Model training method and device, electronic equipment and storage medium

Abstract

The invention discloses a model training method and device, electronic equipment and a storage medium, and relates to the technical field of model training, and the method comprises the steps: obtaining a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matched with each first control signal, and taking the first control signals and the first motion state as first test data; an initial model is trained based on the first test data to obtain an initial dynamic model, the initial model is used for analyzing the obtained control signal based on a data driving algorithm to obtain the motion state of a target vehicle, and the target vehicle is a real vehicle corresponding to the virtual vehicle; acquiring a plurality of second control signals of the target vehicle running in the real test scene and a second motion state matched with each second control signal, and taking the second control signals and the second motion state as second test data; and training the initial dynamical model based on the second test data to obtain a target dynamical model. The test data collected in a real test scene required by a training model can be reduced, and the cost of model training is reduced.

Description

Technical Field

[0001] The present application relates to the field of model training technology, and more specifically, to a model training method, an apparatus electronic device and a storage medium. Background Art

[0002] At present, the vehicle's dynamic model can be used to obtain the vehicle's state information based on the input control signal, and the dynamic model needs to be trained through test data. Although, currently, measured data can be obtained through road sampling, and then the dynamic model can be trained through the measured data. However, it is costly to train the dynamic model through measured data. Summary of the invention

[0003] The present application proposes a model training method, an electronic device and a storage medium.

[0004] In a first aspect, an embodiment of the present application provides a model training method, comprising: obtaining a plurality of first control signals of a virtual vehicle running in a virtual test scenario and a first motion state matching each first control signal as first test data; training an initial model based on the first test data to obtain an initial dynamic model, wherein the initial model is used to obtain the motion state of a target vehicle based on the control signal obtained by analyzing the data-driven algorithm, and the target vehicle is a real vehicle corresponding to the virtual vehicle; obtaining a plurality of second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal as second test data; training the initial dynamic model based on the second test data to obtain a target dynamic model.

[0005] Optionally, for a possible implementation, the initial dynamic model is trained based on the second test data to obtain a target dynamic model, including: obtaining the driving intensity of the target vehicle represented by each second motion state; taking the second motion state with a driving intensity less than or equal to a threshold level and the second control signal corresponding to the second motion state as the first sub-test data; taking the second motion state with a driving intensity greater than a threshold level and the second control signal corresponding to the second motion state as the second sub-test data; training the initial dynamic model based on the first sub-test data to obtain an intermediate dynamic model; training the intermediate dynamic model based on the second sub-test data to obtain the target dynamic model.

[0006] Optionally, for a possible implementation, the initial dynamic model is trained based on the first sub-test data to obtain an intermediate dynamic model, including: obtaining a first initial motion state corresponding to the second control signal of each of the first sub-test data through the initial dynamic model; obtaining a first difference between each of the first initial motion state and a second motion state corresponding to the first initial motion state; and when it is detected that at least a specified number of first differences are greater than a first threshold, training the initial dynamic model based on the first sub-test data to obtain an intermediate dynamic model.

[0007] Optionally, for a possible implementation, the intermediate dynamic model is trained based on the second sub-test data to obtain the target dynamic model, including: obtaining the second initial motion state corresponding to the second control signal of each second sub-test data through the intermediate dynamic model; obtaining the second difference between each second initial motion state and the second motion state corresponding to the second initial motion state; taking the second difference greater than a second threshold as the target second difference; and training the intermediate dynamic model based on the second sub-test data corresponding to the target second difference to obtain the target dynamic model.

[0008] Optionally, for a possible implementation, the method of obtaining multiple first control signals of a virtual vehicle running in a virtual test scenario and a first motion state matching each first control signal as first test data includes: obtaining a third control signal of the target vehicle running in a real test scenario, third state information corresponding to the third control signal, and chassis parameters of a chassis system of the target vehicle; adjusting parameters of a vehicle model based on the third control signal and the chassis parameters to obtain a target vehicle model matching the target vehicle, wherein the vehicle model is a mechanism-driven dynamic model; in the virtual test scenario, obtaining each first motion state matching the first control signal for the virtual vehicle through the target vehicle model; and taking each first control signal and each first motion state matching the first control signal as the first test data.

[0009] Optionally, for a possible implementation, before obtaining the first motion state that matches each first control signal for the virtual vehicle through the target vehicle model in the virtual test scenario, it also includes: generating multiple first control signals for the virtual vehicle in the virtual test scenario based on a simulation application.

[0010] Optionally, for a possible implementation, the initial model is trained based on the first test data to obtain an initial dynamic model, including: obtaining a third initial motion state corresponding to the first control signal through the initial model; obtaining a third difference between the third initial motion state and the first motion state corresponding to the third initial motion state; training the initial model based on the third difference so that the third difference obtained by the trained initial model is less than a third threshold; and using the trained initial model as the initial dynamic model.

[0011] In the second aspect, the embodiment of the present application also provides a model training device, including: a first acquisition unit, a first training unit, a second acquisition unit and a second training unit. The first acquisition unit is used to acquire multiple first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as first test data; the first training unit is used to train the initial model based on the first test data to obtain an initial dynamic model, wherein the initial model is used to obtain the motion state of a target vehicle based on the control signal acquired by the data-driven algorithm analysis, and the target vehicle is a real vehicle corresponding to the virtual vehicle; the second acquisition unit is used to acquire multiple second control signals of the target vehicle running in a real test scene and a second motion state matching each second control signal as second test data; the second training unit is used to train the initial dynamic model based on the second test data to obtain a target dynamic model.

[0012] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: one or more processors; a memory; and one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method described in the first aspect.

[0013] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, in which a program code is stored. The program code can be called by a processor to execute the method described in the first aspect above.

