For these drivers with relatively high risk, choosing drivers with relatively low risk as reference objects will be an effective way to improve driving safety. At the same time, the model will also assign a specific weight to each reference object. The size of this weight directly reflects the reference value of the reference object for a driver with high risk. The larger the weight, the more valuable the reference object can provide for the driver with high risk in improving driving behavior and reducing risk.

[0065] Table 5 lists in detail the "peer benchmark" drivers of five drivers with high risk and the weight values of these benchmark drivers. This information provides valuable improvement direction and related insights for drivers. Taking driver Chuan A**756 as an example, drivers Chuan AE**92, Chuan AY**25, Chuan A4**M1 and Chuan AF**68 are the best reference objects for preventing driving risk. Among them, Chuan AE**92 has the highest weight, indicating that the driving behavior and risk control strategy of Chuan AE**92 have a very high reference value and significance for this driver. Through such comparative analysis, drivers can more clearly recognize their shortcomings in driving safety and improve them accordingly to continuously improve their driving safety.

[0066] Table 5 Reference objects of drivers with high risk

[0067] Based on the relaxation variable results produced by the SE-DEA model, this embodiment analyzes the factors affecting the safety efficiency of drivers. Through systematic investigation of the relaxation variables of each index of 91 drivers, it is found that different categories of risk factors have significant differences in the influence on safety efficiency. Table 6 shows the average value of the relaxation variables of each input index, among which the relaxation values of vehicle tonnage (5.856), factory month (6.089), and daily driving time (15.64) are more prominent, indicating that there is a large improvement space for these factors in the current resource allocation.

[0068] Table 6 Average value of relaxation variable of each index

[0069] By performing principal component analysis on the slack variables of each indicator, three main influencing factors were extracted, with the cumulative variance contribution rate reaching 78.6%. Among them, the eigenvalue of the vehicle-environment adaptation factor is 4.32, explaining 34.7% of the variance change. Its main load indicators include vehicle tonnage (0.82), visibility (0.76), and temperature (0.69). The eigenvalue of the driving intensity factor is 3.15, explaining 25.4% of the variance change, which is mainly reflected in indicators such as average daily driving time (0.91) and the proportion during peak hours (0.85). The eigenvalue of the route complexity factor is 2.31, explaining 18.5% of the variance change, which is mainly reflected in indicators such as road complexity (0.88) and number of lanes (0.79).

[0070] Taking driver Chuan AD**79 with the lowest comprehensive safety index ( =0.663) as an example for in-depth diagnosis, it was found that this driver had obvious efficiency losses in indicators in multiple dimensions. In terms of vehicle configuration, the slack value of its tonnage reached 8.23, exceeding the sample mean by 40.6%; the slack value of vehicle age was 9.15, exceeding the optimal value by 152%. In terms of driving time allocation, the slack value of the proportion of driving during the evening peak was 6.84, and the slack value of morning driving was 0.45 (Z = 2.67, p < 0.01), showing an obvious imbalance in time allocation. In terms of environmental adaptation, the average visibility slack value was 5.32 (exceeding the industry mean by 52.0%), the proportion of overcast, haze, and fog slack value was 7.15 (exceeding the industry mean by 24.8%), and the road complexity slack value was 6.08 (exceeding the industry mean by 57.1%). These data all indicate that this driver has obvious deficiencies in adaptability in complex environments.

[0071] As a non-parametric method, SE-DEA can avoid the bias caused by subjective weight assignment through a data-driven weight assignment mechanism and is applicable to the evaluation of complex systems with multiple inputs and outputs. In addition, SE-DEA can further distinguish decision-making units (DMUs) with a comprehensive safety index of 1 in the traditional DEA model, providing a more refined safety assessment result.

[0072] Specific features: (1) The SE-DEA model can automatically generate weights through data envelopment analysis, avoiding the problem of subjective weight assignment, and can further distinguish the differences between efficient DMUs.

