Vehicle kinetic model establishment method and device and storage medium
By collecting and training a vehicle dynamics model based on recursive neural networks, the problem of inconsistent response of the vehicle dynamics model in the prior art is solved, and the prediction accuracy of the vehicle control state is improved.
Patent Information
- Application Number
- CN202411044774.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2025-04-18
AI Technical Summary
In the prior art, the modeling method of vehicle dynamics model has the problem that the response is inconsistent with the actual operating state, resulting in large errors between the calculation results and the actual operation, and the error accumulated over time will become larger and larger.
By collecting dynamic data of different control states of the vehicle in different driving scenarios, training is performed using a model based on recursive neural networks to establish a more accurate dynamic model. This method combines the powerful learning ability of recurrent neural networks to learn dynamic responses of vehicles under different conditions.
The accuracy of the dynamic model for vehicle control state prediction is improved, and the model overfitting certain specific situations during the training process is reduced, thereby achieving more comprehensive vehicle behavior learning and prediction.
Smart Images
Figure CN119849281A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of artificial intelligence technology, and in particular, relates to a method, device and storage medium for establishing a vehicle dynamics model. Background Art
[0002] Before the automatic driving assistance function can meet the requirements of mass production of actual vehicles, a lot of parameter adjustment and road testing are required. If simulation can be used to help with algorithm tuning and iteration, it will save a lot of time for R&D and actual vehicle matching. However, this will rely on a high-fidelity vehicle dynamics model, that is, the dynamics model response of the simulation after the control command is input should be as consistent as possible with the motion state of the actual vehicle.
[0003] In the related art, the modeling methods for vehicle dynamics models include rule-based dynamics models, which describe the dynamic behavior of the vehicle through a series of predefined rule parameters, such as physical parameters, environmental parameters, etc. In practical applications, there is still a large error between the theoretical calculation results and the actual operation of the vehicle, especially as time accumulates, the cumulative error will become larger and larger. Summary of the invention
[0004] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a method, device and storage medium for establishing a dynamic model of a vehicle to improve the accuracy of the dynamic model in predicting the vehicle control state.
[0005] In a first aspect, the present application provides a method for establishing a dynamics model of a vehicle, comprising:
[0006] Collect dynamic data of the vehicle in different control states in different driving scenarios;
[0007] The prediction of the control state of the vehicle is taken as a learning task, and the dynamics data is used as a training set to train a model built based on a recursive neural network to obtain a dynamics model of the vehicle.
[0008] According to the method for establishing a dynamic model of a vehicle of the present application, the dynamic data of the vehicle in different control states are collected; the control state of the vehicle is predicted as a learning task, and the dynamic data is used as a training set to train the model built based on the recursive neural network to obtain the dynamic model of the vehicle. The embodiment of the present application provides a wealth of training data for the model by collecting the performance of the vehicle under various control states rather than regularized parameters, and combines the powerful learning ability of the recursive neural network, so that the model can learn the dynamic response of the vehicle under different conditions, thereby improving the accuracy of the dynamic model in predicting the vehicle control state.
[0009] According to one embodiment of the present application, the dynamic data includes at least one of speed, acceleration, steering wheel angle, yaw rate, throttle depth, and brake depth.
[0010] According to one embodiment of the present application, before using the kinetic data as a training set to train a model built based on a recursive neural network, the method includes:
[0011] monitoring the distribution of said kinetic data;
[0012] In the case where the distribution of the dynamics data does not satisfy a preset condition, the control state of the vehicle is adjusted based on the distribution of the dynamics data so that the distribution of the dynamics data satisfies the preset condition.
[0013] In this embodiment, the collected vehicle dynamics data is distributed and monitored to evaluate whether it meets preset conditions. If the data distribution is uneven, the data under certain control states may be too concentrated or lacking. By optimizing the data distribution, the model can obtain more balanced data support, reduce deviations, and avoid overfitting of the model to certain specific situations during training, thereby further improving the accuracy of the dynamics model's prediction of vehicle control states.
[0014] According to an embodiment of the present application, adjusting the control state of the vehicle based on the distribution of the dynamics data includes:
[0015] Determine a target control state according to the distribution of the kinetic data; wherein the target control state is a control state corresponding to a data amount of the kinetic data in the kinetic data distribution being lower than an average data amount;
[0016] The control state of the vehicle is adjusted to the target control state to increase the data volume of the dynamics data corresponding to the target control state.
[0017] In this embodiment, by analyzing the distribution of dynamic data, specific control states where the data volume is below the average level are identified, and then by adjusting the vehicle's control state so that the vehicle operates under these target control states, the data volume of the dynamic data under the corresponding state is increased, so that the model can obtain more balanced data support.
[0018] According to one embodiment of the present application, the preset conditions include:
[0019] The difference in the data volume of the dynamics data corresponding to different control states is lower than a preset value.
[0020] In this embodiment, by limiting the difference in data volume under different control states to no more than a preset value, data overload or shortage under certain specific conditions or operating modes can be avoided, thereby improving the balance of the data and helping the model learn more comprehensive vehicle behavior.
[0021] According to an embodiment of the present application, the preset condition further satisfies: the difference in the amount of dynamic data corresponding to different driving scenarios is lower than a preset value.
[0022] In this embodiment, by limiting the difference in data volume under different driving scenarios to no more than a preset value, data overload or shortage under certain specific conditions or operating modes can be avoided, thereby improving the balance of the data and helping the model learn more comprehensive vehicle behavior.
[0023] According to an embodiment of the present application, the control state includes lateral control and / or longitudinal control of the vehicle.
[0024] According to an embodiment of the present application, the lateral control includes a steering wheel angle and / or a yaw rate; and the longitudinal control includes acceleration.
