Vehicle torque control model training method and device, equipment and storage medium

Abstract

The application provides a training method and device of a vehicle torque control model, equipment and a storage medium. The method comprises: outputting a plurality of stage features corresponding to state information through a plurality of stage networks of a residual network respectively, and outputting a first torque feature corresponding to a first control torque through a last stage network of the residual network; processing the plurality of stage features through a feature fusion network to obtain a fusion feature; processing the fusion feature and the first torque feature through a self-recurrent recurrent network to obtain a regression feature; processing the regression feature through a deep neural network to obtain a processing result corresponding to training data; calculating a classification loss between the processing result corresponding to the training data and a label through a cross-entropy loss function, and updating network parameters of the vehicle torque control model according to the classification loss to complete the training of the vehicle torque control model. The above technical means can solve the problem of unstable vehicle torque control in the prior art.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, and in particular to a training method, apparatus, device, and storage medium for a vehicle torque control model. Background Technology

[0002] Existing vehicle torque control algorithms primarily rely on engine speed and vehicle speed for control. This approach only considers a portion of the vehicle's state parameters. In reality, other state parameters such as vehicle acceleration should also be taken into account. Furthermore, since the control torque from the previous moment is related to the control torque to be applied at the current moment, the current vehicle torque control should also consider the control torque from the previous moment. Because existing technologies do not consider all vehicle state parameters and the control torque from the previous moment, they result in unstable vehicle torque control. Summary of the Invention

[0003] In view of this, embodiments of this application provide a training method, apparatus, device, and storage medium for a vehicle torque control model to solve the problem of unstable vehicle torque control in the prior art.

[0004] A first aspect of this application provides a training method for a vehicle torque control model, comprising: constructing a vehicle torque control model using a residual network, a feature fusion network, an autoregressive recurrent network, and a deep neural network; acquiring training data, wherein the training data includes the vehicle's state information at a target time and the vehicle's first control torque at the previous time before the target time, and the label corresponding to the training data is the vehicle's second control torque at the target time; inputting the training data into the vehicle torque control model: outputting multiple stage features corresponding to the state information through multiple stage networks of the residual network, and outputting the first torque feature corresponding to the first control torque through the last stage network of the residual network; processing the multiple stage features through a feature fusion network to obtain fused features; processing the fused features and the first torque feature through an autoregressive recurrent network to obtain regression features; processing the regression features through a deep neural network to obtain the processing result corresponding to the training data; calculating the classification loss between the processing result corresponding to the training data and the label through a cross-entropy loss function, and updating the network parameters of the vehicle torque control model according to the classification loss, thereby completing the training of the vehicle torque control model.

[0005] A second aspect of this application provides a training apparatus for a vehicle torque control model, comprising: a construction module configured to construct a vehicle torque control model using a residual network, a feature fusion network, an autoregressive recurrent network, and a deep neural network; an acquisition module configured to acquire training data, wherein the training data includes the vehicle's state information at a target time and the vehicle's first control torque at the previous time before the target time, and the label corresponding to the training data is the vehicle's second control torque at the target time; and a first processing module configured to input the training data into the vehicle torque control model: outputting multiple stage features corresponding to the state information through multiple stage networks of the residual network, and processing the residual network through the most... The next stage network outputs the first torque feature corresponding to the first control torque; the second processing module is configured to process the features from multiple stages through a feature fusion network to obtain fused features; the third processing module is configured to process the fused features and the first torque feature through an autoregressive recurrent network to obtain regression features; the fourth processing module is configured to process the regression features through a deep neural network to obtain the processing result corresponding to the training data; the training module is configured to calculate the classification loss between the processing result corresponding to the training data and the label through the cross-entropy loss function, and update the network parameters of the vehicle torque control model according to the classification loss to complete the training of the vehicle torque control model.

[0006] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0007] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the methods described above.

