Model training method, driving data processing method, device, medium and equipment

The feature extraction model processes the driving data of the autonomous driving vehicle, judges and discards the redundant data based on similarity, solving the problem of low data storage efficiency of autonomous driving vehicle and realizing the rational utilization of storage resources.

CN112541515BActive Publication Date: 2024-12-03BEIJING JINGDONG QIANSHITECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201910900628.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2019-09-23
Publication Date
2024-12-03
Estimated Expiration
2039-09-23

AI Technical Summary

Technical Problem

When autonomous driving vehicles process and store a large amount of driving data, there is a problem of excessive redundant data and failure to rationally utilize storage resources.

Method used

By acquiring multiple sets of sample data based on driving scenarios, using the feature extraction model to calculate the similarity of the sample data, and using the similarity to train the feature extraction model, and then processing the current driving data, determining the data with high similarity is discarded, Save storage space.

Benefits of technology

It reduces the storage of redundant data, improves the utilization rate of storage resources, and expands the application scenarios of similarity calculation methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112541515B_ABST
    Figure CN112541515B_ABST
Patent Text Reader

Abstract

The present invention discloses a model training method, a driving data processing method, a model training device, a driving data processing device, a computer-readable storage medium and an electronic device, and relates to the field of intelligent driving. The model training method comprises: obtaining multiple groups of first sample data, second sample data and target sample data based on driving scenes; inputting the first sample data, the second sample data and the target sample data into a feature extraction model respectively, determining a first feature vector corresponding to the first sample data, a second feature vector corresponding to the second sample data, and a target feature vector corresponding to the target sample data; calculating the similarity between the first feature vector and the target feature vector as the first similarity, and calculating the similarity between the second feature vector and the target feature vector as the second similarity; and training the feature extraction model using the first similarity and the second similarity. The present invention can remove redundant driving data and promote the rational use of storage resources.
Need to check novelty before this filing date? Find Prior Art

Description

Background Art

[0002] With the rapid development of the Internet of Things and artificial intelligence, the field of automotive technology has also entered a stage of rapid development. For example, in the smart car industry, as people's pursuit of quality of life improves, autonomous driving vehicles have also become a research hotspot in the industry.

[0003] An autonomous vehicle is a smart car that is unmanned through a computer system. Autonomous driving uses various sensors to sense the vehicle's surroundings and integrates data from various sensors, such as road, vehicle location, and obstacle information, to control the vehicle's steering and speed, so that the vehicle can travel safely and reliably on the road.

[0004] However, the driving scenarios in various environments are very complex, and the above-mentioned computer system not only needs to receive a large amount of sensor response data, but also needs to store a large amount of vehicle driving data. Therefore, how to process a large amount of data has become an important issue.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of the present disclosure is to provide a model training method, a driving data processing method, a model training device, a driving data processing device, a computer-readable storage medium and an electronic device, thereby at least to a certain extent overcoming the problems of excessive redundant data and unreasonable use of storage resources due to the limitations and defects of related technologies.

[0007] According to a first aspect of the present disclosure, a model training method is provided, comprising: obtaining multiple groups of sample data based on driving scenarios, each group of sample data comprising first sample data, second sample data, and target sample data; wherein the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data; inputting the first sample data, the second sample data, and the target sample data into a feature extraction model, respectively, to determine a first feature vector corresponding to the first sample data, a second feature vector corresponding to the second sample data, and a target feature vector corresponding to the target sample data; calculating the similarity between the first feature vector and the target feature vector as the first similarity, and calculating the similarity between the second feature vector and the target feature vector as the second similarity; and training the feature extraction model using the first similarity and the second similarity.

[0008] According to a second aspect of the present disclosure, a driving data processing method is provided, comprising: obtaining current driving data, processing the current driving data using a trained feature extraction model, and determining a current feature vector corresponding to the current driving data; wherein the trained feature extraction model is trained by the above-mentioned model training method; obtaining each historical feature vector corresponding to each historical driving data, and respectively calculating the similarity between each historical feature vector and the current feature vector; if there is a target feature vector in each historical feature vector whose similarity with the current feature vector is greater than or equal to a preset threshold, determining target driving data corresponding to the target feature vector; deleting the target driving data from a database and storing the current driving data in the database, or discarding the current driving data.

