Driving behavior prediction method and driving behavior prediction model training method and device

Abstract

The invention relates to a driving behavior prediction method and a driving behavior prediction model training method and device, and the method comprises the steps: inputting a to-be-processed vehicle parameter into a target driving behavior prediction model, and obtaining a predicted driving behavior of a vehicle; the target driving behavior prediction model is obtained by the following method: deleting a vehicle parameter corresponding to an Nth target vehicle parameter category determined according to an importance degree value of the vehicle parameter category from an Nth vehicle parameter sample to obtain an (N + 1) th vehicle parameter sample; obtaining a first target vehicle parameter sample having the same vehicle parameter category as the vehicle parameter sample to train a driving behavior prediction model, and obtaining corresponding prediction precision; the second target vehicle parameter sample having the same vehicle parameter category as the first target vehicle parameter sample corresponding to the preset prediction precision is acquired to train the driving behavior prediction model to obtain the target driving behavior prediction model, and the driving behavior can be accurately predicted based on the target driving behavior prediction model.

Description

Technical Field

[0001] The present invention relates to the field of system technologies, and particularly to a driving behavior prediction method, a driving behavior prediction model training method, and a device thereof. Background Art

[0002] Driving behavior prediction plays an important role in multiple fields such as the intelligent transportation field and the autonomous driving field. In the related art, driving parameters such as vehicle speed and acceleration collected by vehicle sensors are obtained, and these driving parameters are directly used to train a driving behavior prediction model to obtain a trained driving behavior prediction model, and driving behavior prediction is performed based on the trained driving behavior prediction model.

[0003] However, the prediction result accuracy of the driving behavior prediction model obtained by the related technology method is low. Correspondingly, the accuracy of the predicted driving behavior obtained by performing driving behavior prediction based on this driving behavior prediction model is low. Summary of the Invention

[0004] One of the purposes of the present invention is to provide a driving behavior prediction method, a driving behavior prediction model training method, and a device thereof, so as to solve the problems in the driving behavior prediction method in the related technology that the driving behavior prediction model needs to process a large amount of data and the accuracy of the driving behavior prediction result is low; the second purpose is to provide a training method for a driving behavior prediction model; the third purpose is to provide a driving behavior prediction device; the fourth purpose is to provide a training device for a driving behavior prediction model; the fifth purpose is to provide an electronic device; the sixth purpose is to provide a vehicle.

[0005] To achieve the above purposes, the technical solutions adopted by the present invention are as follows: A driving behavior prediction method includes: Obtaining the vehicle parameters to be processed of a vehicle; Inputting the vehicle parameters to be processed into a target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; Among them, the target driving behavior prediction model is obtained through the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample; where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample, and using the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; obtaining second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to the preset prediction accuracy, and using the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0006] A driving behavior prediction method is applied to the cloud, and the method includes: Receiving the vehicle parameters to be processed sent by the vehicle; Inputting the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; Among them, the target driving behavior prediction model is obtained through the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample, and using the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; obtaining second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to the preset prediction accuracy, and using the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0007] A driving behavior prediction method is applied to the vehicle, and the method includes: Obtaining the vehicle parameters to be processed; Sending the vehicle parameters to be processed to the cloud for the cloud to input the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; Among them, the target driving behavior prediction model is obtained through the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as those of each vehicle parameter sample, and using the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; obtaining second target vehicle parameter samples with the same vehicle parameter categories as those of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and using the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0008] A method for training a driving behavior prediction model includes: Deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; Respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as those of each vehicle parameter sample, and using the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; Obtaining second target vehicle parameter samples with the same vehicle parameter categories as those of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and using the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0009] A driving behavior prediction device includes: A first acquisition module, configured to acquire the vehicle parameters to be processed of the vehicle; A second acquisition module, configured to input the vehicle parameters to be processed into a target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample, and using the first target vehicle parameter samples to train a driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; obtaining second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to a preset prediction accuracy, and using the second target vehicle parameter samples to train a preset driving behavior prediction model to obtain a target driving behavior prediction model.

[0010] A driving behavior prediction device, applied to the cloud, the device includes: A third acquisition module, configured to receive the vehicle parameters to be processed sent by the vehicle; A fourth acquisition module, configured to input the vehicle parameters to be processed into a target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample, and using the first target vehicle parameter samples to train a driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; obtaining second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to a preset prediction accuracy, and using the second target vehicle parameter samples to train a preset driving behavior prediction model to obtain a target driving behavior prediction model.

[0011] A driving behavior prediction device, applied to the vehicle, the device includes: A fifth acquisition module, configured to acquire the vehicle parameters to be processed; A sending module, configured to send the to-be-processed vehicle parameters to the cloud, so that the cloud inputs the to-be-processed vehicle parameters into a target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, wherein the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as those of each vehicle parameter sample, and using the first target vehicle parameter samples to train a driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; obtaining second target vehicle parameter samples with the same vehicle parameter categories as those of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and using the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0012] A driving behavior prediction model training device, comprising: A sixth obtaining module, configured to delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, wherein the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; A seventh obtaining module, configured to respectively obtain first target vehicle parameter samples with the same vehicle parameter categories as those of each vehicle parameter sample, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; An eighth obtaining module, configured to obtain second target vehicle parameter samples with the same vehicle parameter categories as those of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and use the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0013] An electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein, the processor is configured to execute the instructions to implement at least one of the above-mentioned driving behavior prediction method and driving behavior prediction model training method.

[0014] A computer-readable storage medium, when the instructions in the storage medium are executed by a processor of a mobile terminal, enable the mobile terminal to execute the above-mentioned driving behavior prediction method and driving behavior prediction model training method.

[0015] A vehicle, the vehicle includes the above-mentioned electronic device.

[0016] Advantages of the present application: In this embodiment, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], n is an integer greater than or equal to 2, then N = 1, 2... n. Based on this embodiment, n loop operations are realized. Among them, in the Nth loop, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. Thus, through n loop operations, n vehicle parameter samples with vehicle parameters corresponding to the deleted target vehicle parameter category can be obtained. For these n vehicle parameter samples and the vehicle parameter sample before vehicle parameter deletion in the first loop operation, first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample are respectively obtained, and the driving behavior prediction model is trained using the first target vehicle parameter samples to obtain the prediction accuracy corresponding to the first target vehicle parameter samples. Second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to the preset prediction accuracy are obtained. The second target vehicle parameter samples are comprehensively determined according to the importance value of the vehicle parameter categories and the corresponding prediction accuracy. Compared with the methods of the related art, the target driving behavior prediction model obtained based on the second target vehicle parameter samples of this embodiment has high prediction result accuracy, and the accuracy of the predicted driving behavior obtained by inputting the to-be-processed parameters of the vehicle into the target driving behavior prediction model is high. Description of the Drawings

[0017] Figure 1 It is a step flow chart of a driving behavior prediction method provided in an embodiment of the present invention; Figure 2 It is a step flow chart of a method for obtaining the prediction accuracy corresponding to the first target vehicle parameter sample provided in an embodiment of the present invention; Figure 3 It is a step flow chart of a driving behavior prediction method provided in an embodiment of the present invention; Figure 4 It is a step flow chart of a driving behavior prediction method provided in an embodiment of the present invention; Figure 5It is a flowchart of steps of a method for training a driving behavior prediction model provided in an embodiment of the present invention; Figure 6 It is a flowchart of steps of another method for training a driving behavior prediction model provided in an embodiment of the present invention; Figure 7 It is a flowchart of steps of a method for obtaining an updated feature set provided in an embodiment of the present invention; Figure 8 It is a flowchart of a method for predicting a driving behavior based on a driving behavior prediction system provided in an embodiment of the present invention; Figure 9 It is a schematic structural diagram of a driving behavior prediction device provided in an embodiment of the present invention; Figure 10 It is a schematic structural diagram of another driving behavior prediction device provided in an embodiment of the present invention; Figure 11 It is a schematic structural diagram of yet another driving behavior prediction device provided in an embodiment of the present invention; Figure 12 It is a schematic structural diagram of a driving behavior prediction model training device provided in an embodiment of the present invention; Figure 13 It is a schematic structural diagram of an electronic device provided in an embodiment of the present invention. Detailed implementation manners

[0018] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for explaining the present invention, rather than for limiting the protection scope of the present invention.

[0019] Refer to Figure 1 , which shows a flowchart of steps of a driving behavior prediction method provided in an embodiment of the present invention. The vehicle is equipped with at least one sensor and component, and specifically includes the following steps: Step 101, obtain the vehicle parameters to be processed of the vehicle.

[0020] The vehicle parameters to be processed include vehicle parameters related to the driving behavior prediction result. By way of example, the vehicle parameters may include at least one of the following: vehicle operation parameters, in-vehicle environment parameters, and out-of-vehicle environment parameters.

[0021] Exemplarily, the vehicle operation parameters may include vehicle operation parameters collected by vehicle body sensors such as vehicle speed, engine speed, engine coolant temperature, transmission oil temperature, etc. The in-vehicle environment parameters may include: driver's driving duration, number of people in the vehicle during driving, whether music is played during driving, whether there is communication and conversation among the people in the vehicle during driving, etc. The out-of-vehicle environment parameters may include weather, time during driving, road conditions of the road on which the vehicle is traveling, traffic congestion of the road on which the vehicle is traveling, etc. The road conditions of the road on which the vehicle is traveling may include that the road on which the vehicle is traveling is an urban road, a rural road, an expressway, or other roads.

[0022] Step 102: Input the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle.

[0023] Exemplarily, the target driving behavior prediction model may be a classification model, and the target driving behavior prediction model is used to obtain the predicted driving behavior based on the input vehicle parameters to be processed. Exemplarily, the target driving behavior prediction model may be a random forest model.

[0024] Among them, the target driving behavior prediction model is obtained through the following method: Step A1: Delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample.

[0025] Among them, the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2.

[0026] Among them, the Nth target vehicle parameter category may include one or more vehicle parameter categories. The importance value is used to characterize the importance of the vehicle parameter category to the driving behavior prediction result.

[0027] For example, if n is equal to 2, then this step includes: deleting the vehicle parameters corresponding to the 1st target vehicle parameter category from the 1st vehicle parameter sample to obtain the 2nd vehicle parameter sample; then deleting the vehicle parameters corresponding to the 2nd target vehicle parameter category from the 2nd vehicle parameter sample to obtain the 3rd vehicle parameter sample.

[0028] For another example, if n is equal to 3, then this step includes: deleting the vehicle parameters corresponding to the 1st target vehicle parameter category from the 1st vehicle parameter sample to obtain the 2nd vehicle parameter sample; then deleting the vehicle parameters corresponding to the 2nd target vehicle parameter category from the 2nd vehicle parameter sample to obtain the 3rd vehicle parameter sample; then deleting the vehicle parameters corresponding to the 3rd target vehicle parameter category from the 3rd vehicle parameter sample to obtain the 4th vehicle parameter sample.

[0029] The vehicle parameter categories may include at least one of the following: vehicle operation parameters, in-vehicle environment parameters, and out-of-vehicle environment parameters.