[0073] (2) SE-DEA can comprehensively consider complex systems with multiple inputs and outputs and is applicable to the integrated analysis of multi-dimensional indicators in driving safety assessment.

[0074] (3) Result stability: The stability and consistency of the model can be tested through sensitivity analysis to ensure the scientificity and practicality of the evaluation results.

[0075] Since both DEA and TOPSIS belong to the category of unsupervised comprehensive evaluation rather than predictive algorithm, they are not applicable to traditional model validation techniques such as K-fold cross, independent test set, etc. However, in order to ensure the robustness of the evaluation results, existing researches have explored and adopted sensitivity analysis to test the reliability of the model output. Sensitivity analysis is to further test the robustness of the model evaluation results, especially to explore whether these results will be significantly fluctuated by the increase or decrease of sample size. That is, when facing the change of sample size, the driver ranking given by the model can still maintain its inherent stability and reliability.

[0076] To evaluate the robustness of the model results, the embodiment uses the ranking difference test strategy. By systematically perturbing the data set and analyzing the ranking fluctuation, the stability of the model is quantified. The specific steps are as follows: (1) Data subset generation Apply five-fold cross validation, assume that the original data set contains N decision making units (DMU), randomly remove DMU in each iteration to generate subset 。

[0077] (2) Value calculation and ranking For each subset ， Recalculate the corresponding index of DMU and sort to get the ranking of the i-th driver in the k-th calculation 。

[0078] (3) Rank difference statistics For each DMU, calculate its ranking fluctuation in 5 iterations. Mean Rank Deviation (MRD):

[0079] Rank Standard Deviation (RSD):

[0080] Where, represents the average ranking of the i-th driver.

[0081] Further, the embodiment analyzes the distribution characteristics of the average difference and standard deviation of all DMUs in the data set to evaluate the stability performance of different models in generating scores and rankings. Through this series of sensitivity analysis, the stability of the model output is confirmed, thereby providing a more solid basis for driving risk assessment.

[0082] S3, according to the comprehensive safety index, the driver is classified and managed, and a hierarchical monitoring mechanism and targeted management strategy are established; DEA efficiency is the deviation degree of each decision unit projected on the effective production frontier, mainly including comprehensive technical efficiency (TE), which can be further decomposed into pure technical efficiency (PTE) and scale efficiency (SE). The comprehensive technical efficiency (TE) is the ratio of the distance from the origin to the current evaluated driver ( ) to the projection point on the frontier, i.e. the efficiency reference object ( ).

[0083] Let the projection point on the production frontier have the same production scale as the evaluated decision unit, and the deviation of the evaluated decision unit from the production frontier is the pure technical efficiency (PTE), i.e. the efficiency calculated by the BBC model is the pure technical efficiency (PTE).

[0084] Let the projection point on the production frontier have the same production scale as the evaluated decision unit, and the deviation of the evaluated decision unit from the production frontier is the pure technical efficiency (PTE), i.e. the efficiency calculated by the BBC model is the pure technical efficiency (PTE).

[0085]

[0086]

[0087] Scale efficiency (SE) is equal to the ratio of comprehensive technical efficiency (TE) to pure technical efficiency (PTE), SE=TE / PTE. According to the comprehensive technical efficiency, technical effect and scale efficiency of SE-DEA, all efficiency levels can be sorted.

[0088] For high-efficiency and safe drivers, the main focus is on normalized monitoring and positive incentives to maintain their safe driving state; for technical improvement type drivers, the focus is on skill training and behavior correction to improve their safe driving ability; for resource deficiency type drivers, vehicle equipment and dispatch management need to be optimized to eliminate safety hazards.

[0089] The core of the management system is: (1) Classification evaluation: relying on quantitative indicators such as comprehensive safety index and technical safety index, to realize the scientific classification of drivers; (2) Hierarchical control: according to the risk level to implement differentiated monitoring, and dynamically adjust the management intensity; (3) Closed-loop optimization: through regular evaluation feedback, continuously improve management strategies and technical means.

