Vehicle high-speed toll evasion credit scoring method based on large model and controller

Abstract

The invention relates to the technical field of data analysis, particularly provides a vehicle high-speed toll evasion credit scoring method based on a large model and a controller, and aims to solve the problem of how to score the toll evasion credit of a high-speed vehicle. In order to achieve the purpose, the method provided by the invention comprises the steps of obtaining historical high-speed driving data of at least one target vehicle; determining the high-speed toll evasion probability of each target vehicle according to the historical high-speed driving data of each target vehicle and a pre-trained high-speed toll evasion probability prediction model, wherein the high-speed toll evasion probability prediction model is a large model obtained by training based on an integrated algorithm; and determining a credit score of the target vehicle according to the high-speed toll evasion probability. According to the invention, high-speed toll evasion probability prediction and credit scoring can be carried out on a plurality of target vehicles, the prediction precision of vehicle high-speed toll evasion is improved, bad toll evasion behaviors can be effectively standardized, the toll road network environment is purified, a more accurate vehicle toll evasion verification basis is provided for high-speed workers, and the working efficiency of the high-speed workers is improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically to a credit scoring method and controller for vehicle highway toll evasion based on a large model. Background Technology

[0002] With the widespread adoption of the Electronic Toll Collection System (ETC) during highway toll collection, highway traffic efficiency has been greatly improved. However, driven by economic interests, toll evasion has become increasingly common.

[0003] Existing technologies typically combine big data analytics and artificial intelligence to improve the accuracy of highway toll collection through vehicle trajectory tracking and payment behavior analysis. However, without a robust vehicle credit system, it's impossible to accurately predict whether a vehicle will evade tolls, hindering the provision of quality service to vehicles with good credit and the effective regulation of toll evasion, thus preventing a cleaner toll network environment. Therefore, optimizing the scoring and classification of vehicle toll credit ratings is crucial. This not only provides highway workers with more scientific and accurate risk assessments for vehicle collection but also has significant practical implications for the informatization and intelligentization of highways.

[0004] Accordingly, a new technical solution is needed in this field to solve the above problems. Summary of the Invention

[0005] To overcome the above-mentioned shortcomings, this invention is proposed to solve, or at least partially solve, the technical problem of how to conduct toll evasion credit scoring for highway vehicles.

[0006] In a first aspect, a credit scoring method for vehicle highway toll evasion based on a large model is provided, the method comprising:

[0007] Acquire historical high-speed driving data for at least one target vehicle;

[0008] Based on the historical high-speed driving data of each target vehicle and the pre-trained high-speed toll evasion probability prediction model, the high-speed toll evasion probability of each target vehicle is determined. The high-speed toll evasion probability prediction model is a large model trained based on an ensemble algorithm.

[0009] Based on the probability of highway toll evasion, a credit score is determined for each target vehicle.

[0010] In one technical solution of the above-mentioned vehicle toll evasion credit scoring method based on a large model, the method further includes obtaining the toll evasion probability prediction model according to the following method:

[0011] Obtain a training dataset, which is a dataset of high-speed driving data containing multiple samples of high-speed vehicles labeled with either toll evasion or non-toll evasion tags.

[0012] Construct an initial ensemble model, which is an ensemble model composed of multiple learners;

[0013] The high-speed toll evasion probability prediction model is trained based on the training dataset and the initial ensemble model.

[0014] In one technical solution of the above-mentioned vehicle highway toll evasion credit scoring method based on a large model, the step of training the highway toll evasion probability prediction model based on the training dataset and the initial ensemble model includes:

[0015] Based on the gradient boosting tree algorithm, the importance value of each feature variable in the training dataset is determined;

[0016] The training dataset is grouped according to each feature variable, and the information value of each feature variable is determined based on the data grouping results.

[0017] Based on the information value and the importance value, the target feature variable is determined from all feature variables in the training dataset;

[0018] Based on the target feature variables and the training dataset, the initial ensemble model is trained to obtain a high-speed toll evasion probability prediction model.

[0019] In one technical solution of the above-mentioned vehicle highway toll evasion credit scoring method based on a large model, the step of grouping the training dataset according to each feature variable and determining the information value of each feature variable based on the data grouping results includes:

[0020] Obtain the total number of highway vehicle samples corresponding to the toll evasion label and the total number of highway vehicle samples corresponding to the non-toll evasion label in the training dataset;

[0021] Based on the values ​​of the feature variables in the training dataset, the training dataset is grouped to obtain multiple data groups;

[0022] Obtain the number of highway vehicle samples corresponding to the toll evasion label and the number of highway vehicle samples corresponding to the non-toll evasion label in each data group, and determine the information value of the feature variable in each data group according to the following formula:

[0023]

[0024] Among them, IV iThe information value of the feature variable in the i-th data group, Bad i Let Good be the number of highway vehicle samples corresponding to the toll evasion label in the i-th data group. i Let Bad be the number of highway vehicle samples corresponding to the "no toll evasion" label in the i-th data group. T Good represents the total number of highway vehicle samples corresponding to the toll evasion label in the training dataset. T The total number of high-speed vehicle sample data corresponding to the "no toll evasion" label in the training dataset;

[0025] The information value of the feature variable is determined by summing the information values ​​of the feature variable in each data set.

[0026] In one technical solution of the above-mentioned vehicle highway toll evasion credit scoring method based on a large model, the step of determining the target feature variable from all feature variables in the training dataset based on the information value and the importance value includes:

[0027] The information value and importance value are processed, and the sum of the information value and importance value corresponding to each feature variable is determined.

[0028] The feature variable whose sum of information value and importance value is greater than a preset threshold is determined as the target feature variable.