[0030] Exemplarily, the importance degree value of the deleted Nth target vehicle parameter category is less than the importance degree values of other vehicle parameter categories in the Nth vehicle parameter sample except the Nth target vehicle parameter category; Exemplarily, the importance degree value is used to characterize the importance degree of the vehicle parameter category corresponding to the importance degree value for the driving behavior prediction result.

[0031] Exemplarily, input the Nth vehicle parameter sample into the driving behavior prediction model to obtain the importance degree values of each vehicle parameter category in the Nth vehicle parameter sample respectively; Exemplarily, the driving behavior prediction model may be a random forest model.

[0032] Exemplarily, the parameter n can be determined according to the vehicle parameter categories in the first vehicle parameter sample and the preset number of remaining vehicle parameter categories. For example, if the number of vehicle parameter categories in the first vehicle parameter sample is k and the number of the Nth target vehicle parameter categories deleted each time is 1, then n = k - 1. Thus, after the operation of this step is completed, the number of remaining vehicle parameter categories in the (n + 1)th vehicle parameter sample is 1.

[0033] Exemplarily, the parameter n can also be set according to user requirements under the constraint condition that the number of remaining vehicle parameter categories in the (n + 1)th vehicle parameter sample is greater than zero. For example, it can be set to 2, 3 or other values.

[0034] Exemplarily, the driving behavior prediction model in this embodiment may be a random forest model. The random forest model can not only be used for classification and regression processing, but also perform feature importance analysis on the vehicle parameters input into the model. Specifically, when using the random forest model for importance analysis, it mainly conducts importance analysis from two aspects: accuracy rate and node impurity. Further, when analyzing importance, the random forest model uses the bootstrap sampling method to sample the parameter features of the vehicle parameters, and calculates its Out-Of-Bag (OOB) value. Then, according to the OOB value, the parameter features are replaced, and the OOB value of the random forest is calculated again based on the replaced parameter features. By comparing the OOB values of the parameter features of each vehicle parameter, the importance degree value of the vehicle parameter is obtained.

[0035] Exemplarily, the importance degree value of the Nth target vehicle parameter category is less than the importance degree values of other vehicle parameter categories in the Nth vehicle parameter sample except the Nth target vehicle parameter category.

[0036] Correspondingly, in this step, what is deleted from the Nth vehicle parameter sample is the vehicle parameters of the Nth target vehicle parameter category whose importance degree value is smaller than the importance degree values of other vehicle parameter categories.

[0037] Exemplarily, there may be one or more Nth target vehicle parameter categories. According to the importance degree values, the importance degree values of each vehicle parameter category in the Nth vehicle parameter sample are sorted, and then a preset number of vehicle parameter categories with smaller importance degree values are determined as the Nth target vehicle parameter categories. For example, if there is one Nth target vehicle parameter category, the vehicle parameter category with the smallest importance degree value can be determined as the Nth target vehicle parameter sample.

[0038] For example, the vehicle parameter categories include vehicle speed, engine speed, engine coolant temperature, transmission oil temperature, road condition of the vehicle's driving road, number of passengers in the vehicle, and weather. The Nth vehicle parameter sample including these vehicle parameters is input into the driving behavior prediction model, and the obtained importance degrees from high to low are: vehicle speed, engine speed, transmission oil temperature, road condition of the vehicle's driving road, number of passengers in the vehicle, weather, and engine coolant temperature. Then, the engine coolant temperature can be determined as the Nth target vehicle parameter category and deleted to update the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. The vehicle parameters included in the (N + 1)th vehicle parameter sample are vehicle speed, engine speed, transmission oil temperature, road condition of the vehicle's driving road, number of passengers in the vehicle, and weather respectively.

[0039] This step realizes n loop operations. For each parameter N, N ∈ [1, n], where n is an integer greater than or equal to 2; in the Nth loop operation, the importance degree values of each vehicle parameter category in the Nth vehicle parameter sample are respectively obtained, and a preset number of vehicle parameter categories are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. If the importance degree value of the Nth target vehicle parameter category is less than the importance degree values of other vehicle parameter categories in the Nth vehicle parameter sample except the Nth target vehicle parameter category, then in each loop operation, the vehicle parameters of the Nth target vehicle parameter category with an importance degree value smaller than that of other vehicle parameter categories are deleted. In other words, the importance degree value of the vehicle parameter types in the (N + 1)th vehicle parameter sample obtained based on this embodiment is greater than the importance degree value of the deleted vehicle parameter types. Based on this, subsequently, the prediction accuracy is calculated for the first target vehicle parameter sample with the same parameter types as each vehicle parameter sample, and the second target vehicle parameter sample is determined according to the prediction accuracy. The prediction accuracy of the vehicle parameter categories in the second target vehicle parameter sample is equal to the preset prediction accuracy and the importance degree value is relatively large. Using the second target vehicle parameter sample including these vehicle parameter categories to train the driving behavior prediction model can avoid the influence of vehicle parameters with relatively small importance degree values on the prediction result and ensure the prediction accuracy.

[0040] This embodiment realizes the recursive deletion of vehicle parameter categories in vehicle parameter samples. For example, if the driving behavior prediction model is a random forest model, based on this step, recursive feature elimination can be combined with the random forest model to delete redundant features with low importance values, and obtain features with high importance values for the driving behavior prediction results. Among them, the redundant features are the vehicle parameters with low importance values in this embodiment.

[0041] Step A2: Respectively obtain first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample, and use the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples.

[0042] For example, if the vehicle parameter samples include vehicle speed, engine speed, transmission oil temperature, road conditions of the vehicle's driving road, number of passengers in the vehicle, and weather, then the first target vehicle parameter samples also include vehicle parameters of these vehicle parameter categories.

[0043] Respectively obtain first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample, and use the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples. Further, use the first target vehicle parameter samples to train the driving behavior prediction model to obtain the trained first driving behavior prediction model, and determine the prediction accuracy of the trained first driving behavior prediction model as the prediction accuracy corresponding to the first target vehicle parameter samples.

[0044] Among them, during each loop, while performing the loop operation, first target vehicle parameter samples that are the same as the vehicle parameter categories of the vehicle parameter samples can be obtained, and the first target vehicle parameter samples can be used to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; alternatively, after the loop ends, first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample can be obtained respectively, and the first target vehicle parameter samples can be used to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples.

[0045] Step A3: Obtain second target vehicle parameter samples that are the same as the vehicle parameter categories of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and use the second target vehicle parameter samples to train the driving behavior prediction model to obtain the target driving behavior prediction model.

[0046] Among them, the second target vehicle parameter sample can be a vehicle parameter sample selected from the vehicle parameter samples and having the same vehicle parameter category as the first target vehicle parameter sample; or it can be another obtained parameter sample having the same vehicle parameter category as the first target vehicle parameter sample. In other words, in this step, it is only necessary to ensure that the vehicle parameter category of the second target vehicle parameter sample is the same as that of the first target vehicle parameter sample corresponding to the preset prediction accuracy.

[0047] Driving behavior prediction is of great significance for improving traffic safety, optimizing traffic flow, and assisting intelligent driving. In related technologies, driving parameters such as vehicle speed and acceleration collected by vehicle sensors are obtained, and these driving parameters are directly used to train a driving behavior prediction model to obtain a trained driving behavior prediction model, and driving behavior prediction is performed based on the trained driving behavior prediction model. However, there may be parameters among these driving parameters that have a negative impact on the accuracy of the model prediction results. According to the related technical methods, directly using these driving parameters to train the driving behavior prediction model results in a low accuracy of the prediction results of the driving behavior prediction model. Correspondingly, based on this driving behavior prediction model for driving behavior prediction, the accuracy of the predicted driving behavior obtained is low.

[0048] Exemplarily, the method of this embodiment can be applied to the cloud, can be applied to the vehicle, or can be applied to a system composed of the cloud and the vehicle. When the method is applied to the cloud, the vehicle's to-be-processed vehicle parameters obtained by the cloud are data collected by the vehicle and sent to the cloud.

[0049] In this embodiment, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2, so N = 1, 2... n. Based on this embodiment, n loop operations are implemented. Among them, in the Nth loop, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. Thus, through n loop operations, n vehicle parameter samples with vehicle parameters corresponding to the deleted target vehicle parameter category can be obtained. For these n vehicle parameter samples and the vehicle parameter sample before vehicle parameter deletion in the first loop operation, first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample are respectively obtained, and the driving behavior prediction model is trained using the first target vehicle parameter samples to obtain the prediction accuracy corresponding to the first target vehicle parameter samples. Second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to the preset prediction accuracy are obtained. The second target vehicle parameter samples are comprehensively determined according to the importance value of the vehicle parameter categories and the corresponding prediction accuracy. Compared with the method of the related art, the target driving behavior prediction model obtained based on the second target vehicle parameter samples of this embodiment has high prediction result accuracy, and when the to-be-processed parameters of the vehicle are input into the target driving behavior prediction model, the accuracy of the predicted driving behavior obtained is high.

[0050] Exemplarily, feature screening for training the model can be performed by the Embedded method or the Filter method. Specifically, the Embedded method screens the feature subset by adjusting the weights of the feature variables, and the applicability of this method to different algorithm models is different. The Filter method evaluates the importance of individual feature variables and then filters out features with importance lower than the threshold according to the set threshold, thereby achieving feature screening. However, the threshold is usually set according to experience, and the feature subset selected may not necessarily have the best effect. However, these feature screening methods are relatively single, and the obtained feature subsets may not necessarily be the best, and the model prediction accuracy obtained by training the model based on these feature subsets may be relatively low. The second target vehicle parameter samples of this embodiment are obtained by combining the importance of vehicle parameters to the driving behavior prediction results and the prediction accuracy. Compared with the method of screening parameters according to the importance of the parameters and training the driving behavior prediction model with the selected parameters, the second target vehicle parameter samples of this embodiment can not only meet the importance requirements but also meet the prediction accuracy requirements. Based on the second target vehicle parameter samples, the target driving behavior prediction model can be accurately obtained, and based on the target driving behavior prediction model, the predicted driving behavior can be accurately obtained.

[0051] A method of obtaining the importance degree value of vehicle parameter features, then screening out the features that meet the threshold conditions according to the set importance degree value threshold, constructing a feature subset based on the screened features, and training a model according to the feature subset. This application does not need to use the importance degree value threshold, which is a parameter that needs to be set according to subjective experience. In this embodiment, vehicle parameter samples of multiple combinations of vehicle parameter categories are obtained according to the importance degree value, and then, according to the prediction accuracy calculated from the first target vehicle parameter samples, multiple combinations of vehicle parameter category samples are screened to obtain second target vehicle parameter samples corresponding to the preset prediction accuracy. The second target vehicle parameter samples obtained in this way are not affected by subjective experience, have higher usability and accuracy. Correspondingly, the prediction result of the target driving behavior prediction model obtained based on the second target vehicle parameter samples has higher accuracy.