[0025] In this embodiment, the control state covers the lateral control and / or longitudinal control of the vehicle. This comprehensive control state coverage enables the model to predict the dynamic behavior of the vehicle, thereby constructing a more comprehensive and accurate dynamic model.
[0026] According to one embodiment of the present application, the step of using the dynamics data as a training set to train a model built based on a recursive neural network to obtain a dynamics model of the vehicle includes:
[0027] Sampling from the kinetic data at a preset time step to obtain a training set;
[0028] The data of the past k time steps in the training set are used as input to train the model built based on the recursive neural network to obtain the dynamic model of the vehicle.
[0029] According to one embodiment of the present application, the method further includes:
[0030] The time step is determined according to the delay between the issuance of the control command and the response of the vehicle chassis.
[0031] In this embodiment, the recursive neural network model is trained by using data of k time steps, taking into account the delay between the issuance of the control command and the response of the vehicle chassis and the actual sampling frequency of the vehicle body signal, so that the model can process and learn data corresponding to the actual operating conditions of the vehicle, effectively capture and learn the time dependency and sequence information in the vehicle dynamics behavior, and further improve the accuracy of the dynamics model in predicting the vehicle control state.
[0032] According to an embodiment of the present application, determining the time step according to the delay between the issuance of the control command and the response of the vehicle chassis includes:
[0033] According to the time when the control command is issued and the time when the chassis responds, the delay between the issuance of the control command and the response of the vehicle chassis is calculated;
[0034] The time step corresponding to the delay is matched from a preset association relationship; wherein the association relationship represents the correspondence between the delay and the time step.
[0035] In this embodiment, by testing the issuance of control commands and the actual response of the chassis, the exact delay between the two is calculated, and the most appropriate time step is matched according to the preset correlation between the delay and the time step, so that the update frequency of the model can match the actual response speed of the vehicle, so that the update of the control command is synchronized with the dynamic changes of the vehicle, thereby reducing control errors and improving the real-time and accuracy of control.
[0036] According to one embodiment of the present application, the model constructed based on the recursive neural network includes: a long short-term memory network containing a target number of hidden units, a nonlinear activation function connected after the long short-term memory network, and a fully connected layer.
[0037] In this embodiment, by adopting a long short-term memory network containing a target number of hidden units as the core structure of the model, combined with a nonlinear activation function and a fully connected layer, this model configuration provides a powerful nonlinear expression capability for the dynamic model, which not only improves the model's fit to complex dynamic behaviors, but also enhances its generalization ability, enabling the model to more accurately predict the control state of the vehicle in various driving scenarios.
[0038] According to one embodiment of the present application, the method further includes:
[0039] Collect target dynamics data of the target vehicle model during driving in different driving scenarios with different control states;
[0040] The target dynamics data is used to train the dynamics model so that the trained dynamics model is suitable for predicting the control state of the target vehicle model.
[0041] In this embodiment, for a specific vehicle model, it is only necessary to collect the target dynamics data of the vehicle model and train the model to quickly adapt to and accurately predict the control state of the vehicle model. There is no need to train the entire model from scratch to achieve the reuse of simulation models for different vehicle models, which significantly reduces the need for separate road testing and debugging for each vehicle model, thereby saving a lot of development time and resources.
[0042] According to one embodiment of the present application, the method further includes:
[0043] Inputting vehicle dynamics data from actual vehicle road tests into the dynamics model to obtain a control state predicted by the dynamics model output;
[0044] The prediction performance of the dynamics model is evaluated based on the actual control state and the predicted control state in the real vehicle road test.
[0045] In this embodiment, the model is evaluated by feeding the vehicle dynamics data from the road test back into the dynamics model, so that the accuracy and reliability of the model in predicting the control state can be quickly identified without conducting actual vehicle testing, avoiding repeatedly sending the vehicle with optimized control parameters to the test road for verification, thereby significantly reducing the R&D cycle and cost.
[0046] In a second aspect, the present application provides a vehicle dynamics model establishment device, comprising:
[0047] The acquisition module is used to collect dynamic data of different control states of the vehicle;
[0048] The training module is used to take the prediction of the control state of the vehicle as a learning task, and use the dynamic data as a training set to train a model built based on a recursive neural network to obtain a dynamic model of the vehicle.
[0049] According to the vehicle dynamics model building device of the present application, the dynamics data of the vehicle in different control states are collected; the control state of the vehicle is predicted as a learning task, and the dynamics data is used as a training set to train the model built based on the recursive neural network to obtain the vehicle's dynamics model. The embodiment of the present application provides a wealth of training data for the model by collecting the performance of the vehicle under a variety of control states, rather than regularized parameters, and combines the powerful learning ability of the recursive neural network, so that the model can learn the dynamic response of the vehicle under different conditions, thereby improving the accuracy of the dynamic model in predicting the vehicle's control state.
[0050] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for establishing a dynamic model of a vehicle as described in the first aspect above is implemented.
[0051] In a fourth aspect, the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for establishing a dynamic model of a vehicle as described in the first aspect above is implemented.
[0052] In a fifth aspect, the present application provides a chip, comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the method for establishing a dynamic model of a vehicle as described in the first aspect above.
[0053] In a sixth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the method for establishing a vehicle dynamics model as described in the first aspect above.
[0054] The above one or more technical solutions in the embodiments of the present application have at least one of the following technical effects:
[0055] According to the method for establishing a dynamic model of a vehicle of the present application, the dynamic data of the vehicle in different control states are collected; the control state of the vehicle is predicted as a learning task, and the dynamic data is used as a training set to train the model built based on the recursive neural network to obtain the dynamic model of the vehicle. The embodiment of the present application provides a wealth of training data for the model by collecting the performance of the vehicle under various control states rather than regularized parameters, and combines the powerful learning ability of the recursive neural network, so that the model can learn the dynamic response of the vehicle under different conditions, thereby improving the accuracy of the dynamic model in predicting the vehicle control state.
