Operating vehicle identification method based on high-speed data

Abstract

The invention relates to the field of road traffic, and provides a commercial vehicle identification method based on highway data so as to realize accurate identification of commercial vehicles. The method comprises the following steps: S1, collecting vehicle passing data and vehicle operation state data; s2, constructing a vehicle portrait according to the multi-dimensional features of the vehicle passing data; s3, constructing a training set and a test set, training the vehicle identification model by using the training set, and performing inspection and evaluation by using the test set; and S4, carrying out operation identification on the vehicle. According to the method, whether the vehicle is an operating vehicle or not is judged in an auxiliary manner by using the traffic data of the expressway, so that a low-cost data acquisition channel is increased, the recognition accuracy and efficiency are improved, and the expressway traffic data of the vehicle are deeply mined and analyzed through a big data analysis and machine learning technology; accurate identification of commercial vehicles is realized, and powerful support is provided for risk judgment of insurance companies.

Description

Technical Field

[0001] The invention relates to the field of road traffic, and in particular to an operating vehicle identification method based on high-speed data. Background Art

[0002] In the insurance industry, the phenomenon of "private cars running for business" and "commercial trucks being insured as non-commercial vehicles" is common, which brings great risks to insurance companies. The reason is that some private car owners engage in business activities in violation of regulations in order to obtain extra income, while some commercial trucks deliberately insure as non-commercial vehicles in order to reduce premiums. This information asymmetry causes insurance companies to face huge risks in the underwriting process. Traditional vehicle identification methods mainly rely on vehicle registration information and manual review. However, this method is difficult to effectively identify the actual operating status of the vehicle, thereby exacerbating the underwriting risk. In order to deal with this problem, the insurance industry usually adopts the method of purchasing enterprise data or public platform data to make risk judgments. However, these data sources have certain limitations, such as incomplete data, untimely updates, and difficulty in ensuring accuracy.

[0003] In general, traditional vehicle identification methods have the following defects and shortcomings:

[0004] 1. Data source limitations: Existing solutions mainly rely on purchasing enterprise data or public platform data, which usually have certain limitations and are difficult to fully and accurately reflect the actual operating status of vehicles. At the same time, there are also problems with the update speed and accuracy of these data, resulting in greater uncertainty in insurance companies' risk judgment.

[0005] 2. High identification cost: Purchasing enterprise data or public platform data requires high costs, which is a considerable expense for insurance companies. At the same time, due to data limitations, even if these data are purchased, the expected risk identification effect may not be achieved.

[0006] 3. Lack of real-time performance: Existing identification methods usually rely on static vehicle registration information and manual review, which makes it difficult to achieve real-time monitoring and dynamic adjustment of vehicle operating status. This leads to a large lag in risk judgment for insurance companies, making it impossible to respond to potential risks in a timely manner.

[0007] 4. Difficulty in dealing with complex situations: In actual operation, the operating status of vehicles is often complex and changeable. For example, some private cars only engage in operating activities during specific time periods, while some commercial trucks may occasionally drive as non-commercial vehicles. Existing recognition methods are often unable to cope with such complex situations, resulting in doubts about the accuracy and reliability of recognition results.

[0008] 5. Lack of intelligent means: Existing identification methods lack advanced intelligent means, such as big data analysis and machine learning, and cannot deeply mine the potential information in vehicle data to achieve accurate identification of operating vehicles. This limits the efficiency and accuracy of insurance companies in risk judgment. Summary of the invention

[0009] The purpose of the present invention is to overcome the deficiencies in the above-mentioned background technology and provide a method for identifying operating vehicles based on high-speed data to achieve accurate identification of operating vehicles.

[0010] The technical solution of the present invention is:

[0011] A method for identifying an operating vehicle based on high-speed data comprises the following steps:

[0012] S1. Collect vehicle traffic data and vehicle operation status data;

[0013] S2, constructing a vehicle profile based on the multi-dimensional features of vehicle traffic data;

[0014] S3, build training set and test set, use the training set to train the vehicle recognition model, and use the test set for inspection and evaluation;

[0015] S4. Identify the vehicle for operation.