[0029] In one technical solution of the above-mentioned vehicle highway toll evasion credit scoring method based on a large model, the learner in the initial ensemble model includes a meta-learner and at least two base learners;

[0030] The step of training the initial ensemble model based on the target feature variables and the training dataset to obtain a high-speed toll evasion probability prediction model includes:

[0031] Based on the K-fold cross-validation algorithm, the training dataset and the base learner are used to generate a weighted training dataset, which is a training dataset after weighting the training data in the training dataset according to the output of the base learner.

[0032] The meta-learner is trained based on the weighted training dataset to obtain the trained meta-learner;

[0033] Based on the trained meta-learner, a high-speed toll evasion probability prediction model is obtained.

[0034] In one technical solution of the above-mentioned vehicle highway toll evasion credit scoring method based on a large model, obtaining the training dataset includes:

[0035] Based on oversampling technology, high-speed driving data of multiple high-speed vehicles labeled with toll evasion tags or not toll evasion tags are obtained to obtain the original training dataset, which includes multiple feature variables.

[0036] The high-speed driving data is processed to obtain the final training dataset; the data processing includes handling missing and outlier values ​​of feature variables in the high-speed driving data and / or normalizing the high-speed driving data based on the deviation standardization method;

[0037] The process of handling missing and outlier values ​​for feature variables in the high-speed driving data includes:

[0038] If the number of missing values ​​of the feature variable does not meet a preset range, the feature variable is deleted; if the number of missing values ​​of the feature variable meets the preset range, a random forest classifier is used to determine the missing values ​​of the feature variable; and / or,

[0039] Based on the univariate outlier detection method, the outliers of the feature variables are obtained and corrected.

[0040] In one technical solution of the above-mentioned vehicle highway toll evasion credit scoring method based on a large model, determining the credit score of each target vehicle based on the highway toll evasion probability includes:

[0041] The probability of the target vehicle not evading highway tolls is determined based on the probability of the target vehicle evading highway tolls, and the ratio between the probability of evading highway tolls and the probability of not evading highway tolls is determined.

[0042] Based on the probability of highway toll evasion and the following preset credit scoring function, the credit score of the target vehicle is determined:

[0043] Score = A - Bln(Odds)

[0044] Wherein, Score is the credit score of the target vehicle, A and B are constants, and Odds is the ratio between the probability of highway toll evasion and the probability of highway toll non-evasion.

[0045] In one technical solution of the above-mentioned vehicle highway toll evasion credit scoring method based on a large model, the method further includes determining constants A and B in the credit scoring function according to the following steps:

[0046] Obtain a preset vehicle score, a preset probability ratio, and a preset vehicle change value; the preset vehicle change value is the change value of the preset vehicle score corresponding to twice the preset probability ratio.

[0047] Based on the preset vehicle score, preset probability ratio, preset vehicle variation value, and the following formula, constants A and B are determined:

[0048] A = p0 + Bln(Odds0)

[0049]

[0050] Where p0 is the preset vehicle rating, Odds0 is the preset probability ratio, PDO is the preset vehicle variation value, and A and B are constants.

[0051] In a second aspect, a controller is provided, comprising at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program that, when executed by the at least one processor, implements the method described in any of the above-described technical solutions of the vehicle highway toll evasion credit scoring method based on a large model.

[0052] In a third aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored therein, the program codes being adapted to be loaded and run by a processor to perform the method described in any of the above-described technical solutions of the vehicle highway toll evasion credit scoring method based on a large model.

[0053] The above-described technical solutions of the present invention have at least one or more of the following beneficial effects:

[0054] In implementing the technical solution provided by this invention, the invention can acquire historical highway driving data of at least one target vehicle; based on the historical highway driving data of each target vehicle and a pre-trained highway toll evasion probability prediction model, the highway toll evasion probability of each target vehicle is determined, wherein the highway toll evasion probability prediction model is a large model trained based on an ensemble algorithm; and based on the highway toll evasion probability, the credit score of each target vehicle is determined. In this way, highway toll evasion probability prediction and credit scoring can be performed on multiple target vehicles, which is conducive to establishing a sound vehicle credit system, thereby improving the accuracy of predicting whether a vehicle is evading highway tolls, effectively regulating improper toll evasion behavior, and purifying the toll road network environment. At the same time, the credit scores of target vehicles can also provide highway workers with more accurate basis for verifying vehicle toll evasion, improving the work efficiency of highway workers.

[0055] Furthermore, by calculating the importance and information value of each feature variable in the training dataset, the influence of data distribution on feature variables can be reduced, and the target feature variable with the greatest correlation to the prediction result can be selected. This can improve the training effect and accuracy of the initial ensemble model, while also avoiding overfitting of the initial ensemble model and improving the model's adaptability to new data.

[0056] Furthermore, the base learners in the initial ensemble learner can be used to weight the training dataset through K-fold cross-validation to obtain a weighted training dataset. This weighted training dataset can then be used as input data for the meta-learner to further optimize the model training process of the meta-learner, thereby improving the performance of the final high-speed toll evasion probability prediction model. Attached Figure Description

[0057] The disclosure of this invention will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Wherein:

[0058] Figure 1 This is a schematic diagram of the main steps of a vehicle highway toll evasion credit scoring method based on a large model according to an embodiment of the present invention;

[0059] Figure 2 This is a flowchart illustrating the main steps of training a high-speed toll evasion probability prediction model according to one embodiment of the present invention.

[0060] Figure 3 This is a schematic diagram of the main steps in establishing a credit scoring function according to one embodiment of the present invention.