[0056] In addition, the dynamic model established in the embodiment of the present application is trained using a wide range of vehicle dynamics data to capture the basic behaviors and patterns of the vehicle in different driving scenarios. For a specific vehicle model, it is only necessary to collect the target dynamics data of the vehicle model in a specific scenario and train the model to quickly adapt to and accurately predict the control state of the vehicle model, thereby realizing the reuse of simulation models of different vehicle models, reducing road test debugging, and saving development time.
[0057] Furthermore, in some embodiments, the collected vehicle dynamics data is distributed and monitored to evaluate whether it meets the preset conditions. If the data distribution is uneven, the data under certain control states may be too concentrated or lacking. By optimizing the data distribution, the model can obtain more balanced data support, reduce deviations, avoid overfitting of the model to certain specific situations during the training process, and further improve the accuracy of the dynamics model in predicting the vehicle control state.
[0058] Furthermore, in some embodiments, by analyzing the distribution of dynamic data, specific control states where the data volume is below the average level are identified, and then by adjusting the control state of the vehicle so that the vehicle operates under these target control states, the data volume of the dynamic data under the corresponding states is increased, so that the model can obtain more balanced data support.
[0059] Furthermore, in some embodiments, by limiting the difference in the amount of data under different control states to no more than a preset value, data overload or shortage under certain specific conditions or operating modes can be avoided, thereby improving the balance of the data and helping the model learn more comprehensive vehicle behavior.
[0060] Furthermore, in some embodiments, by limiting the difference in data volume under different driving scenarios to no more than a preset value, data overload or shortage under certain conditions or operating modes can be avoided, thereby improving the balance of the data and helping the model learn more comprehensive vehicle behavior.
[0061] Furthermore, in some embodiments, the control state covers the lateral control and / or longitudinal control of the vehicle. Such comprehensive control state coverage enables the model to predict the dynamic behavior of the vehicle, thereby constructing a more comprehensive and accurate dynamic model.
[0062] Furthermore, in some embodiments, by using data of k time steps to train the recursive neural network model, the delay between the issuance of the control command and the response of the vehicle chassis and the actual sampling frequency of the vehicle body signal are taken into account, so that the model can process and learn data corresponding to the actual operating conditions of the vehicle, effectively capture and learn the time dependency and sequence information in the vehicle dynamics behavior, and further improve the accuracy of the dynamics model in predicting the vehicle control state.
[0063] Furthermore, in some embodiments, by testing the issuance of control commands and the actual response of the chassis, the exact delay between the two is calculated, and the most appropriate time step is matched according to the preset correlation between the delay and the time step, so that the update frequency of the model can be matched with the actual response speed of the vehicle, so that the update of the control command is synchronized with the dynamic changes of the vehicle, thereby reducing control errors and improving the real-time and accuracy of control.
[0064] Furthermore, in some embodiments, by adopting a long short-term memory network containing a target number of hidden units as the core structure of the model, combined with a nonlinear activation function and a fully connected layer, this model configuration provides a powerful nonlinear expression capability for the dynamic model, which not only improves the model's fit to complex dynamic behaviors, but also enhances its generalization ability, enabling the model to more accurately predict the control state of the vehicle in various driving scenarios.
[0065] Furthermore, in some embodiments, for a specific vehicle model, it is only necessary to collect the target dynamics data of the vehicle model and train the model to quickly adapt to and accurately predict the control state of the vehicle model. There is no need to train the entire model from scratch to achieve the reuse of simulation models for different vehicle models, which significantly reduces the need for separate road testing and debugging of each vehicle model, thereby saving a lot of development time and resources.
[0066] Furthermore, in some embodiments, the model is evaluated by feeding the vehicle dynamics data from the road test back into the dynamics model, so that the accuracy and reliability of the model in predicting the control state can be quickly identified without conducting actual vehicle testing, avoiding repeatedly sending the vehicle with optimized control parameters to the test road for verification, thereby significantly reducing the R&D cycle and cost.
[0067] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:
[0069] Figure 1 It is a flow chart of a method for establishing a dynamics model of a vehicle provided in an embodiment of the present application;
[0070] Figure 2 It is a schematic diagram of the architecture of the kinetic model provided in the embodiment of the present application;
[0071] Figure 3 is a schematic diagram of the kinetic model evaluation process provided in the embodiments of the present application;
[0072] Figure 4 It is one of the schematic diagrams of the model performance evaluation process provided in the embodiment of the present application;
[0073] Figure 5 This is the second schematic diagram of the model performance evaluation process provided in the embodiment of the present application;
[0074] Figure 6 It is a structural schematic diagram of a vehicle dynamics model establishment device provided in an embodiment of the present application;
[0075] Figure 7 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0076] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.
[0077] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.
[0078] In the following, in conjunction with the accompanying drawings, the vehicle dynamics model establishment method, device and storage medium provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.
[0079] Among them, the method for establishing the vehicle's dynamic model can be applied to a terminal, and can be specifically executed by hardware or software in the terminal.
[0080] The terminal includes, but is not limited to, a portable communication device such as a mobile phone or tablet computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad). It should also be understood that in some embodiments, the terminal may not be a portable communication device, but a desktop computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad).
[0081] In the following various embodiments, a terminal including a display and a touch-sensitive surface is described. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, a mouse and a joystick.
[0082] The method for establishing a dynamic model of a vehicle provided in an embodiment of the present application may be executed by an electronic device or a functional module or functional entity in the electronic device that can implement the method for establishing a dynamic model of the vehicle. The electronic devices mentioned in the embodiment of the present application include but are not limited to mobile phones, tablet computers, computers, cameras, and wearable devices. The method for establishing a dynamic model of a vehicle provided in an embodiment of the present application is described below using an electronic device as an example of an execution subject.