[0052] In an embodiment of the present application, the method of using the first target vehicle parameter samples to train the driving behavior prediction model in step A2 to obtain the prediction accuracy corresponding to the first target vehicle parameter samples may include sub-steps A21 to A24: Sub-step A21: Obtain multiple groups of sample combinations including a training set and a test set according to the first target vehicle parameter samples.

[0053] There are multiple first target vehicle parameter samples, and multiple divisions can be performed on the multiple first target vehicle parameter samples to obtain multiple groups of sample combinations including a training set and a test set.

[0054] Sub-step A22: Train the driving behavior prediction model according to the sample combination to obtain the first prediction accuracy corresponding to the sample combination.

[0055] Specifically, for each sample combination, train the driving behavior prediction model according to the training set and the test set in the sample combination to obtain the trained first driving behavior prediction model, and obtain the first prediction accuracy of the first driving behavior prediction model.

[0056] Sub-step A23: Obtain the average value of the prediction accuracies of the multiple first prediction accuracies.

[0057] Sub-step A24: Obtain the prediction accuracy corresponding to the first target vehicle parameter samples according to the average value of the prediction accuracies.

[0058] The average value of the prediction accuracies can be directly determined as the prediction accuracy corresponding to the first target vehicle parameter samples. It is also possible to obtain a preset correction coefficient and correct the average value of the prediction accuracies according to the preset correction coefficient to obtain the prediction accuracy corresponding to the first target vehicle parameter samples.

[0059] In this embodiment, multiple groups of sample combinations including training sets and test sets are obtained based on the first target vehicle parameter samples. For each sample combination including a training set and a test set, a driving behavior prediction model is trained respectively to obtain a first prediction accuracy corresponding to the sample combination. These first prediction accuracies are prediction accuracies corresponding one by one to the sample combinations. According to the average value of these first prediction accuracies, the prediction accuracy that can accurately reflect the overall prediction effect of multiple groups of sample combinations can be obtained, and it is used as the prediction accuracy corresponding to the first target vehicle parameter samples, which has the characteristic of high accuracy of the obtained first target vehicle parameter samples.

[0060] In an embodiment of the present application, there are multiple first target vehicle parameter samples. In sub-step A21, obtaining multiple groups of sample combinations of training sets and test sets based on the first target vehicle parameter samples may include: traversing the first target vehicle parameter samples. For each traversal operation, the traversed first target vehicle parameter sample is determined as the test set, and the other first target vehicle parameter samples are determined as the training set, so as to obtain a sample combination corresponding to the traversal operation.

[0061] For example, referring to Figure 2 , if there are k first target vehicle parameter samples, in the first traversal operation, the 1st first target vehicle parameter sample can be determined as the test set, and the other k - 1 first target vehicle parameter samples are determined as the training set, so as to obtain a sample combination 1 including a test set and a training set; in the second traversal operation, the 2nd first target vehicle parameter sample is determined as the test set, and the other k - 1 first target vehicle parameter samples are determined as the training set, so as to obtain a sample combination 2 including a test set and a training set. And so on, k sample combinations including training sets and test sets are obtained. Among them, in the kth sample combination, the kth first target vehicle parameter sample is used as the test set, and the other second vehicle test samples are used as the training set.

[0062] After obtaining k sample combinations including training sets and test sets, each sample combination is used to train a preset driving behavior prediction module respectively to obtain a first prediction accuracy corresponding to the sample combination, and then the average of the k first prediction accuracies is calculated to obtain the prediction accuracy corresponding to the first target vehicle parameter samples.

[0063] According to the foregoing embodiments, a sample combination including a training set and a test set is used to determine the calculation of the prediction accuracy of the first target vehicle parameter sample. In this embodiment, multiple sample combinations including a training set and a test set are obtained through a traversal method. Then, each first target vehicle parameter sample is sequentially used as the test set, and the other first target vehicle parameter samples except the traversed first target vehicle parameter sample are used as the training set, thereby obtaining multiple groups of sample combinations. Based on the sample combination, the prediction accuracy of the first target vehicle parameter sample is obtained, realizing the cross-validation of the prediction accuracy corresponding to the vehicle parameter category in the first target vehicle parameter sample, and the prediction accuracy when using the vehicle parameters of the vehicle parameter category in the first target vehicle parameter sample for driving behavior prediction can be dynamically and accurately obtained.

[0064] Updating the Nth vehicle parameter sample according to the importance degree of the vehicle parameter category realizes the recursive feature elimination of the vehicle parameters in the Nth vehicle parameter sample. Combining the method of this embodiment, the combination of the recursive feature elimination and the cross-validation method is realized. Based on the combined processing of the recursive feature elimination and the cross-validation method, the prediction accuracy corresponding to the vehicle parameter category can be dynamically evaluated, and the stability of the obtained prediction accuracy can be ensured. Based on the combined processing of the recursive feature elimination and the cross-validation method, the second target vehicle parameter sample including the optimal vehicle parameter category can be screened out. Based on the second target vehicle parameter sample, a feature subset with optimal prediction performance can be obtained, and when training a driving behavior prediction model based on the second target vehicle parameter sample, the obtained target driving behavior prediction model has high prediction accuracy.

[0065] In an embodiment of the present application, sub-step A22 may include sub-steps A221 to A222: Sub-step A221, training a driving behavior prediction model according to the sample combination to obtain a first driving behavior prediction model, as well as the precision rate and recall rate of the first driving behavior prediction model.

[0066] The precision rate of the first driving behavior prediction model can reflect the accuracy of the first driving behavior prediction model in predicting positive examples and is used to characterize the proportion of true positive examples. Among them, a positive example is a prediction result predicted as a driving behavior, and a true positive example is a prediction result correctly predicted as a driving behavior. The recall rate of the first driving behavior prediction model is used to reflect the ability of the first driving behavior prediction model to capture positive examples.

[0067] Sub-step A222, obtaining a first prediction accuracy corresponding to the sample combination according to the precision rate and the recall rate.

[0068] The first prediction accuracy is used to evaluate the performance of the first driving behavior prediction model and can comprehensively reflect the magnitudes of the precision rate and the recall rate. For example, the first prediction accuracy F corresponding to the sample combination can be obtained according to the following method:

[0069] Precision is the proportion of true positive samples among the samples predicted as positive; Recall is the proportion of actual positive samples correctly predicted as positive.

[0070] Among them, the higher the first prediction accuracy F value, the better the balance between precision and recall of the first driving behavior prediction model, and the better the performance of the first driving behavior prediction model. For example, if the F value is 1, it means that both the precision and recall of the model reach 100%; if the F value is low, it indicates that the precision or recall of the first driving behavior prediction model is low.

[0071] In this embodiment, a driving behavior prediction model is trained according to a sample combination to obtain a first driving behavior prediction model, as well as the precision and recall of the first driving behavior prediction model. According to the precision and recall, a first prediction accuracy corresponding to the sample combination is obtained, and this first prediction accuracy can accurately reflect the prediction performance of the first driving behavior prediction model obtained when training the driving behavior prediction model using the training set and validation set in the sample combination.

[0072] In an embodiment of the present application, during the loop operation of step A1, when N = 1, the Nth vehicle parameter sample can be obtained through sub-steps A11 to A15: Sub-step A11, obtain the driving behavior time series and at least one vehicle operation parameter time series.

[0073] Among them, when N = 1, the Nth vehicle parameter sample is the vehicle parameter sample processed in the first loop operation.

[0074] Among them, the types of vehicle operation parameters in different vehicle operation parameter time series are different from each other.

[0075] The vehicle operation parameters can be parameters collected by vehicle body sensors. For example, the vehicle operation parameters can include vehicle speed, engine speed, engine coolant temperature, transmission oil temperature, etc.

[0076] The vehicle operation parameter time series includes multiple vehicle operation parameters sorted in chronological order; the driving behavior time series includes multiple driving behaviors arranged in chronological order. Among them, the collection time of the vehicle operation parameters in the vehicle operation parameter time series corresponds to the collection time of the driving behaviors in the driving behavior time series. For example, the vehicle operation parameter time series includes vehicle operation parameters at time T1, time T2, time T3 to time Tn, and correspondingly, the driving behavior time series includes driving behaviors at time T1, time T2, time T3 to time Tn.

[0077] Sub-step A12: Obtain the time difference correlation coefficients between the driving behavior time series and each vehicle operation parameter time series respectively.

[0078] The time difference correlation coefficient is used to characterize that the change of the vehicle operation parameter in the vehicle operation parameter time series is advanced or lagged behind the change of the driving behavior in the driving behavior time series, or changes simultaneously with the driving behavior in the driving behavior time series.

[0079] Sub-step A13: According to the time difference correlation coefficients, obtain at least one target vehicle operation parameter time series from at least one vehicle operation parameter time series.

[0080] Among them, the time when the vehicle operation parameter in the target vehicle operation parameter time series changes is earlier than the time when the driving behavior in the driving behavior time series changes.

[0081] The time when the vehicle operation parameter in the target vehicle operation parameter time series changes is earlier than the time when the driving behavior in the driving behavior time series changes, indicating that the change of the vehicle operation parameter in the target vehicle operation parameter time series causes the change of the driving behavior in the time series. Therefore, these vehicle operation parameters are the vehicle operation parameters that have a greater impact on the driving behavior. Training the driving behavior prediction model based on these vehicle operation parameters can obtain a higher accuracy of the model prediction result.

[0082] Sub-step A14: Obtain vehicle operation parameter samples with the same vehicle parameter category as the target vehicle operation parameter time series.

[0083] For example, if the vehicle operation parameters in the target vehicle operation parameter sequence include vehicle speed and engine speed, then obtain the vehicle speed and engine speed, and construct vehicle operation parameter samples based on this.

[0084] Sub-step A15: Construct the Nth vehicle parameter sample according to the vehicle operation parameter samples.

[0085] Other vehicle parameter samples related to the driving behavior can be obtained, and the sample combination of these vehicle parameter samples and the vehicle operation parameter samples is determined as the Nth vehicle parameter sample.

[0086] In the related art, vehicle sensor data can be obtained, and then driving behavior prediction can be performed based on a graph convolutional network model (Graph Convolutional Network, GCN) and a least squares regularized multi-manifold neural network (Least Squares Regularized Multi-Manifold Neural Network, LSTMNN) model. However, such prediction methods all have problems of high requirements for data and sensitivity to noise, and these prediction models are equivalent to a black box, with the problem that the prediction results are difficult to interpret. In addition, the methods in the related art do not judge the time series of vehicle sensor data and driving behavior, but directly perform predictions based on all sensor data. However, vehicle sensor data that is consistent with or lags behind the change in driving behavior is meaningless for the prediction result, which not only affects the accuracy of the prediction result but also increases the amount of data that the model needs to process.

[0087] In this embodiment, by calculating the cross-correlation coefficient of time difference between the driving behavior time series and each vehicle operation parameter time series, and screening the vehicle operation parameters according to the cross-correlation coefficient of time difference, the vehicle operation parameters leading to the change in driving behavior can be screened out. Based on this, the problems in the related art of low prediction result accuracy and large amount of data that the model needs to process caused by the failure to screen sensor data can be overcome.