[0014] The model training method, device electronic device and storage medium provided in the embodiment of the present application first obtain multiple first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as the first test data; then train the initial model based on the first test data to obtain an initial dynamic model, wherein the initial model is used to obtain the motion state of a target vehicle based on the control signal obtained by data-driven algorithm analysis, and the target vehicle is a real vehicle corresponding to the virtual vehicle; then obtain multiple second control signals of the target vehicle running in a real test scene and a second motion state matching each second control signal as the second test data; thereby training the initial dynamic model based on the second test data to obtain a target dynamic model. If the dynamic model is trained entirely through simulation data in a virtual scene, the trained model may not be suitable for the actual scene in the application, and the usability is low; and if the model is trained entirely through measured data obtained through road sampling, the demand for measured data will be too large due to the need for a large number of measured data, resulting in a high acquisition cost. In the present application, the initial model can first be trained based on the first test data obtained in the virtual test scene to obtain the initial dynamic model, and then the initial dynamic model can be trained by the second test data obtained in the real test scene to obtain the target dynamic model. It can be seen that the object trained by the second test data in the real test scene is the initial model that has been trained with the first test data. Therefore, without affecting the availability of the target dynamic model obtained later in the actual scene, the amount of the second test data required for the training model can be reduced to reduce the acquisition cost, thereby reducing the cost of model training.

[0015] Other features and advantages of the embodiments of the present application will be described in the subsequent description, and partly become apparent from the description, or can be understood by practicing the embodiments of the present application. The purposes and other advantages of the embodiments of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 An application scenario diagram of the model training method provided in an embodiment of the present application is shown;

[0018] Figure 2 A method flow chart of the model training method provided in an embodiment of the present application is shown;

[0019] Figure 3 A method flow chart of a model training method provided by another embodiment of the present application is shown;

[0020] Figure 4 A method flow chart of a model training method provided by yet another embodiment of the present application is shown;

[0021] Figure 5 A structural block diagram of a model training device provided in an embodiment of the present application is shown;

[0022] Figure 6 A structural block diagram of an electronic device provided in an embodiment of the present application is shown;

[0023] Figure 7 A structural block diagram of a computer-readable storage medium provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] In order to make those skilled in the art better understand the present application scheme, the technical scheme in the present application embodiment will be clearly and completely described below in conjunction with the drawings in the present application embodiment. Obviously, the described embodiment is only a part of the present application embodiment, rather than all the embodiments. The components of the present application embodiment usually described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application for protection, but merely represents the selected embodiment of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0025] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0026] At present, the vehicle's dynamic model can be used to obtain the vehicle's state information based on the input control signal, and the dynamic model needs to be trained with test data. Although, currently, measured data can be obtained through road sampling, and then the dynamic model can be trained with the measured data. However, it is costly to train the dynamic model with measured data. How to reduce the cost of dynamic model training is an urgent problem to be solved.

[0027] At present, the vehicle's on-board system can be configured with functions such as assisted driving and intelligent driving, and these functions need to rely on the dynamic model to analyze the input control signal to obtain the vehicle's status information. It can be seen that the dynamic model is crucial for the vehicle. At present, the model can be trained by obtaining test data through road sampling in a real environment to obtain a trained dynamic model. It is also possible to perform simulation tests in virtual scenes to meet the test requirements of large-scale mileage, thereby obtaining test data based on simulation tests and then training the model.

[0028] However, the inventors found in their research that if the dynamic model is trained entirely through simulation data in virtual scenes, the trained model may not be suitable for actual scenes during application and have low usability; and if the model is trained entirely through measured data obtained through road sampling, the large amount of measured data required will lead to excessive demand for measured data, resulting in high collection costs.

[0029] Therefore, in order to solve or partially solve the above problems, the present application provides a model training method, an electronic device and a storage medium.

[0030] See also Figure 1 , Figure 1 An application scenario diagram of the model training method provided in an embodiment of the present application is shown, namely, a model training scenario 100, which may include an electronic device 110.

[0031] The electronic device 110 can access the Internet, for example, through wireless communication technologies such as Wi-Fi and Bluetooth; the electronic device 110 can also access the Internet in a wired manner, for example, through an Rj45 network cable or optical fiber.

[0032] The user can control the electronic device 110 so that the electronic device executes the model training method. For example, the user can directly operate the electronic device 110 to control the electronic device to execute the model training method, wherein the electronic device 110 can be locally deployed with an initial model for training. For a detailed description, please refer to the subsequent implementation. The user can also control the electronic device 110 through another electronic device that is also connected to the Internet, thereby controlling the electronic device 110 to execute the model training method.

[0033] See also Figure 2 , Figure 2 A method flow chart of a model training method provided in an embodiment of the present application is shown. The model training method can be applied to Figure 1The electronic device in the video generation scene shown can specifically use the processor of the electronic device as the execution subject of the model training method. The model training method can include steps S110 to S140.

[0034] Step S110: Acquire a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as first test data.

[0035] The virtual test scene is different from the real vehicle test scene. The virtual test scene is a virtual scene constructed by digital modeling and other related software based on some virtual target objects acquired in advance. For example, the virtual target objects can be imported into the vehicle test scene constructed by digital modeling related software to obtain the virtual test scene. Among them, the virtual target objects can include virtual vehicles, virtual lanes, etc. Therefore, the real vehicle test scene can be simulated more conveniently and concisely through the virtual test scene, so as to realize the simulation test of the real vehicle in reality through the virtual test scene.

[0036] The virtual vehicle may be a digital model corresponding to a real vehicle in reality, and the virtual vehicle may move according to a control signal in a virtual test scene, thereby having a motion state that matches the control signal. For some embodiments, the relevant parameters of the virtual vehicle may be adjusted and set by the relevant parameters of the real vehicle in reality, so that the virtual vehicle has the same or similar motion state as the real vehicle when the virtual vehicle obtains the same control signal as the real vehicle. For example, the above-mentioned relevant parameters may include parameters of the powertrain system, parameters of the suspension system, parameters of the body system, etc.

[0037] In the virtual test scenario, multiple first control signals may be generated. Exemplarily, the virtual test scenario may be constructed by a simulation application such as Carsim or Carmaker. In addition, multiple first control signals may be generated by Carsim or Carmaker. Among them, the first control signal may be used to input into the virtual vehicle to control the virtual vehicle, for example, the first control signal may include a steering angle, an accelerator pedal signal, a brake pedal signal, and a vehicle gear position.