[0016] In step S1, the vehicle traffic data includes the exit flow data and traffic trajectory data of trucks and buses, and the vehicle operation status data is the operation status of trucks and buses.

[0017] In step S2, the vehicle profile includes a license plate recognition tag, a charging data derived tag, a traffic feature tag, a traffic period distribution tag, a common entry / exit gate tag, a common OD tag, an airport route driving tag, a travel discreteness tag, a cargo volume analysis tag, and a speeding feature tag.

[0018] In the step S2: the license plate recognition label and the toll data derived label of the vehicle portrait are extracted according to the exit flow data; the traffic feature label, the traffic period distribution label, the common entry / exit label, the common OD label, the airport route driving label, the travel discreteness label, the cargo volume analysis label and the speeding feature label of the vehicle portrait are calculated according to the exit flow data and the traffic trajectory data.

[0019] The calculation of the airport route travel label includes:

[0020] ① Airport route identification: According to the geographic information system data, the longitude and latitude of all airports in the jurisdiction are obtained, and the distance to the nearest toll station of airport i is calculated, which is recorded as l i , the distance from airport i to iAll toll stations within are recorded as airport toll stations, and α is the distance coefficient;

[0021] ② Airport route marking: traverse each driving record and mark the driving records whose end points or passing points include airport toll booths as airport route driving records;

[0022] ③ Feature calculation: Count the number of times a vehicle drives to the airport toll station within the selected time period, and calculate the proportion of the number of airport route driving records within the selected time period to the total number of driving records.

[0023] The calculation of the travel dispersion label includes variance and frequent travel OD ratio;

[0024]

[0025] The speeding feature label includes the number of normal gantry section passes and the speeding ratio. The normal gantry section passes are the total number of normal gantry sections passed by the vehicle, and the number of gantries whose travel speed exceeds the speed limit standard accounts for the ratio of the normal gantry section passes.

[0026] In step S3, the vehicle identification model is a CatBoost model, including a passenger vehicle identification model and a truck vehicle identification model.

[0027] The step S3 includes matching the vehicle portrait with the operating status data, dividing the vehicle portrait into a training set and a test set in proportion, using the bus training set to train the bus operating vehicle recognition model, and using the bus test set to test and evaluate the trained bus operating vehicle recognition model, using the truck training set to train the truck operating vehicle recognition model, and using the truck test set to test and evaluate the trained truck operating vehicle recognition model.

[0028] The step S4 includes: collecting the traffic data of the vehicle to be identified, constructing a vehicle portrait and inputting it into a vehicle identification model, and obtaining the operating status of the vehicle.

[0029] The beneficial effects of the present invention are:

[0030] 1. Data comprehensiveness and accuracy: By using highway traffic data, the present invention can obtain dynamic information about vehicles in actual road operation, including travel time, speed, route, etc. These data are closer to the actual operation status of the vehicle. Compared with the traditional method that relies on vehicle registration information and manual review, highway traffic data has higher accuracy and comprehensiveness, and can more accurately reflect the nature of vehicle operation.

[0031] 2. Low cost and high efficiency: The present invention does not require additional purchase of enterprise data or public platform data, thus reducing the cost of data acquisition. In addition, through big data analysis and machine learning technology, the present invention can efficiently process and analyze a large amount of data, quickly identify operating vehicles, improve recognition efficiency, and reduce the cost of manual review.

[0032] 3. Real-time and dynamic: The present invention can obtain highway traffic data in real time, and realize real-time monitoring and dynamic adjustment of vehicle operation status. This enables insurance companies to discover potential risks in a timely manner and take corresponding risk management measures, thus improving the timeliness and effectiveness of risk control.

[0033] 4. Ability to cope with complex situations: The present invention can identify changes in the operating status of vehicles in different time periods and on different routes through deep learning and machine learning algorithms. This enables the present invention to cope with complex situations such as some private cars only operating in specific time periods, and some commercial trucks occasionally driving as non-commercial vehicles, thereby improving the accuracy and reliability of identification.