[0088] In the related art, driving behavior can also be predicted based on a weighted graph convolutional neural network (Weighted Graph Convolutional Networks, WGCN) model. Specifically, a feature matrix and a local map of each vehicle are generated, the weighted feature matrix and the local map encoded by a convolutional neural network model (Convolutional Neural Network) are constructed into a graph, and then input into an edge-enhanced graph convolutional neural network GCN. Then, the GCN extracts the features of the input data, and the interaction features output by the GCN are input into a Transformer model for model training, and the prediction result of the vehicle driving behavior is obtained through the fully connected layer of the model. This method needs to complete the prediction based on the GCN model, requires a large amount of data to be processed, has extremely high requirements for computing power, and it is very difficult to interpret the prediction result. In addition, this method does not consider the time-delay relationship between vehicle data and driving behavior. For example, the random forest model has lower requirements for computing power than the GCN. In this embodiment, the random forest model can be used for driving behavior prediction to reduce the computing power requirements.

[0089] In practical applications, there is a temporal relationship between the changes in vehicle body sensing data and driving behavior. In the entire time series, there are leading, concurrent, or lagging relationships between two or more sequences. In this embodiment, the vehicle operation parameters may be data collected by vehicle sensors. Based on the cross-correlation coefficient of this embodiment, sensor data that leads the change in driving behavior can be obtained. By using the cross-correlation coefficient to obtain the vehicle operation parameters that lead the change in driving behavior, the vehicle operation parameters whose change moments lag behind the driving behavior or change simultaneously with the driving behavior are filtered out, which can avoid the interference of these vehicle operation parameters on the prediction results.

[0090] Analyze the driving behavior time series and the vehicle operation parameter time series according to the cross-correlation coefficient. Take the vehicle operation parameter time series as the reference sequence and the driving behavior time series as the analysis sequence. By quantifying the time relationship between the two sequences through the cross-correlation coefficient, it can be accurately determined whether the analysis sequence leads, lags, or is concurrent with the reference sequence. This processing method has the advantages of easy-to-understand principle, simple calculation, and accurate and reliable calculation results.

[0091] During the training process of the driving behavior prediction model, filter out the vehicle operation parameters that lead the change in driving behavior. Based on the filtered vehicle operation parameters, construct a training sample for training the driving behavior prediction model. Train the driving behavior prediction model based on this training sample, and the prediction accuracy of the obtained target driving behavior prediction model is relatively high. In other words, by screening the vehicle operation parameters obtained by methods such as sensor collection based on the method of this embodiment, sensor data that leads the change in driving behavior can be screened out. Based on the screened sensor data, the prediction result accuracy of the target driving behavior prediction model can be improved.

[0092] Further, according to the foregoing analysis, in this embodiment, by obtaining the time difference correlation coefficient between the driving behavior time series and each vehicle operation parameter time series, and based on the time difference correlation coefficient, at least one target vehicle operation parameter time series is obtained from at least one vehicle operation parameter time series. For the obtained target vehicle operation parameter time series, the time when the vehicle operation parameter changes in the target vehicle operation parameter time series is earlier than the time when the driving behavior changes in the driving behavior time series. Therefore, the vehicle operation parameters in the target vehicle operation parameter time series are the parameters that cause changes in the driving behavior, and these vehicle operation parameters have an important impact on the driving behavior. According to the obtained vehicle operation parameter samples, the Nth vehicle parameter samples that have an important impact on the driving behavior can be obtained, where N = 1. Therefore, through this embodiment, the vehicle parameter samples that have an important impact on the driving behavior processed in the first loop operation can be obtained. Based on this vehicle parameter sample, subsequent n loop operations are performed. The vehicle parameter samples obtained in each loop operation are all parameter samples that have an important impact on the driving behavior. Based on this, the second target vehicle parameter sample is obtained, and the driving behavior prediction model is trained according to the second target vehicle parameter sample, and the prediction result of the obtained target driving behavior prediction model has high accuracy. In addition, based on this embodiment, it can be known which vehicle operation parameters of which vehicle parameter categories change earlier than the driving behavior, and these parameters are the vehicle operation parameters that have an important impact on the driving behavior prediction. Based on the vehicle operation parameters of these vehicle parameter categories, the Nth vehicle parameter sample is constructed and the subsequent training process of the driving behavior prediction model is performed, which can realize the interpretability of the model.

[0093] In addition, compared with the method of obtaining the importance degree value of features, then screening out the features that meet the threshold conditions according to the set importance degree value threshold, constructing a feature subset according to the screened features, and training a model according to the feature subset, this embodiment avoids the problem that after calculating the importance of all features at one time and screening features according to the importance degree value threshold, the screened features cannot reflect the correlation between features, which will affect the effectiveness of feature selection.

[0094] In an embodiment of the present application, obtaining the time difference correlation coefficient between the driving behavior time series and each vehicle operation parameter time series in sub-step A12 may include sub-step A121: Sub-step A121, for each vehicle operation parameter time series, obtain the time difference correlation coefficient between the driving behavior time series and the vehicle operation parameter time series according to the following method:

[0095] where is the order of time lag The corresponding time difference correlation coefficient is used to characterize the time interval in which the driving behavior changes in the driving behavior time series of the time interval and the vehicle operation parameters change in the vehicle operation parameter time series; is the vehicle operation parameter at the moment in the vehicle operation parameter time series, is the average value of all vehicle operation parameters in the vehicle operation parameter time series; is the eigenvalue of the driving behavior at the moment in the driving behavior time series, is the average value of the eigenvalues of all driving behaviors in the driving behavior time series, is the total moment.

[0096] In this embodiment, the time difference correlation coefficient can reflect the correlation degree between the driving behavior in the driving behavior time series and the vehicle operation parameters in the vehicle operation parameter time series; based on the calculation method for obtaining the time difference correlation coefficient in this embodiment, the time difference correlation coefficient corresponding to different time intervals can be accurately obtained, and then the time interval corresponding to different correlation degrees between the driving behavior in the driving behavior time series and the vehicle operation parameters in the vehicle operation parameter time series can be accurately obtained. According to this time interval, the vehicle operation parameters that are earlier than the change of the driving behavior in the driving behavior time series can be accurately obtained.

[0097] In an embodiment of the present application, if there are multiple time lag orders in sub-step A13, obtaining at least one target vehicle operation parameter time series from at least one vehicle operation parameter time series according to the time difference correlation coefficient may include sub-steps A131 to A133.

[0098] Sub-step A131: For each vehicle operation parameter time series, obtain the time difference correlation coefficient corresponding to each time lag order. By way of example, the value range of the time lag order can be , where is the maximum delay number, and the maximum delay number is equal to the preset prediction duration. The prediction duration can be set according to user needs. For example, if it is necessary to predict the driving behavior after 1 minute, the preset duration can be set to 1 minute. Further, within the value range of the time lag order , multiple time lag orders can be obtained according to the preset value interval.

[0099] Sub-step A132: Obtain the maximum time difference correlation coefficient corresponding to each vehicle operation parameter time series, and the time lag order corresponding to the maximum time difference correlation coefficient.

[0100] Among them, the time difference correlation coefficient can reflect the degree of correlation between the driving behavior in the driving behavior time series and the vehicle operation parameters in the vehicle operation parameter time series. The maximum time difference correlation coefficient is the maximum value of the degree of correlation between the driving behavior in the driving behavior time series and the vehicle operation parameters in the vehicle operation parameter time series, and the corresponding time lag order , is the time lag order in the case of the maximum degree of correlation.

[0101] The time lag order is used to characterize the time interval between the change in the driving behavior in the driving behavior time series and the change in the vehicle operation parameters in the vehicle operation parameter time series. If is less than zero, it means that the driving behavior is advanced relative to the change in the vehicle operation parameters. If is greater than zero, it means that the driving behavior is lagging behind the change in the vehicle operation parameters. If is equal to 0, it means that the two changes are basically the same. In other words, if the time lag order is greater than zero, it indicates that the change in the vehicle operation parameters is ahead of the change in the driving behavior, and the vehicle operation parameters corresponding to the time lag order greater than zero are the vehicle operation parameters that have a great influence on the change in the driving behavior Sub-step A133, determine the vehicle operation parameter time series with a time lag order greater than zero corresponding to the maximum time difference correlation coefficient as the target vehicle operation parameter time series.

[0102] In this embodiment, if the time difference correlation coefficient is the largest, it indicates that the correlation degree between the driving behavior and the vehicle operation parameters is the largest. The order of time lag corresponding to the largest time difference correlation coefficient is the order of time lag when the correlation degree between the driving behavior and the vehicle operation parameters is the largest. If this order of time lag is greater than zero, it indicates that the change in the vehicle operation parameters in the case of the largest correlation degree is earlier than the change in the driving behavior. Further, the time series of the target vehicle operation parameters with the order of time lag corresponding to the largest time difference correlation coefficient greater than zero is the time series of the vehicle operation parameters in which the vehicle operation parameters are earlier than the change in the driving behavior when the correlation degree with the driving behavior is the largest. Constructing a vehicle operation parameter sample based on this time series of the target vehicle operation parameters can ensure that the vehicle operation parameters in the vehicle operation parameter sample are the vehicle operation parameters that cause changes in the vehicle driving behavior. These vehicle operation parameters are the vehicle operation parameters that are earlier than the change in the driving behavior, are the vehicle operation parameters that remove the vehicle operation parameters that lag behind the change in the driving behavior or change in unison with the driving behavior, and are the vehicle operation parameters that remove the interference of vehicle sensor data on the prediction result. These vehicle operation parameters have a great influence on the driving behavior. Based on these vehicle operation parameters, the Nth vehicle parameter sample is constructed, and based on the Nth vehicle parameter sample, a target driving behavior prediction model is obtained, and the prediction result of the target driving behavior prediction model has high accuracy. In addition, based on the time series of the target vehicle operation parameters obtained in this embodiment, it can be determined which vehicle operation parameters have a great influence on the prediction result of the driving behavior, and the interpretability of the driving behavior prediction model can be realized.

[0103] In an embodiment of the present application, in sub-step A15, constructing the Nth vehicle parameter sample according to the vehicle operation parameter sample may include sub-steps A151 to A152: Sub-step A151, obtaining the in-vehicle environment parameters and the out-of-vehicle environment parameters.

[0104] For example, the in-vehicle environment parameters may include: the driving duration of the driver, the number of people in the vehicle during driving, whether music is played during driving, whether there is communication and conversation among the people in the vehicle, etc. The out-of-vehicle environment parameters may include the weather, the time during driving, the road conditions of the road on which the vehicle is traveling, the congestion condition of the road on which the vehicle is traveling, etc. The road conditions of the road on which the vehicle is traveling may include whether the road on which the vehicle is traveling is an urban road, a rural road, a highway, or other roads.

[0105] Sub-step A152, constructing the Nth vehicle parameter sample according to the vehicle operation parameter sample, the in-vehicle environment parameters, and the out-of-vehicle environment parameters.