[0083] like Figure 1As shown, the method for establishing the vehicle dynamics model includes: step 110 and step 120.
[0084] Step 110: Collect dynamic data of the vehicle in different control states in different driving scenarios.
[0085] In the embodiment of the present application, the control state may be data related to vehicle control, such as vehicle speed, steering wheel angle, throttle depth, brake depth, etc. Dynamic data is data on vehicle motion characteristics collected during vehicle driving, including speed, acceleration, steering wheel angle, yaw rate, throttle depth, brake depth, etc.
[0086] In some embodiments, the control state includes the lateral control and / or longitudinal control of the vehicle. Lateral control involves the motion control of the vehicle perpendicular to its forward direction, such as steering wheel angle, yaw rate, etc.; longitudinal control involves the motion control of the vehicle along its forward direction (i.e., the direction of travel of the vehicle), such as acceleration. In this embodiment, the control state covers the lateral control and / or longitudinal control of the vehicle. This comprehensive coverage of the control state enables the model to predict the dynamic behavior of the vehicle, thereby building a more comprehensive and accurate dynamic model.
[0087] To ensure the richness of the data, as shown in Table 1, the responses of the vehicle in different control states, such as different speeds, steering wheel angles, throttle depths, brake depths, etc., can be used.
[0088] Table 1
[0089]
[0090] In the embodiment of the present application, the vehicle can be driven on an actual road by running a real vehicle, and the control state can be adjusted during driving to obtain dynamic data. For example, the vehicle speed can be obtained by a wheel speed sensor or a global positioning system (GPS) and / or a real-time kinematic (RTK) system, the yaw rate and acceleration can be obtained by an inertial measurement unit (IMU), the steering wheel angle can be obtained by a steering wheel angle sensor, and the throttle depth, brake depth and other data can be obtained by a controller area network (CAN) bus. Of course, the vehicle's dynamic data can also be obtained by other means, which is not limited in the embodiment of the present application.
[0091] Step 120 , taking the prediction of the control state of the vehicle as the learning task, using the dynamics data as a training set to train the model built based on the recursive neural network, and obtaining the dynamics model of the vehicle.
[0092] In the embodiment of the present application, a recurrent neural network (RNN) can be used to build a model. In the process of building the model, it is necessary to determine the number of layers of the RNN, the number of neurons in each layer, etc., define the damage function, select an optimizer for updating the network weights, etc.
[0093] In an embodiment of the present application, predicting the control state of the vehicle can be used as a learning task, that is, given dynamic data, such as vehicle speed, acceleration, steering wheel angle, yaw rate, throttle, and brake, the output is a control state, such as lateral control state and / or longitudinal control state.
[0094] In the embodiment of the present application, the kinetic data can be used as a training set to train the model built based on the recursive neural network. During the training process, the loss function is calculated based on the difference between the predicted value and the true value of the model, and the parameters of the model are updated by back propagation.
[0095] In some embodiments, the data in the training set may also be preprocessed, such as standardized or normalized, to eliminate the influence of different dimensions.
[0096] According to the method for establishing a dynamic model of a vehicle of the present application, the dynamic data of the vehicle in different control states are collected; the prediction of the control state of the vehicle is used as a learning task, and the dynamic data is used as a training set to train the model built based on the recursive neural network to obtain the dynamic model of the vehicle. The embodiment of the present application provides a wealth of training data for the model by collecting the performance of the vehicle under various control states rather than regularized parameters, and combines the powerful learning ability of the recursive neural network, so that the model can learn the dynamic response of the vehicle under different conditions, thereby improving the accuracy of the dynamic model in predicting the vehicle control state.
[0097] In some embodiments, before using the kinetic data as a training set to train a model based on a recurrent neural network, the method includes:
[0098] Monitor the distribution of kinetic data;
[0099] When the distribution of the dynamics data does not satisfy the preset condition, the control state of the vehicle is adjusted based on the distribution of the dynamics data so that the distribution of the dynamics data satisfies the preset condition.
[0100] In this embodiment, it is possible to predetermine which data needs to be collected and the importance of each type of data. For example, it may be necessary to collect more data at different vehicle speeds because the vehicle speed has a great impact on the dynamics model; it may be necessary to collect more data in scenarios where the road surface is smooth or wet; and it may be necessary to keep the amount of data collected in different scenarios similar. According to the demand for dynamics data, the requirements of the preset conditions can be defined.
[0101] In this embodiment, the distribution of the dynamic data can be monitored, for example, the distribution of the currently collected dynamic data can be displayed in real time in a visual manner. When a sufficient amount of dynamic data is collected, the control state of the vehicle can be adjusted according to the distribution of the currently collected dynamic data, or in the process of collecting dynamic data, the control state of the vehicle can be adjusted according to the distribution of the currently collected dynamic data. According to the distribution of the dynamic data, it is determined which data is missing or which data is excessive, and then the control state of the vehicle is adjusted so that the collected dynamic data meets the preset conditions.
[0102] By monitoring the distribution of the collected vehicle dynamics data, we can evaluate whether it meets the preset conditions. If the data distribution is uneven, the data under certain control states may be too concentrated or lacking. By optimizing the data distribution, the model can obtain more balanced data support, reduce deviations, and avoid overfitting of certain specific situations during the training process, thereby further improving the accuracy of the dynamics model's prediction of vehicle control states.
[0103] In some embodiments, adjusting the control state of the vehicle based on the distribution of dynamics data includes:
[0104] Determine a target control state according to the distribution of the kinetic data; wherein the target control state is a control state corresponding to a data amount of the kinetic data in the kinetic data distribution being lower than an average data amount;
[0105] The control state of the vehicle is adjusted to a target control state to increase the data volume of the dynamics data corresponding to the target control state.