[0061] Figure 4 This is a schematic diagram of the main steps in training a high-speed toll evasion probability prediction model according to another embodiment of the present invention. Detailed Implementation

[0062] Some embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0063] In the description of this invention, "module" and "processor" can include hardware, software, or a combination of both. A module can include hardware circuitry, various suitable sensors, communication ports, memory, and may also include software components, such as program code, or a combination of software and hardware. A processor can be a central processing unit, microprocessor, image processor, digital signal processor, or any other suitable processor. The processor has data and / or signal processing capabilities. The processor can be implemented in software, in hardware, or a combination of both. Computer-readable storage media include any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The terms "at least one A or B" or "at least one of A and B" have a similar meaning to "A and / or B" and can include only A, only B, or A and B. The singular terms "a" or "this" can also include plural forms.

[0064] The relevant user personal information involved in the various embodiments of the present invention is processed in strict accordance with the requirements of laws and regulations, following the principles of legality, legitimacy, and necessity, based on the reasonable purpose of the business scenario, and includes personal information that users actively provide or generate during the use of the product / service, as well as personal information obtained with user authorization.

[0065] The personal information processed by this invention may vary depending on the specific product / service scenario and should be determined based on the specific scenario in which the user uses the product / service. This may involve the user's account information, device information, driving information, vehicle information, or other related information. This invention will treat the user's personal information and its processing with the utmost diligence.

[0066] This invention places great importance on the security of users' personal information and has adopted reasonable and feasible security protection measures that meet industry standards to protect users' information and prevent unauthorized access, disclosure, use, modification, damage or loss of personal information.

[0067] See appendix Figure 1 , Figure 1 This is a schematic flowchart illustrating the main steps of a vehicle highway toll evasion credit scoring method based on a large model according to an embodiment of the present invention. Figure 1 As shown, the vehicle highway toll evasion credit scoring method based on a large model in this embodiment of the invention mainly includes the following steps S101 to S103.

[0068] Step S101: Obtain historical high-speed driving data for at least one target vehicle.

[0069] In this embodiment, the target vehicle refers to a motor vehicle traveling on a highway, which may include cars, buses, and other motor vehicles. This method is primarily applied to scenarios involving toll evasion prediction and credit scoring of target vehicles at highway toll booths.

[0070] In one implementation, historical high-speed driving data may refer to real-time operating parameters generated by the vehicle while it is traveling at high speed, mainly including vehicle data such as speed, acceleration, and braking system status.

[0071] Historical highway driving data can include multiple feature variables, which are input into a pre-trained highway toll evasion probability prediction model to determine the probability of toll evasion for the target vehicle. These feature variables can include vehicle license plate identifier, highway passage sign, number of axles, vehicle type, vehicle trajectory, vehicle speed, highway communication costs, and other data variables.

[0072] In one implementation, historical highway driving data of multiple target vehicles can be obtained through the Electronic Toll Collection (ETC) system at the toll gate of the target highway. Alternatively, historical highway driving data can be obtained based on information such as the license plate number and vehicle type of the target vehicle by the local highway administration or traffic management department.

[0073] Step S102: Determine the probability of toll evasion for each target vehicle based on its historical highway driving data and the pre-trained highway toll evasion probability prediction model. In this embodiment, the high-speed toll evasion probability prediction model can be a large model trained based on an ensemble algorithm. An ensemble algorithm is a machine learning method that improves the overall model performance by combining the prediction results of multiple weak learners. A large model refers to a deep learning model with a large number of parameters (usually in the billions or more), trained on massive amounts of data, possessing strong generalization and multi-task processing capabilities.

[0074] In this embodiment, the probability of highway toll evasion refers to the possibility that a vehicle may evade its toll payment obligation by illegal means while traveling on a highway.

[0075] In one implementation, step S102 may specifically include step S1021:

[0076] Step S1021: For each target vehicle, input the corresponding historical highway driving data into the highway toll evasion probability prediction model, and use the output of the highway toll evasion probability prediction model as the predicted highway toll evasion probability for the target vehicle.

[0077] In one implementation, see Appendix Figure 2 , attached Figure 2 This is a schematic flowchart illustrating the main steps of training a high-speed toll evasion probability prediction model according to one embodiment of the present invention. Figure 2 As shown, before step S101, a high-speed toll evasion probability prediction model can be trained through the following steps, specifically steps S201 to S203:

[0078] Step S201: Obtain the training dataset.

[0079] Step S202: Construct the initial integration model.

[0080] Step S203: Train a high-speed toll evasion probability prediction model based on the training dataset and the initial ensemble model.

[0081] In this embodiment, the training dataset is a dataset of high-speed driving data containing multiple high-speed vehicle samples labeled with either toll evasion or non-toll evasion tags, and the initial ensemble model is an ensemble model composed of multiple learners.

[0082] In one implementation, the training dataset in step S201 may include a training set, a test set, and a validation set. The training set can be used to train an initial ensemble model, the validation set can be used to determine the model structure of the initial ensemble model and adjust its parameters, and the test set can be used to evaluate the performance of the final high-speed toll evasion probability prediction model.

[0083] In one implementation, the initial ensemble model in step S202 can be constructed using a stacking ensemble learning algorithm. This involves training multiple base learners based on different algorithms, then training a meta-learner, and finally fusing the outputs of the base learners through the meta-learner to improve the overall generalization performance of the model.

[0084] In one embodiment, step S201 may further include steps S2011 and S2012:

[0085] Step S2011: Based on oversampling technology, obtain high-speed driving data of multiple high-speed vehicles labeled with toll evasion tags or not toll evasion tags to obtain the original training dataset.

[0086] Step S2012: Process the high-speed driving data to obtain the final training dataset.

[0087] In this embodiment, high-speed driving data may include multiple feature variables.