[0106] Exemplarily, the Nth vehicle parameter sample includes both the vehicle driving data collected by the vehicle-mounted sensor, the out-of-vehicle environment parameters obtained from the publicly available natural driving dataset, and the in-vehicle environment parameters.

[0107] In the related art, the training samples used for driving behavior prediction can be vehicle operation parameters such as vehicle speed. However, driving behavior is not only affected by vehicle operation parameters, but also by external environment parameters such as weather and traffic flow on the driving road, and is closely related to internal environment parameters such as the environment inside the vehicle where the driver is located. The related art ignores the influence of the external environment and the internal environment on driving behavior. For example, the related art conducts model training based on the data collected by in-vehicle sensors without considering the data information of the internal environment during driving. The related art can also perform driving behavior prediction based on the GCN model, and convert the vehicle position into an image format for driving behavior prediction. However, the driving scenario is variable. For example, in a multi-vehicle scenario, the changes in external environment parameters such as the relative speed between vehicles have a greater impact on driving behavior, but this method of the related art does not consider external environment factors, which will lead to inaccurate prediction results. In this embodiment, vehicle operation parameters, external environment parameters, and internal environment parameters are considered, and the Nth vehicle parameter sample obtained based on the method of this embodiment can improve the prediction accuracy of the target driving behavior prediction model.

[0108] In this embodiment, according to the vehicle operation parameter sample, the internal environment parameter, and the external environment parameter, the Nth vehicle parameter sample is constructed. The vehicle parameters in the Nth vehicle parameter sample not only include vehicle operation parameters that change earlier than driving behavior and have a great influence on the driving behavior prediction result, but also include internal environment parameters and external environment parameters. This Nth vehicle parameter sample can accurately reflect the influence of the internal and external environments on driving behavior. Compared with the method of constructing samples using vehicle operation parameters such as vehicle speed in the related art, the vehicle parameters included in the Nth vehicle parameter sample of this embodiment are more comprehensive, and the prediction result accuracy of the target driving behavior prediction model obtained based on this is higher.

[0109] In an embodiment of the present application, there are multiple second target vehicle parameter samples. Using the second target vehicle parameter samples to train the driving behavior prediction model in sub-step A3 to obtain the target driving behavior prediction model may include sub-steps A31 to A33: Sub-step A31: Train the driving behavior prediction model using each second target vehicle parameter sample respectively to obtain a driving behavior prediction model corresponding to each second target vehicle parameter sample respectively.

[0110] Sub-step A32: Obtain the determination coefficient of the driving behavior prediction model corresponding to each second target vehicle parameter sample respectively.

[0111] Among them, the determination coefficient is used to characterize the model goodness of fit of the driving behavior prediction model.

[0112] Sub-step A33: Determine the driving behavior prediction model corresponding to the preset determination coefficient as the target driving behavior prediction model.

[0113] Obtain the driving behavior prediction model corresponding to the preset determination coefficient from the driving behavior prediction models corresponding to the determination coefficients.

[0114] In this embodiment, the determination coefficients of the driving behavior prediction models corresponding to each second target vehicle parameter sample are obtained respectively; the determination coefficient is used to characterize the goodness of fit of the driving behavior prediction model; optimizing the model parameters of the driving behavior prediction model according to the determination coefficient can prevent the driving behavior prediction model from overfitting. Determining the driving behavior prediction model corresponding to the preset determination coefficient as the target driving behavior prediction model can avoid the overfitting phenomenon of the target driving behavior prediction model and ensure the usability of the target driving behavior prediction model and the accuracy of the prediction results.

[0115] In an embodiment of the present application, the vehicle parameters to be processed are of the same vehicle parameter category as the second target vehicle parameter samples. Thus, it can be ensured that the vehicle parameters to be processed input into the target driving behavior prediction model are of the same vehicle parameter category as the second target vehicle parameter samples used in the model training process, and the accuracy of the driving behavior prediction results can be ensured. In addition, since the vehicle parameter category of the second target vehicle parameter samples is obtained by screening multiple vehicle parameter categories according to the importance value and prediction accuracy, therefore, obtaining the vehicle parameters to be processed with the same vehicle parameter category as the second target vehicle parameter samples and performing driving behavior prediction based on the vehicle parameters to be processed requires fewer types and amounts of vehicle parameters to be collected and processed, which can ensure the accuracy of the prediction results and improve the processing efficiency.

[0116] Refer to Figure 3 , a driving behavior prediction method according to an embodiment of the present application, which is applied to the cloud, may include the following steps: Step 201: Receive the vehicle parameters to be processed sent by the vehicle.

[0117] Among them, the vehicle parameters to be processed sent by the vehicle may include vehicle operation parameters and in-vehicle environment parameters collected by vehicle sensors, or may also include vehicle operation parameters, in-vehicle environment parameters, and out-of-vehicle environment parameters.

[0118] Step 202: Input the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; among them, the target driving behavior prediction model is obtained through the following method: Delete the vehicle parameters corresponding to the target vehicle parameter category from the vehicle parameter samples to obtain the vehicle parameter samples, where the The vehicle parameter sample consists of vehicle parameters of multiple vehicle parameter categories. The target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the vehicle parameter sample; , is an integer greater than or equal to 2; Obtain the first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample respectively, and use the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; Obtain the second target vehicle parameter samples that are the same as the vehicle parameter categories of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and use the second target vehicle parameter samples to train the driving behavior prediction model to obtain the target driving behavior prediction model.

[0119] Wherein, if the to-be-processed parameters collected by the vehicle include vehicle operation parameters and in-vehicle environment parameters, the cloud can obtain the out-of-vehicle environment parameters, and then input the to-be-processed vehicle parameters including the vehicle operation parameters, in-vehicle environment parameters and out-of-vehicle environment parameters into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle.

[0120] In this embodiment, the vehicle obtains the to-be-processed vehicle parameters and sends them to the cloud, and then the cloud performs driving behavior prediction according to the target driving behavior prediction model. According to the foregoing analysis, the target driving behavior prediction model obtained in this embodiment has the characteristic of high prediction result accuracy, and the predicted driving behavior obtained by inputting the to-be-processed parameters of the vehicle into the target driving behavior prediction model has high accuracy. In addition, the computing power of the cloud is usually higher than that of the vehicle. By performing the above operations on the cloud to obtain the predicted driving behavior, it has the advantages of high processing efficiency and can save the computing power resources of the vehicle.

[0121] Referring to Figure 4 , a driving behavior prediction method according to an embodiment of the present application, applied to a vehicle, may include the following steps: Step 301, obtain the to-be-processed vehicle parameters.

[0122] The to-be-processed vehicle parameters may include vehicle operation parameters and in-vehicle environment parameters collected by vehicle sensors, or may include vehicle operation parameters, in-vehicle environment parameters and out-of-vehicle environment parameters.

[0123] Step 302, send the to-be-processed vehicle parameters to the cloud for the cloud to input the to-be-processed vehicle parameters into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: Delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; , where n is an integer greater than or equal to 2; Respectively obtain the first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample, and use the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; Obtain the second target vehicle parameter samples that are the same as the vehicle parameter categories of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and use the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0124] Wherein, if the to-be-processed parameters collected by the vehicle include vehicle operation parameters and in-vehicle environment parameters, the cloud can obtain the out-of-vehicle environment parameters, and then input the to-be-processed vehicle parameters composed of vehicle operation parameters, in-vehicle environment parameters and out-of-vehicle environment parameters into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle.

[0125] In this embodiment, the vehicle obtains the to-be-processed vehicle parameters and sends them to the cloud, and then the cloud performs driving behavior prediction according to the target driving behavior prediction model. According to the foregoing analysis, the target driving behavior prediction model obtained in this embodiment has the characteristic of high prediction result accuracy, and the predicted driving behavior obtained by inputting the to-be-processed parameters of the vehicle into the target driving behavior prediction model has high accuracy. In addition, the vehicle obtains the to-be-processed vehicle parameters and the cloud performs driving behavior prediction. Since the computing power of the cloud is usually higher than that of the vehicle, performing the above operations through the cloud to obtain the predicted driving behavior has the advantages of high processing efficiency and can save the computing power resources of the vehicle.

[0126] Refer to Figure 5 , a method for training a driving behavior prediction model according to an embodiment of the present application may include the following steps: Step 401, delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], where n is an integer greater than or equal to 2.

[0127] The importance value is used to characterize the importance of the vehicle parameter category to the driving behavior prediction result.

[0128] Step 402: respectively obtain first target vehicle parameter samples that have the same vehicle parameter categories as each vehicle parameter sample, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples.

[0129] Among them, the importance value of the deleted vehicle parameter category is less than the importance values of other vehicle parameter categories except the deleted vehicle parameter category; the importance value is used to represent the importance of the vehicle parameter category to the driving behavior prediction result.

[0130] Step 403: obtain second target vehicle parameter samples that have the same vehicle parameter categories as the first target vehicle parameter samples corresponding to a preset prediction accuracy, and use the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain a target driving behavior prediction model.

[0131] In this embodiment, the vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; , where n is an integer greater than or equal to 2, then N = 1, 2... n. Based on this embodiment, n loop operations are implemented. Among them, in the Nth loop, the vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. Thus, through n loop operations, n vehicle parameter samples with the vehicle parameters corresponding to the deleted target vehicle parameter category can be obtained. For these n vehicle parameter samples and the vehicle parameter sample before deleting the vehicle parameters in the first loop operation, respectively obtain first target vehicle parameter samples that have the same vehicle parameter categories as each vehicle parameter sample, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples, and obtain second target vehicle parameter samples that have the same vehicle parameter categories as the first target vehicle parameter samples corresponding to a preset prediction accuracy. The second target vehicle parameter samples are comprehensively determined according to the importance value of the vehicle parameter category and the corresponding prediction accuracy. Compared with the method of the related art, the target driving behavior prediction model obtained based on the second target vehicle parameter samples of this embodiment has a high prediction result accuracy, and when the to-be-processed parameters of the vehicle are input into the target driving behavior prediction model, the accuracy of the predicted driving behavior obtained is high.

[0132] With the development of vehicle networking technology and artificial intelligence, traditional driving behavior prediction has been difficult to meet the requirements of driving behavior prediction accuracy. To solve this problem, driving behavior prediction based on artificial intelligence has emerged. In related technologies, vehicle driving information can be collected based on in-vehicle sensors, and according to the driving information and the system's lane-changing trajectory prediction requirements, a random forest model can be trained and the vehicle's lane-changing trajectory can be predicted based on the trained model. However, in this method of related technologies, only the ranking results of the feature importance of the random forest itself are used to extract features, and a prediction model is trained based on the extracted features. This processing method may not necessarily select the optimal feature subset, and the accuracy of the driving behavior prediction model trained based on the extracted features is not high. The driving behavior prediction model training method of this embodiment can overcome the problem of low accuracy of the driving behavior prediction model in related technologies. To enable those skilled in the art to better understand the embodiments of the present invention, the following further illustrates the training method of the driving behavior prediction model of this application by way of further examples.