[0090] Through the above measures, a driving safety management system covering the whole chain of "evaluation-monitoring-intervention-improvement" is formed, and the level of enterprise traffic safety management is effectively improved.

[0091] In this embodiment, first, a three-level classification standard is established: the first level is high-efficiency and safe driver, whose comprehensive safety index reaches the production front; the second level is technical improvement type driver, which is characterized by insufficient technical safety; and the third level is resource defect type driver, whose main feature is that the resource allocation is not optimal. This classification method not only can identify the safety state of the driver, but also can accurately locate the risk source. The classification conditions are shown in Table 7.

[0092] Table 7 Classification standard and distribution of driver safety level

[0093] Secondly, a hierarchical monitoring mechanism is constructed. According to the classification results of the driver, a three-level monitoring system is implemented: for high-efficiency and safe drivers, basic monitoring is adopted, mainly through monthly spot checks and positive incentives for management; for resource defect type drivers, standard level monitoring is implemented, focusing on monitoring the vehicle resource allocation; and for technical improvement type drivers, strict level monitoring is implemented, including real-time behavior monitoring and regular examination. On the technical level, an intelligent monitoring system composed of vehicle-mounted OBD equipment, driving behavior analysis algorithm and mobile early warning platform is deployed to realize real-time collection and analysis of monitoring data.

[0094] In terms of management strategy, different measures are formulated for different types of drivers. The management of high-efficiency and safe drivers aims to maintain the existing safety level, and adopts the combination of spot check mechanism and score reward; the management of technical improvement type drivers focuses on improving driving skills, and intervenes through forced training, follow-up guidance and behavior target management; and the management of resource defect type drivers focuses on optimizing vehicle scheduling and equipment configuration to eliminate safety hazards from the hardware conditions.

[0095] To ensure the effective implementation of the management system, a comprehensive guarantee mechanism needs to be established. In terms of organizational guarantee, a special safety management committee is established, which is divided into three special groups of evaluation, training and equipment management; in terms of system guarantee, supporting system documents such as "Driver Safety Management Measures" are formulated; in terms of technical guarantee, a technical support system including safety management information system, driving behavior database and risk assessment model is constructed. At the same time, a PDCA cycle improvement mechanism is established to realize the self-improvement of the management system through annual plan, quarterly evaluation and continuous optimization. Finally, the implementation effect is verified through safety performance verification, accident rate monitoring and economic benefit evaluation.

[0096] Based on this, the classification standard established in this embodiment classifies the sample drivers into the following three categories after systematic classification: Highly safe and efficient drivers: 66 drivers, accounting for 72.5%. The typical characteristics are that the average of the comprehensive safety index is 1.623 (standard deviation 0.656), the technical safety index and the resource allocation index are both 1, and the slack variable is zero. Representative samples such as Sichuan AE**00 (comprehensive safety index 5.167) and Sichuan AY**25 (3.335) show that their driving behavior is standardized and resource allocation is reasonable, and they are on the production frontier.

[0097] Technically improved drivers: 3 drivers, accounting for 3.3%. The average of the comprehensive safety index is 0.755 (standard deviation 0.062), and the average of the technical safety index is 0.779, which is significantly lower than the effective threshold. Typical samples include Sichuan AC**10 (technical safety index 0.773) and Sichuan AG**37 (0.715), which have obvious characteristics of dangerous driving behavior.

[0098] Resource-deficient drivers: 22 drivers, accounting for 24.18%. The average of the comprehensive safety index is 0.901 (standard deviation 0.083), and the average of the resource allocation index is 0.985, which is close to but not optimal. Representative samples such as Sichuan AE**08 (0.977) and Sichuan AE**56 (0.973) show that their safety performance is mainly constrained by external resource allocation factors.

[0099] Based on the evaluation results of the DEA model, the specific data is shown in Table 8. It can be observed that the efficiency performance of vehicles with different license plates shows significant differences. The driver of the vehicle with license plate Sichuan AD**79 has the lowest comprehensive safety index (0.663), and the risk is the highest.