[0088] In this embodiment, the oversampling technique in step S2011 refers to a data sampling technique that balances the data distribution by increasing the number of minority class samples, mainly used to solve the class imbalance problem. When the actual number of toll-evading vehicles is much smaller than the number of non-toll-evading vehicles, the proportion of minority class samples can be increased by replicating minority class samples or generating synthetic samples, reducing the model's excessive reliance on majority class samples. For example, oversampling techniques can include Synthetic Minority Over-sampling Technique (SMOTE).

[0089] In this embodiment, the data processing in step S2012 may include processing missing and outlier values ​​of feature variables in the high-speed driving data and normalizing the high-speed driving data based on the deviation standardization method.

[0090] Min-Max standardization is a data standardization method that linearly transforms the original data to the [0, 1] interval, and can be used to normalize the collected high-speed driving data.

[0091] In one implementation, processing the missing and outlier values ​​of feature variables in the high-speed driving data may further include steps S20121 to S20123:

[0092] Step S20121: Delete the feature variable when the number of missing values ​​of the feature variable does not meet the preset range.

[0093] Step S20122: When the number of missing values ​​of the feature variables meets the preset range, a random forest classifier is used to determine the missing values ​​of the feature variables.

[0094] Step S20123: Based on the univariate outlier detection method, obtain and correct the outliers of the feature variables.

[0095] In this embodiment, the missing values ​​in step S20121 refer to the missing values ​​of the feature variables in the high-speed driving data. When the number of missing values ​​is within a preset range, a random forest classifier can be used to determine the missing values ​​of the feature variables. Specifically, the random forest classifier can use the prediction results of multiple decision trees to jointly determine the final missing values ​​of the feature variables.

[0096] In this embodiment, the univariate outlier detection method in step S20123 can be a statistical method for determining outliers based on a single variable, mainly used to identify observations that significantly deviate from the overall level of the data.

[0097] In one implementation, a univariate outlier detection method can be used to plot box plots of the feature variables to identify outliers of all feature variables, and to determine correction values ​​for the outliers based on the plotted box plots.

[0098] In one embodiment, step S203 may further include:

[0099] Step S2031: Based on the gradient boosting tree algorithm, determine the importance value of each feature variable in the training dataset.

[0100] Step S2032: Group the training dataset according to each feature variable, and determine the information value of each feature variable based on the data grouping results.

[0101] Step S2033: Based on the information value and importance value, determine the target feature variable from all feature variables in the training dataset.

[0102] Step S2034: Based on the target feature variables and the training dataset, train the initial ensemble model to obtain the high-speed toll evasion probability prediction model.

[0103] In this embodiment, the gradient boosting algorithm in step S2031 is an algorithm that constructs a strong learner by iteratively training multiple weak learners (such as decision trees).

[0104] In this embodiment, the importance value in step S2031 is an indicator that can be used to measure the importance of a feature variable. For example, the importance value can be the Gini coefficient. During the decision tree generation process, when each feature variable is used to segment the data, the decrease in the Gini coefficient reflects the degree to which that feature variable improves the model's classification or regression performance. The smaller the Gini coefficient, the higher the importance of the feature variable.

[0105] In this implementation, Information Value (IV) is an indicator that can be used to measure the predictive power of a feature variable. In some possible implementations, the feature variable can be binned, the Weight of Evidence (WOE) of each bin can be calculated, and the IV value can be obtained by weighted summation. The higher the IV value, the stronger the predictive power of the feature variable on the probability of vehicle toll evasion.

[0106] In one embodiment, step S2032 may further include steps S20321 to S20324:

[0107] Step S20321: Obtain the total number of highway vehicle samples corresponding to the toll evasion label and the total number of highway vehicle samples corresponding to the non-toll evasion label in the training dataset.

[0108] Step S20322: Based on the values ​​of the feature variables in the training dataset, group the training dataset into multiple data groups.

[0109] Step S20323: Obtain the number of highway vehicle samples corresponding to the toll evasion label and the number of highway vehicle samples corresponding to the non-toll evasion label in each data group, and determine the information value of the feature variable in each data group according to the following formula (1):

[0110]

[0111] Among them, IV i Bad represents the information value of a feature variable in the i-th data set. i Let Good be the number of highway vehicle samples corresponding to the toll evasion label in the i-th data group. i Let Bad be the number of highway vehicle samples corresponding to the "no toll evasion" label in the i-th data group. T To determine the total number of highway vehicle samples corresponding to the toll evasion labels in the training dataset, Good T This represents the total number of high-speed vehicle sample data corresponding to the "no toll evasion" label in the training dataset.

[0112] Step S20324: Determine the information value of the feature variables based on the sum of their information values ​​in each data set.

[0113] In this embodiment, the number of data groups in step S20322 can be determined based on the characteristics or values ​​of the feature variables.

[0114] As an example, when the feature variable is vehicle type, the number of data groups can be determined based on the value of the vehicle type in the target vehicle. For example, if the vehicle type includes trucks, cars and buses, the number of data groups can be 3.

[0115] In one implementation, step S20324 can determine the information value of the feature variable according to the following formula (2):

[0116]

[0117] Where IV represents the information value of the feature variable. i Let be the information value of the i-th data group. Let n be the number of data groups.

[0118] In one implementation, step S2033 may further include steps S20331 and S20332:

[0119] Step S20331: Perform data processing on the information value and importance values, and determine the sum of the information value and importance values ​​corresponding to each feature variable.

[0120] Step S20332: The feature variable whose sum of information value and importance value is greater than a preset threshold is determined as the target feature variable.

[0121] In this embodiment, step S20331 may specifically include: normalizing the information value and importance values, then adding the two together to obtain a comprehensive score, and when the comprehensive score of the feature variable is greater than a preset threshold, determining the corresponding feature variable as the target feature variable.