[0038] Thus, the virtual vehicle in the virtual test scene can have different first motion states under the control of the first control signal. In some embodiments, the acquired first control signal can be analyzed by the target vehicle model to obtain the corresponding motion state. Please refer to the subsequent embodiments for detailed description.

[0039] The first motion state can be used to describe the vehicle state data of the virtual vehicle under the control of the first control signal in the virtual test scene. For example, the first motion state can include the position of the virtual vehicle, the speed, angular velocity, acceleration and angular acceleration of the vehicle. It should be noted that the first motion state is the vehicle state data corresponding to the virtual vehicle obtained by calculation and simulation in the virtual test scene.

[0040] Exemplarily, the first control signal may be provided by a simulation application, and the first motion state of the virtual vehicle under the action of the first control signal may also be monitored by the simulation application.

[0041] It is understandable that in reality, even if the same control signal is input, the matching motion state obtained by the real car in different types of real vehicle test scenarios is different. For example, for the same initial speed and brake pedal information, the motion state obtained on sunny days and rainy days is different. Therefore, the virtual test scene can also be divided into different types according to different working conditions. For example, the virtual test scene can be divided into test scenes such as highways, urban roads, national roads, and rural roads according to the road type; for example, the virtual test scene can be divided into test scenes such as straight roads and curves according to road characteristics; for example, the virtual test scene can be divided into test scenes such as sunny days, rainy days, and cloudy days according to the weather; for example, the virtual test scene can be divided into test scenes such as following, meeting, overtaking, changing lanes, emergency avoidance, and traffic lights according to driving conditions. Thus, the first control signal under various types of virtual test scenarios can be obtained, and then the first motion state matched by the first control signal under different types of virtual test scenarios can be obtained.

[0042] Furthermore, the obtained multiple first control signals and the first motion state matching each first control signal can be used as first test data, and the first test data can be used for subsequent training of the initial model.

[0043] Step S120: training the initial model based on the first test data to obtain an initial dynamics model, wherein the initial model is used to obtain the motion state of a target vehicle based on a control signal obtained by analyzing a data-driven algorithm, and the target vehicle is a real vehicle corresponding to the virtual vehicle.

[0044] After the first test data is obtained, the initial model can be trained based on the first test data to obtain an initial dynamics model. The initial model is used to obtain the motion state of the target vehicle based on the control signal obtained by the data-driven algorithm analysis. In other words, the initial model can output the motion state according to the input control signal obtained.

[0045] For some implementations, the first control signal in the first test data may be used as the input of the initial model, and the first motion state matched by each first test data may be used as the label corresponding to the first test data, so as to train the initial model.

[0046] Specifically, the acquired first control signal can be analyzed by the initial model, and specifically, the first control signal can be analyzed based on a data-driven algorithm to obtain a corresponding output. The obtained output is then compared with the label matching the first control signal, that is, compared with the first motion state matching the first control signal, to determine the difference between the output and the first motion state matching the first control signal, and then the initial model is trained by the difference. Please refer to the subsequent embodiments for a detailed description.

[0047] The target vehicle is a real vehicle corresponding to the virtual vehicle, that is, for the same control signal, the motion state corresponding to the target vehicle should be the same or similar to the motion state corresponding to the virtual vehicle.

[0048] Step S130: Acquire multiple second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal as second test data.

[0049] Furthermore, the road sampling data of the target vehicle can be obtained and the initial dynamic model can be further trained. For some embodiments, multiple second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal can be obtained.

[0050] Similarly, real test scenarios can also be divided into different types according to working conditions. For example, real test scenarios can be divided into highways, urban roads, national roads, rural roads and other test scenarios according to road types; for example, real test scenarios can be divided into straight roads, curves and other test scenarios according to road characteristics; for example, real test scenarios can be divided into sunny days, rainy days, cloudy days and other scenes according to weather; for example, real test scenarios can be divided into following, meeting, overtaking, lane changing, emergency avoidance, traffic lights and other test scenarios according to driving conditions.

[0051] The target vehicle may be placed in different types of real test scenarios, and then multiple second control signals may be input to the target vehicle, thereby acquiring the second motion state matched by each second control signal.

[0052] Exemplarily, the target vehicle can monitor the input control signal through the vehicle system as the second control signal. For example, the steering angle input by the user can be monitored by the vehicle system and then used as the second control signal; for another example, the accelerator pedal signal input by the user can also be monitored by the vehicle system and then used as the second control signal. Similar to the first control signal, the second control signal can also include a steering angle, an accelerator pedal signal, a brake pedal signal, and a vehicle gear position.

[0053] Furthermore, the target vehicle may also be provided with a sensor module, and the motion state of the target vehicle may be collected through the sensor module. Specifically, for each second signal, the collected motion state may be used as a second motion state matching the second motion state. Exemplarily, the sensor module may include a wheel speed sensor, an inertial measurement unit, etc. Similar to the first motion state, the second motion state may include the position of the target vehicle, the speed, angular velocity, acceleration, and angular acceleration of the vehicle, etc.

[0054] Therefore, the acquired multiple second control signals and the second motion state matched with each second control signal can be used as the second test data, and the second test data can be used for subsequent training of the initial dynamics model.

[0055] Step S140: training the initial dynamics model based on the second test data to obtain a target dynamics model.

[0056] After the second test data is acquired, the initial dynamics model may be trained based on the second test data to obtain a target dynamics model.

[0057] For some embodiments, similar to the method of training the initial model by the first test data in the aforementioned steps, the second motion state matched by the second control signal can be used as the label of the second control signal. Thus, the second control signal in the second test data can be used as the input of the target dynamic model to obtain the corresponding output. The obtained output is then compared with the label matched by the second control signal, that is, compared with the second motion state matched by the second control signal, and the difference between the output and the second motion state matched by the second control signal is determined, and then the initial dynamic model is trained based on the difference.

[0058] For some other implementations, the second test data may also be divided first. For example, the second test data may be divided based on the second motion state characterizing the driving intensity of the target vehicle, and then the initial dynamic model may be gradually trained through the test data corresponding to different driving intensities. For a detailed introduction, please refer to the subsequent embodiments.