[0100] In terms of technical efficiency, Sichuan AG**37 has a low technical efficiency of 0.715, reflecting that it has insufficient resource utilization at the output level, which may be due to driving skills or operational standardization. Drivers such as Sichuan AE**56, Sichuan AC**82, and Sichuan AB**19 have high technical efficiency, indicating that they have high efficiency in utilizing input resources in driving tasks and can ensure their driving safety. In terms of scale efficiency, Sichuan AF**97 has the highest scale efficiency (1.326), indicating that it performs outstandingly in task allocation and resource allocation and can fully utilize the advantages of roads and environmental factors. Sichuan AD**79 has the lowest scale efficiency (0.663), indicating that it may have unreasonable task allocation and needs to optimize the configuration of vehicle factors and road factors.

[0101] Table 8: Efficiency index values of drivers

[0102] The scale returns represent the proportion of output increase when the resource input increases. As can be seen from the table, the scale returns of different drivers are different: the scale returns of drivers such as Chuan AF**97 are constant, indicating that their current resource allocation and task allocation have reached the optimal state, and increasing resource input will not bring additional efficiency improvement. The scale returns of drivers such as Chuan AE**56, Chuan AC**82 and Chuan AB**19 are increasing, indicating that their current resource utilization has not yet reached the optimal state, and increasing resource input may bring higher efficiency improvement. In particular, for drivers with low scale efficiency (such as Chuan AD**79), optimizing resource allocation and task allocation will help improve their overall efficiency. The scale returns of drivers such as Chuan AB**63 and Chuan AC**10 are decreasing, and their current resource input has already exceeded the optimal level, and increasing resource input may lead to efficiency decline. Therefore, these drivers may need to reduce resource input or readjust task allocation to achieve higher efficiency. In summary, drivers with low technical efficiency (such as "Chuan AC**10" and "Chuan AG**37") have more dangerous driving behaviors in resource utilization and need to strengthen safety supervision and behavior regulation; drivers with low scale efficiency (such as "Chuan AF**53" and "Chuan AD**79") should improve resource utilization efficiency by reasonably arranging tasks, while paying attention to controlling dangerous driving behaviors.

[0103] Based on the evaluation results of the SE-DEA model and the analysis of the slack variables, this embodiment constructs a systematic and differentiated driver safety management strategy system.

[0104] For the group of high-efficiency and safe drivers, the management strategy focuses on benchmarking and experience promotion. A safety driving case library is established, and the excellent practices and typical experiences of benchmark drivers such as Chuan AE**00 and Chuan AY**25 are systematically extracted. Regular experience exchange activities are organized, and the best practices are promoted through forms such as "Safety Driving Lecture Hall". A safety performance point reward system is implemented, which quantitatively assesses indicators such as safety mileage and standard operation, and is linked to salary and welfare. In addition, outstanding drivers are selected to participate in safety management decision-making, and their professional advantages are utilized in route planning and equipment selection.

[0105] For the technical improvement type driver group, precise intervention measures based on relaxation variable analysis are taken. Taking three drivers, Sichuan AB**63, Sichuan AC**10 and Sichuan AG**37, as examples, through the analysis of relaxation variables, it is found that Sichuan AB**63 has obvious deficiencies in speed variance (relaxation value 4.2) and lane width adaptability; Sichuan AC**10 performs poorly in lane keeping (relaxation value 3.1) and complex road condition handling; Sichuan AG**37 mainly has problems such as fatigue driving (relaxation value 5.3). Therefore, individualized training programs are developed, including simulation driving training, real vehicle guidance, work and rest adjustment, etc., focusing on improving their weak links. A follow-up guidance system is established, with experienced drivers demonstrating on site, setting stage improvement targets and regularly evaluating.