[0009] According to a third aspect of the present disclosure, a driving data processing device is provided, comprising: a feature determination module, used to obtain current driving data, process the current driving data using a trained feature extraction model, and determine a current feature vector corresponding to the current driving data; wherein the trained feature extraction model is trained by the above-mentioned model training method; a similarity calculation module, used to obtain each historical feature vector corresponding to each historical driving data, and respectively calculate the similarity between each historical feature vector and the current feature vector; a similarity judgment module, used to determine target driving data corresponding to the target feature vector if there is a target feature vector in each historical feature vector whose similarity with the current feature vector is greater than or equal to a preset threshold, and used to store the current driving data in a database if the similarity between each historical feature vector and the current feature vector is less than the preset threshold; a data processing module, used to delete the target driving data from the database and store the current driving data in the database, or discard the current driving data.

[0010] According to a fourth aspect of the present disclosure, a model training device is provided, including: a sample acquisition module, used to acquire multiple groups of sample data based on driving scenarios, each group of sample data includes first sample data, second sample data, and target sample data; wherein the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data; a feature extraction module, used to input the first sample data, the second sample data, and the target sample data into a feature extraction model, respectively, to determine a first feature vector corresponding to the first sample data, a second feature vector corresponding to the second sample data, and a target feature vector corresponding to the target sample data; a similarity calculation module, used to calculate the similarity between the first feature vector and the target feature vector as the first similarity, and calculate the similarity between the second feature vector and the target feature vector as the second similarity; and a model training module, used to train the feature extraction model using the first similarity and the second similarity.

[0011] Optionally, the feature extraction module includes a first feature extraction submodule and a second feature extraction submodule; the first feature extraction submodule is used to input the first sample data, the second sample data and the target sample data into the first feature extraction unit respectively, and determine the first feature matrix corresponding to the first sample data, the second feature matrix corresponding to the second sample data, and the target feature matrix corresponding to the target sample data; the second feature extraction submodule is used to input the first feature matrix, the second feature matrix and the target feature matrix into the second feature extraction unit respectively, and determine the first feature vector corresponding to the first feature matrix, the second feature vector corresponding to the second feature matrix, and the target feature vector corresponding to the target feature matrix.

[0012] Optionally, the first feature extraction submodule includes: a data embedding unit, used to perform embedding operations on the first sample data, the second sample data and the target sample data, respectively, to determine a first intermediate matrix corresponding to the first sample data, a second intermediate matrix corresponding to the second sample data, and a target intermediate matrix corresponding to the target sample data; a feature determination unit, used to determine a first feature matrix by combining the first intermediate matrix with a specified parameter matrix, determine a second feature matrix by combining the second intermediate matrix with a specified parameter matrix, and determine a target feature matrix by combining the target intermediate matrix with a specified parameter matrix.

[0013] Optionally, the model training module includes: a loss function determination unit, used to determine a loss function based on the first similarity and the second similarity; and a model training unit, used to train a feature extraction model using the loss function.

[0014] According to a fifth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the model training method or the driving data processing method as described above is implemented.

[0015] According to the sixth aspect of the present disclosure, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, which, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the model training method as described above or the driving data processing method as described above.

[0016] The exemplary embodiments of the present disclosure have the following beneficial effects:

[0017] In the technical solutions provided by some embodiments of the present disclosure, after obtaining driving data, a feature extraction model is used to process the driving data to obtain a feature vector corresponding to the driving data. In the training process of the feature extraction model, first, multiple groups of sample data based on driving scenarios are obtained, each group of sample data includes first sample data, second sample data, and target sample data; then, the first sample data, the second sample data, and the target sample data are respectively input into the feature extraction model to determine the first feature vector corresponding to the first sample data, the second feature vector corresponding to the second sample data, and the target feature vector corresponding to the target sample data; next, the similarity between the first feature vector and the target feature vector is calculated as the first similarity, and the similarity between the second feature vector and the target feature vector is calculated as the second similarity; then, the feature extraction model is trained using the first similarity and the second similarity. On the one hand, the similarity between the acquired driving data to be processed and the target driving data is determined by the feature extraction model, and according to the determined similarity, the driving data with high similarity to the target driving data is discarded, thereby reducing redundant driving data and saving storage space. Moreover, the saved storage space can be used to store driving data with low similarity, that is, the method also promotes the rational use of storage resources. On the other hand, the similarity between driving data obtained in different scenarios and target driving data can be determined through the feature extraction model, that is, the application scenarios of the similarity calculation method are increased, and the utilization rate of the similarity calculation method is improved.