[0059] The model training method provided in the embodiment of the present application first obtains a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as the first test data; then trains the initial model based on the first test data to obtain an initial dynamic model, wherein the initial model is used to obtain the motion state of a target vehicle based on the control signal obtained by data-driven algorithm analysis, and the target vehicle is a real vehicle corresponding to the virtual vehicle; then obtains a plurality of second control signals of the target vehicle running in a real test scene and a second motion state matching each second control signal as the second test data; thereby training the initial dynamic model based on the second test data to obtain a target dynamic model. If the dynamic model is obtained by training all simulation data in the virtual scene, the trained model may not be suitable for the actual scene in the application, and the usability is low; and if the model is trained all by the measured data obtained by road sampling, the demand for measured data will be too large due to the need for a large number of measured data, resulting in a high acquisition cost. In the present application, the initial model can first be trained based on the first test data obtained in the virtual test scene to obtain the initial dynamic model, and then the initial dynamic model can be trained by the second test data obtained in the real test scene to obtain the target dynamic model. It can be seen that the object trained by the second test data in the real test scene is the initial model that has been trained with the first test data. Therefore, without affecting the availability of the target dynamic model obtained later in the actual scene, the amount of the second test data required for the training model can be reduced to reduce the acquisition cost, thereby reducing the cost of model training.

[0060] See also Figure 3 , Figure 3 A method flow chart of a model training method provided in an embodiment of the present application is shown. The model training method can be applied to Figure 1 The electronic device in the video generation scene shown can specifically use the processor of the electronic device as the execution subject of the model training method. The model training method can include steps S210 to S2110.

[0061] Step S210: Acquire a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as first test data.

[0062] Among them, step S210 has been introduced in detail in the above embodiment and will not be repeated here.

[0063] Step S220: obtaining a third initial motion state corresponding to the first control signal through the initial model.

[0064] Step S230: Obtain a third difference between the third initial motion state and the first motion state corresponding to the third initial motion state.

[0065] Step S240: training the initial model based on the third difference so that the third difference obtained by the trained initial model is smaller than a third threshold.

[0066] Step S250: using the trained initial model as the initial dynamics model.

[0067] The initial model is trained based on the first test data. Specifically, the first control signal in the first test data can be input into the initial model as an input quantity. Thus, the third initial motion state corresponding to the first control signal is obtained through the initial model. That is, the output quantity generated by the initial model for the input first control signal is used as the third initial motion state.

[0068] It can be understood that the first test data also includes the first motion state that matches the first control signal, and the initial model is trained to enable the initial model to analyze the input first control signal. Specifically, the first control signal is analyzed through a data-driven algorithm, so that the output generated for the first control signal is as identical or similar as possible to the first motion state that matches the first control signal.

[0069] Thus, a third difference between the third initial motion state and the first motion state corresponding to the third initial motion state can be further obtained. It should be noted that the first motion state corresponding to the third initial motion state is the first motion state matched by the first control signal corresponding to the third initial motion state. It can be understood that the larger the third difference is, the lower the accuracy of the result output after analyzing the first control signal through the initial model is, and the smaller the third difference is, the higher the accuracy of the result output after analyzing the first control signal through the initial model is.

[0070] Therefore, the initial model can be trained based on the third difference to reduce the third difference. In the implementation provided in the present application, the initial model can be trained until the third difference is less than a third threshold, wherein the third threshold can be pre-set and is not specifically limited in the embodiments of the present application.

[0071] When the initial model is trained so that the third difference obtained by the trained initial model is smaller than the third threshold, the trained initial model may be used as the initial dynamics model.

[0072] Optionally, the initial model may be trained after satisfying the number of iterations, and the trained initial model may be used as the initial dynamics model.

[0073] Optionally, the first test data may be divided into training data and test data, wherein the training data is used to train the initial model to obtain the trained initial model; and the test data is used to evaluate the trained initial model to obtain the initial dynamics model.

[0074] Exemplarily, the first test data may be randomly divided into training data and test data according to a specified ratio, for example, the training data accounts for 70% and the test data accounts for 30%.

[0075] In another exemplary embodiment, when the first test data is randomly divided according to a specified ratio, the division can also be carried out in combination with the working condition distribution of the first test data, so that the divided training data and test data include the type corresponding to each working condition. Among them, the working condition distribution of the first test data refers to the proportion of the number corresponding to each type in the first test data, and each type is divided based on the working condition. For example, the first test data includes 10% of highways, 80% of urban roads, 5% of national highways, and 5% of rural roads. Therefore, continue to use the specified proportion of 70% for training data and 30% for test data as an example for explanation. 70% of the expressways accounting for 10%, 70% of the urban roads accounting for 80%, 70% of the national highways accounting for 5%, and 70% of the rural roads accounting for 5% in the first test data can be divided into training data; and 30% of the expressways accounting for 10%, 30% of the urban roads accounting for 80%, 30% of the national highways accounting for 5%, and 30% of the rural roads accounting for 5% in the first test data can be divided into test data.

[0076] It should be noted that each test data and each training data after division includes a first control signal and a first motion state matched by the first control signal. Then, the initial model is trained by the training data to obtain a trained initial model. The method of training the initial model by the training data can be referred to the above introduction, which will not be repeated here.

[0077] The trained initial model is then evaluated using the test data. Exemplarily, the first control signal in the test data can be input into the trained initial model to obtain the output of the trained initial model, and then the output is compared with the first motion state matched with the first control signal as the input to determine the difference. If the difference is less than or equal to a specified threshold, the trained initial model can be determined as the initial dynamic model. If the above difference is less than or equal to the specified threshold, it can be regarded that the trained initial model has a higher degree of fit.

[0078] For some embodiments, the initial model can also be trained by training data to obtain multiple trained initial models, and then each trained initial model can be evaluated by test data to select the trained initial model with the highest model fit as the initial dynamic model. For example, the difference corresponding to each trained initial model can be determined based on the aforementioned method for obtaining the difference, so that the trained initial model with the smallest difference is the model with the highest model fit.

[0079] Step S260: Acquire multiple second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal as second test data.

