Identification method of non-business vehicles engaged in business activities based on big data
By acquiring the operational feature vectors of commercial vehicles, training a logistic regression model, and using a big data platform to calculate and score features, the problems of data uniformity and hardware modification risks in existing technologies are solved, enabling rapid and accurate identification of business activities.
Patent Information
- Application Number
- CN202311357806.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-19
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2043-10-19
AI Technical Summary
In existing technologies, methods for identifying business activities based on vehicle trajectories suffer from data uniformity and the risk of hardware modification, leading to an increased probability of identification errors and failures to identify vehicles.
By acquiring the operational feature vectors of commercial vehicles, a logistic regression model is trained, and the Star-rocks big data platform is used to calculate features such as daily mileage, duration, number of charging times, and trajectory dispersion. Combined with the logistic regression model score, the commercial activities of non-commercial vehicles can be quickly identified.
It enables rapid and accurate identification of vehicles engaged in commercial activities, reduces the probability of false identification, and improves identification efficiency and accuracy.
Smart Images

Figure CN117422596B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vehicle behavior recognition, and in particular to a method for identifying non-business vehicles engaging in business behavior based on big data. Background Art
[0002] With the gradual popularization of vehicles engaged in commercial activities, more and more people are engaged in working with vehicles engaged in commercial activities. In order to evade supervision and reduce auto insurance costs, a large number of private cars are illegally engaged in the operation of vehicles engaged in commercial activities. In order to standardize the market management of vehicles engaged in commercial activities and reduce the vehicle underwriting risks of auto insurance institutions, vehicles engaged in commercial activities are increasingly widely used through vehicle trajectory identification.
[0003] The current method for identifying vehicles engaged in business activities mainly identifies vehicles engaged in business activities through the vehicle's driving trajectory; it does not include data such as the vehicle's SOC, mileage, and driving time.
[0004] This method has a certain recognition effect, but it is only based on vehicle trajectory and the data is relatively simple. At the same time, this data acquisition method is to add hardware to the vehicle, which may lead to risks of human modification of hardware data and damage to hardware, resulting in the probability of misidentification and failure to identify vehicles engaged in business activities. Summary of the invention
[0005] The purpose of the present invention is to provide a method for identifying non-business vehicles engaged in business activities based on big data, so as to solve the problems raised in the above-mentioned background technology.
[0006] In order to achieve the above-mentioned purpose, a method for identifying non-business vehicles engaged in business activities based on big data is provided, which includes the following method steps:
[0007] S1. Obtain the operation characteristic vector of the operating vehicle;
[0008] S2, training a logistic regression model based on the operational feature vector of the operating vehicle;
[0009] S3, obtaining the running feature vector of the vehicle to be identified;
[0010] S4. The running feature vector of the vehicle to be identified is passed into the logistic regression model for scoring, where:
[0011] The logistic regression model sets the score scale for the vehicle to be identified as an illegal vehicle by expressing the score as a linear expression of the logarithm of the ratio, that is:
[0012] Score = AB × log (Odds);
[0013] Where, Score is the score of the vehicle to be identified as an illegal vehicle; A is the score compensation, B is the score scale, A and B are both constants; Odds is the occurrence ratio.
[0014] As a further improvement of the present technical solution, the S1 uses the usage characteristics of the operating vehicle as an index to obtain the operating data of the operating vehicle, and extracts and processes the operating data into a feature vector, the characteristic attributes of the feature vector include the average daily mileage, the average daily driving time, the average daily charging times, the dispersion of driving time and the dispersion of vehicle trajectory.
[0015] As a further improvement of the present technical solution, the operating characteristic vector α of the operating vehicle is [M, T, C, D, L], where M represents the average daily mileage, T represents the average daily driving time, C represents the average daily charging times, D represents the dispersion of driving time, and L represents the dispersion of trajectory.
[0016] As a further improvement of this technical solution, the running feature vector α of the vehicle to be identified t =[M,T,C,D,L], where M represents the average daily mileage, T represents the average daily driving time, C represents the average daily charging times, D represents the dispersion of driving time, and L represents the dispersion of trajectories.