[0106] In this embodiment, the data volume of the kinetic data corresponding to different control states can be determined based on the distribution of the kinetic data, and the average data volume of the kinetic data can be calculated. The control state with a data volume below the average data volume is determined as the target control state. The low data volume of the kinetic data of the target control state may affect the training effect of the model, so it is necessary to increase the data volume of the kinetic data of the target control state.
[0107] For example, assuming that during the monitoring of the distribution of dynamic data, it is found that the data at a steering wheel angle of 15 degrees is less than the average data volume, the vehicle can be instructed to perform more tests at a steering wheel angle of 15 degrees, such as increasing the number of driving times at this speed and angle, thereby increasing the data volume.
[0108] In this embodiment, by analyzing the distribution of dynamic data, specific control states where the data volume is below the average level are identified, and then by adjusting the vehicle's control state so that the vehicle operates under these target control states, the data volume of the dynamic data under the corresponding state is increased, so that the model can obtain more balanced data support.
[0109] In some embodiments, the preset conditions include:
[0110] The difference in the data volume of the dynamics data corresponding to different control states is lower than a preset value.
[0111] In this embodiment, the corresponding dynamic data under different control states are very important. Therefore, in order to make the data more balanced, the data amounts under different control states can be basically the same, that is, the difference in data amounts under different control states is less than a preset value.
[0112] In this embodiment, by limiting the difference in data volume under different driving scenarios or control states to no more than a preset value, data overload or shortage under certain specific conditions or operating modes can be avoided, thereby improving the balance of the data and helping the model learn more comprehensive vehicle behavior.
[0113] In some embodiments, the preset condition is also satisfied that the difference in the amount of dynamic data corresponding to different driving scenarios is lower than a preset value.
[0114] In this embodiment, the driving scene may include road scenes and road surface conditions during vehicle driving. Road scenes may include different types of roads such as straight roads, small curvature curves, uphill, downhill, right-angle bends, and U-shaped bends; road surface conditions may include wet, dry, smooth, etc. The vehicle may be driven in different driving environments by running on a real vehicle, and the control state may be adjusted accordingly according to the driving environment, and dynamic data may be obtained during vehicle driving.
[0115] In this embodiment, the dynamic data corresponding to different driving scenarios are very important, so in order to make the data more balanced, the data volume of different driving scenarios can be basically the same, that is, the difference in the data volume of different driving scenarios is less than a preset value. For example, if the data on a slippery road is insufficient, it can be arranged on rainy days or by artificial sprinkling to simulate a slippery road, so as to increase data collection in such a scenario.
[0116] When the difference in the amount of dynamic data corresponding to different control states is lower than a preset value, in order to maintain the balance of data under the same driving scenario, the amount of dynamic data corresponding to different control states under the same driving scenario can also be made basically the same, that is, the difference in the amount of dynamic data corresponding to different control states under the same driving scenario is smaller than a preset value.
[0117] In this embodiment, by limiting the difference in data volume under different driving scenarios to no more than a preset value, data overload or shortage under certain specific conditions or operating modes can be avoided, thereby improving the balance of the data and helping the model learn more comprehensive vehicle behavior.
[0118] In some embodiments, the dynamics data is used as a training set to train a model built based on a recursive neural network to obtain a dynamics model of the vehicle, including:
[0119] Sampling from the kinetic data at a preset time step to obtain a training set;
[0120] The data of the past k time steps in the training set are used as input to train the model built based on the recurrent neural network to obtain the vehicle's dynamic model.
[0121] In this embodiment, the vehicle's dynamic data, such as speed, acceleration, steering wheel angle, etc., and the corresponding control state, change continuously over time and can be regarded as a time series. Since there is a delay between the control command being issued and the chassis response, this chassis delay characteristic is crucial for accurately predicting vehicle dynamics. Considering the influence of this chassis delay characteristic and the influence of the actual sampling frequency of the vehicle body signal, in the dynamic data with time series characteristics, the time step can be determined according to the delay between the control command being issued and the chassis response, and the dynamic data is sampled at this time step to obtain a series of data with a time sequence relationship. For example, if the speed of the vehicle is recorded every 0.1 second, then the time step is 0.1 second.
[0122] Then, the data of the past k time steps are used as input to train the model built based on the recurrent neural network, so that during the training process, the model learns to remember the information of the past period of time and considers the data points within k time steps from the current moment. For example, if the time step is 0.1 seconds and k=5, the model will consider the data within the current moment and the previous 0.4 seconds (i.e. 4 time steps). Figure 2 As shown, during the model training process, the data for each time step can include the kinetic data {t 1,ctrl ,t 2,ctrl ,t …,ctrl ,t k-1,ctrl}, and the corresponding control state {t1,state ,t 2,state ,t …,state ,t k-1,state}, the output is the lateral control state and / or longitudinal control state at the current moment. The optimization goal is to minimize the gap between the predicted value and the true value, which is specifically expressed as follows:
[0123]
[0124] in, represents the predicted value of the model, s t Represents the true value.
[0125] In this embodiment, the recursive neural network model is trained by using data of k time steps, taking into account the delay between the issuance of the control command and the response of the vehicle chassis and the actual sampling frequency of the vehicle body signal, so that the model can process and learn data corresponding to the actual operating conditions of the vehicle, effectively capture and learn the time dependency and sequence information in the vehicle dynamics behavior, and further improve the accuracy of the dynamics model in predicting the vehicle control state.
[0126] In some embodiments, determining the time step according to the delay between the issuance of the control command and the response of the vehicle chassis includes:
[0127] According to the time when the control command is issued and the time when the chassis responds, the delay between the issuance of the control command and the response of the vehicle chassis is calculated;
[0128] The time step corresponding to the delay is matched from a preset association relationship; wherein the association relationship represents the correspondence between the delay and the time step.