[0034] 5. Application of intelligent means: The present invention makes full use of advanced technologies such as big data analysis and machine learning to conduct in-depth mining and analysis of vehicle data. By building an efficient vehicle identification model, the present invention can achieve accurate identification of operating vehicles and improve the intelligence and automation level of risk judgment.

[0035] 6. Scalability and flexibility: The technical solution of the present invention is scalable and can continuously optimize and improve the recognition effect as the highway traffic data is continuously enriched and improved. At the same time, the present invention is also flexible and can be customized and adjusted according to the business needs and risk preferences of different insurance companies.

[0036] In summary, the present invention demonstrates significant advantages in terms of data comprehensiveness, cost-effectiveness, real-time performance, ability to cope with complex situations, application of intelligent means, scalability and flexibility, and provides insurance companies with a more accurate, efficient and low-cost operating vehicle identification solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the present invention.

[0038] Figure 2 is a bar chart showing the prediction explanations for the truck operation vehicle identification model.

[0039] Figure 3 It is a bee swarm diagram of the prediction explanation of the truck operation vehicle identification model.

[0040] Figure 4 It is a bar chart showing the prediction explanations of the bus operation vehicle identification model.

[0041] Figure 5 It is the bee swarm diagram of the prediction explanation of the bus operation vehicle identification model. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0043] like Figure 1 As shown, a method for identifying an operating vehicle based on high-speed data comprises the following steps:

[0044] S1. Collect vehicle traffic data and vehicle operation status data;

[0045] Collect historical vehicle traffic data and vehicle operation status data.

[0046] Vehicle traffic data includes the exit flow data and traffic trajectory data of trucks and buses. Vehicle operation status data refers to the operation status of trucks and buses.

[0047] The operation status data of trucks include commercial freight and non-commercial freight. The operation status data of passenger cars include family use and commercial rental.

[0048] The exit flow data is extracted from the exit toll station, and the passage trajectory data is extracted from the ETC gantry. Operation status data (partially manually annotated data is obtained from the outside).

[0049] Preprocess vehicle traffic data: clean the data and remove duplicate data (keep the latest data) and invalid data (such as missing key information, abnormal location data, etc.) to ensure data quality.

[0050] S2, constructing a vehicle profile based on the multi-dimensional features of vehicle traffic data;

[0051] The vehicle portrait includes each vehicle’s license plate recognition label, toll data derived label, traffic characteristic label, traffic period distribution label, common entry / exit gate label, common OD label, airport route driving label, travel dispersion degree label, cargo volume analysis label, and speeding characteristic label.

[0052] S2.1. Extract the license plate recognition tag of the vehicle portrait and the toll data derived tag based on the export flow data.

[0053] 1) License Plate Recognition Tag

[0054] Specifically include: license plate number (the unique identifier for each vehicle), license plate color (an important feature for distinguishing vehicle types), vehicle ownership location (information about the vehicle registration location, e.g., "Zhejiang A" represents Hangzhou, Zhejiang).

[0055] 2) Derived tags from toll data

[0056] Specifically include: vehicle type and category (such as Class I passenger vehicle, Class II freight vehicle, etc.), number of axles (the number of axles of the vehicle), approved passenger capacity / number of seats (the maximum number of passengers allowed by the vehicle design), approved load weight (the maximum load weight allowed for the vehicle), vehicle dimensions (height, length, width of the vehicle), specific types of vehicles (new energy vehicles, semi-trailers, container trucks, official vehicles).

[0057] S2.2. Calculate the traffic feature tags, traffic time period distribution tags, frequently entered / exited toll gates tags, frequently traveled OD tags, airport route driving tags, travel dispersion degree tags, cargo capacity analysis tags, and speeding feature tags of the vehicle portrait based on the exit transaction data and traffic trajectory data.

[0058] 1) Traffic feature tags include: overall traffic features, ETC traffic features, green channel traffic features, official vehicle traffic features, container traffic features, holiday traffic features, night traffic features, late-night traffic features.

[0059] ① Overall traffic features include: number of trips, travel duration, travel mileage.

[0060] Number of trips: The total number of times the vehicle passes within the selected time period.