[0017] As a further improvement of this technical solution, the calculation process of M, T, C, D, L is as follows:
[0018] Calculate the average daily mileage M:
[0019] Through the national platform's big data storage platform Star-rocks, the mileage data of the past month for the training model, test model and verification model are retrieved according to the vehicle frame number vin, and the data is further processed to calculate the average daily mileage M.
[0020] Calculate the average daily driving time T:
[0021] Through the national platform's big data storage platform Star-rocks, the driving time data of the past month for the training model, test model and verification model are retrieved according to the vehicle frame number vin, and the data is further processed to calculate the average daily driving time T.
[0022] Calculate the average daily charging times C:
[0023] Through the national platform's big data storage platform Star-rocks, the charging data of the past month for the training model, test model and verification model are retrieved according to the vehicle, and the data is further processed to calculate the average daily charging times C.
[0024] Calculate the dispersion D of driving time:
[0025] Through the national platform's big data storage platform Star-rocks, the driving time data of the training model, test model and verification model in the past month are retrieved according to the chassis number; first, the time is divided into 24 driving time periods, each time period is 1 hour, and then the vehicle is labeled according to whether it has driven in the time period every day. The label of driving behavior is 1, and the label of no driving behavior is 0. The number of days with driving behavior in each time period in the past month is counted. If the number of driving days in this time period exceeds 20 days, the label f of this time period is recorded as 1. The sum of the labels of the 24 time periods is the dispersion D of the driving time, which is expressed as:
[0026]
[0027] Calculate the dispersion L of vehicle trajectories:
[0028] Through the national platform's big data storage platform Star-rocks, the vehicle location data of the training model, test model and verification model in the past month are retrieved according to the frame number. Among all the location data, the easternmost, westernmost, southernmost and northernmost positions are found to form a rectangular area. The positions of each day are placed in the rectangular area to form a trajectory image. The RMSE is used to compare the similarity of the trajectory images in the past month. RMSE is the dispersion L of the vehicle trajectory. The expression of RMSE is:
[0029]
[0030] n is the number of days in a month.
[0031] As a further improvement of this technical solution, α is put into the logistic regression model for training, so that the logistic regression model has the five vehicle identification features M, T, C, D, and L engaged in business behavior and the corresponding thresholds. t By comparing with the threshold, the probability that the vehicle is used for illegal operation is obtained.
[0032] As a further improvement of the technical solution, the probability that the vehicle is used for illegal operation is expressed as p, and the normal probability is 1-p;
[0033]
[0034] Odds indicates the occurrence ratio.
[0035] The p of using the vehicle for illegal operation can be expressed as:
[0036]
[0037] Among them, the higher the Score value, the lower the possibility that the vehicle to be identified is used for illegal operation.
[0038] Compared with the prior art, the present invention has the following beneficial effects:
[0039] In the method for identifying non-business vehicles engaged in business activities based on big data, the characteristic data of vehicles engaged in business activities are explored in the data of operating vehicles to determine the identification characteristics of vehicles engaged in business activities and the corresponding thresholds, and the characteristic weights of vehicles engaged in business activities under different identification characteristics are calculated; then, the dispersion of the driving trajectory of the vehicle to be identified during the working day, whether the mileage reaches the threshold, whether the driving time reaches the threshold, and whether the number of charging times reaches the threshold are judged; based on the characteristic data thresholds of vehicles engaged in business activities and the corresponding weights, it is finally determined whether the vehicle to be identified is a vehicle engaged in business activities, thereby achieving the effect of quickly identifying vehicles engaged in business activities. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 The present invention is a schematic flow chart of the steps of a method for identifying a non-business vehicle engaged in business activities. DETAILED DESCRIPTION
[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0042] The present invention provides a method for identifying non-commercial vehicles engaged in business activities based on big data, and uses a combination of Internet of Vehicles big data cleaning, feature analysis and other technologies. It is mainly aimed at the problem that auto insurance institutions and relevant functional departments are unable to effectively identify non-commercial vehicles as vehicles engaged in business operations; based on Internet of Vehicles big data (GB / T 32960 standard) and known driving behavior data of vehicles engaged in business activities, through big data analysis technology, vehicles engaged in business activities can be quickly identified, the commercial vehicle market can be standardized, and an effective commercial vehicle management and auto insurance underwriting system can be established.