[0129] In this embodiment, a series of control tests can be performed on the vehicle, such as actual vehicle tests or simulation tests, to record the time point when the control command is issued and the response time point when the chassis receives the control command and generates the action corresponding to the control command, and calculate the difference between the time point when the control command is issued and the response time point to obtain the delay from the issuance of the control command to the actual response of the chassis.
[0130] In this embodiment, the preset association relationship represents the correspondence between delay and time step, that is, each delay or delay range corresponds to a time step, and after the delay is determined, the corresponding time step can be matched through the association relationship.
[0131] If the time step is set too small, the control system may update the control command too frequently, and the response of the vehicle chassis may not keep up with this high-frequency update, thus affecting the control effect. If the time step is set too large, the control system may not be able to respond to the dynamic changes of the chassis in time, reducing the control accuracy and response speed.
[0132] In some embodiments, the relationship between the delay and the time step can be determined by referring to industry standards and combining empirical values, or the relationship between the delay and the time step can be pre-calibrated through experiments, simulations, or actual tests, so that the selected time step can meet the training requirements of the model. In some embodiments, the time step should be slightly larger than the chassis delay so that the actuator of the vehicle has enough time to respond to the control input, for example, the time step may be 1.1 times, 1.5 times, or higher than the chassis delay.
[0133] In this embodiment, by testing the issuance of control commands and the actual response of the chassis, the exact delay between the two is calculated, and the most appropriate time step is matched according to the preset correlation between the delay and the time step, so that the update frequency of the model can match the actual response speed of the vehicle, so that the update of the control command is synchronized with the dynamic changes of the vehicle, thereby reducing control errors and improving the real-time and accuracy of control.
[0134] In some embodiments, the model constructed based on the recursive neural network includes: a long short-term memory network containing a target number of hidden units, a nonlinear activation function connected after the long short-term memory network, and a fully connected layer.
[0135] like Figure 2 As shown, the model constructed by the recurrent neural network can include a long short-term memory network (LSTM) with 16 hidden units, the input is the data of the past k time steps, and the output is the lateral control state yaw (t k ) and the longitudinal control state a(t k ). Of course, the target number of hidden units may also be 12, 15, 18, 20, etc. Those skilled in the art may set it according to actual needs, and the embodiments of the present application are not limited thereto.
[0136] In this embodiment, the LSTM network can learn long-term dependencies due to its unique gating mechanism, effectively solving the gradient vanishing problem of traditional RNN when processing long sequence data. By introducing nonlinear activation functions after LSTM, such as the Rectified Linear Unit (ReLU), the model can capture more complex data patterns and enhance the ability to express nonlinear features in dynamic data. These nonlinear features may include the dynamic response of the vehicle at different speeds, steering angles, or road conditions. The fully connected layer (FC) further maps the features learned by the LSTM layer to the output space to generate a prediction of the vehicle control state.
[0137] In this embodiment, by adopting a long short-term memory network containing a target number of hidden units as the core structure of the model, combined with a nonlinear activation function and a fully connected layer, this model configuration provides a powerful nonlinear expression capability for the dynamic model, which not only improves the model's fit to complex dynamic behaviors, but also enhances its generalization ability, enabling the model to more accurately predict the control state of the vehicle in various driving scenarios.
[0138] In some embodiments, the method further comprises:
[0139] Collect target dynamics data of the target vehicle model during driving in different driving scenarios with different control states;
[0140] The dynamics model is trained using the target dynamics data so that the trained dynamics model is suitable for predicting the control state of the target vehicle model.
[0141] In this embodiment, the dynamic model can meet the prediction requirements in most cases. However, since the training data used by the dynamic model is not collected for different vehicle models, the relationship between the dynamic data and control state of different vehicle models may be different. For example, for different types of vehicles, the throttle response, brake response, etc. may be different, the dynamic model may be lacking in predicting specific vehicle models.
[0142] In order for the dynamics model to accurately predict the control state of a specific vehicle model, it is necessary to collect the dynamics data of the vehicle model during actual driving, and use the dynamics data to train the dynamics model so that the dynamics model is more suitable for the specific vehicle model. For example, if the requirement is to predict the control state of vehicle model A, the vehicle model A can be controlled to run on the road, and the control state can be adjusted during driving to obtain dynamics data.
[0143] In some embodiments, after determining the vehicle model, the different driving scenarios in which the vehicle model will be tested can be determined, such as urban roads, highways, mountain roads, rainy and snowy weather, etc., and the control states that the vehicle may adopt in each scenario can be defined. The inertial measurement unit (IMU), real-time kinematic measurement (RTK), controller area network (CAN), etc. in the vehicle are used to collect the vehicle's dynamic data in various driving scenarios and various control states.
[0144] After the target dynamics data is collected, the target dynamics data can be preprocessed, such as normalization and standardization, to meet the input requirements of the dynamics model. The dynamics model is then trained using the target dynamics data so that the dynamics model can learn the unique dynamic characteristics of the target vehicle model.
[0145] In this embodiment, since the dynamic model itself is a trained model, the model has learned some common features in the control state prediction task. In the process of training the dynamic model using the target dynamic data, the dynamic model can directly apply the previously learned features to the new learning task without starting to learn again. Therefore, during the training process, most of the parameters of the dynamic model remain unchanged, and only some parameters will be adjusted to adapt to the characteristics of the target dynamic data, thereby significantly reducing the time and data required for training. In this embodiment, the new learning task is a task of predicting the control state of the target vehicle model.
[0146] In this embodiment, for a specific vehicle model, it is only necessary to collect the target dynamics data of the vehicle model in a specific scenario and train the model to quickly adapt to and accurately predict the control state of the vehicle model. There is no need to train the entire model from scratch to achieve the reuse of simulation models for different vehicle models, which significantly reduces the need for separate road testing and debugging for each vehicle model, thereby saving a lot of development time and resources.