[0106] For the resource defect type driver group, the management strategy focuses on systematic optimization of working conditions. Taking Sichuan AD**79 with the lowest comprehensive safety index (0.663) as an example, its tonnage relaxation value reaches 8.23 (40.6% higher than the average value), and the vehicle age relaxation value is 9.15 (152% higher than the optimal value), indicating a significant imbalance in resource allocation. Based on relaxation variable analysis, measures such as vehicle tonnage-task matching optimization and peak period task adjustment are taken. At the same time, the working environment is improved, standardized rest areas are set up in major freight hubs; the working system of "4 hours driving + 30 minutes rest" is implemented; and an intelligent route planning system is developed to avoid high-risk road sections.

[0107] At the same time, for the technical improvement type driver and the resource defect type driver, based on the analysis results of relaxation variables, this embodiment establishes a hierarchical improvement recommendation system. For the priority improvement items with relaxation values greater than the average value plus twice the standard deviation, for the key improvement items with relaxation values between the average value and the average value plus twice the standard deviation, and for the general improvement items with relaxation values less than the average value, it is recommended to continuously optimize through improving vehicle equipment configuration and optimizing rest site layout, etc. At the same time, attention should be paid to limiting such drivers to undertake high-risk tasks, real-time monitoring, and regular data analysis of risk drivers to adjust management measures to reduce the possibility of accidents.

[0108] In this embodiment, based on the evaluation results of the SE-DEA model, a complete driver safety management guarantee and continuous improvement mechanism is established. Through the synergistic effect of the four-dimensional guarantee system of organization, technology, system and resources, combined with the PDCA cycle management framework, the whole process closed-loop management from theoretical model to practical application is realized, providing a systematic solution to improve the safety management level of operating trucks.

[0109] A four-dimensional synergistic guarantee mechanism is innovatively established to ensure the effective implementation of the management system: (1) Organization guarantee aspect, the "decision-making-execution-operation" three-level management framework is built. The decision-making layer is composed of the safety management committee of enterprise executives, responsible for approving the annual safety budget (such as setting up special improvement fund for technology improvement type driver); the execution layer sets up safety operation center, equipped with professional data analyst n, trainer n, full-time responsible for high-risk driver management; The grass-roots set up safety supervisor, daily check the DSM monitoring report of technology improvement type driver (such as tracking the fatigue driving early warning record of chuan AG**37 driver).

[0110] (2) Technology guarantee level, research and development related intelligent safety management platform, integrated three core function modules: real-time monitoring module, realize 100% operation vehicle OBD data real-time acquisition, implement millisecond level abnormal behavior monitoring for technology improvement type driver; Intelligent scheduling module, develop customized algorithm for resource defect type driver, comprehensive consideration of vehicle environment parameters such as load coefficient, route complexity; Case database system, record the excellent operation record of high efficiency and safety type driver, support intelligent retrieval according to 15 dimensions such as weather, road conditions.

[0111] (3) System guarantee aspect, the dynamic adjustment mechanism is established: in the aspect of evaluation system, the technology improvement type driver is implemented "biweekly-monthly-quarterly" ladder evaluation, and the non-standard person triggers forced training; For resource defect type driver, set resource allocation optimization rate and other new KPI.

[0112] Incentive mechanism, implement safety performance points exchange priority order right (such as chuan AE**00 driver annual points reach 1500 points, obtain 20% high quality order tilt); Establish "safety mileage bank", every thousand kilometers of safe driving can be converted into 0.5 days of paid leave.

[0113] (4) Resource guarantee aspect, focus on optimizing physical environment: such as building standardized rest area in site, building training room that can simulate complex road conditions scene, and updating vehicle equipment for resource defect type driver, installing adaptive cruise control system and precise load monitoring device.

[0114] In order to ensure the continuous optimization of management system, based on the PDCA cycle framework, the data driven improvement mechanism is established: (1) Plan (Plan) stage Update SE-DEA model parameters every quarter, dynamically adjust input-output index system (such as can add "night lighting adaptability" index); According to the analysis result of slack variable, revise the improvement priority rule (such as set the task redistribution is automatically triggered when the tonnage slack value is greater than 5).