[0080] Among them, step S260 has been introduced in detail in the above embodiment and will not be repeated here.

[0081] Step S270: Obtain the driving intensity of the target vehicle represented by each second motion state.

[0082] Step S280: taking the second motion state in which the driving intensity is less than or equal to the threshold and the second control signal corresponding to the second motion state as the first sub-test data.

[0083] Step S290: taking the second motion state in which the driving intensity is greater than the threshold and the second control signal corresponding to the second motion state as the second sub-test data.

[0084] It is understandable that the driving intensity of the target vehicle can be characterized by the second motion state in the second test data. As can be seen from the foregoing introduction, the second motion state may include the position of the target vehicle, the speed, angular velocity, acceleration and angular acceleration of the vehicle, etc. Thus, for example, if the second motion state includes the speed of the vehicle, the driving intensity can be determined by the speed of the vehicle; for another example, if the second motion state includes the speed and acceleration of the vehicle, the driving intensity can be determined by the speed and acceleration of the vehicle. The driving intensity program is further compared with the threshold level to achieve the division of the second test data by driving intensity.

[0085] Exemplarily, a parameter threshold can be set for each parameter that may be included in the second motion state, and then each parameter in the second motion state is compared with the corresponding parameter threshold. If it is greater than the parameter threshold, a higher score is set for the parameter; if it is less than or equal to the parameter threshold, a lower score is set for the parameter. The sum of the scores of each parameter is then taken as the driving intensity corresponding to the second motion state.

[0086] Continuing with the example of the second motion state including the vehicle's speed and acceleration, if the parameter threshold corresponding to the vehicle's speed is 150km / h, and the parameter threshold corresponding to the acceleration is 0.5g, the second motion state includes a vehicle speed of 160km / h and an acceleration of 0.2g. It can be determined that the vehicle's speed is greater than the corresponding parameter threshold, and a score of 1 can be set for the vehicle's speed; it can be determined that the acceleration is less than the corresponding parameter threshold, and a score of 0.2 can be set for the acceleration, thereby obtaining 1+0.2=1.2 as the driving intensity represented by the second motion state.

[0087] Optionally, different parameter thresholds can be set for each parameter in the second motion state for different types. The types are divided according to working conditions. For example, for the speed of a vehicle in the second motion state of an urban road type, the parameter threshold corresponding to the speed of the vehicle can be set to 90km / h; for the speed of a vehicle in the second motion state of a high-speed type, the parameter threshold corresponding to the speed of the vehicle can be set to 150km / h. Thus, the second state information can be used to more accurately characterize the intensity of driving, so as to achieve a more accurate division of the second test data based on the intensity of driving.

[0088] It should be noted that the specific values ​​of the parameter thresholds corresponding to the parameters in the second motion states of different types are only examples, and can be flexibly set according to experience, needs, etc. in practical applications, and are not specifically limited in the embodiments of the present application.

[0089] The threshold value is further described as 1. In the case where the driving intensity of the second motion state is 1.2, it can be known that 1.2 is greater than 1, so the second motion state and the second control signal corresponding to the second motion state can be used as the second sub-test data. For another example, if the driving intensity of some second motion states is less than or equal to 1, the second motion state with a driving intensity less than or equal to 1 and the second control signal corresponding to the second motion state can be used as the first sub-test data.

[0090] Thus, the second test data is divided into first sub-test data and second sub-test data through the above steps, wherein the second motion state in the first sub-test data represents a lower driving intensity, while the second motion state in the second sub-test data represents a higher driving intensity. For some embodiments, the first sub-test data can be regarded as test data corresponding to normal working conditions, and the second sub-test data can be regarded as test data corresponding to extreme working conditions.

[0091] Step S2100: training the initial dynamics model based on the first sub-test data to obtain an intermediate dynamics model.

[0092] It is understandable that in daily vehicle driving, most of the time is spent in a situation where the driving intensity is low. Therefore, the initial dynamics model can be first trained with the first sub-test data to obtain an intermediate dynamics model, so as to improve the applicability of the intermediate dynamics model to the first sub-test data with low driving intensity.

[0093] For some implementations, after obtaining the first sub-test data, the initial kinetic model may be directly trained using the first sub-test data, and the trained initial kinetic model may be used as an intermediate kinetic model.

[0094] For some other embodiments, before training the initial kinetic model with the first sub-test data, it is also possible to first verify whether the initial kinetic model is applicable to the first sub-test data, and only when the initial kinetic model is less applicable to the first sub-test data, the initial kinetic model is trained with the first sub-test data. Specifically, step S2100 may also include steps S2101 to S2103.

[0095] Step S2101: obtaining a first initial motion state corresponding to the second control signal of each of the first sub-test data through the initial dynamics model.

[0096] Step S2102: Obtain a first difference between each of the first initial motion states and a second motion state corresponding to the first initial motion state.

[0097] Step S2103: When it is detected that at least a specified number of first differences are greater than a first threshold, the initial kinetic model is trained based on the first sub-test data to obtain an intermediate kinetic model.

[0098] First, the first initial motion state corresponding to the second control signal of each of the first sub-test data can be obtained through the initial dynamic model. Then, the first difference between each of the first initial motion state and the second motion state corresponding to the first initial motion state is obtained. It can be understood that the larger the first difference is, the worse the applicability of the initial dynamic model to the first sub-test data is; the smaller the first difference is, the better the applicability of the initial dynamic model to the first sub-test data is.

[0099] Therefore, when it is detected that at least a specified number of first differences are greater than a first threshold, the initial kinetic model can be trained based on the first subtest data to obtain an intermediate kinetic model. The specified number can be flexibly set according to demand and is not specifically limited in the embodiments of the present application.

[0100] Among them, the introduction of training the initial dynamics model through the first sub-test data is similar to the aforementioned method of training the initial model through the first test data, and will not be repeated here.

[0101] In the case that it is not detected that at least a specified number of first differences are greater than the first threshold, the initial kinetic model can be directly used as the intermediate kinetic model.

[0102] Therefore, not only can some training processes be reduced and the efficiency of obtaining the intermediate dynamics model be improved, but also the applicability of the obtained intermediate dynamics model to the first sub-test data will hardly be reduced.