[0018] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0020] Figure 1 A flowchart of a model training method according to an exemplary embodiment of the present disclosure is schematically shown;

[0021] Figure 2 A flowchart schematically shows a driving data processing method according to an exemplary embodiment of the present disclosure;

[0022] Figure 3 A block diagram schematically shows a driving data processing device according to an exemplary embodiment of the present disclosure;

[0023] Figure 4 A block diagram of a model training device according to an exemplary embodiment of the present disclosure is schematically shown;

[0024] Figure 5 Schematically shows a block diagram of a feature extraction module according to an exemplary embodiment of the present disclosure;

[0025] Figure 6 Schematically shows a block diagram of a first feature extraction module according to an exemplary embodiment of the present disclosure;

[0026] Figure 7 Schematically shows a block diagram of a model training module according to an exemplary embodiment of the present disclosure;

[0027] Figure 8 A block diagram of an electronic device according to an exemplary embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0028] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0029] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0030] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the steps. For example, some steps can be decomposed, and some steps can be combined or partially combined, so the actual execution order may change according to the actual situation. In addition, the terms "first" and "second" used in the present disclosure are only for the purpose of distinction and should not be used as limitations of the present disclosure.

[0031] With the advent of the intelligent era, the automotive industry has introduced smart cars to improve people's quality of life. Smart cars are also called self-driving cars. The self-driving cars judge the steering, speed, etc. of the car by integrating the driving data collected by various sensors. Therefore, this process requires the storage and processing of a large amount of driving data. In view of the above situation, in order to improve the utilization rate of storage space, redundant data with high similarity in the driving data can be discarded. However, the existing similarity calculation method is artificially designed and cannot process a large amount of driving data. To this end, the present disclosure proposes a model training method and a driving data processing method to solve the above problems.

[0032] It should be noted that in the exemplary embodiments of the present disclosure, the model training method described below can generally be implemented by a server, that is, each step of the model training method can be performed by the server, in which case the model training device can be configured in the server. However, the model training method can also be implemented by a terminal device (such as a mobile phone, a tablet, a personal computer, etc.), and the present disclosure does not impose any special restrictions on this.

[0033] In addition, the driving data processing method described below can usually be implemented by a vehicle-mounted terminal device so that the driving data obtained by the vehicle-mounted terminal device can be processed in real time. The vehicle-mounted terminal device can be a vehicle-mounted central control system, a vehicle-mounted computer system, etc. In this case, the driving data processing device can be configured in the vehicle-mounted terminal device. However, in scenarios where real-time requirements are not high, the driving data processing method can also be implemented by a server, and the present disclosure does not impose any special restrictions on this.

[0034] Figure 1 The flowchart of the model training method according to the exemplary embodiment of the present disclosure is schematically shown. Figure 1 , the model training method may include the following steps:

[0035] S102. Acquire multiple groups of sample data based on driving scenarios, each group of sample data includes first sample data, second sample data, and target sample data; wherein the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data.

[0036] In an exemplary embodiment of the present disclosure, the sample data based on the driving scenario may be sensor data acquired during the data analysis process of driving the vehicle. For example, the information of the sample data may include information analyzed by vehicle positioning data obtained by a lidar, information analyzed by image information of the vehicle's surroundings captured by a camera, information analyzed by computational data of processes such as sensing the vehicle's surroundings and planning the vehicle's next operation, and so on.

[0037] The server may obtain multiple sets of driving data in actual driving scenarios according to manually marked driving behaviors, or may obtain multiple sets of driving data from experimental driving scenarios. The experimental driving scenarios may be specially conducted for training the feature extraction model.

[0038] For example, in an experimental driving scenario, the first sample data is driving data of a vehicle driving in a straight line at a speed of 3000 rpm, the second sample data is driving data of a vehicle turning right, and the target sample data is driving data of a vehicle driving in a straight line at a speed of 700 rpm.

[0039] The first sample data and the target sample data are both driving data of the vehicle driving in a straight line, and the similarity between the two is high; the second sample data and the target sample data are completely different driving data, and the similarity between the two is low. Therefore, the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data.

[0040] S104. Input the first sample data, the second sample data and the target sample data into the feature extraction model respectively, and determine the first feature vector corresponding to the first sample data, the second feature vector corresponding to the second sample data, and the target feature vector corresponding to the target sample data.

[0041] The feature extraction model may include a first feature extraction unit and a second feature extraction unit. First, the server inputs the first sample data, the second sample data, and the target sample data into the first feature extraction unit, respectively, and determines a first feature matrix corresponding to the first sample data, a second feature matrix corresponding to the second sample data, and a target feature matrix corresponding to the target sample data; then, the first feature matrix, the second feature matrix, and the target feature matrix are input into the second feature extraction unit, respectively, and determines a first feature vector corresponding to the first feature matrix, a second feature vector corresponding to the second feature matrix, and a target feature vector corresponding to the target feature matrix.

[0042] The first feature extraction unit may be a recurrent neural network model, and the second feature extraction unit may be a fully connected neural network. The input of the second feature extraction unit may be a matrix, and the output may be a vector of a fixed length, and the fixed length may be a dynamic length adjusted according to actual conditions. The fully connected neural network may include a three-layer neural network, for example, an input layer, an activation layer, and an output layer.