[0043] A method for identifying non-business vehicles engaged in business activities based on big data, such as Figure 1 As shown, it includes the following method steps:
[0044] S1. Obtain the operation characteristic vector of the operating vehicle;
[0045] S2, training a logistic regression model based on the operational feature vector of the operating vehicle;
[0046] S3, obtaining the running feature vector of the vehicle to be identified;
[0047] S4. The running feature vector of the vehicle to be identified is passed into the logistic regression model for scoring, where:
[0048] The logistic regression model sets the score scale for the vehicle to be identified as an illegal vehicle by expressing the score as a linear expression of the logarithm of the ratio, that is:
[0049] Score = AB × log (Odds);
[0050] Where, Score is the score of the vehicle to be identified as an illegal vehicle; A is the score compensation, B is the score scale, A and B are both constants; Odds is the occurrence ratio.
[0051] The operation feature vector of the acquired operating vehicle is compatible with the operation feature vector of the vehicle to be identified, that is, the acquired feature data is the same, which are both thermal data of the vehicle, including vehicle trajectory, vehicle position, SOC, number of charging times, mileage, driving time, etc. Specifically, the slice reporting information of the Internet of Vehicles big data (based on the GB / T32960 standard) is obtained, and then the Internet of Vehicles slice data of the target vehicle is cleaned and calculated to obtain multiple feature data of the vehicle; then, the characteristic data of the vehicle engaged in business behavior is explored for the operating vehicle (that is, the known vehicle engaged in business behavior) data, and the identification features and corresponding thresholds of the vehicle engaged in business behavior are determined, and the characteristic weights of the vehicle engaged in business behavior under different identification features are calculated; then, the dispersion degree of the driving trajectory of the vehicle to be identified during the working day, whether the mileage reaches the threshold, whether the driving time reaches the threshold, and whether the number of charging times reaches the threshold are judged, and based on the characteristic data threshold and corresponding weight of the vehicle engaged in business behavior, it is finally determined whether the vehicle to be identified is a vehicle engaged in business behavior, so as to achieve the effect of quickly identifying the vehicle engaged in business behavior.
[0052] S1. Obtain the operating characteristic vector of the operating vehicle:
[0053] It uses the usage characteristics of the operating vehicle as an index to obtain the operating data of the operating vehicle, and extracts and processes the operating data into a feature vector. The characteristic attributes of the feature vector include the average daily mileage, the average daily driving time, the average daily charging times, the dispersion of driving time and the dispersion of vehicle trajectory;
[0054] The operating characteristic vector of the operating vehicle α = [M, T, C, D, L], where M represents the average daily mileage of the operating vehicle, T represents the average daily driving time of the operating vehicle, C represents the average daily number of charging times, D represents the dispersion of driving time, and L represents the dispersion of trajectory.
[0055] The calculation process of M, T, C, D, and L is as follows:
[0056] (1) Calculate the average daily mileage M:
[0057] Through the national platform's big data storage platform Star-rocks, the mileage data of the past month for the training model, test model and verification model are retrieved according to the vehicle frame number vin, and the data is further processed to calculate the average daily mileage M.
[0058] (2) Calculate the average daily driving time T:
[0059] Through the national platform's big data storage platform Star-rocks, the driving time data of the past month for the training model, test model and verification model are retrieved according to the vehicle frame number vin, and the data is further processed to calculate the average daily driving time T.
[0060] (3) Calculate the average daily charging times C:
[0061] Through the national platform's big data storage platform Star-rocks, the charging data of the past month for the training model, test model and verification model are retrieved according to the vehicle, and the data is further processed to calculate the average daily charging times C.