[0061] Travel duration: The total travel duration, average travel duration, maximum travel duration, minimum travel duration, as well as the standard deviation and median of the travel duration of the vehicle within the selected time period.

[0062] Travel mileage: The total travel mileage, average travel mileage, maximum travel mileage, minimum travel mileage, as well as the standard deviation and median of the travel mileage of the vehicle within the selected time period.

[0063] ② ETC traffic features include: the number of times, duration, and mileage of the vehicle using ETC for passing within the selected time period.

[0064] ③ Green channel traffic features: The number of times, duration, and mileage of the vehicle passing through the green channel within the selected time period.

[0065] ④ Official vehicle traffic features: The number of times, duration, and mileage of the vehicle passing as an official vehicle within the selected time period.

[0066] ⑤ Container traffic features: The number of times, duration, and mileage of the vehicle transporting containers for passing within the selected time period.

[0067] ⑥Holiday traffic characteristics: the number, duration and mileage of vehicles traveling during national statutory holidays in the selected time period.

[0068] The selected time period is one or more combinations of 30 days, 90 days, 365 days, and the entire historical period.

[0069] 2) Traffic period distribution label

[0070] Taking into account the differences in travel period characteristics between operating vehicles and non-operating vehicles, statistics are collected on the number of times, duration, mileage and earliest travel time of vehicles in the first half of the night, the second half of the night and the noon.

[0071] Defined time periods: The first half of the night is from 20:00 to 24:00, the second half of the night is from 00:00 to 05:00, and the midday is from 10:30 to 14:30.

[0072] Handling travel that spans multiple days: When the start and end times of a trip span multiple days, it is necessary to split the trip by day for ease of calculation to ensure that the start and end times after the split are on the same day.

[0073] For example: if the trip starts at 18:00 on October 1, 2024 and ends at 03:00 on October 3, 2024, it needs to be divided into three trips, namely: 18:00 to 24:00 on October 1, 2024, 00:00 to 24:00 on October 2, 2024, and 00:00 to 03:00 on October 3, 2024.

[0074] Calculation of time for passage in different time periods:

[0075] The duration of the first half of the night = max(0,min(24:00,t end )-max(20:00,t start ))

[0076] The duration of the second half of the night = max(0,min(05:00,t end )-max(00:00,t start ))

[0077] Noon traffic time = max(0,min(14:30,t end )-max(10:30,t start ))

[0078] Among them, t start is the travel start time, t end The end time of the trip.

[0079] Earliest travel time; the time in the earliest travel trajectory data (accurate to the month); used to evaluate the age of the vehicle.

[0080] 3) Frequent entry / exit labels

[0081] Indicates the most commonly used entrances and exits for vehicles during the selected time period, as well as the corresponding frequencies and proportions, which helps to understand the vehicles' entry habits and preferences.

[0082] 4) Common OD label

[0083] Indicates the most common starting and ending points of vehicles during the selected time period as well as the corresponding frequency and proportion, which helps to understand the main travel routes and destinations of vehicles.

[0084] 5) Airport route driving label

[0085] It indicates the frequency and proportion of vehicles driving to the airport toll station during the selected time period, which helps to analyze the frequency of vehicles passing through the airport.

[0086] The specific steps include:

[0087] ① Airport route identification

[0088] According to the geographic information system (GIS) data, the latitude and longitude of all airports in the jurisdiction are obtained, and the distance to the nearest toll station of airport i is calculated, which is recorded as l i , the distance from airport i to i All toll stations within are recorded as airport toll stations, and α is the distance coefficient, which is generally 1.5.

[0089] ②Airport route signs

[0090] Traverse each driving record to determine whether its destination or waypoint contains an airport toll station. If so, mark it as an airport route driving record.

[0091] The driving record is a complete trip information, including the exit flow data generated by the exit toll station of the trip, and the passage trajectory data generated by all ETC gantries passed by the trip.

[0092] ③Feature calculation

[0093] Frequency calculation: Count the number of times a vehicle drives to the airport toll station during the selected time period, recorded as F i,T , i represents the vehicle ID, and T represents the selected time period.