[0122] For example, the preset threshold can be specifically 0.8.

[0123] In one implementation, the learners in the initial ensemble model include a meta-learner and at least two base learners.

[0124] In this implementation, a base learner can be a single weak learner that constitutes the initial ensemble model, typically a simple model such as a decision tree or linear regression. A meta-learner can be a component used to integrate the predictions of multiple base learners, typically a logistic regression model.

[0125] In this embodiment, step S2034 may further include steps S20341 to S20343:

[0126] Step S20341: Based on the K-fold cross-validation algorithm, generate a weighted training dataset from the training dataset and the base learner.

[0127] Step S20342: Train the meta-learner based on the weighted training dataset to obtain the trained meta-learner.

[0128] Step S20343: Based on the trained meta-learner, obtain the trained high-speed toll evasion probability prediction model.

[0129] In this embodiment, the weighted training dataset can be a training dataset that is weighted according to the output of the base learner.

[0130] In this embodiment, K-fold cross-validation is a statistical method used to evaluate the generalization ability of machine learning models on finite datasets. Specifically, K-fold cross-validation involves dividing the training dataset into k subsets, using one subset as the validation set in turn, and using the rest as the training set to train the base learner.

[0131] Step S103: Determine the credit score for each target vehicle based on the probability of highway toll evasion.

[0132] In this embodiment, different levels of credit scores can be determined based on the probability of highway toll evasion, so that the credit scores can be used to determine the toll evasion prediction results of the target vehicle and the highway toll collection service.

[0133] In one embodiment, step S103 may further include steps S1031 to S1032:

[0134] Step S1031: Determine the probability of the target vehicle not evading tolls on the highway based on the probability of the target vehicle evading tolls on the highway, and determine the ratio between the probability of evading tolls on the highway and the probability of not evading tolls on the highway.

[0135] Step S1032: Determine the credit score of the target vehicle based on the probability of highway toll evasion and the preset credit scoring function of the following formula (3):

[0136] Score = A - Bln(Odds) (3)

[0137] Where Score is the credit score of the target vehicle, A and B are constants, and Odds is the ratio between the probability of evading highway tolls and the probability of not evading highway tolls.

[0138] In this embodiment, constants A and B are proportional constants used to measure the strength of the correlation between the probability of highway toll evasion and the vehicle credit score. Constants A and B can be adjusted by user-defined values ​​such as preset vehicle scores, preset probability ratios, and preset vehicle change values.

[0139] In one embodiment, this embodiment may further include determining constants A and B in the credit scoring function according to the following steps, including step S10321:

[0140] Step S10321: Obtain the preset vehicle score, preset probability ratio, and preset vehicle variation value. Step S10322: Based on the preset vehicle score, preset probability ratio, preset vehicle variation value, and the following formulas (4) and (5), determine constant A and constant B:

[0141] A = p0 + Bln(Odds0) (4)

[0142]

[0143] Where p0 is the preset vehicle rating, Odds0 is the preset probability ratio, PDO is the preset vehicle variation value, and A and B are constants.

[0144] In this embodiment, the preset vehicle change value is the change value of the preset vehicle score corresponding to twice the preset probability ratio.

[0145] In one application scenario according to an embodiment of the present invention, please refer to the appendix. Figure 3 , attached Figure 3 This is a schematic diagram illustrating the main steps of establishing a credit scoring function according to one embodiment of the present invention. Figure 3 As shown, this method may further include determining a credit scoring function according to the following steps, including steps S301 to S304:

[0146] Step S301: Obtain the preset credit scoring function.

[0147] In this embodiment, the credit scoring function is the same as formula (3) in the above embodiment.

[0148] Step S302: Determine the ratio of non-toll evasion samples to toll evasion samples.

[0149] In this embodiment, we can assume that the probability of a vehicle evading highway tolls is p, then the probability of a vehicle not evading highway tolls is 1-p. The ratio of non-toll-evading vehicle samples to toll-evading vehicle samples can then be set according to the following formula (6):

[0150]

[0151] Where p is the probability of a vehicle evading highway tolls, and Odds is the ratio of non-toll-evading vehicle samples to toll-evading vehicle samples.

[0152] Step S303: Determine constants A and B.

[0153] In this embodiment, step S302 may specifically include:

[0154] Step S3031: Obtain the given preset score p0 and the corresponding preset ratio Odd s0 And the score PDO after doubling the preset ratio, the obtained value is substituted into formula (3) to obtain the following formulas (7) and (8):

[0155] p0 = A - Bln(Odd) s0 (7)

[0156] p0-PDO=A-Bln(2*Odd s0 (8)

[0157] Where p0 is the given preset score, Odds0 is the preset ratio corresponding to the preset score, and PDO is the score corresponding to the preset ratio after doubling.

[0158] Step S3032: Based on formulas (7) and (8), obtain formulas (4) and (5) in the above embodiments to determine the final constants A and B.

[0159] Step S304: Determine the final credit scoring function based on constants A and B and the ratio of non-toll evasion samples to toll evasion samples.

[0160] The probability of a vehicle evading highway tolls can be directly substituted into the credit scoring function to obtain the vehicle's credit score. A lower probability of highway toll evasion results in a higher credit score, corresponding to a higher credit rating and lower toll evasion risk.

[0161] In one application scenario according to an embodiment of the present invention, please refer to the appendix. Figure 4 , attached Figure 4 This is a schematic flowchart illustrating the main steps of training a high-speed toll evasion probability prediction model according to another embodiment of the present invention. Figure 4 As shown, this method trains a high-speed toll evasion probability prediction model according to the following steps, including steps S401 to S404:

[0162] Step S401: Collect high-speed vehicle data and perform data preprocessing on the high-speed vehicle data to obtain a training dataset.