[0103] Step S2110: training the intermediate dynamics model based on the second sub-test data to obtain the target dynamics model.

[0104] After obtaining the second sub-test data, the intermediate dynamics model can be further trained based on the second sub-test data to obtain the target dynamics model. Thus, the target dynamics model obtained after training can not only better adapt to the first sub-test data representing a lower driving intensity, but also better adapt to the second sub-test data representing a higher driving intensity.

[0105] For some implementations, after the second sub-test data are acquired, the intermediate dynamics model may be trained by each second sub-test data to obtain the target dynamics model.

[0106] For some other embodiments, before the intermediate kinetic model is trained using the second sub-test data, the second sub-test data may be further screened to extract part of the test data with lower applicability of the intermediate kinetic model to the second sub-test data to train the intermediate kinetic model. Specifically, step S2110 may also include steps S2111 to S2114.

[0107] Step S2111: obtaining a second initial motion state corresponding to a second control signal of each second sub-test data through the intermediate dynamics model.

[0108] Step S2112: Obtain a second difference between each of the second initial motion states and a second motion state corresponding to the second initial motion state.

[0109] Step S2113: taking the second difference greater than the second threshold as the target second difference.

[0110] Step S2114: training the intermediate dynamics model based on the second sub-test data corresponding to the target second difference to obtain the target dynamics model.

[0111] Similar to the above steps, the second initial motion state corresponding to the second control signal of each second sub-test data can first be obtained through the intermediate dynamic model. Then, the second difference between each second initial motion state and the second motion state corresponding to the second initial motion state is obtained. It can be understood that the larger the second difference is, the worse the applicability of the intermediate dynamic model to the second sub-test data is; the smaller the second difference is, the better the applicability of the intermediate dynamic model to the second sub-test data is.

[0112] Therefore, the second sub-test data can be screened by the second difference. Specifically, the second difference greater than the second threshold value can be used as the target second difference. Then, the intermediate dynamic model is trained based on the second sub-test data corresponding to the target second difference, so as to obtain the target dynamic model. Among them, the method of training the intermediate dynamic model by the second sub-test data corresponding to the target second difference is similar to the method of training the initial model by the first test data, which will not be repeated here.

[0113] Exemplarily, a model optimization data set may be constructed, thereby adding the second sub-test data corresponding to the target second difference into the model optimization data set, and subsequently the intermediate dynamics model may be trained directly through the model optimization data set.

[0114] Therefore, the intermediate dynamics model can be trained more specifically through the second sub-test data with lower applicability, thereby improving the efficiency of training the intermediate dynamics model.

[0115] The model training method provided in the embodiment of the present application trains the initial kinetic model based on the first sub-test data to obtain an intermediate kinetic model; and trains the intermediate kinetic model based on the second sub-test data to obtain the target kinetic model. Not only can some training processes be reduced and the efficiency of obtaining the intermediate kinetic model be improved, but also the applicability of the obtained intermediate kinetic model to the first sub-test data is hardly reduced. Moreover, the intermediate kinetic model can be trained more specifically through the second sub-test data with lower applicability, which also improves the efficiency of training the intermediate kinetic model.

[0116] See also Figure 4 , Figure 4 A method flow chart of a model training method provided in an embodiment of the present application is shown. The model training method can be applied to Figure 1 The electronic device in the video generation scene shown can specifically use the processor of the electronic device as the execution subject of the model training method. The model training method can include steps S310 to S380.

[0117] Step S310: Acquire a third control signal of the target vehicle running in a real test scenario, third state information corresponding to the third control signal, and chassis parameters of a chassis system of the target vehicle.

[0118] Step S320: adjusting the parameters of the vehicle model based on the third control signal and the chassis parameters to obtain a target vehicle model that matches the target vehicle, wherein the vehicle model is a dynamic model based on mechanism drive.

[0119] It is understandable that, for a real vehicle of a certain model, even if the same control signal is input, the corresponding motion state of the real vehicle may be different. Therefore, in order to improve the dynamic characteristics of the virtual vehicle simulating the target vehicle and improve the accuracy of the simulation by using the virtual vehicle as the corresponding target vehicle in reality, the virtual vehicle can be first calibrated with parameters by the target vehicle. Among them, the virtual vehicle simulates the target vehicle through the target vehicle model, that is, the virtual vehicle converts the input control signal into the motion state through the target vehicle model, thereby achieving the effect of simulating the target vehicle. Thus, the virtual vehicle is calibrated with parameters, which is essentially the adjustment of the parameters of the target vehicle model.

[0120] For some implementations, the third control signal of the target vehicle running in the real test scenario and the third state information corresponding to the third control signal may be first obtained. The method of obtaining the third control signal and the third state information is similar to the method of collecting the second control signal and the second state information of the target vehicle running in the real scenario in the aforementioned embodiment, and will not be repeated here.

[0121] Furthermore, chassis parameters of the chassis system of the target vehicle may also be obtained. Exemplarily, the chassis system may include a powertrain system, a suspension system, a body system, etc., and thus the chassis parameters may include parameters of the powertrain system, parameters of the suspension system, parameters of the body system, etc.

[0122] Among them, the chassis parameters of the chassis system of the target vehicle can be obtained through supplier provision or through actual measurement, and the embodiments of the present application do not make specific limitations.

[0123] Furthermore, the parameters of the vehicle model can be adjusted based on the third control signal and the chassis parameters to obtain a target vehicle model that matches the target vehicle, wherein the vehicle model is a dynamic model based on mechanism drive. For example, the vehicle model can be a high-precision dynamic model established based on mechanism drive, and the mechanism drive can be a drive model established based on calculus data. The chassis-related parameters in the vehicle model can be adjusted by the chassis parameters obtained above, so that the adjusted vehicle model can better simulate the dynamic characteristics of the target vehicle.

[0124] Then, the adjusted vehicle model can be trained by the third control signal and the third state information corresponding to the third control signal, so as to obtain a target vehicle model matching the target vehicle. The method of training the adjusted vehicle model by the third control signal and the third state information corresponding to the third control signal is similar to the method of training the initial model based on the first test data in the aforementioned embodiment to obtain the initial dynamics model, and will not be repeated here.