[0147] In some embodiments, the method further comprises:
[0148] The vehicle dynamics data from the actual vehicle road test is input into the dynamics model to obtain the control state predicted by the dynamics model output;
[0149] The prediction performance of the dynamics model is evaluated based on the actual control state and the predicted control state in the real vehicle road test.
[0150] In this embodiment, the vehicle can be tested under actual road conditions to collect vehicle dynamics data, including vehicle speed, acceleration, steering wheel angle, yaw rate, throttle depth, brake depth, etc. The model simulation performance verification can be performed using Simulink or other simulation software, such as Figure 3As shown, you can run a simulation in Simulink, use the pack re-injection method of message recording, and feed a large amount of data from the actual vehicle road scene into the dynamics model through playback, compare the control state predicted by the simulation model with the actual control state in the actual vehicle road test, and analyze the difference between the two. For example, you can calculate error indicators such as mean squared error (MSE) or mean absolute error (MAE) to analyze and compare the difference between the control state predicted by the simulation model and the actual control state in the actual vehicle road test, thereby evaluating the prediction performance of the dynamics model. If the model prediction performance does not meet expectations, the model can be optimized based on the evaluation results, including adjusting the network structure, hyperparameters or training process. Figure 4 and Figure 5 As shown in the figure, an example of the model performance evaluation process is given, including nearly 1 hour of real car road running data, with a speed range of 0-120km / h, including straight roads, S-curves, and uphill and downhill scenes. Figure 4 Compare the evaluation results of the true value Result and the predicted value Lable of acceleration. Figure 4 As can be seen from the lower part of the figure, the difference between the true value Result and the predicted value Label of the acceleration is mostly close to 0, and only a few frames have jumps in the difference; Figure 5 Compare the evaluation results of the true value Result and the predicted value Lable of the yaw angular rate. Figure 5 It can be seen from the lower part that the difference between the true value Result and the predicted value Lable of acceleration is mostly close to 0, and only a few frames have a jump in the difference.
[0151] In this embodiment, the model is evaluated by feeding the vehicle dynamics data from the road test back into the dynamics model, so that the accuracy and reliability of the model in predicting the control state can be quickly identified without conducting actual vehicle testing, avoiding repeatedly sending the vehicle with optimized control parameters to the test road for verification, thereby significantly reducing the R&D cycle and cost.
[0152] The vehicle dynamics model establishment method provided in the embodiment of the present application can be executed by a vehicle dynamics model establishment device. In the embodiment of the present application, the vehicle dynamics model establishment method is executed by a vehicle dynamics model establishment device as an example to illustrate the vehicle dynamics model establishment device provided in the embodiment of the present application.
[0153] An embodiment of the present application also provides a device for establishing a dynamic model of a vehicle.
[0154] like Figure 6 As shown, the vehicle dynamics model building device includes:
[0155] The acquisition module 610 is used to acquire dynamic data of different control states of the vehicle;
[0156] The training module 620 is used to train the model built based on the recursive neural network by taking the prediction of the control state of the vehicle as the learning task and using the dynamics data as the training set to obtain the dynamics model of the vehicle.
[0157] According to the vehicle dynamics model building device of the present application, the dynamics data of the vehicle in different control states are collected; the control state of the vehicle is predicted as a learning task, and the dynamics data is used as a training set to train the model built based on the recursive neural network to obtain the dynamics model of the vehicle. The embodiment of the present application provides rich training data for the model by collecting the performance of the vehicle under various control states rather than regularized parameters, and combines the powerful learning ability of the recursive neural network, so that the model can learn the dynamic response of the vehicle under different conditions, thereby improving the accuracy of the dynamic model in predicting the vehicle control state.
[0158] In some embodiments, the dynamics data includes at least one of speed, acceleration, steering wheel angle, yaw rate, throttle depth, and brake depth.
[0159] In some embodiments, the acquisition module 610 is further configured to:
[0160] Monitor the distribution of kinetic data;
[0161] When the distribution of the dynamics data does not satisfy the preset condition, the control state of the vehicle is adjusted based on the distribution of the dynamics data so that the distribution of the dynamics data satisfies the preset condition.
[0162] In some embodiments, the acquisition module 610 is further configured to:
[0163] Determine a target control state according to the distribution of the kinetic data; wherein the target control state is a control state corresponding to a data amount of the kinetic data in the kinetic data distribution being lower than an average data amount;
[0164] The control state of the vehicle is adjusted to a target control state to increase the data volume of the dynamics data corresponding to the target control state.
[0165] In some embodiments, the preset conditions include:
[0166] The difference in the data volume of the dynamics data corresponding to different control states is lower than a preset value.
[0167] In some embodiments, the preset condition is also satisfied that the difference in the amount of dynamic data corresponding to different driving scenarios is lower than a preset value.
[0168] In some embodiments, the control state includes lateral control and / or longitudinal control of the vehicle.
[0169] In some embodiments, lateral control includes steering wheel angle and / or yaw rate; longitudinal control includes acceleration.
[0170] In some embodiments, the training module 620 is further configured to:
[0171] Sampling from the kinetic data at a preset time step to obtain a training set;
[0172] The data of the past k time steps in the training set are used as input to train the model built based on the recurrent neural network to obtain the vehicle's dynamic model.
[0173] In some embodiments, the training module 620 is further configured to:
[0174] The time step is determined based on the delay between the control command being issued and the chassis responding.
[0175] In some embodiments, the training module 620 is further configured to:
[0176] According to the time when the control command is issued and the time when the chassis responds, the delay between the issuance of the control command and the response of the vehicle chassis is calculated;
[0177] The time step corresponding to the delay is matched from a preset association relationship; wherein the association relationship represents the correspondence between the delay and the time step.