[0163] In this embodiment, the training dataset may include toll evasion samples labeled with toll evasion tags and non-toll evasion samples labeled with non-toll evasion tags.

[0164] In this embodiment, the collected highway vehicle data is typically of low quality, containing missing and outlier values. Furthermore, since the number of toll-evading vehicles in real life is far less than the number of non-toll-evading vehicles, the collected data on toll-evading and non-toll-evading vehicles exhibits an imbalanced distribution. Therefore, data preprocessing is necessary for the collected highway vehicle data.

[0165] Specifically, data preprocessing can include: directly deleting the corresponding feature variables when the proportion of missing values ​​is very high or very low (i.e., when the preset quantity range is not met); using a random forest model to predict the values ​​of feature variables when the proportion of missing values ​​is moderate (i.e., when the preset quantity range is met) in order to fill in the missing values; and using a univariate outlier detection method to draw box plots of feature variables to identify and correct outliers of feature variables.

[0166] In other implementations, the Min-Max method can be used to normalize the original high-speed vehicle data in order to accelerate model convergence during model training. When there is an uneven distribution of label categories in the training dataset, the SMOTE sampling method can be used to oversample the high-speed vehicle toll evasion samples to increase the number of minority class samples, thereby improving the model's ability to identify minority class data during model training.

[0167] Step S402: From all feature variables in the training dataset, select feature variables that meet the preset conditions for correlation with the probability of highway toll evasion and determine them as target feature variables.

[0168] In this embodiment, each feature variable can be evaluated from both model and statistical dimensions by calculating its importance (i.e., the importance value in the above embodiments) and information value (i.e., the IV value in the above embodiments). This avoids the influence of data distribution on the feature variable selection process and identifies feature variables that are highly correlated with the probability of toll evasion by high-speed vehicles (i.e., the prediction result of the high-speed vehicle toll evasion probability prediction model).

[0169] In one embodiment, step S402 may specifically include steps S4021 to S4023:

[0170] Step S4021: Determine the importance of the feature variables.

[0171] In this implementation, the importance of feature variables can be obtained using the gradient boosting tree algorithm. The gradient boosting decision tree (GBDT) can be composed of multiple classification and regression trees (CART), with each subsequent tree fitting the residuals of the previous tree to obtain the final model.

[0172] In one implementation, step S4021 may specifically include: summing up the gains (i.e. minimizing the mean squared error (MSE)) of the feature variables after splitting at the nodes of the CART regression tree, comprehensively evaluating all CART regression trees, selecting the feature variables with the highest correlation and lowest redundancy without changing the feature variables, constructing a feature subset, inputting the preprocessed training dataset into the gradient boosting tree model for training, and obtaining the importance score of each feature variable.

[0173] The importance score corresponding to the j-th feature variable is C. j, j is greater than 1 and less than or equal to n, where n is the total number of feature variables. The importance score can be an evaluation of the feature variables at the model level.

[0174] In other implementations, a CART classification tree can be used to obtain the Gini coefficient of each feature variable as a score indicating the importance of that feature variable.

[0175] Step S4022: Determine the information value of the feature variables.

[0176] In this implementation, the information value of a feature variable can be determined based on its weight of evidence in the training dataset. The weight of evidence (WOE) can be an indicator used in credit scoring models to quantify the ability of a categorical variable to distinguish a target variable. Its risk prediction capability can be measured by calculating the difference in the proportion of evaded payment samples to non-evaded payment samples in each bin.

[0177] In this embodiment, firstly, a binning strategy can be used to group the data in the training dataset according to the values ​​of the feature variables to determine the WOE value of the feature variables. Since the feature variables can include continuous and discrete variables, continuous variables can be binned at equal frequencies, and discrete variables can be binned by setting thresholds or intervals to obtain multiple data groups. Then, based on the binning results and the following formula (9), the WOE value of the feature variables in each data group is determined:

[0178]

[0179] Among them, WOE i Let WOE value be the feature variable in the i-th group, Bad i Let Good be the number of highway vehicle samples corresponding to the toll evasion label in the i-th data group. i Let Bad be the number of highway vehicle samples corresponding to the "no toll evasion" label in the i-th data group. T To determine the total number of highway vehicle samples corresponding to the toll evasion labels in the training dataset, Good T This represents the total number of high-speed vehicle sample data corresponding to the "no toll evasion" label in the training dataset.

[0180] Then, after determining the WOE value of the feature variable in each data group, the IV value of the feature variable in each data group can be determined based on the WOE value and the following formula (10):

[0181]

[0182] Among them, IV i WOE represents the information value of the feature variable in the i-th group. i Let WOE value be the feature variable in the i-th group, Badi Let Good be the number of highway vehicle samples corresponding to the toll evasion label in the i-th data group. i Let Bad be the number of highway vehicle samples corresponding to the "no toll evasion" label in the i-th data group. T To determine the total number of highway vehicle samples corresponding to the toll evasion labels in the training dataset, Good T This represents the total number of high-speed vehicle sample data corresponding to the "no toll evasion" label in the training dataset.

[0183] Finally, the information value of the feature variable is determined based on its IV value in each data group.

[0184] In this embodiment, the IV value corresponding to the j-th feature variable is D. j , j is greater than 1 and less than or equal to n. Here, the IV value can be an evaluation of the feature variable in a statistical dimension.

[0185] Step S4023: Select target feature variables based on their importance and information value.