[0062] (4) Calculate the dispersion D of driving time:
[0063] Through the national platform's big data storage platform Star-rocks, the driving time data of the training model, test model and verification model in the past month are retrieved according to the chassis number; first, the time is divided into 24 driving time periods, each time period is 1 hour, and then the vehicle is labeled according to whether it has driven in the time period every day. The label of driving behavior is 1, and the label of no driving behavior is 0. The number of days with driving behavior in each time period in the past month is counted. If the number of driving days in this time period exceeds 20 days, the label f of this time period is recorded as 1. The sum of the labels of the 24 time periods is the dispersion D of the driving time, which is expressed as:
[0064]
[0065] (5) Calculate the dispersion L of vehicle trajectories:
[0066] Through the national platform's big data storage platform Star-rocks, the vehicle location data of the training model, test model and verification model in the past month are retrieved according to the frame number. Among all the location data, the easternmost, westernmost, southernmost and northernmost positions are found to form a rectangular area. The positions of each day are placed in the rectangular area to form a trajectory image. The RMSE is used to compare the similarity of the trajectory images in the past month. RMSE is the dispersion L of the vehicle trajectory. The expression of RMSE is:
[0067]
[0068] n is the number of days in a month.
[0069] S2. Training logistic regression model based on the operating feature vector of the operating vehicle:
[0070] It trains the logistic regression model through machine learning and puts the constructed feature vector α into the logistic regression model for training. At this time, the logistic regression model has the five vehicle identification features M, T, C, D, and L engaged in business activities and the corresponding thresholds. The α subsequently passed into the model t Compare with the threshold value to obtain the probability that the vehicle is used for illegal operation;
[0071] The probability that a vehicle is used for illegal operation is expressed as p, and the normal probability is 1-p;
[0072]
[0073] Odds indicates the occurrence ratio.
[0074] The p of using the vehicle for illegal operation can be expressed as:
[0075]
[0076] The logistic regression model sets the score scale for the vehicle to be identified as an illegal vehicle by expressing the score as a linear expression of the logarithm of the ratio, that is:
[0077] Score = AB × log (Odds);
[0078] The higher the Score value, the lower the possibility that the vehicle to be identified is used for illegal operation.
[0079] The values of constants A and B in the formula can be obtained by substituting two known or assumed scores into the calculation, setting two assumptions:
[0080] The expected score of the probability that the vehicle is used for illegal operation is P0;
[0081] The score PDO that doubles the probability that the vehicle is used for illegal operation;
[0082]
[0083] The logistic regression model calculates the ratio as follows:
[0084] log(Odds)=β0+β1x1+β2x2+...+β n x n .
[0085] After the model goes live, it is necessary to effectively monitor the model's performance and stability. The monitoring content includes the PSI (Population Stability Index) of the score distribution of illegal operation vehicles and the PSI of all variables involved in the model.
[0086] Model score distribution stability: Monitor whether there are changes in the distribution of the model's scoring results. Mainly compare the sample Real Data after the model goes live with the sample Train Data during model building.
[0087] Variable stability: Monitor whether there are changes in the distribution of the input variables of the model. Mainly compare the sample Real Data after the model goes live with the sample Train Data during model building.
[0088] The main indicator used is mainly PSI, which is used to measure whether there are significant differences in the partial scores or distributions of two groups of samples. The PSI calculation formula:
[0089]
[0090] The following reference judgment criteria can usually be used:
[0091] PSI ≤ 0.1, no difference
[0092] 0.1 < PSI ≤ 0.25, further judgment is required
[0093] 0.25 ≤ PSI, there is a significant shift, and the model needs to be adjusted.
[0094] S3. Obtain the operation feature vector of the vehicle to be recognized:
[0095] According to the calculation process (1), (2), (3), (4), (5), extract the characteristic daily driving mileage M, daily driving duration T, daily charging times C, dispersion degree D of driving time, and dispersion degree L of vehicle trajectory of the vehicle to be recognized, and convert them into the feature vector α t = [M, T, C, D, L], α t represents the operation feature vector of the vehicle to be recognized.
[0096] Pass α t into the model. The higher the result score, the lower the possibility that the vehicle to be recognized is an illegal operation vehicle.
[0097] The beneficial effects of the present invention are as follows: fast query speed, high accuracy, and batch recognition can be carried out.