[0133] Referring to Figure 6 , the training method of the driving behavior prediction model of the embodiment of this application may include the following steps: Step S1, obtain body sensor data and driving behavior data.

[0134] Among them, the body sensor data is equivalent to the vehicle operation parameters in the foregoing embodiment, and the driving behavior data includes the in-vehicle environment data in the foregoing embodiment.

[0135] Step S2, perform time difference correlation analysis on the body sensor data and the driving behavior data to obtain a time difference correlation coefficient.

[0136] The method of this step has been described in the foregoing steps A11 to A15 and will not be elaborated here.

[0137] Step S3, according to the time difference correlation coefficient, obtain the body sensor data leading to the change in driving behavior. Then enter step S6.

[0138] The body sensor data leading to the change in driving behavior is the body sensor data earlier than the change in driving behavior. The body sensor data leading to the change in driving behavior is equivalent to the vehicle operation parameters in the time series of the target vehicle parameters in the foregoing embodiment.

[0139] Step S4, obtain in-vehicle environment parameters. Then enter step S6.

[0140] Step S5, obtain out-of-vehicle environment parameters.

[0141] Step S6: Construct a vehicle parameter sample based on vehicle body sensor data, in-vehicle environment data, and out-vehicle environment data that lead the driving behavior changes. Combine the analysis of the importance values of vehicle parameter categories and the verification of the prediction accuracy of predicting driving behaviors using the vehicle parameters corresponding to the vehicle parameter categories, and screen the vehicle parameter categories in the Nth vehicle parameter sample.

[0142] In this step, use the vehicle body sensor data that leads the driving behavior changes, the in-vehicle environment parameters, and the out-vehicle environment parameters obtained by screening to jointly construct the Nth vehicle parameter sample. Then, based on multiple loop operations, delete and update the vehicle parameter categories of the vehicle parameters according to the importance values corresponding to each vehicle parameter category, realizing the recursive feature elimination of the vehicle parameter categories.

[0143] Exemplarily, the driving behavior prediction model is a random forest model. Based on this step, the combination of the recursive feature elimination method, the random forest model processing method, and the cross-validation method is realized. Based on the combination of these methods, the optimal feature subset can be obtained, and the optimal feature subset is equivalent to the second target vehicle parameter sample in the foregoing embodiment.

[0144] Further, this step is implemented through loop operations. Among them, the input data set is obtained in each loop operation: , , where M is the total number of features, the feature is the vehicle parameter category, Xi is the i-th parameter feature; y is the predicted driving behavior category. Among them, the initial feature set is F={}. The initial feature set F={} is the Mth vehicle parameter sample input into the driving behavior prediction model in the first loop operation.

[0145] In each loop operation, sort the types of vehicle parameters according to the importance values to obtain the sorted feature set O{}. The method of this step is further described by way of example below: Refer to Figure 7 , in this step, the method for obtaining the prediction accuracy may include sub-steps B1 to B4: Sub-step B1: Obtain the feature set.

[0146] The feature set in this step is the Nth vehicle parameter sample in the foregoing embodiment.

[0147] Sub-step B2: Use the random forest model to obtain the sorting result of the importance values of each vehicle parameter category in the feature set. According to the sorting result of the importance values, delete the vehicle parameters of the vehicle parameter category with the smallest importance value to obtain the updated feature set.

[0148] This embodiment is only an exemplary illustration. In this embodiment, the vehicle parameter with the smallest importance value is deleted. It is also possible to delete the vehicle parameters of a preset number of vehicle parameter categories with relatively small importance values.

[0149] Sub-step B3: Determine whether the updated feature meets the preset loop termination condition. If so, end the loop; otherwise, return to step B1.

[0150] Sub-step B4: Respectively verify the prediction accuracy of using vehicle parameter samples that are the same as the vehicle parameter categories of each updated feature set for driving behavior prediction.

[0151] Exemplarily, step S6 may further include sub-steps C1 to C3: Sub-step C1: Use random forest to perform feature importance ranking on various vehicle parameters in the initial feature set F{}, and obtain the sorted feature set O{}.

[0152] Sub-step C2: Remove the feature with the smallest importance value from the sorted feature set O{} to obtain an updated feature set.

[0153] Remove the feature with the smallest importance value from the sorted feature set O{}, then obtain the importance value of the feature through the random forest model, and update the feature sorting result of the set O{} according to the importance value.

[0154] Sub-step C3: Repeat sub-step C2 until all features are traversed, and select the feature subset with the highest prediction accuracy.

[0155] A feature subset is a feature set that includes a part of the data in the feature set.

[0156] In this embodiment, by combining cross-validation, recursive feature elimination through loop operations, and the random forest model, a feature subset with the highest prediction accuracy is obtained.

[0157] Step S7: According to the selected vehicle parameter categories, obtain the second target vehicle parameter samples for training the driving behavior prediction model.

[0158] Step S8: Input the second target vehicle parameter samples into the driving behavior prediction model, adjust the parameters of the driving behavior prediction model, and obtain the target driving behavior prediction model.

[0159] The prediction model in this embodiment can be a random forest model. The random forest model has two model parameters that affect its prediction accuracy and generalization error. These two model parameters include the number of decision trees in the random forest and the number of features selected by the decision trees. Specifically, these two model parameters affect the classification accuracy of the decision trees and the correlation between the decision trees. Further, by reducing the correlation between the decision trees and increasing the classification accuracy of a single decision tree, the generalization error of the random forest can be reduced. For example, the larger the value of the number of features K selected by the decision tree, the greater the classification accuracy of a single decision tree and the greater the correlation between the decision trees.

[0160] In this embodiment, the grid search method can be used to optimize the model parameters of the driving behavior prediction model. Among them, grid search is an exhaustive search method used to train the driving behavior prediction model by traversing all the values of the vehicle parameter samples and obtain the optimal solution of the model parameters according to the model training effect.

[0161] Specifically, when searching for the optimal solution of the model parameters by the grid search method, the following method can be used to obtain the values of the vehicle parameters: obtain the value range and the preset step size of the vehicle parameter r, and according to the preset step size and the value range, obtain the values of the parameter samples to be traversed. For example, if the value range of the vehicle parameter r is [0, 10] and the preset step size is 5, then the values of the vehicle parameter r are 0, 5, and 10. Use the grid search method to solve the optimal model parameters corresponding to the vehicle parameter r. Further, respectively obtain the training effects of the models obtained by training the models with these three vehicle parameters of 0, 5, and 10, and then obtain the optimal solution by comparing the training effects of the models. This optimal solution is the model parameter of the target driving behavior prediction model.

[0162] Exemplarily, the training effects of the driving behavior prediction models can be compared by the coefficient of determination. The coefficient of determination is an index used to evaluate the regression goodness of fit of the driving behavior prediction model, and its value range is between 0 and 1. The calculation formula of the coefficient of determination is

[0163]

[0164] Among them, represents the coefficient of determination, is the i-th driving behavior observation value or actual value, is the total number of driving behaviors; is the driving behavior predicted value, is the mean value of driving behavior observation values. For example, driving behavior can be represented by preset feature values, and the mean value of driving behavior observation values can be the mean value of feature values corresponding to driving behavior prediction results. The numerator part in this calculation formula is the sum of squares of the differences between the observation values and the predicted values, that is, the residual sum of squares of the observation values and the predicted values, which is used to characterize the error degree of model prediction; the denominator part is the sum of squares of the differences between the observation values and the mean value, which is used to characterize the total fluctuation degree of the observation values. Among them, the higher the value of the coefficient of determination, the better the fitting effect of the driving behavior prediction model. When constructing the vehicle parameter sample, the influence factors of the in-vehicle environment on the driving behavior of the driver are considered. Specifically, in-vehicle environment information, out-vehicle environment information, and vehicle body sensor data are obtained, and then based on the random forest model, vehicle parameter categories are screened according to the importance degree, and the prediction accuracy of the screened vehicle parameter categories is cross-validated. Furthermore, the optimal feature subset (the second target vehicle parameter sample) is screened from the vehicle parameters. Training the driving behavior prediction model based on the optimal feature subset can improve the prediction result accuracy of the target driving behavior prediction model. Specifically, vehicle parameters corresponding to a preset number of vehicle parameter categories are deleted according to the importance degree values of the vehicle parameters, realizing recursive feature elimination. In the preprocessing process, through loop operations, features are repeatedly eliminated and models are established until all features are traversed. Based on the loop operations, features will be eliminated and feature subsets with different numbers will be generated. Finally, the optimal feature subset is selected according to the prediction effects of the models established based on different feature subsets.

[0165] In this embodiment, the importance rankings of all features in the initial feature set are obtained by using random forest importance evaluation, and then the feature with the smallest importance degree value is removed from the feature set to generate a new feature set. All features are traversed, and the foregoing operations are repeatedly executed in a loop. Finally, the feature subset with the highest classification accuracy is obtained. Thus, it can be ensured that in each round of the loop, the operation of re-evaluating the importance degree of the current feature set is performed. Based on this, the importance degree values of the features will be continuously adjusted. Based on this embodiment, the importance degree values of the feature subsets can be adjusted and evaluated dynamically and in a timely manner to ensure that the final optimal feature subset has better prediction performance and more stable data. At the same time, to ensure the stability of the processing results, the entire screening process of vehicle parameter categories is executed in the loop of cross-validation. Further, based on the screened optimal feature subset, grid search is used to train the model parameters to obtain a random forest prediction model for driving behavior prediction, and this model has the characteristic of high prediction result accuracy.

[0166] For example, the embodiment of the present application also provides a driving behavior prediction system. Refer to Figure 8 , the driving behavior prediction system includes a cloud end and a vehicle end. The following is an exemplary description of the method for the cloud end and the vehicle end to cooperate in processing to achieve driving behavior prediction: A vehicle for obtaining vehicle parameters to be processed and sending them to the cloud; The cloud for inputting the vehicle parameters to be processed into a target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: Delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; Respectively obtain the first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; Obtain the second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to the preset prediction accuracy, and use the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0167] Specifically, the cloud sends a data collection instruction to the vehicle terminal. In response to the received data collection instruction, the vehicle terminal collects vehicle operation parameters and in-vehicle environment data and returns them to the cloud. Among them, the data collection instruction is used to instruct the vehicle terminal to collect the optimal vehicle parameters, and the vehicle parameters may include body sensor data features and in-vehicle environment features.

[0168] The cloud receives the vehicle parameters collected by the vehicle terminal and inputs them into the target driving behavior prediction model to obtain the predicted driving behavior.

[0169] Through the interaction between the vehicle terminal and the cloud, the target driving behavior prediction model is obtained and the driving behavior is predicted. Based on this embodiment, the rich computing power resources of the cloud can be fully utilized. Specifically, the optimal feature subset is screened on the cloud, and the training of the driving behavior prediction model and the optimization of the model parameters are carried out. Because the cloud has rich computing power resources, the processing speed of training and constructing the driving behavior prediction model is faster.