[0186] In this embodiment, the feature importance C can be... j and eigenvalues ​​D j After normalization, the two scores are added together to obtain the overall score F. j Among them, F j This is used to represent the comprehensive score corresponding to the j-th feature variable, where j is greater than 1 and less than or equal to n.

[0187] In one implementation, a comprehensive score greater than or equal to 80% can be used as a screening criterion for feature variables, that is, feature variables with a comprehensive score greater than or equal to 0.8 are determined as target feature variables.

[0188] Step S403: Train the model based on the training dataset and the target feature variables.

[0189] In this embodiment, an initial ensemble model can be constructed and trained based on the optimized Stacking algorithm. The initial ensemble model may include a first-stage learner and a second-stage learner. The first-stage learner may include N base learners, and the second-stage learner may include a meta-learner, where N is a constant.

[0190] In one implementation, the Stacking algorithm can be optimized using the K-fold cross-validation algorithm.

[0191] In this embodiment, the training process of the first-stage learner can specifically include: First, performing K-fold cross-validation on N base learners. For example, the training dataset is divided into K equal parts. Each fold cross-validation uses K-1 parts of the data as the training set, and the remaining part as the validation set. The model is trained using the K-1 training sets. Then, after each training set is used to complete the training of the first-stage learner, the trained first-stage learner is used to predict the validation set and a preset test set.

[0192] Wherein, the prediction result of the validation set after the i-th training set is trained can be P. i The maximum difference between positive and negative samples (Kolmogorov-Smirnov, KS) in the prediction results of the validation set after training the i-th training set is W. i The prediction result of the test set after the i-th training set is trained can be R. i Where i is greater than 1 and less than or equal to K.

[0193] Finally, the validation sets are weighted and integrated into a new training dataset based on the prediction results of each validation set, i.e., P = {P1*W1, P2*W2, ..., P...} K *W K The prediction results of the K test sets are then weighted and averaged to obtain a new test set sample.

[0194] In this embodiment, the training process of the second-stage learner may specifically include: inputting the new training dataset P obtained from the training process of the first-stage learner into the meta-learner of the second-stage learner for training, and obtaining the final high-speed toll evasion probability prediction model.

[0195] Step S404: Based on the training results, perform performance testing on the obtained high-speed toll evasion probability prediction model to obtain the final high-speed toll evasion probability prediction model.

[0196] In this embodiment, the test set R obtained from the training process of the first-stage learner can be input into the second-stage learner for prediction to obtain the prediction result M(R). The performance of the high-speed toll evasion probability prediction model is analyzed based on the prediction result M(R). If the performance meets the preset conditions, the final high-speed toll evasion probability prediction model is obtained. If the performance does not meet the preset conditions, the process can return to step S403 and retrain the model.

[0197] Based on the methods described in steps S101 to S103 above, this invention can predict the probability of toll evasion and assign credit scores to multiple target vehicles, which is beneficial for establishing a sound vehicle credit system. This improves the accuracy of predicting whether a vehicle will evade tolls, effectively regulating toll evasion behavior and purifying the toll road network environment. Simultaneously, the credit scores of target vehicles can provide highway workers with more accurate information for toll evasion verification, improving their work efficiency.

[0198] Furthermore, by calculating the importance and information value of each feature variable in the training dataset, the influence of data distribution on feature variables can be reduced, and the target feature variable with the greatest correlation to the prediction result can be selected. This can improve the training effect and accuracy of the initial ensemble model, while also avoiding overfitting of the initial ensemble model and improving the model's adaptability to new data.

[0199] Furthermore, the base learners in the initial ensemble learner can be used to weight the training dataset through K-fold cross-validation to obtain a weighted training dataset. This weighted training dataset can then be used as input data for the meta-learner to further optimize the model training process of the meta-learner, thereby improving the performance of the final high-speed toll evasion probability prediction model.

[0200] It should be noted that although the steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effects of the present invention, different steps do not necessarily have to be executed in such an order. They can be executed simultaneously (in parallel) or in other orders. These adjusted solutions are equivalent to the technical solutions described in the present invention and therefore will also fall within the protection scope of the present invention.

[0201] Those skilled in the art will understand that all or part of the processes in the method of the above embodiment of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium can include any entity or device capable of carrying the computer program code, a medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0202] Another aspect of the present invention provides a computer-readable storage medium.

[0203] In one embodiment of the computer-readable storage medium according to the present invention, the computer-readable storage medium may be configured to store a program that executes the large-model-based vehicle toll evasion credit scoring method of the above-described method embodiments. This program may be loaded and run by a processor to implement the above-described large-model-based vehicle toll evasion credit scoring method. For ease of explanation, only the parts related to the embodiments of the present invention are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of the present invention. The computer-readable storage medium may be a storage device comprising various electronic devices. Optionally, in the embodiments of the present invention, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0204] Another aspect of the present invention provides a controller.

[0205] In one embodiment of a controller according to the present invention, the controller may include at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program, which, when executed by the at least one processor, implements the method described in any of the above embodiments. The controller of the present invention may include devices such as driving equipment, intelligent vehicles, and robots.

[0206] In some embodiments of the invention, the controller may further include at least one sensor for sensing information. The sensor is communicatively connected to any type of processor mentioned in the invention.

[0207] The technical solution of the present invention has been described above with reference to one embodiment shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions resulting from such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A credit scoring method for vehicle highway toll evasion based on a large model, characterized in that, The method includes: Acquire historical high-speed driving data for at least one target vehicle; Based on the historical high-speed driving data of each target vehicle and the pre-trained high-speed toll evasion probability prediction model, the high-speed toll evasion probability of each target vehicle is determined. The high-speed toll evasion probability prediction model is a large model trained based on an ensemble algorithm. Based on the probability of highway toll evasion, a credit score is determined for each target vehicle.