[0098] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A method for identifying non-business vehicles engaged in business activities based on big data, characterized in that: It includes the following method steps: S1. Obtain the operation characteristic vector of the operating vehicle; S2, training a logistic regression model based on the operational feature vector of the operating vehicle; S3, obtaining the running feature vector of the vehicle to be identified; S4. The running feature vector of the vehicle to be identified is passed into the logistic regression model for scoring, where: The logistic regression model sets the score scale for the vehicle to be identified as an illegal vehicle by expressing the score as a linear expression of the logarithm of the ratio, that is: ; In the formula, is the score of the vehicle to be identified as an illegal vehicle; A is the score compensation, B is the score scale, and both A and B are constants; is the occurrence ratio; In S1, the operating vehicle usage property is used as an index to obtain the operating data of the operating vehicle, and the operating data is extracted and processed into a feature vector, the characteristic attributes of which include the average daily mileage, the average daily driving time, the average daily charging times, the dispersion of driving time and the dispersion of vehicle trajectory; The operating characteristic vector of the operating vehicle is α=[M,T,C,D,L], where M represents the average daily mileage, T represents the average daily driving time, C represents the average daily charging times, D represents the dispersion of driving time, and L represents the dispersion of trajectory; The running feature vector α of the vehicle to be identified t = [M,T,C,D,L], where M is the average daily mileage, T is the average daily driving time, C is the average daily charging times, D is the dispersion of driving time, and L is the dispersion of trajectories; The calculation process of M, T, C, D, L is as follows: Calculate the average daily mileage M: Through the national platform's big data storage platform Star-rocks, the mileage data of the past month in the training model, test model and verification model are retrieved according to the vehicle frame number vin, and the data is further processed to calculate the average daily mileage M; Calculate the average daily driving time T: Through the national platform's big data storage platform Star-rocks, the driving time data of the past month in the training model, test model and verification model are retrieved according to the vehicle frame number vin, and the data is further processed to calculate the average daily driving time T; Calculate the average daily charging times C: Through the national platform's big data storage platform Star-rocks, the charging data of the past month for the training model, test model and verification model are retrieved according to the vehicle, and the data is further processed to calculate the average daily charging times C; Calculate the dispersion D of driving time: Through the national platform's big data storage platform Star-rocks, the driving time data of the training model, test model and verification model in the past month are retrieved according to the vehicle frame number; first, the time is divided into 24 driving time periods, each time period is 1 hour, and then the vehicle is labeled according to whether it has driven in the time period every day. The label of driving behavior is 1, and the label of no driving behavior is 0. The number of days with driving behavior in each time period in the past month is counted. If the number of driving days in this time period exceeds 20 days, the label f of this time period is recorded as 1. The sum of the labels of the 24 time periods is the dispersion D of the driving time, which is expressed as: ; Calculate the dispersion L of vehicle trajectories: Through the national platform's big data storage platform Star-rocks, the vehicle location data of the training model, test model and verification model in the past month are retrieved according to the frame number. Among all the location data, the easternmost, westernmost, southernmost and northernmost positions are found to form a rectangular area. The positions of each day are placed in the rectangular area to form a trajectory image. The RMSE is used to compare the similarity of the trajectory images in the past month. RMSE is the dispersion L of the vehicle trajectory. The expression of RMSE is: ; n is the number of days in a month; Put α into the logistic regression model for training, so that the logistic regression model has five identification features of vehicles engaged in business activities, namely M, T, C, D, and L, and corresponding thresholds, and calculates the feature weights of vehicles engaged in business activities under different identification features; The α of the subsequent logistic regression model t Compared with the threshold, based on the characteristic data threshold and corresponding weight of the vehicle engaged in business activities, the probability of the vehicle being used for illegal operation is obtained; The probability that a vehicle is used for illegal operation is expressed as p, and the normal probability is 1−p; ; Odds indicates the occurrence ratio; The p of using the vehicle for illegal operation can be expressed as: ; in, The higher the score, the lower the possibility that the vehicle to be identified is used for illegal operation.