[0115] (2) Execution (Do) stage For the improved type of driver, the "3+1" improvement plan is implemented: 3 days of intensive training (including typical error case review) + 1 month of behavior correction tracking; develop a resource allocation optimization digital twin system to simulate the expected efficiency improvement of drivers such as AE**56 under different scheduling schemes.

[0116] (3) Check phase Malmquist index is used to analyze quarterly efficiency changes, and productivity improvement index targets for the technology improvement group and scale efficiency improvement targets for the resource allocation group are set; control charts are used to monitor key indicator fluctuations.

[0117] (4) Act phase Establish an abnormal case backtracking mechanism: for example, for drivers with a comprehensive safety index decrease of >15%, initiate a cross-department diagnosis process; update case library content every six months and eliminate outdated cases.

[0118] Scientifically evaluate the implementation effect of the management system. This example designs a systematic verification scheme, as shown in Table 9. Verification work mainly focuses on safety performance, accident rate monitoring, and economic benefits. A combination of quantitative and qualitative methods is used to ensure the comprehensiveness and reliability of the evaluation results.

[0119] Table 9 Verification methods for implementation effects

[0120] In terms of safety performance verification, the improvement rate of the comprehensive safety index, technical efficiency improvement rate, and scale efficiency improvement rate are used to analyze the differences before and after the implementation of the management system using paired sample t-test (confidence interval 95%). Data collection frequency is monthly, and quarterly moving averages are calculated to eliminate short-term fluctuations and ensure the stability of trend analysis.

[0121] For accident rate monitoring, an accident classification reporting system is established to calculate key indicators such as monthly average accidents, accident rate per thousand kilometers, and year-on-year decline rate. Chi-square test is used to analyze the significance of accident rate changes, and GIS spatial analysis technology is used to identify the correlation between accident-prone areas and management measures, providing a basis for optimizing resource allocation.

[0122] In terms of economic benefit evaluation, cost-benefit analysis is used to calculate direct cost savings, insurance cost reduction, and input-output ratio, and financial models are used to quantify net present value (NPV) to evaluate the long-term economic feasibility of the management system.

[0123] The present application is described in reference to the flowchart and / or block diagram of the method, apparatus (system) and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus generate a means for implementing the functions specified in the flowchart and / or block diagram. Figure 1 one or more flows and / or blocks. Figure 1 one or more blocks.

[0124] These computer program instructions can also be stored in a computer readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer readable memory produce a product including instruction apparatus, which implements the functions specified in the flowchart and / or block diagram. Figure 1 one or more flows and / or blocks. Figure 1 one or more blocks.

[0125] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a means for implementing the functions specified in the flowchart and / or block diagram. Figure 1 one or more flows and / or blocks. Figure 1 one or more blocks.

[0126] The principles and implementation of the present application are described in the specific embodiments, and the above description of the embodiments is only for the purpose of helping the reader to understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation and application range, and the above description of the present application should not be understood as a limitation.

[0127] Those skilled in the art will realize that the embodiments described herein are for the purpose of helping the reader to understand the principles of the present application, and should be understood as not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspiration disclosed in the present application without departing from the essence of the present application, and these modifications and combinations are still within the scope of protection of the present application.

Claims

1. A truck driver safe driving evaluation method based on SE_DEA, characterized in that, The method comprises the following steps: S1, constructing a multi-source traffic risk database, integrating vehicle characteristics, driving behavior characteristics, road characteristics and environmental characteristic data; S2, based on the multi-source traffic risk database, applying an SE-DEA model to calculate a comprehensive safety index of the driver; S3, according to the comprehensive safety index of the driver, classifying and managing the driver, and establishing a hierarchical monitoring mechanism and a targeted management strategy.

2. The SE-DEA based truck driver safe driving evaluation method according to claim 1, characterized in that, The S1 comprises: S11, collecting multi-source data, including truck GPS data, vehicle information data, weather data and street view data; S12, preprocessing the data, including de-duplication, deleting abnormal data and trajectory division; S13, extracting feature indexes, including vehicle characteristics, driving behavior characteristics, road characteristics and environmental characteristics, to form a risk database.