[0170] The cloud issues a data acquisition instruction to the vehicle terminal according to the vehicle parameter categories in the optimal feature subset. According to the data acquisition instruction, the vehicle terminal sends the body sensor data and the in-vehicle environment information to the cloud. The cloud obtains the out-of-vehicle environment information according to the vehicle parameter categories in the optimal feature subset, assembles it with the body sensor data and the in-vehicle environment information, and inputs the assembled data to be processed into the target driving behavior prediction model. Because the cloud has rich computing power resources, it can realize fast and accurate prediction of the driving behavior.

[0171] Train a driving behavior prediction model in the cloud, analyze the vehicle parameters that need to be collected, and send a data collection instruction to the vehicle terminal. Finally, complete the driving behavior prediction, reducing the requirement of the vehicle terminal for computing resources.

[0172] It should be noted that this embodiment is only an exemplary illustration of the driving behavior prediction system. In this embodiment, the driving behavior prediction is completed through the cooperation of the cloud and the vehicle terminal. In some embodiments, the driving behavior prediction method in the foregoing embodiment can also be executed independently by the vehicle terminal.

[0173] As Figure 9 shown, the present invention discloses a driving behavior prediction device, which includes: a first acquisition module 501 for acquiring the vehicle parameters to be processed of the vehicle; a second acquisition module 502 for inputting the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively acquire the first target vehicle parameter samples with the same vehicle parameter categories as those of each vehicle parameter sample, and use the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; acquire the second target vehicle parameter samples with the same vehicle parameter categories as those of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and use the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0174] Optionally, using the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples includes: obtaining multiple sample combinations including a training set and a test set according to the first target vehicle parameter samples; training the driving behavior prediction model according to the sample combinations to obtain the first prediction accuracy corresponding to the sample combinations; obtaining the average value of the prediction accuracies of multiple first prediction accuracies; and obtaining the prediction accuracy corresponding to the first target vehicle parameter samples according to the average value of the prediction accuracies.

[0175] Optionally, there are multiple first target vehicle parameter samples; obtaining multiple sample combinations including a training set and a test set according to the first target vehicle parameter samples includes: traversing the first target vehicle parameter samples, and for each traversal operation, determining the traversed first target vehicle parameter sample as the test set and determining the other first target vehicle parameter samples as the training set to obtain the sample combination corresponding to the traversal operation.

[0176] Optionally, training a driving behavior prediction model based on a sample combination to obtain a first prediction accuracy corresponding to the sample combination, including: training a driving behavior prediction model based on the sample combination to obtain a first driving behavior prediction model, as well as the precision rate and recall rate of the first driving behavior prediction model; obtaining the first prediction accuracy corresponding to the sample combination according to the precision rate and the recall rate.

[0177] Optionally, in the case of N = 1, the Nth vehicle parameter sample is obtained by the following method: obtaining a driving behavior time series and at least one vehicle operation parameter time series; the vehicle parameter categories of the vehicle operation parameters in different vehicle operation parameter time series are different from each other; respectively obtaining the time difference correlation coefficients between the driving behavior time series and each vehicle operation parameter time series; according to the time difference correlation coefficients, obtaining at least one target vehicle operation parameter time series from at least one vehicle operation parameter time series; the time when the vehicle operation parameter in the target vehicle operation parameter time series changes is earlier than the time when the driving behavior in the driving behavior time series changes; obtaining a vehicle operation parameter sample with the same vehicle parameter category as that in the target vehicle operation parameter time series; Constructing the Nth vehicle parameter sample according to the vehicle operation parameter sample.

[0178] Optionally, constructing the Nth vehicle parameter sample according to the vehicle operation parameter sample, including: obtaining the in-vehicle environment parameters and the out-vehicle environment parameters; constructing the Nth vehicle parameter sample according to the vehicle operation parameter sample, the in-vehicle environment parameters, and the out-vehicle environment parameters.

[0179] Optionally, the importance degree value of the Nth target vehicle parameter category is less than the importance degree values of other vehicle parameter categories in the Nth vehicle parameter sample.

[0180] In this embodiment, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], n is an integer greater than or equal to 2, so N = 1, 2... n. Based on this embodiment, n loop operations are implemented. Among them, in the Nth loop, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. Thus, through n loop operations, n vehicle parameter samples with vehicle parameters corresponding to the deleted target vehicle parameter category can be obtained. For these n vehicle parameter samples and the vehicle parameter sample before vehicle parameter deletion in the first loop operation, first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample are respectively obtained, and the driving behavior prediction model is trained using the first target vehicle parameter samples to obtain the prediction accuracy corresponding to the first target vehicle parameter samples. Second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to the preset prediction accuracy are obtained. The second target vehicle parameter samples are comprehensively determined according to the importance value of the vehicle parameter categories and the corresponding prediction accuracy. Compared with the methods of related technologies, the target driving behavior prediction model obtained based on the second target vehicle parameter samples of this embodiment has high prediction result accuracy, and when the to-be-processed parameters of the vehicle are input into the target driving behavior prediction model, the accuracy of the predicted driving behavior is high.

[0181] Such as Figure 10As shown in the figure, the present invention discloses a driving behavior prediction device, which is applied to the cloud. The device includes: a third acquisition module 601, configured to receive vehicle parameters to be processed sent by a vehicle; a fourth acquisition module 602, configured to input the vehicle parameters to be processed into a target driving behavior prediction model to obtain a predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: deleting vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining a first target vehicle parameter sample with the same vehicle parameter category as each vehicle parameter sample, and using the first target vehicle parameter sample to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter sample; obtaining a second target vehicle parameter sample with the same vehicle parameter category as the first target vehicle parameter sample corresponding to a preset prediction accuracy, and using the second target vehicle parameter sample to train a preset driving behavior prediction model to obtain a target driving behavior prediction model.

[0182] In this embodiment, the vehicle acquires vehicle parameters to be processed and sends them to the cloud, and then the cloud performs driving behavior prediction according to the target driving behavior prediction model. According to the foregoing analysis, the target driving behavior prediction model obtained in this embodiment has the characteristic of high prediction result accuracy, and the accuracy of the predicted driving behavior obtained by inputting the vehicle parameters to be processed into the target driving behavior prediction model is high. In addition, the computing power of the cloud is usually higher than that of the vehicle. By performing the above operations on the cloud to obtain the predicted driving behavior, it has the advantages of high processing efficiency and can save the computing power resources of the vehicle.

[0183] As Figure 11 shown in the figure, the present invention discloses a driving behavior prediction device, which is applied to a vehicle. The device includes: The fifth acquisition module 701 is configured to acquire vehicle parameters to be processed; the sending module 702 is configured to send the vehicle parameters to be processed to the cloud, so that the cloud inputs the vehicle parameters to be processed into a target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, wherein the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples with the same vehicle parameter categories as each vehicle parameter sample, and using the first target vehicle parameter samples to train a driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; obtaining second target vehicle parameter samples with the same vehicle parameter categories as the first target vehicle parameter samples corresponding to the preset prediction accuracy, and using the second target vehicle parameter samples to train the preset driving behavior prediction model to obtain the target driving behavior prediction model.

[0184] In this embodiment, the vehicle acquires the vehicle parameters to be processed and sends them to the cloud, and then the cloud performs driving behavior prediction according to the target driving behavior prediction model. According to the foregoing analysis, the target driving behavior prediction model obtained in this embodiment has the characteristic of high prediction result accuracy, and the accuracy of the predicted driving behavior obtained by inputting the vehicle parameters to be processed into the target driving behavior prediction model is high. In addition, the vehicle acquires the vehicle parameters to be processed, and the cloud performs driving behavior prediction. Since the computing power of the cloud is usually higher than that of the vehicle, performing the above operations through the cloud to obtain the predicted driving behavior has the advantages of high processing efficiency and can save the computing power resources of the vehicle.

[0185] Such as Figure 12As shown in the figure, the present invention discloses a driving behavior prediction model training device. The driving behavior prediction device includes: a sixth acquisition module 801, configured to delete vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample, where the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; a seventh acquisition module 802, configured to respectively acquire a first target vehicle parameter sample that is the same as the vehicle parameter category of each vehicle parameter sample, and use the first target vehicle parameter sample to train the driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter sample; an eighth acquisition module 803, configured to acquire a second target vehicle parameter sample that is the same as the vehicle parameter category of the first target vehicle parameter sample corresponding to the preset prediction accuracy, and use the second target vehicle parameter sample to train the preset driving behavior prediction model to obtain a target driving behavior prediction model.

[0186] In this embodiment, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. The Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2, then N = 1, 2... n. Based on this embodiment, n loop operations are realized. Among them, in the Nth loop, vehicle parameters corresponding to the Nth target vehicle parameter category are deleted from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample. Thus, through n loop operations, n vehicle parameter samples with vehicle parameters corresponding to the deleted target vehicle parameter category can be obtained. For these n vehicle parameter samples, as well as the vehicle parameter sample before vehicle parameter deletion in the first loop operation, a first target vehicle parameter sample that is the same as the vehicle parameter category of each vehicle parameter sample is respectively acquired, and the driving behavior prediction model is trained using the first target vehicle parameter sample to obtain a prediction accuracy corresponding to the first target vehicle parameter sample, and a second target vehicle parameter sample that is the same as the vehicle parameter category of the first target vehicle parameter sample corresponding to the preset prediction accuracy is acquired. The second target vehicle parameter sample is comprehensively determined according to the importance value of the vehicle parameter category and the corresponding prediction accuracy. Compared with the method of the related art, the target driving behavior prediction model obtained based on the second target vehicle parameter sample of this embodiment has a high prediction result accuracy, and when the to-be-processed parameters of the vehicle are input into the target driving behavior prediction model, the accuracy of the predicted driving behavior obtained is high.

[0187] For the apparatus embodiments, since they are basically similar to the method embodiments, they are described relatively simply. For related parts, please refer to the descriptions in the method embodiments. An embodiment of the present invention further provides an electronic device, such as Figure 13 shown, including a processor 901, a device interface 902, a memory 903 and a bus 904; The memory 903 is used to store computer programs; When the processor 901 executes the programs stored on the memory 903, it implements the steps of any of the foregoing embodiments.

[0188] The bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience in representation, only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0189] The memory may include a Random Access Memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0190] The aforementioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0191] The present invention also provides a storage medium. When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can execute the driving behavior prediction method and the driving behavior prediction model training method of the foregoing embodiments.

[0192] The present application also provides a vehicle, including the electronic device or apparatus in the foregoing embodiments.

[0193] For the apparatus embodiments, since they are substantially similar to the method embodiments, the description is relatively simple. For the relevant parts, reference may be made to the partial description of the method embodiments.

[0194] The algorithms and displays provided herein are not inherently related to any particular computer, virtual apparatus, or other device. From the above description, the structures required to construct such an apparatus are obvious. In addition, the present invention is not directed to any particular programming language. It should be understood that the content of the present invention described herein can be implemented using various programming languages, and the description of a particular language above is for disclosing the best mode of the present invention.