2. The vehicle highway toll evasion credit scoring method based on a large model according to claim 1, characterized in that, The method also includes obtaining the highway toll evasion probability prediction model according to the following method: Obtain a training dataset, which is a dataset of high-speed driving data containing multiple samples of high-speed vehicles labeled with either toll evasion or non-toll evasion tags. Construct an initial ensemble model, which is an ensemble model composed of multiple learners; The high-speed toll evasion probability prediction model is trained based on the training dataset and the initial ensemble model.

3. The vehicle highway toll evasion credit scoring method based on a large model according to claim 2, characterized in that, The step of training the high-speed toll evasion probability prediction model based on the training dataset and the initial ensemble model includes: Based on the gradient boosting tree algorithm, the importance value of each feature variable in the training dataset is determined; The training dataset is grouped according to each feature variable, and the information value of each feature variable is determined based on the data grouping results. Based on the information value and the importance value, the target feature variable is determined from all feature variables in the training dataset; Based on the target feature variables and the training dataset, the initial ensemble model is trained to obtain a high-speed toll evasion probability prediction model.

4. The vehicle highway toll evasion credit scoring method based on a large model according to claim 3, characterized in that, The step of grouping the training dataset according to each feature variable and determining the information value of each feature variable based on the data grouping results includes: Obtain the total number of highway vehicle samples corresponding to the toll evasion label and the total number of highway vehicle samples corresponding to the non-toll evasion label in the training dataset; Based on the values ​​of the feature variables in the training dataset, the training dataset is grouped to obtain multiple data groups; Obtain the number of highway vehicle samples corresponding to the toll evasion label and the number of highway vehicle samples corresponding to the non-toll evasion label in each data group, and determine the information value of the feature variable in each data group according to the following formula: Among them, IV i The information value of the feature variable in the i-th data group, Bad i Let Good be the number of highway vehicle samples corresponding to the toll evasion label in the i-th data group. i Let Bad be the number of highway vehicle samples corresponding to the "no toll evasion" label in the i-th data group. T Good represents the total number of highway vehicle samples corresponding to the toll evasion label in the training dataset. T The total number of high-speed vehicle sample data corresponding to the "no toll evasion" label in the training dataset; The information value of the feature variable is determined by summing the information values ​​of the feature variable in each data set.

5. The vehicle highway toll evasion credit scoring method based on a large model according to claim 3, characterized in that, The step of determining the target feature variable from all feature variables in the training dataset based on the information value and the importance value includes: The information value and importance value are processed, and the sum of the information value and importance value corresponding to each feature variable is determined. The feature variable whose sum of information value and importance value is greater than a preset threshold is determined as the target feature variable.

6. The vehicle highway toll evasion credit scoring method based on a large model according to claim 3, characterized in that, The learners in the initial ensemble model include a meta-learner and at least two base learners; The step of training the initial ensemble model based on the target feature variables and the training dataset to obtain a high-speed toll evasion probability prediction model includes: Based on the K-fold cross-validation algorithm, the training dataset and the base learner are used to generate a weighted training dataset, which is a training dataset after weighting the training data in the training dataset according to the output of the base learner. The meta-learner is trained based on the weighted training dataset to obtain the trained meta-learner; Based on the trained meta-learner, a high-speed toll evasion probability prediction model is obtained.

7. The vehicle highway toll evasion credit scoring method based on a large model according to claim 2, characterized in that, The acquisition of the training dataset includes: Based on oversampling technology, high-speed driving data of multiple high-speed vehicles labeled with toll evasion tags or not toll evasion tags are obtained to obtain the original training dataset, which includes multiple feature variables. The high-speed driving data is processed to obtain the final training dataset; the data processing includes handling missing and outlier values ​​of feature variables in the high-speed driving data and / or normalizing the high-speed driving data based on the deviation standardization method; The process of handling missing and outlier values ​​for feature variables in the high-speed driving data includes: If the number of missing values ​​of the feature variable does not meet a preset range, the feature variable is deleted; if the number of missing values ​​of the feature variable meets the preset range, a random forest classifier is used to determine the missing values ​​of the feature variable; and / or, Based on the univariate outlier detection method, the outliers of the feature variables are obtained and corrected.

8. The vehicle highway toll evasion credit scoring method based on a large model according to claim 1, characterized in that, The step of determining the credit score of each target vehicle based on the highway toll evasion probability includes: The probability of the target vehicle not evading highway tolls is determined based on the probability of the target vehicle evading highway tolls, and the ratio between the probability of evading highway tolls and the probability of not evading highway tolls is determined. Based on the probability of highway toll evasion and the following preset credit scoring function, the credit score of the target vehicle is determined: Score = A - Bln(Odds) Wherein, Score is the credit score of the target vehicle, A and B are constants, and Odds is the ratio between the probability of highway toll evasion and the probability of highway toll non-evasion.

9. The vehicle highway toll evasion credit scoring method based on a large model according to claim 8, characterized in that, The method further includes determining constants A and B in the credit scoring function according to the following steps: Obtain a preset vehicle score, a preset probability ratio, and a preset vehicle change value; the preset vehicle change value is the change value of the preset vehicle score corresponding to twice the preset probability ratio. Based on the preset vehicle score, preset probability ratio, preset vehicle variation value, and the following formula, constants A and B are determined: A = p0 + Bln(Odds0) Where p0 is the preset vehicle rating, Odds0 is the preset probability ratio, PDO is the preset vehicle variation value, and A and B are constants.

10. A controller, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores a computer program, which, when executed by the at least one processor, implements the vehicle highway toll evasion credit scoring method based on a large model as described in any one of claims 1 to 9.

Images