[0043] For example: It is known that the input is a driving data in, It can be expressed as i The driving data measured by each of the m sensors at a time. In other words, the input driving data can be represented as a T×m matrix.

[0044] First, the data X i Input to the first feature extraction unit to calculate the feature matrix W i , see formula 1:

[0045] W i =f(X i ) Formula (1)

[0046] Among them, W i ∈R d×r , f(·) represents the formula for calculating the feature matrix by the first feature extraction unit.

[0047] Then, the obtained feature matrix W i Input to the second feature extraction unit, and calculate the feature vector H according to formula 2 i :

[0048] H i =g(W i ) Formula (2)

[0049] Among them, H i ∈R k×1 , g(·) represents the formula for calculating the feature vector by the second feature extraction unit.

[0050] In addition, in an exemplary embodiment of the present disclosure, the server may also perform embedding operations on the first sample data, the second sample data, and the target sample data, respectively, to determine a first intermediate matrix corresponding to the first sample data, a second intermediate matrix corresponding to the second sample data, and a target intermediate matrix corresponding to the target sample data; the first intermediate matrix is ​​combined with a specified parameter matrix to determine a first characteristic matrix, the second intermediate matrix is ​​combined with a specified parameter matrix to determine a second characteristic matrix, and the target intermediate matrix is ​​combined with a specified parameter matrix to determine a target characteristic matrix.

[0051] The number of specified parameter matrices may be a dynamic value that can be adjusted according to model calculations. For example, the server may introduce two specified parameter matrices into the model.

[0052] For example, if the input is a driving data

[0053] First, the driving data is embedded using the bidirectional long short-term memory network (LSTM) technology to obtain the intermediate matrix B = [b 1 ,b 2 ,…,b t ,…,b T ] T , then, introduce two parameter matrices and Next, the feature matrix W is obtained according to Formula 3 i :

[0054]

[0055] Among them, W i ∈R d×r , the element in the pth row and qth column is represented by W i (p,q).

[0056] Then, the obtained feature matrix W i Input the second feature extraction unit and map it according to formula 4 to obtain the feature vector H i For each element in H i =[h i,1 ,h i,2 ,…,h i,j ,…,h i,k ]∈R k×1 Therefore, H i The calculation of each element in is shown in Formula 4:

[0057] h i,j =reLU(∑ p ∑ q a p,q,j w(p,q)+b j ) Formula (4)

[0058] Among them, a p,q,j , p∈[1,d], q∈[1,r], b j ,j∈[1,k] are the coefficients of Formula 4, and these coefficients are dynamic values ​​that can be adjusted according to model calculations.

[0059] S106. Calculate the similarity between the first feature vector and the target feature vector as the first similarity, and calculate the similarity between the second feature vector and the target feature vector as the second similarity.

[0060] In an exemplary embodiment of the present disclosure, the server can calculate the similarity between the first feature vector and the target feature vector, and can use the Euclidean distance between the first feature vector and the target feature vector to determine the similarity between the two, that is, the reciprocal of the Euclidean distance between the first feature vector and the target feature vector can be used as the similarity between the two. The server can calculate the similarity between the second feature vector and the target feature vector, and can use the Euclidean distance between the second feature vector and the target feature vector to characterize the similarity between the two. It should be noted that in an exemplary embodiment of the present disclosure, the Euclidean distance can be used, but is not limited to this, and can include but is not limited to the following methods: Manhattan distance, Chebyshev distance, angle cosine distance, chi-square distance, etc.

[0061] If the Euclidean distance between the first feature vector and the target feature vector is smaller than the Euclidean distance between the second feature vector and the target feature vector, it can be represented that the first similarity is higher than the second similarity; if the Euclidean distance between the first feature vector and the target feature vector is larger than the Euclidean distance between the second feature vector and the target feature vector, it can be represented that the first similarity is lower than the second similarity.

[0062] For example, given a set of driving data, including left-turn driving data X i , right-angle left turn driving data and straight driving data First, calculate according to Formula 5 and X i The similarity is calculated according to Formula 6 and X i Similarity:

[0063]

[0064]

[0065] Among them, φ(·) can represent the formula after the cascade of f(·) and g(·), that is, the calculation formula of the eigenvector. Then, according to the Euclidean distance calculation method, we can get and X i The distance, X i and Then, if and X i The distance is less than X i and The distance indicates that the left turn driving data X i and right-angle left turn driving data The similarity is high, the left turn driving data X i and straight driving data The similarity is low. and X iThe distance is greater than X i and The distance indicates that the left turn driving data X i and right-angle left turn driving data The similarity is low, the left turn driving data X i and straight driving data High similarity.