[0094] Proportion calculation: Calculate the ratio of the number of airport route driving records in the selected time period to the total number of driving records, recorded as N i,T is the total number of trips of vehicle i in the selected time period T.

[0095] 6) Trip dispersion label

[0096] The degree of dispersion of vehicle travel is measured by the variance of all travel OD times and the proportion of frequent travel OD.

[0097] The larger the variance of the OD times of all vehicle trips, the greater the difference in the trip times of different ODs, that is, the more uneven the distribution of vehicle trips is, the higher the degree of dispersion is; otherwise, the lower the degree of dispersion is.

[0098] However, in the process of data statistics, it was found that a large number of vehicles had multiple ODs and the number of ODs was 1, which distorted the variance rule. Therefore, the frequent OD ratio indicator was introduced to jointly judge the degree of travel dispersion. The higher the frequent OD ratio, the more concentrated the vehicle travel is on a few ODs, and the degree of dispersion is relatively low. On the contrary, it means that the degree of dispersion is high. OD is defined as the combination of the entrance toll station and the exit toll station, such as "Zhejiang Airport Station-Zhejiang Tongxiang Station".

[0099] ① Variance of travel dispersion:

[0100]

[0101] ② The OD ratio of frequent trips in terms of travel dispersion (rate):

[0102]

[0103] 7) Cargo capacity analysis

[0104] Represents the average, minimum, maximum, standard deviation and median of the cargo loads of vehicles passing through during the selected time period.

[0105] 8) Speeding feature label

[0106] Since there is no direct speed detection for high-speed traffic data, it can only be calculated by dividing the distance between adjacent gantries by the travel time:

[0107]

[0108] in, represents the travel speed of vehicle c between gantries i and j, l i and l j is the position stake number of gantry i and j, and is the time when vehicle c passes through gantries i and j.

[0109] In the above formula: For the sake of uniformity, only the travel speed between adjacent gantries is calculated; l i and l jIt is the position stake number of gantries i and j. Strictly speaking, it is not the position where vehicle c is detected by gantries i and j. According to the specific gantry layout and the power difference of the detection antenna, the position where the vehicle is detected by the gantry may be offset by several hundred meters at most. This offset has little effect on gantries with larger intervals, but has a greater impact on gantries with a shorter distance of only 2 or 3 kilometers, which will significantly affect the speed detection result.

[0110] Therefore, an improvement strategy is proposed: In the process of comprehensive evaluation of the travel speed distribution of the gantry section, two core judgment criteria are set based on industry experience and actual road conditions to define the normal gantry section: First, the 15th percentile of the speed distribution (speed_15) must be higher than the threshold of 40 kilometers per hour; second, for the 85th percentile of the speed distribution (speed_85), it must be kept below 120 kilometers per hour on regular roads, and for special roads with a speed limit of 100 kilometers per hour, the value of speed_85 must be strictly controlled at 100 kilometers per hour. Only when the speed data of the gantry section meets the above two strict standards at the same time can it be identified as a gantry section that meets normal operation requirements.

[0111] Ultimately, the speeding situation characteristics correspond to two indicators:

[0112] Indicator 1 is the number of normal gantry sections passed: the total number of normal gantry sections that vehicles pass through, referred to as the "number of passes";

[0113] The second indicator is the speeding ratio: it measures the ratio of the number of times vehicle c passes through gantries with a travel speed exceeding the speed limit standard among all the normal gantry sections passed by the vehicle c, referred to as the "speeding ratio".

[0114] S3, build training set and test set, use the training set to train the vehicle recognition model, and use the test set for inspection and evaluation;

[0115] The vehicle portrait of each vehicle is matched with the operating status data, and divided into training set and test set in proportion. The training set is used to train the vehicle recognition model, and the test set is used for inspection and evaluation.

[0116] The vehicle identification model includes the passenger vehicle identification model and the truck vehicle identification model. The vehicle identification model uses the CatBoost model as the core model of this project. CatBoost has excellent ability to process classification features and is highly robust to noise and outliers in the data, which is suitable for the complex and changeable data environment in this method.