[0125] Step S330: generating a plurality of first control signals for the virtual vehicle in the virtual test scenario based on a simulation application.

[0126] Step S340: in the virtual test scenario, obtaining, through the target vehicle model, a first motion state that matches each first control signal for the virtual vehicle.

[0127] Step S350: taking each of the first control signals and each of the first motion states matched by the first control signals as the first test data.

[0128] From the foregoing introduction, it can be seen that a virtual test scenario can be constructed through a simulation application, so that multiple first control signals for the virtual vehicle can be generated in the virtual test scenario based on the simulation application. For example, the generated first control signal can include a steering angle, an accelerator pedal signal, a brake pedal signal, and a vehicle gear position, etc.

[0129] Thus, in the virtual test scenario, the first motion state matched by each first control signal for the virtual vehicle is obtained through the target vehicle model, and the first control signal and the first motion state matched by the first control signal can be used as the first test data.

[0130] Step S360: Training the initial model based on the first test data to obtain an initial dynamic model, wherein the initial model is used to obtain the motion state of a target vehicle based on a control signal obtained by analyzing the data-driven algorithm, and the target vehicle is a real vehicle corresponding to the virtual vehicle.

[0131] Step S370: Acquire multiple second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal as second test data.

[0132] Step S380: training the initial dynamics model based on the second test data to obtain a target dynamics model.

[0133] Among them, steps S360 to S380 have been introduced in detail in the above embodiments and will not be repeated here.

[0134] The model training method provided in the embodiment of the present application obtains the third control signal of the target vehicle running in the real test scene, the third state information corresponding to the third control signal, and the chassis parameters of the chassis system of the target vehicle; adjusts the parameters of the vehicle model based on the third control signal and the chassis parameters to obtain a target vehicle model matching the target vehicle, wherein the vehicle model is a dynamic model driven by a mechanism; then generates multiple first control signals for the virtual vehicle in the virtual test scene based on the simulation application; in the virtual test scene, obtains the first motion state matching each first control signal for the virtual vehicle through the target vehicle model; thereby taking each first control signal and each first motion state matching the first control signal as the first test data. That is to say, the first test data obtained is obtained in the virtual test scene by the target vehicle model after adjustment and training with the measured data of the target vehicle and the chassis parameters, so that the virtual vehicle represented by the target vehicle model simulates the dynamic characteristics of the target vehicle in reality with high accuracy, so that the first test data obtained in the virtual test scene can better represent the dynamic characteristics of the target vehicle in reality, thereby improving the accuracy of the initial dynamic model obtained by subsequent training of the initial model based on the first test data. At the same time, because the first test data are all obtained in the virtual test scene, there is no need to conduct actual road sampling through the target vehicle, which can reduce the cost of training the initial model.

[0135] See also Figure 5 , Figure 5 A structural block diagram of a model training device provided in an embodiment of the present application is shown. The model training device 500 includes: a first acquisition unit 510, a first training unit 520, a second acquisition unit 530 and a second training unit 540.

[0136] The first acquisition unit 510 is used to acquire a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as first test data.

[0137] Optionally, the first acquisition unit 510 can also be used to obtain a third control signal of the target vehicle running in a real test scenario, third state information corresponding to the third control signal, and chassis parameters of the chassis system of the target vehicle; adjust the parameters of the vehicle model based on the third control signal and the chassis parameters to obtain a target vehicle model that matches the target vehicle, wherein the vehicle model is a mechanism-driven dynamic model; in the virtual test scenario, obtain the first motion state that matches each first control signal for the virtual vehicle through the target vehicle model; and use each first control signal and each first motion state that matches the first control signal as the first test data.

[0138] Optionally, the first acquisition unit 510 may also be configured to generate a plurality of first control signals for the virtual vehicle in the virtual test scenario based on a simulation application.

[0139] The first training unit 520 is used to train the initial model based on the first test data to obtain an initial dynamic model, wherein the initial model is used to obtain the motion state of the target vehicle based on the control signal obtained by the data-driven algorithm analysis, and the target vehicle is a real vehicle corresponding to the virtual vehicle.

[0140] Optionally, the first training unit 520 can also be used to obtain a third initial motion state corresponding to the first control signal through the initial model; obtain a third difference between the third initial motion state and the first motion state corresponding to the third initial motion state; train the initial model based on the third difference so that the third difference obtained by the trained initial model is less than a third threshold; and use the trained initial model as the initial dynamic model.

[0141] The second acquisition unit 530 is used to acquire a plurality of second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal as second test data.

[0142] The second training unit 540 is used to train the initial dynamics model based on the second test data to obtain a target dynamics model.

[0143] Optionally, the second training unit 540 can also be used to obtain the driving intensity of the target vehicle represented by each second motion state; use the second motion state with a driving intensity less than or equal to a threshold level and the second control signal corresponding to the second motion state as the first sub-test data; use the second motion state with a driving intensity greater than a threshold level and the second control signal corresponding to the second motion state as the second sub-test data; train the initial dynamic model based on the first sub-test data to obtain an intermediate dynamic model; and train the intermediate dynamic model based on the second sub-test data to obtain the target dynamic model.

[0144] Optionally, the second training unit 540 can also be used to obtain the first initial motion state corresponding to the second control signal of each of the first sub-test data through the initial dynamic model; obtain the first difference between each of the first initial motion state and the second motion state corresponding to the first initial motion state; and when it is detected that at least a specified number of first differences are greater than a first threshold, train the initial dynamic model based on the first sub-test data to obtain an intermediate dynamic model.

[0145] Optionally, the second training unit 540 can also be used to obtain the second initial motion state corresponding to the second control signal of each second sub-test data through the intermediate dynamic model; obtain the second difference between each second initial motion state and the second motion state corresponding to the second initial motion state; take the second difference greater than the second threshold as the target second difference; and train the intermediate dynamic model based on the second sub-test data corresponding to the target second difference to obtain the target dynamic model.