[0008] The beneficial effects of this application embodiment compared with the prior art include at least the following: by using a residual network to determine multiple stage features corresponding to the state information and a first torque feature corresponding to the first control torque, the multiple stage features are processed by a feature fusion network to obtain fused features; by using an autoregressive recurrent network to process the fused features and the first torque feature to obtain regression features; and by using a deep neural network to process the regression features to obtain the processing result corresponding to the training data. Therefore, by using the above technical means, the problem of unstable vehicle torque control in the prior art can be solved, thereby ensuring smooth vehicle driving and improving user comfort and satisfaction. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a training method for a vehicle torque control model provided in an embodiment of this application;

[0011] Figure 2 This is a flowchart illustrating another method for training a vehicle torque control model provided in an embodiment of this application.

[0012] Figure 3 This is a schematic diagram of the structure of a training device for a vehicle torque control model provided in an embodiment of this application;

[0013] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0014] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0015] Figure 1 This is a flowchart illustrating a training method for a vehicle torque control model provided in an embodiment of this application. Figure 1 The training method for the vehicle torque control model can be executed by a computer or server, a processor set on a computer or server, or software on a computer or ordinary server. The training method for the vehicle torque control model includes:

[0016] S101, a vehicle torque control model is constructed using residual networks, feature fusion networks, autoregressive recurrent networks and deep neural networks;

[0017] S102, acquire training data, wherein the training data includes the vehicle's state information at the target time and the vehicle's first control torque at the previous time before the target time, and the label corresponding to the training data is the vehicle's second control torque at the target time.

[0018] S103, input the training data into the vehicle torque control model: output the state information corresponding to multiple stage features through multiple stage networks of the residual network, and output the first torque feature corresponding to the first control torque through the last stage network of the residual network.

[0019] S104, The features of multiple stages are processed by a feature fusion network to obtain fused features;

[0020] S105, the fused features and the first torque features are processed by an autoregressive recurrent network to obtain the regression features;

[0021] S106, The regression features are processed through a deep neural network to obtain the processing results corresponding to the training data;

[0022] S107 calculates the classification loss between the processing results and labels corresponding to the training data using the cross-entropy loss function, and updates the network parameters of the vehicle torque control model based on the classification loss to complete the training of the vehicle torque control model.

[0023] Torque is the force produced by an engine or electric motor when it rotates, measured in Newton-meters (N·m) or pound-feet (lb-ft). In automobiles, torque is a crucial parameter determining a vehicle's acceleration, climbing ability, and load-bearing capacity. Higher torque allows the vehicle to generate greater driving force during start-up, acceleration, and hill climbing, thus better handling complex driving conditions.

[0024] It is understood that the training data can be multiple, and each training data point consists of a state information at the target time and a control torque information from the previous time (the first control torque is obtained by collecting vehicle information). The label corresponding to each training data point is the control torque that should be used at the target time (the second control torque is obtained through labeling). This application defines the time of each change in the vehicle's control torque as one time point, and the previous time point is the time of the last change in the vehicle's control torque.

[0025] The status information includes: vehicle speed, vehicle acceleration, steering wheel angle, steering wheel angular velocity and steering wheel angular acceleration, engine speed, and yaw rate. Yaw rate refers to the vehicle's deflection about its vertical axis, and its magnitude represents the vehicle's stability. Existing technologies only consider vehicle speed and engine speed, neglecting vehicle acceleration, steering wheel angle, steering wheel angular velocity, steering wheel angular acceleration, and yaw rate. This application considers all of the above vehicle status information, thus ensuring smoother vehicle torque output.

[0026] The residual network can be ResNet50, the autoregressive recurrent network (ARN) is an autoregressive recurrent network (ARN), and the deep neural network (DNN) is a deep neural network (DNN). The feature fusion network uses common feature fusion methods to fuse the multiple stage features output by the residual network according to certain weights. The ARN is used to determine the regression features at the current time step (treating the target time step as the current time step) based on the fused features and the first torque feature from the previous time step. The DNN is used to determine the control torque at the current time step based on the regression features. The residual network, feature fusion network, ARN, and DNN are connected sequentially. Additionally, the residual network and ARN need to be connected separately to obtain the vehicle torque control model.

