Driving risk level classification model construction method, system and device and storage medium

Abstract

A driving risk level classification model construction method, system and device and a storage medium relate to the technical field of automobile driving active safety, and comprise the steps of training a historical classification model by using a historical data set to obtain a target classification model, the historical data set comprising historical driving data with risk level classification labels; screening out a high-value data set from a target data set based on an active learning method, wherein the target data set comprises target data corresponding to the newly added driving scene; adding the high-value data set marked with the risk level classification label into a historical data set to form a new data set; and training the target classification model based on the new data set to generate a target driving risk level classification model. Through the method and the device, the technical problem of low risk prediction accuracy of newly increased driving data in related technologies can be solved.

Description

Technical Field

[0001] The present application relates to the field of active safety technology for automobile driving, and in particular to a method, system, device and storage medium for constructing a driving risk level classification model. Background Art

[0002] With the development of science and technology, intelligent connected cars have become an important part of people's daily travel. However, the accompanying driving behavior risk problem has become increasingly prominent and has become a key factor restricting the healthy and sustainable development of the automotive industry. Therefore, in-depth analysis and research on driving behavior risks and the proposal of an effective and accurate driving behavior risk prediction model are of great significance for preventing driving behavior risks, improving road traffic safety, and protecting people's lives and property.

[0003] In the related technology, most of the existing driving behavior risk analysis methods are based on existing data for model training, and the obtained driving behavior risk model is deployed as the final model to perform risk level classification warnings.

[0004] However, in real life, there are often many new driving situations and scenarios. Existing technical models classify and predict the risk levels of existing historical data. This can achieve high accuracy for existing historical data, but it often has the problem of low accuracy for newly added driving scenario data. Summary of the invention

[0005] The present application provides a method, system, device and storage medium for constructing a driving risk level classification model, which can solve the technical problem in the prior art of low accuracy in risk prediction of newly added driving data.

[0006] In a first aspect, an embodiment of the present application provides a method for constructing a driving risk level classification model, the method for constructing a driving risk level classification model comprising:

[0007] Using a historical data set to train a historical classification model to obtain a target classification model, the historical data set comprising historical driving data with risk level classification labels;

[0008] Screening out a high-value data set from a target data set based on an active learning method, wherein the target data set includes target data corresponding to a newly added driving scenario;

[0009] Add high-value datasets annotated with risk level classification labels to historical datasets to form new datasets;

[0010] The target classification model is trained based on the new data set to generate a target driving risk level classification model.

[0011] In combination with the first aspect, in one implementation, before the step of using the historical data set to train the historical classification model, the method further includes:

[0012] Factor analysis is performed on historical driving data to obtain target factors, which include driver behavior factors, vehicle operating conditions factors, driver status factors, vehicle traffic conditions factors and vehicle environmental conditions factors. The driving data corresponding to the target factors constitute a historical data set.

[0013] In combination with the first aspect, in one implementation, the active learning method is an entropy query strategy.

[0014] In combination with the first aspect, in one implementation, the entropy value calculation formula of the entropy value query strategy is:

[0015]

[0016] Among them, Entropy_value is the entropy value of the new data in the new data set, x is the target data, and i is the risk level category. is the probability that the target data x is classified as the i-th risk, and θ is the parameter space.

[0017] In a second aspect, an embodiment of the present application provides a driving risk level classification model construction system, the driving risk level classification model construction system comprising:

[0018] A first processing module, which is used to train a historical classification model using a historical data set to obtain a target classification model, wherein the historical data set includes historical driving data with risk level classification labels;

[0019] A second processing module, which is used to screen out a high-value data set from the target data set based on an active learning method, wherein the target data set includes target data corresponding to the newly added driving scenario;

[0020] A third processing module is used to add the high-value data set marked with risk level classification labels to the historical data set to form a new data set;

[0021] A fourth processing module is used to train the target classification model based on the new data set to generate a target driving risk level classification model.