[0066] S108. Train the feature extraction model using the first similarity and the second similarity.

[0067] First, according to the first similarity and the second similarity obtained in step S106, the server can determine the size relationship between the first similarity and the second similarity, and then, combined with step S102, the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data, and the feature extraction model is trained.

[0068] In an exemplary embodiment of the present disclosure, the server can determine a loss function based on the first similarity and the second similarity in feature extraction model training, and train the feature extraction model using the loss function. It should be noted that the similarity used to determine the loss function is consistent with the definition of the similarity in step S106.

[0069] For example, given a set of driving data, including X i , and

[0070] First, under the premise of using Euclidean distance to determine the similarity, the loss function of the feature extraction model can be calculated in combination with Formula 7:

[0071]

[0072] Among them, φ(·) can represent the calculation formula of the feature vector, l can be a hyperparameter, which is not updated during the training process, but its value can be adjusted and then trained. Its value is determined according to the quality of the training results. For example, l can be 1, 2, 4, etc.

[0073] Then, the loss function is adjusted and the feature extraction model is trained.

[0074] In summary, the model training method disclosed in the present invention uses multiple groups of sample data based on driving scenarios to train the model, so that the loss function of the feature extraction model is minimized, so that the feature vector corresponding to each driving data can be obtained, and the distance between each driving data and the target driving data is calculated based on the feature vector, and the similarity between each driving data and the target driving data is obtained, thereby improving the processing efficiency of the similarity between each driving data and the target driving data.

[0075] Furthermore, this example implementation also provides a driving data processing method.

[0076] Figure 2 The flowchart of the driving data processing method according to an exemplary embodiment of the present disclosure is schematically shown. Figure 2 , the driving data processing method may include the following steps:

[0077] S201. Acquire current driving data, process the current driving data using a trained feature extraction model, and determine a current feature vector corresponding to the current driving data.

[0078] In an exemplary embodiment of the present disclosure, the current driving data may be of the same type as the sample data based on the driving scenario in step S102 , and therefore, will not be described in detail herein.

[0079] For the current driving data obtained, driving data preprocessing can be performed to crop the data that does not need to be processed in the driving data, or the driving data of different scenarios can be converted into a form that can be processed by the above-mentioned feature extraction model.

[0080] The trained feature extraction model is obtained by the above model training method. The determined current feature vector may be of a fixed length or a dynamic length that can be adjusted according to actual needs.

[0081] In addition, if the real-time requirement is high, the above training model can be configured on the vehicle-mounted terminal device to process the driving data in real time. If the real-time requirement is not high, the above training model can be configured on the server to process the driving data.

[0082] S203. Obtain each historical feature vector corresponding to each historical driving data, and respectively calculate the similarity between each historical feature vector and the current feature vector.

[0083] Each historical driving data may be stored in a database of the vehicle-mounted terminal device or in a server. Each historical driving data may be driving data acquired in different historical driving scenes. Each historical driving data may be processed using the trained model to determine each historical feature vector corresponding to each historical driving data.

[0084] In an exemplary embodiment of the present disclosure, the above-mentioned trained model can be used to process each historical driving data first, and each historical feature vector corresponding to each historical driving data can be obtained, and then stored in a database. Then, when it is necessary to calculate the similarity between each historical feature vector and the current feature vector, each historical feature vector can be directly extracted from the database for calculation. It is also possible to use the above-mentioned trained model to process each historical driving data, determine each historical feature vector corresponding to each historical driving data, and calculate the similarity between each historical feature vector and the current feature vector when it is necessary to process the current driving data.

[0085] The similarity between each historical feature vector and the current feature vector can be represented by the Euclidean distance between each historical feature vector and the current feature vector. It should be noted that in the exemplary embodiment of the present disclosure, the Euclidean distance can be used, but is not limited to this, and can include but is not limited to the following methods: Manhattan distance, Chebyshev distance, angle cosine distance, chi-square distance, etc.

[0086] S205. If there is a target feature vector in each historical feature vector whose similarity with the current feature vector is greater than or equal to a preset threshold, determine the target driving data corresponding to the target feature vector.

[0087] In an exemplary embodiment of the present disclosure, the target feature vector may be one or more. That is, the database may store multiple historical driving data similar to the current driving data, that is, after calculating the similarity between each historical feature vector and the current feature vector, one or more target feature vectors may appear with a similarity greater than a preset threshold.

[0088] The preset threshold may be a fixed value or a dynamic value adjusted according to actual conditions. For example, the preset threshold may be 0.8, 0.9, etc.