[0146] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0147] In several embodiments provided in the present application, the coupling between the units may be electrical, mechanical or other forms of coupling. In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0148] See also Figure 6 , Figure 6The block diagram of a structure of an electronic device provided in an embodiment of the present application is shown. The electronic device 110 may be a smart phone, a desktop computer, a vehicle-mounted computer, a server or a tablet computer, etc. The electronic device 110 in the present application may include one or more of the following components: a processor 111, a memory 112 and one or more application programs, wherein the processor 111 is electrically connected to the memory 112, and the one or more program configurations are used to execute the methods described in the aforementioned embodiments.

[0149] The processor 111 may include one or more processing cores. The processor 111 uses various interfaces and lines to connect various parts of the entire electronic device 110, and executes various functions and processes data of the electronic device 110 by running or executing instructions, programs, code sets or instruction sets stored in the memory 112, and calling data stored in the memory 112. Optionally, the processor 111 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 111 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and computer programs; the GPU is responsible for rendering and drawing display content; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 111, and it can be implemented separately through a communication chip. Specifically, the method described in the above embodiment can be executed by one or more processors 111.

[0150] For some embodiments, the memory 112 may include a random access memory (RAM) or a read-only memory (ROM). The memory 112 may be used to store instructions, programs, codes, code sets, or instruction sets. The memory 112 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the following various method embodiments, etc. The data storage area may also store data created by the electronic device 110 during use, etc.

[0151] See also Figure 7, which shows a structural block diagram of a computer-readable storage medium provided in an embodiment of the present application. The computer-readable medium 700 stores program codes, which can be called by a processor to execute the method described in the above method embodiment.

[0152] The computer readable storage medium 700 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk, or a ROM. Optionally, the computer readable storage medium 700 includes a non-transitory computer-readable storage medium. The computer readable storage medium 700 has storage space for program code 710 that performs any method step in the above method. These program codes can be read from or written to one or more computer program products. The program code 710 can be compressed, for example, in an appropriate form.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A model training method, characterized in that: include: Acquire a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as first test data; Training an initial model based on the first test data to obtain an initial dynamics model, wherein the initial model is used to obtain a motion state of a target vehicle based on a control signal obtained by analyzing a data-driven algorithm, and the target vehicle is a real vehicle corresponding to the virtual vehicle; Acquire a plurality of second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal as second test data; The initial kinetic model is trained based on the second test data to obtain a target kinetic model.

2. The method according to claim 1, characterized in that The training of the initial kinetic model based on the second test data to obtain a target kinetic model includes: Obtaining a driving intensity of the target vehicle represented by each second motion state; The second motion state in which the driving intensity is less than or equal to the threshold and the second control signal corresponding to the second motion state are used as the first sub-test data; The second motion state in which the driving intensity is greater than the threshold and the second control signal corresponding to the second motion state are used as the second sub-test data; Training the initial kinetic model based on the first sub-test data to obtain an intermediate kinetic model; The intermediate dynamics model is trained based on the second sub-test data to obtain the target dynamics model.

3. The method according to claim 2, characterized in that The step of training the initial kinetic model based on the first sub-test data to obtain an intermediate kinetic model comprises: Acquire a first initial motion state corresponding to a second control signal of each of the first sub-test data through the initial dynamics model; Acquire a first difference between each of the first initial motion states and a second motion state corresponding to the first initial motion state; When it is detected that at least a specified number of first differences are greater than a first threshold, the initial kinetic model is trained based on the first sub-test data to obtain an intermediate kinetic model.

4. The method according to claim 2, characterized in that: The step of training the intermediate dynamics model based on the second sub-test data to obtain the target dynamics model includes: Acquire a second initial motion state corresponding to a second control signal of each second sub-test data through the intermediate dynamics model; Acquire a second difference between each of the second initial motion states and a second motion state corresponding to the second initial motion state; taking a second difference greater than a second threshold as a target second difference; The intermediate dynamics model is trained based on the second sub-test data corresponding to the target second difference to obtain the target dynamics model.

5. The method according to claim 1, characterized in that The step of obtaining a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as first test data includes: Acquire a third control signal of the target vehicle running in a real test scenario, third state information corresponding to the third control signal, and chassis parameters of a chassis system of the target vehicle; adjusting the parameters of the vehicle model based on the third control signal and the chassis parameters to obtain a target vehicle model that matches the target vehicle, wherein the vehicle model is a dynamic model based on mechanism drive; In the virtual test scenario, obtaining, by means of the target vehicle model, a first motion state that matches each first control signal for the virtual vehicle; Each of the first control signals and the first motion state matched by each of the first control signals are used as the first test data.

6. The method according to claim 5, characterized in that Before acquiring each first motion state matched with the first control signal of the virtual vehicle through the target vehicle model in the virtual test scenario, the method further includes: A plurality of first control signals for the virtual vehicle are generated in the virtual test scenario based on a simulation application.

7. The method according to claim 1, characterized in that The step of training the initial model based on the first test data to obtain an initial dynamics model includes: Acquire a third initial motion state corresponding to the first control signal through the initial model; Acquire a third difference between the third initial motion state and a first motion state corresponding to the third initial motion state; Training the initial model based on the third difference so that the third difference obtained by the trained initial model is less than a third threshold; The trained initial model is used as the initial dynamics model.

8. A model training device, characterized in that: include: A first acquisition unit, used to acquire a plurality of first control signals of a virtual vehicle running in a virtual test scene and a first motion state matching each first control signal as first test data; a first training unit, configured to train an initial model based on the first test data to obtain an initial dynamics model, wherein the initial model is used to obtain a motion state of a target vehicle based on a control signal obtained by analyzing a data-driven algorithm, wherein the target vehicle is a real vehicle corresponding to the virtual vehicle; A second acquisition unit, used to acquire a plurality of second control signals of the target vehicle running in a real test scenario and a second motion state matching each second control signal as second test data; The second training unit is used to train the initial dynamics model based on the second test data to obtain a target dynamics model.

9. An electronic device, characterized in that: include: one or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, which can be called by a processor to execute the method according to any one of claims 1 to 7.

Images