[0117] The bus operating vehicle recognition model is trained using the bus training set, and the truck operating vehicle recognition model is trained using the truck training set. The recognition accuracy of the model is continuously improved by performing multiple iterations of training, adjusting model parameters, and optimizing feature combinations. Combined with cross-validation, the stability and generalization ability of the model on different data sets are ensured.

[0118] The trained bus operating vehicle recognition model is tested and evaluated using the passenger bus test set, and the trained truck operating vehicle recognition model is tested and evaluated using the truck test set. The recognition accuracy and practicality of the model are evaluated by comparing the degree of conformity between the model output results and the actual operating status.

[0119] S4. Identify the vehicle for operation;

[0120] Collect the traffic data of vehicles to be identified, build a vehicle portrait and input it into the vehicle identification model to obtain the vehicle's operating status.

[0121] Example

[0122] 1. Collect sample data of Shanghai-Hangzhou-Ningbo Expressway for two years (2022 and 2023). Among them, there are 9,954 vehicles in total, including 1,022 trucks for commercial freight (normal trucks), 3,287 trucks for non-commercial freight (commercial trucks are insured as non-commercial), 3,916 passenger cars for family use (normal passenger cars), and 1,729 passenger cars for commercial rental (private cars for commercial operation).

[0123] 2. After randomly shuffling the order of truck samples and bus samples, take 20% as the test set and the rest as the training set, as shown in Table 1.

[0124]

[0125] Table 1

[0126] 3. Truck operating vehicle identification model

[0127] The truck training set was input into the truck operating vehicle recognition model training. After the model training was completed, the accuracy was verified by the test set and was 81.79%. The specific results are shown in Table 2.

[0128]

[0129] Table 2

[0130] The SHAP (SHapley Additive exPlanations) method is used to interpret the model and measure the contribution of different features to the results. Figure 2As shown in the figure, it can be found that the top 10 features with the greatest contribution include vehicle origin (e.g., "Zhe A"), earliest travel time, license plate color, regular travel OD label, number of trips in the late night, variance of travel dispersion, number of passes through gantry sections, average load, maximum load, and proportion of speeding. It should be noted that the label contribution is measured by calculating the average marginal contribution of all feature combinations involving the label.

[0131] Through further SHAP analysis, a swarm plot is obtained (see Figure 3 ). Each point (i.e., "bee") in the figure represents a sample. The darker the color of the point, the higher the feature value of the corresponding sample; conversely, the bluer the color, the lower the feature value. The position of the point on the X-axis, moving to the left indicates that the feature makes the sample more likely to be judged as a non-operating vehicle, and moving to the right indicates that it is more likely to be judged as an operating vehicle. Accordingly, it can be seen that features such as a high number of appearances in the late night, small variance of travel dispersion, few normal passes through gantry sections, large maximum and average load values, and small proportion of speeding are more likely to make a vehicle be judged as an operating freight vehicle.

[0132] 4. Passenger vehicle operating vehicle recognition model

[0133] The training set of passenger vehicles is input into the passenger vehicle operating vehicle recognition model for training. After the model training is completed, it is verified by the test set, and the accuracy is 93.98%. The specific results are shown in Table 3.

[0134]

[0135] Table 3

[0136] Use the SHAP method to interpret the model and measure the contribution degree of different features to the result, as Figure 4 shown. It can be found that the features with the greatest contribution include license plate color, vehicle origin, proportion of regular travel OD, number of trips in the late night, proportion of speeding, variance of travel dispersion, the second largest regular exit within 90 days, the third largest regular exit within 30 days, number of trips, regular travel OD label, etc.

[0137] Through further SHAP analysis, a swarm plot is obtained (see Figure 5 ). It can be found that features such as a small proportion of regular travel OD, a high number of appearances in the late night, a small proportion of speeding, a small variance of travel dispersion, and a large number of trips are more likely to make a vehicle be judged as an operating taxi vehicle.

[0138] 5. Model test and evaluation

[0139] The data in the model test link includes 2,653 trucks and 3,248 passenger vehicles. The model recognition results are shown in Table 4.