[0027] According to the technical solution provided in this application, a vehicle torque control model is constructed using a residual network, a feature fusion network, an autoregressive recurrent network, and a deep neural network. Training data is acquired, including the vehicle's state information at the target time and the vehicle's first control torque at the previous time. The label corresponding to the training data is the vehicle's second control torque at the target time. The training data is input into the vehicle torque control model: multiple stage features corresponding to the state information are output through multiple stage networks of the residual network, and the first torque feature corresponding to the first control torque is output through the last stage network of the residual network. Multiple stage features are processed through a feature fusion network to obtain fused features. The fused features and the first torque feature are processed through an autoregressive recurrent network to obtain regression features. The regression features are processed through a deep neural network to obtain the processing result corresponding to the training data. The classification loss between the processing result and the label corresponding to the training data is calculated using a cross-entropy loss function, and the network parameters of the vehicle torque control model are updated based on the classification loss to complete the training of the vehicle torque control model. By employing the above technical means, the problem of unstable vehicle torque control in the prior art can be solved, thereby ensuring smooth vehicle driving and improving user comfort and satisfaction.

[0028] Figure 2 This is a flowchart illustrating another method for training a vehicle torque control model provided in this application embodiment, as shown below. Figure 2 As shown, it includes:

[0029] S201, input the labels corresponding to the training data into the vehicle torque control model, and output the second torque feature corresponding to the second control torque through the last stage network of the residual network;

[0030] S202, calculates the supervision loss between the regression features and the second torque features using the mean squared error loss function;

[0031] S203 updates the network parameters of the vehicle torque control model based on the classification loss and the supervision loss to complete the training of the vehicle torque control model.

[0032] Furthermore, the supervision loss between the regression features and the second torque feature is calculated using the following mean squared error loss function:

[0033]

[0034] Where MSE() is the mean squared error function (the mean squared error loss function includes the mean squared error function), W i Z represents the regression feature corresponding to the i-th training data point. i Let i be the second torque feature of the label (second control torque) corresponding to the i-th training data, where i is a natural number and its value is between 1 and N, and N is the number of training data.

[0035] The network parameters of the vehicle torque control model can be updated based on the classification loss and the supervision loss. This can be achieved by weighted summation of the classification loss and the supervision loss, and then updating the network parameters of the vehicle torque control model according to the weighted summation result.

[0036] Furthermore, the network parameters of the vehicle torque control model are updated based on classification loss and supervised loss to complete the training of the vehicle torque control model. This includes: multi-stage training of the vehicle torque control model: freezing the network parameters of the deep neural network, updating the network parameters of the residual network, feature fusion network, and autoregressive recurrent network based on supervised loss to complete the first stage of training of the vehicle torque control model; after completing the first stage of training, unfreezing the network parameters of the deep neural network, freezing the network parameters of the residual network, feature fusion network, and autoregressive recurrent network, and updating the network parameters of the deep neural network based on classification loss to complete the second stage of training of the vehicle torque control model.

[0037] The residual network, feature fusion network, and autoregressive recurrent network can be viewed as an encoding network, while the deep neural network can be seen as a decoding network. In other words, the residual network, feature fusion network, and autoregressive recurrent network perform encoding, while the deep neural network performs decoding. Based on this, the residual network, feature fusion network, and autoregressive recurrent network can be trained in the first stage. After the first stage of training is completed, the deep neural network can be trained in the second stage. This allows the training of the vehicle torque control model to be divided into two parts, thereby improving the targeting and efficiency of the training.

[0038] Furthermore, multiple stage features are output through multiple stage networks of the residual network, including: inputting training data into the residual network; processing the training data through the zero-stage network of the residual network to obtain the zero-stage feature; processing the zero-stage feature through the first-stage network of the residual network to obtain the first-stage feature; processing the first-stage feature through the second-stage network of the residual network to obtain the second-stage feature; processing the second-stage feature through the third-stage network of the residual network to obtain the third-stage feature; and processing the third-stage feature through the fourth-stage network of the residual network to obtain the fourth-stage feature. The residual network comprises a zero-stage network, a first-stage network, a second-stage network, a third-stage network, and a fourth-stage network; and the multiple stage features include the first-stage feature, the second-stage feature, the third-stage feature, and the fourth-stage feature.

[0039] For example, the residual network is ResNet50, which includes a zero-stage network (Stage 0), a first-stage network (Stage 1), a second-stage network (Stage 2), a third-stage network (Stage 3), and a fourth-stage network (Stage 4). Each stage network processes the output of the previous stage network. For example, the zero-stage network processes the state information to obtain the zero-stage feature, the first-stage network processes the zero-stage feature to obtain the first-stage feature, and so on. The last stage network in the residual network is the fourth-stage network, so the first torque feature and the second torque feature are outputs of the fourth-stage network.