[0089] The similarity can be a value between 0 and 1, where 0 can indicate that each historical feature vector has no similarity with the feature vector to be processed, and 1 can indicate that each historical feature vector is exactly the same as the feature vector to be processed. In addition, 0 can also indicate that each historical feature vector is exactly the same as the feature vector to be processed, and 1 can also indicate that each historical feature vector has no similarity with the feature vector to be processed. The value between 0 and 1 can indicate the degree of similarity between each historical feature vector and the feature vector to be processed.

[0090] S207. Delete the target driving data from the database and store the current driving data in the database, or discard the current driving data.

[0091] After executing step S205, one or more target feature vectors greater than or equal to a preset threshold may be determined in the database, that is, one or more target driving data are determined. In an exemplary embodiment of the present disclosure, from the perspective of the database, one or more target driving data may be deleted and the current driving data may be stored in the database; from the perspective of the current driving data, the current driving data may also be directly discarded.

[0092] It should be noted that, from the perspective of the database, the specific driving information contained in the current driving data can be ignored, and the target driving data can be directly deleted and the current driving data can be stored in the database. It can also be considered to compare the driving information contained in the current driving data with the driving information contained in the target driving data to determine whether to delete the target driving data and store the current driving data in the database, or to directly discard the current driving data.

[0093] For example, after determining one or more target driving data similar to the current driving data in the database, semantic analysis can then be used to judge the information of the target driving data and the current driving data, and then the steps of deleting the target driving data from the database and storing the current driving data in the database are determined. In an exemplary embodiment of the present disclosure, it may include but is not limited to using a semantic analysis method to determine whether it is necessary to perform the steps of deleting the target driving data from the database and storing the current driving data in the database. It may also be determined whether it is necessary to perform the steps of deleting the target driving data from the database and storing the current driving data in the database according to the actual scenario.

[0094] According to an exemplary embodiment of the present disclosure, if the similarity between each historical feature vector and the current feature vector is less than a preset threshold, that is, there is no historical driving data similar to the current driving data in the database, that is, the current driving data may not be stored in the database, and the current driving data needs to be stored in the database in preparation for subsequent processing of the driving data.

[0095] In summary, after obtaining driving data, the driving data processing method disclosed in the present invention can be used to process the driving data to obtain a feature vector corresponding to the driving data, and then obtain the similarity between the feature vector and each historical feature vector. The driving data processing method improves the calculation and processing efficiency of the similarity, helps to determine whether the driving data has been stored, reduces the storage of redundant data, and improves the rational use of storage resources.

[0096] It should be noted that although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0097] Furthermore, this example embodiment also provides a driving data processing device.

[0098] Figure 3 FIG. 1 schematically shows a block diagram of a driving data processing device according to an exemplary embodiment of the present disclosure. Figure 3 According to an exemplary embodiment of the present disclosure, the driving data processing device 300 may include: a feature determination module 302 , a similarity calculation module 304 , a similarity judgment module 306 and a data processing module 308 .

[0099] Among them, the feature determination module 302 is used to obtain the current driving data, use a trained feature extraction model to process the current driving data, and determine the current feature vector corresponding to the current driving data; wherein the trained feature extraction model is trained by the above-mentioned model training method; the similarity calculation module 304 is used to obtain each historical feature vector corresponding to each historical driving data, and respectively calculate the similarity between each historical feature vector and the current feature vector; the similarity judgment module 306 is used to determine the target driving data corresponding to the target feature vector if there is a target feature vector in each historical feature vector whose similarity with the current feature vector is greater than or equal to a preset threshold; the data processing module 308 is used to delete the target driving data from the database and store the current driving data in the database, or discard the current driving data.

[0100] In addition, according to an exemplary embodiment of the present disclosure, the similarity judgment module 306 may be configured to execute: if the similarity between each historical feature vector and the current feature vector is less than a preset threshold, then the current driving data is stored in the database.

[0101] Furthermore, this example implementation also provides a model training device.

[0102] Figure 4 Schematically shows a block diagram of a model training device according to an exemplary embodiment of the present disclosure. Figure 4 According to an exemplary embodiment of the present disclosure, a model training device 400 may include: a sample acquisition module 401, a feature extraction module 403, a similarity calculation module 405 and a model training module 407.

[0103] Among them, the sample acquisition module 401 is used to obtain multiple groups of sample data based on driving scenarios, each group of sample data includes first sample data, second sample data, and target sample data; wherein the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data; the feature extraction module 403 is used to input the first sample data, the second sample data and the target sample data into the feature extraction model respectively, and determine the first feature vector corresponding to the first sample data, the second feature vector corresponding to the second sample data, and the target feature vector corresponding to the target sample data; the similarity calculation module 405 is used to calculate the similarity between the first feature vector and the target feature vector as the first similarity, and calculate the similarity between the second feature vector and the target feature vector as the second similarity; the model training module 407 is used to train the feature extraction model using the first similarity and the second similarity.