[0022] In conjunction with the second aspect, in one implementation, the first processing module is specifically configured to:

[0023] Factor analysis is performed on historical driving data to obtain target factors, which include driver behavior factors, vehicle operating conditions factors, driver status factors, vehicle traffic conditions factors and vehicle environmental conditions factors. The driving data corresponding to the target factors constitute a historical data set.

[0024] In conjunction with the second aspect, in one implementation, the active learning method is an entropy query strategy.

[0025] In conjunction with the second aspect, in one implementation, the entropy value calculation formula of the entropy value query strategy is:

[0026]

[0027] Among them, Entropy_value is the entropy value of the new data in the new data set, x is the target data, and i is the risk level category. is the probability that the target data x is classified as the i-th risk, and θ is the parameter space.

[0028] In a third aspect, an embodiment of the present application provides a driving risk level classification model construction device, wherein the driving risk level classification model construction device includes a processor, a memory, and a driving risk level classification model construction program stored in the memory and executable by the processor, wherein when the driving risk level classification model construction program is executed by the processor, the steps of the aforementioned driving risk level classification model construction method are implemented.

[0029] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a driving risk level classification model construction program is stored, wherein when the driving risk level classification model construction program is executed by a processor, the steps of the aforementioned driving risk level classification model construction method are implemented.

[0030] The beneficial effects brought by the technical solution provided in the embodiments of the present application include:

[0031] First, the historical classification model is trained through the historical data set to obtain the target classification model; then, the high-value data is screened in the target data set through the active learning method, and the labeled high-value data is put into the historical data set to form a new data set; finally, the target classification model is trained with the new data set to generate a target driving risk level classification model that can accurately predict the driving risk level of the newly added driving scenario data, which solves the technical problem of low accuracy in risk prediction of newly added driving data in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of an embodiment of a method for constructing a driving risk level classification model of the present application;

[0033] Figure 2 This is a schematic diagram of the label system involved in the embodiment of the present application;

[0034] Figure 3This is a schematic diagram of the driving behavior ID data involved in the embodiment of the present application;

[0035] Figure 4 A schematic diagram of the hardware structure of a device for building a driving risk level classification model involved in an embodiment of the present application. DETAILED DESCRIPTION

[0036] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0037] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0038] In a first aspect, an embodiment of the present application provides a method for constructing a driving risk level classification model.

[0039] In the embodiment, refer to Figure 1 , Figure 1 This is a flow chart of an embodiment of the method for constructing a driving risk level classification model in this application. Figure 1 As shown, the method for constructing a driving risk level classification model includes:

[0040] Step S10: training a historical classification model using a historical data set to obtain a target classification model, wherein the historical data set includes historical driving data with risk level classification labels.

[0041] Before the step of using the historical data set to train the historical classification model, the method further includes:

[0042] Factor analysis is performed on historical driving data to obtain target factors, which include driver behavior factors, vehicle operating conditions factors, driver status factors, vehicle traffic conditions factors and vehicle environmental conditions factors. The driving data corresponding to the target factors constitute a historical data set.

[0043] For example, in this embodiment, historical driving data with risk level classification labels are obtained by performing data cleaning and factor analysis on the acquired labeled data. Specifically, Figure 2 and Figure 3 As shown in the figure, the data of the driver's driving behavior in a certain period of time is collected through the sensors on the vehicle, and the data is marked as a behavior id. The data corresponding to the id is the label sample data. The specific label classification of the behavior id is as follows Figure 2 As shown in the figure, the specific labels of the behavior id (i.e., sample features) are mainly divided into three sections: vehicle attributes, behavior attributes, and environmental attributes. Among them, the vehicle attribute section is mainly divided into driving status, driving mode, vehicle health, and vehicle control tags; the behavior attribute section is mainly divided into driver violations, turning and lane change behavior, following distance, driving style, attention, emotion, and driving time tags; the environmental attribute section is mainly divided into route tags, time period, weather, road conditions, lighting conditions, and traffic conditions tags. The behavior id data collected in this embodiment is shown in Figure 1. Figure 3 As shown, each ID data shows the specific driving situation corresponding to the driving behavior in a certain period of time. For example, the driving status of behavior ID data 1 is driving, the driving mode is ECO, the route label is regular, the violation is speeding, and the vehicle health is normal.