[0178] In some embodiments, the model constructed based on the recursive neural network includes: a long short-term memory network containing a target number of hidden units, a nonlinear activation function connected after the long short-term memory network, and a fully connected layer.
[0179] In some embodiments, the training module 620 is further configured to:
[0180] Collect target dynamics data of the target vehicle model during driving in different driving scenarios with different control states;
[0181] The dynamics model is trained using the target dynamics data so that the trained dynamics model is suitable for predicting the control state of the target vehicle model.
[0182] In some embodiments, the training module 620 is further configured to:
[0183] The vehicle dynamics data from the actual vehicle road test is input into the dynamics model to obtain the control state predicted by the dynamics model output;
[0184] The prediction performance of the dynamics model is evaluated based on the actual control state and the predicted control state in the real vehicle road test.
[0185] The vehicle dynamics model building device in the embodiment of the present application can be an electronic device, or a component in the electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or other devices other than a terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmented reality, AR) / virtual reality (virtual reality, VR) device, a robot, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc., and can also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present application.
[0186] The vehicle dynamics model building device in the embodiment of the present application may be a device having an operating system. The operating system may be a Microsoft (Windows) operating system, an Android (Android) operating system, an IOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.
[0187] In some embodiments, Figure 7 As shown, an embodiment of the present application also provides an electronic device 700, including a processor 701, a memory 702, and a computer program stored in the memory 702 and executable on the processor 701. When the program is executed by the processor 701, each process of the above-mentioned vehicle dynamics model establishment method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.
[0188] It should be noted that the electronic devices in the embodiments of the present application include the above-mentioned mobile electronic devices and non-mobile electronic devices.
[0189] An embodiment of the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the above-mentioned vehicle dynamics model establishment method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
[0190] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.
[0191] An embodiment of the present application also provides a computer program product, including a computer program, which implements the above-mentioned method for establishing a dynamic model of the vehicle when executed by a processor.
[0192] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.
[0193] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned vehicle dynamics model establishment method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0194] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.
[0195] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.
[0196] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0197] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
[0198] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0199] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.
Claims
1. A method for establishing a vehicle dynamics model, characterized in that: include: Collect dynamic data of the vehicle in different control states; The prediction of the control state of the vehicle is taken as a learning task, and the dynamics data is used as a training set to train a model built based on a recursive neural network to obtain a dynamics model of the vehicle.
2. The method according to claim 1, characterized in that The dynamic data includes at least one of speed, acceleration, steering wheel angle, yaw rate, throttle depth, and brake depth.
3. The method according to claim 1, characterized in that Before using the kinetic data as a training set to train a model based on a recurrent neural network, the method includes: monitoring the distribution of said kinetic data; In the case where the distribution of the dynamics data does not satisfy a preset condition, the control state of the vehicle is adjusted based on the distribution of the dynamics data so that the distribution of the dynamics data satisfies the preset condition.
4. The method according to claim 3, characterized in that The adjusting the control state of the vehicle based on the distribution of the dynamics data comprises: Determine a target control state according to the distribution of the kinetic data; wherein the target control state is a control state corresponding to a data amount of the kinetic data in the kinetic data distribution being lower than an average data amount; The control state of the vehicle is adjusted to the target control state to increase the data volume of the dynamics data corresponding to the target control state.
5. The method according to claim 4, characterized in that The preset conditions include: The difference in the data volume of the dynamics data corresponding to different control states is lower than a preset value.
6. The method according to claim 4, characterized in that The preset condition also satisfies that: the difference in the data volume of the dynamics data corresponding to different driving scenarios is lower than a preset value.
7. The method according to claim 1, characterized in that The control state includes lateral control and / or longitudinal control of the vehicle.
8. The method according to claim 7, characterized in that The lateral control includes steering wheel angle and / or yaw rate; the longitudinal control includes acceleration.
9. The method according to claim 1, characterized in that: The method of using the dynamic data as a training set to train a model built based on a recursive neural network to obtain a dynamic model of the vehicle includes: Sampling from the kinetic data at a preset time step to obtain a training set; The data of the past k time steps in the training set are used as input to train the model built based on the recursive neural network to obtain the dynamic model of the vehicle.
10. The method according to claim 9, characterized in that The method further comprises: The time step is determined according to the delay between the issuance of the control command and the response of the vehicle chassis.
11. The method according to claim 10, characterized in that The step of determining the time step according to the delay between the issuance of the control command and the response of the vehicle chassis comprises: According to the time when the control command is issued and the time when the chassis responds, the delay between the issuance of the control command and the response of the vehicle chassis is calculated; The time step corresponding to the delay is matched from a preset association relationship; wherein the association relationship represents the correspondence between the delay and the time step.
12. The method according to claim 1, characterized in that The model constructed based on the recursive neural network includes: a long short-term memory network containing a target number of hidden units, a nonlinear activation function connected to the long short-term memory network, and a fully connected layer.
13. The method according to claim 1, characterized in that The method further comprises: Collect target dynamics data of the target vehicle model during driving in different driving scenarios with different control states; The target dynamics data is used to train the dynamics model so that the trained dynamics model is suitable for predicting the control state of the target vehicle model.
14. The method according to claim 1, characterized in that The method further comprises: Inputting vehicle dynamics data from actual vehicle road tests into the dynamics model to obtain a control state predicted by the dynamics model output; The prediction performance of the dynamics model is evaluated based on the actual control state and the predicted control state in the real vehicle road test.
15. A vehicle dynamics model building device, characterized in that: include: The acquisition module is used to collect dynamic data of different control states of the vehicle; The training module is used to take the prediction of the control state of the vehicle as a learning task, and use the dynamic data as a training set to train a model built based on a recursive neural network to obtain a dynamic model of the vehicle.
16. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 14 when executing the program.
17. A non-transitory computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the method according to any one of claims 1 to 14 is implemented.