[0104] According to an exemplary embodiment of the present disclosure, referring to Figure 5 The feature extraction module 403 may include: a first feature extraction submodule 502 and a second feature extraction submodule 504 .

[0105] Among them, the first feature extraction submodule 502 is used to input the first sample data, the second sample data and the target sample data into the first feature extraction unit respectively, and determine the first feature matrix corresponding to the first sample data, the second feature matrix corresponding to the second sample data, and the target feature matrix corresponding to the target sample data; the second feature extraction submodule 504 is used to input the first feature matrix, the second feature matrix and the target feature matrix into the second feature extraction unit respectively, and determine the first feature vector corresponding to the first feature matrix, the second feature vector corresponding to the second feature matrix, and the target feature vector corresponding to the target feature matrix.

[0106] According to an exemplary embodiment of the present disclosure, referring to Figure 6 The first feature extraction submodule 502 may include: a data embedding unit 601 and a feature matrix determination unit 603 .

[0107] Among them, the data embedding unit 601 is used to perform embedding operations on the first sample data, the second sample data and the target sample data respectively, and determine the first intermediate matrix corresponding to the first sample data, the second intermediate matrix corresponding to the second sample data, and the target intermediate matrix corresponding to the target sample data; the feature determination unit 603 is used to determine the first feature matrix by combining the first intermediate matrix with the specified parameter matrix, determine the second feature matrix by combining the second intermediate matrix with the specified parameter matrix, and determine the target feature matrix by combining the target intermediate matrix with the specified parameter matrix.

[0108] According to an exemplary embodiment of the present disclosure, referring to Figure 7, the model training module 407 may include: a loss function determination unit 702 and a model training unit 704.

[0109] The loss function determination unit 702 is used to determine the loss function based on the first similarity and the second similarity; the model training unit 704 is used to train the feature extraction model using the loss function.

[0110] Since the functional modules of the program running performance analysis device according to the embodiment of the present invention are the same as those in the above-mentioned method invention embodiment, they will not be described in detail here.

[0111] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present invention may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of the present specification.

[0112] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0113] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as a system, method or program product. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as a "circuit", "module" or "system".

[0114] Refer to the following Figure 8 An electronic device 800 according to this embodiment of the present invention will be described. Figure 8 The electronic device 800 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0115] like Figure 8 As shown, the electronic device 800 is in the form of a general computing device. The components of the electronic device 900 may include, but are not limited to: the at least one processing unit 810, the at least one storage unit 820, a bus 830 connecting different system components (including the storage unit 820 and the processing unit 810), and a display unit 840.

[0116] The storage unit stores program codes, which can be executed by the processing unit 810, so that the processing unit 810 performs the steps of various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification. For example, the processing unit 810 can perform the following steps: Figure 1 Steps S102 to S108 shown in FIG. Figure 2 Steps S201 to S207 shown in FIG.

[0117] The storage unit 820 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 8201 and / or a cache memory unit 8202 , and may further include a read-only memory unit (ROM) 8203 .

[0118] The storage unit 820 may also include a program / utility 8204 having a set (at least one) of program modules 8205, such program modules 8205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0119] Bus 830 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0120] The electronic device 800 may also communicate with one or more external devices 900 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to communicate with the electronic device 800, and / or communicate with any device that enables the electronic device 800 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 850. Furthermore, the electronic device 800 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 860. As shown, the network adapter 860 communicates with other modules of the electronic device 800 via a bus 830. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0121] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0122] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0123] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0124] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0125] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A model training method, characterized in that: include: Acquire multiple groups of sample data based on driving scenarios, each group of sample data includes first sample data, second sample data, and target sample data; wherein the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data; Inputting the first sample data, the second sample data, and the target sample data into a feature extraction model respectively, and determining a first feature vector corresponding to the first sample data, a second feature vector corresponding to the second sample data, and a target feature vector corresponding to the target sample data; Calculating a similarity between the first feature vector and the target feature vector as a first similarity, and calculating a similarity between the second feature vector and the target feature vector as a second similarity; A loss function is determined based on the first similarity and the second similarity, and the feature extraction model is trained using the loss function; the trained feature extraction model is used to determine a current feature vector corresponding to the acquired current driving data; when there is target driving data corresponding to a target feature vector whose similarity with the current feature vector is higher than a preset threshold in the historical driving data, the target driving data is deleted from the database, the current driving data is stored in the database, or the current driving data is discarded.