[0140] Vehicle Type Match quantity result truck 2653 1,612 non-commercial trucks and 1,041 commercial freight trucks bus 3248 2,886 vehicles for family use and 362 vehicles for business rental

[0141] Table 4

[0142] The recognition results were tested, and the feedback showed that the accuracy rate of trucks was 72.9% and the accuracy rate of buses was 95.9%, which fully demonstrated the effectiveness and reliability of the model in practical applications.

[0143] The preferred embodiments of the present invention are shown in the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in this specification. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

Claims

1. A method for identifying an operating vehicle based on high-speed data, comprising the following steps: S1. Collect vehicle traffic data and vehicle operation status data; S2, constructing a vehicle profile based on the multi-dimensional features of vehicle traffic data; S3, build training set and test set, use the training set to train the vehicle recognition model, and use the test set for inspection and evaluation; S4. Identify the vehicle for operation.

2. The method for identifying commercial vehicles based on high-speed data according to claim 1, characterized in that: In step S1, the vehicle traffic data includes the exit flow data and traffic trajectory data of trucks and buses, and the vehicle operation status data is the operation status of trucks and buses.

3. The method for identifying commercial vehicles based on high-speed data according to claim 1, characterized in that: In step S2, the vehicle profile includes a license plate recognition tag, a charging data derived tag, a traffic feature tag, a traffic period distribution tag, a common entry / exit gate tag, a common OD tag, an airport route driving tag, a travel discreteness tag, a cargo volume analysis tag, and a speeding feature tag.

4. The method for identifying commercial vehicles based on high-speed data according to claim 3, characterized in that: In the step S2: the license plate recognition label and the toll data derived label of the vehicle portrait are extracted according to the exit flow data; the traffic feature label, the traffic period distribution label, the common entry / exit label, the common OD label, the airport route driving label, the travel discreteness label, the cargo volume analysis label and the speeding feature label of the vehicle portrait are calculated according to the exit flow data and the traffic trajectory data.

5. The method for identifying commercial vehicles based on high-speed data according to claim 4, characterized in that: The calculation of the airport route travel label includes: ① Airport route identification: According to the geographic information system data, the longitude and latitude of all airports in the jurisdiction are obtained, and the distance to the nearest toll station of airport i is calculated, which is recorded as l i , the distance from airport i to i All toll stations within are recorded as airport toll stations, and α is the distance coefficient; ② Airport route marking: traverse each driving record and mark the driving records whose end points or passing points include airport toll booths as airport route driving records; ③ Feature calculation: Count the number of times a vehicle drives to the airport toll station within the selected time period, and calculate the proportion of the number of airport route driving records within the selected time period to the total number of driving records.

6. The method for identifying commercial vehicles based on high-speed data according to claim 4, characterized in that: The calculation of the travel dispersion label includes variance and frequent travel OD ratio; 7. The method for identifying commercial vehicles based on high-speed data according to claim 4, characterized in that: The speeding feature label includes the number of normal gantry section passes and the speeding ratio. The normal gantry section passes are the total number of normal gantry sections passed by the vehicle, and the number of gantries whose travel speed exceeds the speed limit standard accounts for the ratio of the normal gantry section passes.

8. The method for identifying commercial vehicles based on high-speed data according to claim 1, characterized in that: In step S3, the vehicle identification model is a CatBoost model, including a passenger vehicle identification model and a truck vehicle identification model.

9. The method for identifying commercial vehicles based on high-speed data according to claim 8, characterized in that: The step S3 includes matching the vehicle portrait with the operating status data, dividing the vehicle portrait into a training set and a test set in proportion, using the bus training set to train the bus operating vehicle recognition model, and using the bus test set to test and evaluate the trained bus operating vehicle recognition model, using the truck training set to train the truck operating vehicle recognition model, and using the truck test set to test and evaluate the trained truck operating vehicle recognition model.

10. The method for identifying commercial vehicles based on high-speed data according to claim 1, characterized in that: The step S4 includes: collecting the traffic data of the vehicle to be identified, constructing a vehicle portrait and inputting it into a vehicle identification model, and obtaining the operating status of the vehicle.

Images