3. The SE-DEA based truck driver safe driving evaluation method according to claim 2, characterized in that, The vehicle characteristics in S13 include vehicle tonnage and factory time; the driving behavior characteristics include average daily driving time, overspeed frequency and fatigue driving time; the road characteristics include the number of lanes and lane type proportion; and the environmental characteristics include average temperature and weather proportion. The S2 comprises: S21, setting input variables as the feature indexes in the risk database, and setting output variables as the inverse of dangerous driving behaviors; 4. The SE-DEA based truck driver safe driving evaluation method according to claim 1, characterized in that, S22, constructing an SE-DEA linear programming model to output comprehensive efficiency, technical efficiency and scale efficiency indexes, and simultaneously outputting reference objects of non-effective drivers and corresponding index relaxation variables; S23, based on the comprehensive safety index, identifying high-efficiency safe drivers, technical improvement type drivers and resource deficiency type drivers. The SE-DEA model in S22 is represented as: The S3 specifically comprises the following steps:

5. The SE-DEA based truck driver safe driving evaluation method according to claim 4, characterized in that, S31, based on the super-efficiency value of the driver, judging the effectiveness of the decision-making unit DMU and taking values of corresponding index relaxation variables; In the formula, represents the super-efficiency value of the driver j . j o denotes the driver currently being evaluated; denotes a non-Archimedean infinitesimal; denotes the output dataset, the inverse of the abnormal driving behavior parameters of the driver j ; denotes the input dataset, the 4 characteristic indicators of the driver j ; , the slack variables of the input and the slack variables of the output; denotes the input and output shares of the driver being evaluated.

6. The SE-DEA based truck driver safe driving evaluation method according to claim 5, characterized in that, S32, for the driver with a comprehensive safety index less than 1, finding a linear combination of a group of drivers with a comprehensive safety index greater than or equal to 1 to generate the same output as the non-effective decision-making unit, and assigning weights to each reference object driver; The S3 classifies and manages the drivers specifically comprising: A1, defining that the comprehensive safety index, the technical safety index and the resource allocation index respectively correspond to the comprehensive technical efficiency, the pure technical efficiency and the scale efficiency calculated in S33, wherein the comprehensive technical efficiency represents the driver's comprehensive safety index, the technical efficiency represents the driver's technical safety index, and the scale efficiency represents the driver's resource allocation index, and the drivers are sorted according to the comprehensive safety index and the efficiency value; S33, continue to decompose the comprehensive technical efficiency TE into the pure technical efficiency PTE and the scale efficiency SE, the comprehensive technical efficiency TE is the ratio of the distance from the origin to the current evaluated driver (X) and the distance to the frontier projection point, i.e. the efficiency reference object (Y). )​ 7. The SE-DEA based truck driver safe driving evaluation method according to claim 6, characterized in that, A2, defining classification standards: The high-efficiency safe type is the driver with a comprehensive safety index greater than or equal to 1 and a relaxation variable of 0, which is a "peer benchmark" driver; The technical improvement type is the driver with a comprehensive safety index less than 1 and a technical safety index less than 1; The resource deficiency type is the driver with a comprehensive safety index less than 1 and a resource allocation index less than 1; The specific way of establishing a hierarchical monitoring mechanism and a targeted management strategy in S3 is: B1, designing a hierarchical monitoring mechanism: implementing basic monitoring on the high-efficiency safe driver; A3, based on the classification criteria, dividing the driver class, according to its efficiency reference object ( ) and slack variable The value of the safety training of the quantitative basis.

8. The SE-DEA based truck driver safe driving evaluation method according to claim 1, wherein, Implementing strict level monitoring on the technical improvement type driver; Implementing standard level monitoring on the resource deficiency type driver; ​ ​ B2, develop targeted management strategy: high efficiency and safety type with benchmarking and points reward as the main; technology improvement type with behavior correction and mandatory training as the main; Resource deficiency type with resource allocation optimization and intelligent scheduling as the main.

Images