2. The model training method according to claim 1, characterized in that: The feature extraction model includes a first feature extraction unit and a second feature extraction unit; wherein the first sample data, the second sample data, and the target sample data are respectively input into the feature extraction model, and a first feature vector corresponding to the first sample data, a second feature vector corresponding to the second sample data, and a target feature vector corresponding to the target sample data are determined, including: Inputting the first sample data, the second sample data, and the target sample data into the first feature extraction unit respectively, and determining a first feature matrix corresponding to the first sample data, a second feature matrix corresponding to the second sample data, and a target feature matrix corresponding to the target sample data; The first feature matrix, the second feature matrix and the target feature matrix are respectively input into the second feature extraction unit to determine a first feature vector corresponding to the first feature matrix, a second feature vector corresponding to the second feature matrix, and a target feature vector corresponding to the target feature matrix.

3. The model training method according to claim 2, characterized in that: Inputting the first sample data, the second sample data, and the target sample data into the first feature extraction unit respectively, and determining a first feature matrix corresponding to the first sample data, a second feature matrix corresponding to the second sample data, and a target feature matrix corresponding to the target sample data includes: Performing embedding operations on the first sample data, the second sample data, and the target sample data respectively, to determine a first intermediate matrix corresponding to the first sample data, a second intermediate matrix corresponding to the second sample data, and a target intermediate matrix corresponding to the target sample data; The first intermediate matrix is ​​combined with a specified parameter matrix to determine a first characteristic matrix, the second intermediate matrix is ​​combined with the specified parameter matrix to determine a second characteristic matrix, and the target intermediate matrix is ​​combined with the specified parameter matrix to determine a target characteristic matrix.

4. The model training method according to any one of claims 1 to 3, characterized in that: Training the feature extraction model using the first similarity and the second similarity includes: Determine a loss function based on the first similarity and the second similarity; The feature extraction model is trained using the loss function.

5. A driving data processing method, characterized in that: include: Acquire current driving data, process the current driving data using a trained feature extraction model, and determine a current feature vector corresponding to the current driving data; wherein the trained feature extraction model is trained by the model training method according to any one of claims 1 to 4; Obtaining each historical feature vector corresponding to each historical driving data, and respectively calculating the similarity between each historical feature vector and the current feature vector; If there is a target feature vector in each of the historical feature vectors, the similarity between which and the current feature vector is greater than or equal to a preset threshold, determining the target driving data corresponding to the target feature vector; The target driving data is deleted from the database and the current driving data is stored in the database, or the current driving data is discarded.

6. The driving data processing method according to claim 5, characterized in that: The driving data processing method further includes: If the similarity between each historical feature vector and the current feature vector is less than a preset threshold, the current driving data is stored in the database.

7. A driving data processing device, characterized in that: include: a feature determination module, used to obtain current driving data, process the current driving data using a trained feature extraction model, and determine a current feature vector corresponding to the current driving data; wherein the trained feature extraction model is trained by the model training method according to any one of claims 1 to 4; A similarity calculation module, used for obtaining each historical feature vector corresponding to each historical driving data, and respectively calculating the similarity between each historical feature vector and the current feature vector; A similarity judgment module, configured to determine target driving data corresponding to a target feature vector if there is a target feature vector in each of the historical feature vectors whose similarity with the current feature vector is greater than or equal to a preset threshold; A data processing module is used to delete the target driving data from the database and store the current driving data in the database, or discard the current driving data.

8. A model training device, characterized in that: include: A sample acquisition module, used to acquire multiple groups of sample data based on driving scenarios, each group of sample data includes first sample data, second sample data, and target sample data; wherein the similarity between the first sample data and the target sample data is greater than the similarity between the second sample data and the target sample data; a feature extraction module, configured to input the first sample data, the second sample data, and the target sample data into a feature extraction model, respectively, and determine a first feature vector corresponding to the first sample data, a second feature vector corresponding to the second sample data, and a target feature vector corresponding to the target sample data; a similarity calculation module, configured to calculate the similarity between the first feature vector and the target feature vector as a first similarity, and calculate the similarity between the second feature vector and the target feature vector as a second similarity; A model training module is used to determine a loss function based on the first similarity and the second similarity, and use the loss function to train the feature extraction model; the trained feature extraction model is used to determine a current feature vector corresponding to the acquired current driving data; when there is target driving data corresponding to a target feature vector whose similarity with the current feature vector is higher than a preset threshold in the historical driving data, the target driving data is deleted from the database, the current driving data is stored in the database, or the current driving data is discarded.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the model training method as described in any one of claims 1 to 4 or the driving data processing method as described in claim 5 or 6 is implemented.

10. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement the model training method as described in any one of claims 1 to 4 or the driving data processing method as described in claim 5 or 6.