[0044] Perform preliminary data cleaning and processing on the behavior id data; specifically, first check the missing data in the acquired id data, and remove or fill in the data with too high a missing percentage in the sample features or behavior id data. In this embodiment, the median filling method can be preferably used to fill in the missing data, and then the filled data is label-encoded to finally obtain the cleaned historical driving data.

[0045] When factor analysis is performed on historical driving data, it is understood that there will be a certain amount of multicollinearity between various ID data, which will lead to large errors in model prediction when these ID data are used for model training, thereby reducing the accuracy and robustness of the training model. However, this embodiment obtains common factors through factor analysis to replace the original sample features with the common factors, thereby eliminating the impact of multicollinearity on model parameter estimation.

[0046] First, the sample characteristics of each data in the historical driving data are represented by mathematical symbols, namely: driving status (x1), driving mode (x2), vehicle health (x3), vehicle control (x4), violation behavior (x5), turning and lane change behavior (x6), following distance (x7), driving style (x8), attention (x9), emotion (x10), driving style (x11), driving style (x12), driving style (x13), driving style (x14), driving style (x15), driving style (x16), driving style (x17), driving style (x18), driving style (x19), driving style (x20), driving style (x21), driving style (x22), driving style (x23), driving style (x24), driving style (x25), driving style (x26), driving style (x27), driving style (x28), driving style (x29), driving style (x3 ... 10 ), driving time (x 11 ), route label (x 12 ), time period (x 13 )、Weather(x 14 ), road condition (x 15 ), lighting conditions (x 16 )、Traffic conditions(x 17 ).

[0047] Factor analysis is performed on the sample characteristics covered in the historical driving data to obtain the following five common factors f1, f2, f3, f4, and f5:

[0048]

[0049] In the formula, f1 represents the driver's behavior factor, which specifically reflects the driver's behavior status, violations, etc.; f2 represents the vehicle's operating condition factor, which specifically reflects the vehicle's health and vehicle control and other operating condition information; f3 represents the driver's state factor, which specifically reflects the driver's attention and emotions and other information; f4 represents the vehicle's traffic condition factor, which specifically reflects the vehicle's route and traffic conditions and other information; f5 represents the vehicle's environmental condition factor, which specifically reflects the vehicle's weather, lighting conditions, and road conditions and other information; Represents the weight coefficient of different sample characteristics in each factor.

[0050] Among them, the weight coefficient can be calculated by the following formula:

[0051]

[0052] In the formula, x represents the sample feature data, Var(x) represents the variance of the sample feature, and T represents the transpose of the matrix. As the intermediate quantity of this formula, in represents the singular values ​​of matrix A, which represents sample features x1,x2,x3…,…,x 17 The weight coefficient is

[0053] It is understandable that, for the historical data set formed after factor analysis, the historical data set can be divided into a training set and a test set in a ratio of 8:2, and the data of the training set is used to train the existing classification prediction model (i.e., the historical classification model), and the data of the test set is used to test the accuracy of the trained model, wherein the classification prediction model can preferably be a Logistic model; it should be noted that the accuracy of the model can be determined by the following loss function:

[0054]

[0055] In the formula, x is the driving behavior data of the classification model, y is the driving behavior risk level, m is the number of samples in the model, i is the risk level category, θ is the parameter space, and h is the driving behavior risk level. θ is the sigmoid function (i.e., activation function), J θ Represents the loss value of the loss function.

[0056] Step S20: Filter out a high-value data set from the target data set based on an active learning method, where the target data set includes target data corresponding to the newly added driving scenario.

[0057] Among them, the active learning method in this application adopts an entropy query strategy, and the entropy calculation formula of the entropy query strategy is:

[0058]

[0059] In the formula, Entropy_value is the entropy value of the new data in the new data set, x is the target data, i is the risk level category, is the probability that the target data x is classified as the i-th risk, and θ is the parameter space.