[0195] In the specification provided herein, a large number of specific details are set forth. However, it can be understood that the embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures, and technologies are not shown in detail so as not to obscure the understanding of this specification.

[0196] Similarly, it should be understood that, in order to streamline the present invention and assist in understanding one or more of the various inventive aspects, in the foregoing description of the exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, the inventive aspects lie in less than all the features of the preceding disclosed single embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of the present invention.

[0197] Those skilled in the art can understand that the modules in the devices in the embodiments can be adaptively changed and disposed in one or more devices different from the embodiments. The modules or units or component samples in the embodiments can be combined into a module or unit or component, and in addition, they can be divided into multiple sub-modules or sub-units or sub-components. Except that at least some of such features and / or processes or units are mutually exclusive, any sample combination can be used to combine all the features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all the processes or units of any method or device so disclosed. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) can be replaced by an alternative feature that provides the same, equivalent, or similar purpose.

[0198] Each component embodiment of the present invention can be implemented in hardware, or in software modules running on one or more processors, or in a combination of their samples. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the sorting device according to the present invention. The present invention can also be implemented as a device or apparatus program for executing part or all of the methods described herein. Such a program for implementing the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0199] It should be noted that the above embodiments illustrate the present invention rather than limit the present invention, and those skilled in the art can design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim listing several devices, several of these devices can be embodied by the same item of hardware. The use of the words first, second, and third, etc. does not denote any order. These words can be interpreted as names.

[0200] Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, the specific working processes of the above-described devices, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0201] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0202] The above are only the specific implementation manners of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, and all of them should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

[0203] It should be noted that in the embodiments of the present application, all processes related to obtaining various data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where it is located and obtaining the authorization given by the owner of the corresponding device.

Claims

1. A driving behavior prediction method, characterized in that: Including: Obtaining the vehicle parameters to be processed of the vehicle; Inputting the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: Deleting the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample; wherein, the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; Respectively obtaining the first target vehicle parameter samples with the same vehicle parameter categories as those of each vehicle parameter sample, and training the driving behavior prediction model with the first target vehicle parameter samples to obtain the prediction accuracy corresponding to the first target vehicle parameter samples; Obtaining the second target vehicle parameter samples with the same vehicle parameter categories as those of the first target vehicle parameter samples corresponding to the preset prediction accuracy, and training the driving behavior prediction model with the second target vehicle parameter samples to obtain the target driving behavior prediction model.

2. The method according to claim 1, wherein The training the driving behavior prediction model with the first target vehicle parameter samples to obtain the prediction accuracy corresponding to the first target vehicle parameter samples includes: According to the first target vehicle parameter samples, obtaining multiple groups of sample combinations including training sets and test sets; Training the driving behavior prediction model according to the sample combinations to obtain the first prediction accuracy corresponding to the sample combinations; Obtaining the average value of the prediction accuracies of multiple first prediction accuracies; According to the average value of the prediction accuracies, obtaining the prediction accuracy corresponding to the first target vehicle parameter samples.

3. The method according to claim 2, characterized in that, There are multiple first target vehicle parameter samples; the obtaining multiple groups of sample combinations including training sets and test sets according to the first target vehicle parameter samples includes: Traversing the first target vehicle parameter samples, and for each traversal operation, determining the traversed first target vehicle parameter sample as the test set and the other first target vehicle parameter samples as the training set to obtain the sample combination corresponding to the traversal operation.

4. The method according to claim 2, wherein The training the driving behavior prediction model according to the sample combinations to obtain the first prediction accuracy corresponding to the sample combinations includes: Training the driving behavior prediction model according to the sample combinations to obtain the first driving behavior prediction model, as well as the precision rate and recall rate of the first driving behavior prediction model; According to the precision rate and the recall rate, obtaining the first prediction accuracy corresponding to the sample combinations.

5. The method according to claim 1, characterized in that, In the case of N = 1, the Nth vehicle parameter sample is obtained by the following method: Obtaining the driving behavior time series and at least one vehicle operation parameter time series; the vehicle parameter categories of the vehicle operation parameters in different vehicle operation parameter time series are different from each other; Respectively obtaining the time difference correlation coefficients between the driving behavior time series and each vehicle operation parameter time series; Based on the time difference correlation coefficient, at least one target vehicle operation parameter time series is obtained from at least one vehicle operation parameter time series; the time when the vehicle operation parameter in the target vehicle operation parameter time series changes is earlier than the time when the driving behavior in the driving behavior time series changes. Obtain vehicle operation parameter samples with the same vehicle parameter category as the vehicle parameters in the target vehicle operation parameter time series. Construct the Nth vehicle parameter sample according to the vehicle operation parameter samples.

6. The method according to claim 5, characterized in that, The constructing the Nth vehicle parameter sample according to the vehicle operation parameter samples includes: Obtain in-vehicle environment parameters and out-of-vehicle environment parameters. Construct the Nth vehicle parameter sample according to the vehicle operation parameter samples, the in-vehicle environment parameters, and the out-of-vehicle environment parameters.

7. The method according to claim 1, wherein The importance degree value of the Nth target vehicle parameter category is less than the importance degree values of other vehicle parameter categories in the Nth vehicle parameter sample except the Nth target vehicle parameter category.

8. A driving behavior prediction method, characterized in that, When applied to the cloud, the method includes: Receive the vehicle parameters to be processed sent by the vehicle. Input the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: Delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample; wherein, the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance degree value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2. Respectively obtain the first target vehicle parameter samples with the same vehicle parameter category as each vehicle parameter sample, and use the first target vehicle parameter samples to train the driving behavior prediction model to obtain the prediction accuracy corresponding to the first target vehicle parameter samples. Obtain the second target vehicle parameter samples with the same vehicle parameter category as the first target vehicle parameter samples corresponding to the preset prediction accuracy, and use the second target vehicle parameter samples to train the driving behavior prediction model to obtain the target driving behavior prediction model.

9. A driving behavior prediction method, characterized in that, When applied to the vehicle, the method includes: Obtain the vehicle parameters to be processed. Send the vehicle parameters to be processed to the cloud for the cloud to input the vehicle parameters to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: Delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample; wherein, the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance degree value of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2. Obtain first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample respectively, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples; Obtain second target vehicle parameter samples that are the same as the vehicle parameter categories of the first target vehicle parameter samples corresponding to a preset prediction accuracy, and use the second target vehicle parameter samples to train a driving behavior prediction model to obtain a target driving behavior prediction model.

10. A driving behavior prediction model training method, characterized in that: Including: Delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample; wherein, the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance degree values of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; Obtain first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample respectively, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples; Obtain second target vehicle parameter samples that are the same as the vehicle parameter categories of the first target vehicle parameter samples corresponding to a preset prediction accuracy, and use the second target vehicle parameter samples to train a driving behavior prediction model to obtain a target driving behavior prediction model.

11. A driving behavior prediction device, characterized in that, Including: A first acquisition module, configured to acquire the vehicle parameter to be processed of the vehicle; A second acquisition module, configured to input the vehicle parameter to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: Delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the (N + 1)th vehicle parameter sample; wherein, the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined according to the importance degree values of each vehicle parameter category in the Nth vehicle parameter sample; N ∈ [1, n], and n is an integer greater than or equal to 2; Obtain first target vehicle parameter samples that are the same as the vehicle parameter categories of each vehicle parameter sample respectively, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples; Obtain second target vehicle parameter samples that are the same as the vehicle parameter categories of the first target vehicle parameter samples corresponding to a preset prediction accuracy, and use the second target vehicle parameter samples to train a driving behavior prediction model to obtain a target driving behavior prediction model.

12. A driving behavior prediction device, characterized in that, Applied to the cloud, the device includes: A third acquisition module, configured to receive the vehicle parameter to be processed sent by the vehicle; A fourth acquisition module, configured to input the vehicle parameter to be processed into the target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein, the target driving behavior prediction model is obtained by the following method: Deleting vehicle parameters corresponding to an Nth target vehicle parameter category from an Nth vehicle parameter sample to obtain an N+1th vehicle parameter sample; wherein the Nth vehicle parameter sample is composed of vehicle parameters of a plurality of vehicle parameter categories, and the Nth target vehicle parameter category is determined based on the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N∈[1,n], where n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples of the same vehicle parameter category as each vehicle parameter sample, and using the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples; A second target vehicle parameter sample of the same vehicle parameter category as the first target vehicle parameter sample corresponding to a preset prediction accuracy is obtained, and the driving behavior prediction model is trained using the second target vehicle parameter sample to obtain a target driving behavior prediction model.

13. A driving behavior prediction device, characterized in that, Applied to a vehicle, the device comprises: A fifth acquisition module, used to acquire parameters of the vehicle to be processed; A sending module is used to send the vehicle parameters to be processed to the cloud, so that the cloud can input the vehicle parameters to be processed into a target driving behavior prediction model to obtain the predicted driving behavior of the vehicle; wherein the target driving behavior prediction model is obtained by the following method: Deleting vehicle parameters corresponding to an Nth target vehicle parameter category from an Nth vehicle parameter sample to obtain an N+1th vehicle parameter sample; wherein the Nth vehicle parameter sample is composed of vehicle parameters of a plurality of vehicle parameter categories, and the Nth target vehicle parameter category is determined based on the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N∈[1,n], where n is an integer greater than or equal to 2; respectively obtaining first target vehicle parameter samples of the same vehicle parameter category as each vehicle parameter sample, and using the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples; A second target vehicle parameter sample of the same vehicle parameter category as the first target vehicle parameter sample corresponding to a preset prediction accuracy is obtained, and the driving behavior prediction model is trained using the second target vehicle parameter sample to obtain a target driving behavior prediction model.

14. A driving behavior prediction model training device, characterized in that, include: a sixth acquisition module, configured to delete the vehicle parameters corresponding to the Nth target vehicle parameter category from the Nth vehicle parameter sample to obtain the N+1th vehicle parameter sample; wherein the Nth vehicle parameter sample is composed of vehicle parameters of multiple vehicle parameter categories, and the Nth target vehicle parameter category is determined based on the importance value of each vehicle parameter category in the Nth vehicle parameter sample; N∈[1,n], where n is an integer greater than or equal to 2; a seventh acquisition module, configured to respectively acquire first target vehicle parameter samples of the same vehicle parameter category as each vehicle parameter sample, and use the first target vehicle parameter samples to train a driving behavior prediction model to obtain a prediction accuracy corresponding to the first target vehicle parameter samples; An eighth acquisition module, configured to acquire a second target vehicle parameter sample that has the same vehicle parameter category as a first target vehicle parameter sample corresponding to a preset prediction accuracy, and use the second target vehicle parameter sample to train a driving behavior prediction model to obtain a target driving behavior prediction model.

15. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to execute the instructions to implement the method according to any one of claims 1 to 10.

16. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by a processor of a mobile terminal, the mobile terminal is enabled to execute the method according to any one of claims 1 to 10.

Images