[0060] It is understandable that the entropy query strategy is a sample selection method based on uncertainty, which uses entropy to measure the uncertainty or information content of data. In this embodiment, the entropy query strategy can be preferably used as an active learning method to filter out high-value data sets from the target data corresponding to the newly added driving scene, wherein the larger the entropy value of the entropy query data, the greater the uncertainty of the data, the richer the information content, and the higher the value of the data.

[0061] Specifically, this embodiment performs data cleaning and factor analysis on the new data corresponding to the newly added driving scenarios, and standardizes the data so that all sample features are at the same level for easy comparison; for the standardized data, the proportion of each sample feature is calculated; the target data x, the risk level category i, and the probability that the target data x is classified as the i-th risk are calculated. (i.e., the weight of each sample feature, which represents the ratio of the value of the sample in the sample feature to the sum of all sample values ​​of the sample feature) and the parameter space θ are substituted into the entropy value calculation formula: Get the entropy value Entropy_value of the new data in the new data set; calculate the difference coefficient of the new data in the new data set, and use the difference coefficient to calculate the weight of each sample feature. The larger the weight, the higher the importance of the sample feature in the comprehensive evaluation. The higher the importance of the sample feature, the higher the value of the sample data.

[0062] Step S30: adding the high-value dataset annotated with the risk level classification label to the historical dataset to form a new dataset.

[0063] For example, high-value data can be manually labeled by experienced industry experts, and high-value data sets labeled with risk level classification labels can be added to historical data sets to form new data sets, which can enable the classification model to better identify new driving scenarios.

[0064] Step S40: training the target classification model based on the new data set to generate a target driving risk level classification model.

[0065] Exemplarily, the target classification model is repeatedly trained with a new data set until the accuracy of the target classification model meets a preset condition or all new user samples are labeled, thereby obtaining a target driving risk level classification model.

[0066] In this embodiment, by cleaning and factor analyzing the data, training the classification model with historical data, and then using active learning methods to screen high-value data from the new data, the labeled high-value data is placed in the historical data set and the historical classification model is trained, thereby solving the technical problem of low accuracy in risk prediction of new driving data in related technologies.

[0067] In a second aspect, an embodiment of the present application also provides a driving risk level classification model construction system.

[0068] In one embodiment, the driving risk level classification system includes:

[0069] A first processing module, which is used to train a historical classification model using a historical data set to obtain a target classification model, wherein the historical data set includes historical driving data with risk level classification labels;

[0070] A second processing module, which is used to screen out a high-value data set from the target data set based on an active learning method, wherein the target data set includes target data corresponding to the newly added driving scenario;

[0071] A third processing module is used to add the high-value data set marked with risk level classification labels to the historical data set to form a new data set;

[0072] A fourth processing module is used to train the target classification model based on the new data set to generate a target driving risk level classification model.

[0073] Furthermore, in the embodiment, the first processing module is used for:

[0074] Factor analysis is performed on historical driving data to obtain target factors, which include driver behavior factors, vehicle operating conditions factors, driver status factors, vehicle traffic conditions factors and vehicle environmental conditions factors. The driving data corresponding to the target factors constitute a historical data set.

[0075] Furthermore, in an embodiment, the active learning method is an entropy value query strategy.

[0076] Furthermore, in the embodiment, the entropy value calculation formula of the entropy value query strategy is:

[0077]

[0078] Among them, Entropy_value is the entropy value of the new data in the new data set, x is the target data, and i is the risk level category. is the probability that the target data x is classified as the i-th risk, and θ is the parameter space.

[0079] Among them, the functional implementation of each module in the above-mentioned driving risk level classification model construction method device corresponds to the various steps in the above-mentioned driving risk level classification model construction method embodiment, and its functions and implementation processes will not be repeated here one by one.

[0080] In a third aspect, an embodiment of the present application provides a driving risk level classification model construction device, which can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0081] Reference Figure 4 , Figure 4 The hardware structure diagram of the driving risk level classification model building device involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the driving risk level classification model building device may include a processor, a memory, a communication interface, and a communication bus.

[0082] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0083] The communication interface includes an input / output (I / O) interface, a physical interface, and a logical interface, which are used to interconnect the devices inside the driving risk level classification model building device, and an interface used to interconnect the driving risk level classification model building device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, an optical fiber interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0084] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0085] The processor may be a general-purpose processor, and the general-purpose processor may call the driving risk level classification model construction method program stored in the memory, and execute the driving risk level classification model construction method provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). Among them, the method executed when the driving risk level classification model construction method program is called can refer to the various embodiments of the driving risk level classification model construction method of the present application, which will not be repeated here.

[0086] Those skilled in the art will understand that Figure 4 The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0087] In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium.

[0088] The computer-readable storage medium of the present application stores a driving risk level classification model construction method program, wherein when the driving risk level classification model construction method program is executed by a processor, the steps of the driving risk level classification model construction method as described above are implemented.

[0089] Among them, the method implemented when the driving risk level classification model construction method program is executed can refer to the various embodiments of the driving risk level classification model construction method of the present application, and will not be repeated here.

[0090] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0091] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0092] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0093] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0094] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0095] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0096] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for constructing a driving risk level classification model, characterized in that: The driving risk level classification model construction method comprises: Using a historical data set to train a historical classification model to obtain a target classification model, the historical data set comprising historical driving data with risk level classification labels; Screening out a high-value data set from a target data set based on an active learning method, wherein the target data set includes target data corresponding to a newly added driving scenario; Add high-value datasets annotated with risk level classification labels to historical datasets to form new datasets; The target classification model is trained based on the new data set to generate a target driving risk level classification model.

2. The method for constructing a driving risk level classification model according to claim 1, characterized in that: Before the step of using the historical data set to train the historical classification model, the method further includes: Factor analysis is performed on historical driving data to obtain target factors, which include driver behavior factors, vehicle operating conditions factors, driver status factors, vehicle traffic conditions factors, and vehicle environmental conditions factors. The driving data corresponding to the target factors constitute a historical data set.

3. The method for constructing a driving risk level classification model according to claim 1, wherein: The active learning method is an entropy value query strategy.

4. The method for constructing a driving risk level classification model according to claim 3, characterized in that: The entropy value calculation formula of the entropy value query strategy is: Among them, Entropy_value is the entropy value of the new data in the new data set, x is the target data, and i is the risk level category. is the probability that the target data x is classified as the i-th risk, and θ is the parameter space.

5. A driving risk level classification model construction system, characterized in that: The driving risk level classification model construction system includes: A first processing module, which is used to train a historical classification model using a historical data set to obtain a target classification model, wherein the historical data set includes historical driving data with risk level classification labels; A second processing module, which is used to screen out a high-value data set from the target data set based on an active learning method, wherein the target data set includes target data corresponding to the newly added driving scenario; A third processing module is used to add the high-value data set marked with risk level classification labels to the historical data set to form a new data set; A fourth processing module is used to train the target classification model based on the new data set to generate a target driving risk level classification model.

6. The driving risk level classification model construction system according to claim 5, characterized in that: The first processing module is specifically used for: Factor analysis is performed on historical driving data to obtain target factors, which include driver behavior factors, vehicle operating conditions factors, driver status factors, vehicle traffic conditions factors, and vehicle environmental conditions factors. The driving data corresponding to the target factors constitute a historical data set.

7. The driving risk level classification model construction system according to claim 5, characterized in that: The active learning method is an entropy value query strategy.

8. The driving risk level classification model construction system according to claim 7, characterized in that: The entropy value calculation formula of the entropy value query strategy is: Among them, Entropy_value is the entropy value of the new data in the new data set, x is the target data, and i is the risk level category. is the probability that the target data x is classified as the i-th risk, and θ is the parameter space.

9. A driving risk level classification model construction device, characterized in that: The driving risk level classification model building device includes a processor, a memory, and a driving risk level classification model building program stored in the memory and executable by the processor, wherein when the driving risk level classification model building program is executed by the processor, the steps of the driving risk level classification model building method as described in any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a driving risk level classification model construction program, wherein when the driving risk level classification model construction program is executed by the processor, the steps of the driving risk level classification model construction method according to any one of claims 1 to 4 are implemented.

Images