[0040] Furthermore, the features of multiple stages are processed by a feature fusion network to obtain fused features, including: fusing the features of the first stage, the second stage, the third stage, and the fourth stage through the feature fusion network to obtain fused features.

[0041] Furthermore, after updating the network parameters of the vehicle torque control model based on the classification loss to complete the training of the vehicle torque control model, the method also includes: obtaining the target state information of the target vehicle at the current moment and the first target control torque of the target vehicle at the previous moment; inputting the target state information and the first target control torque into the vehicle torque control model and outputting the second target control torque of the target vehicle at the current moment; and controlling the target vehicle according to the second target control torque at the current moment.

[0042] The target state information is the same as the state information; the first target control torque is the same as the first control torque; and the second target control torque is the same as the label corresponding to the training data. The different names are only to distinguish between the training process and the inference process.

[0043] In some embodiments, target state information and a first target control torque are input into the vehicle torque control model: multiple target stage features corresponding to the target state information are output through multiple stage networks of the residual network, and a first target torque feature corresponding to the first target control torque is output through the last stage network of the residual network; multiple target stage features are processed through a feature fusion network to obtain target fusion features; the target fusion features and the first target torque feature are processed through an autoregressive recurrent network to obtain target regression features; and the target regression features are processed through a deep neural network to obtain a second target control torque.

[0044] Furthermore, after inputting the target state information and the first target control torque into the vehicle torque control model and outputting the second target control torque of the target vehicle at the current moment, the method further includes: uploading the target state information and the first target control torque to the cloud, determining the third target control torque of the target vehicle at the current moment through the cloud's autonomous driving service; determining the reward for the vehicle torque control model to predict the second target control torque based on the target state information and the first target control torque according to the difference between the second target control torque and the third target control torque, wherein the smaller the difference between the second target control torque and the third target control torque, the greater the reward; updating the network parameters of the vehicle torque control model according to the reward, so as to optimize the vehicle torque control model through reinforcement learning during the target vehicle's driving process.

[0045] Autonomous driving services are services provided by the manufacturer of the target vehicle or a third party. The manufacturer or third party trains a unified autonomous driving algorithm using its own resources, then stores this algorithm in the cloud to provide autonomous driving services to its customers. The cloud-based autonomous driving service determines the third target control torque that the target vehicle should take at the current moment based on the target state information and the first target control torque. This embodiment of the application is equivalent to using the cloud-based autonomous driving service to constrain the vehicle torque control model. The model size of the cloud-based autonomous driving service is much larger than the model size of the vehicle torque control model.

[0046] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0047] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0048] Figure 3 This is a schematic diagram of a training device for a vehicle torque control model provided in an embodiment of this application.

[0049] like Figure 3As shown, the training device for the vehicle torque control model includes:

[0050] Module 301 is configured to construct a vehicle torque control model using residual networks, feature fusion networks, autoregressive recurrent networks, and deep neural networks.

[0051] The acquisition module 302 is configured to acquire training data, wherein the training data includes the vehicle's state information at the target time and the vehicle's first control torque at the previous time before the target time, and the label corresponding to the training data is the vehicle's second control torque at the target time.

[0052] The first processing module 303 is configured to input training data into the vehicle torque control model: output multiple stage features corresponding to the state information through multiple stage networks of the residual network, and output the first torque feature corresponding to the first control torque through the last stage network of the residual network.

[0053] The second processing module 304 is configured to process features from multiple stages through a feature fusion network to obtain fused features;

[0054] The third processing module 305 is configured to process the fused features and the first torque features through an autoregressive recurrent network to obtain regression features;

[0055] The fourth processing module 306 is configured to process the regression features through a deep neural network to obtain the processing results corresponding to the training data;

[0056] Training module 307 is configured to calculate the classification loss between the processing results and labels corresponding to the training data through the cross-entropy loss function, and update the network parameters of the vehicle torque control model based on the classification loss to complete the training of the vehicle torque control model.

[0057] According to the technical solution provided in this application, a vehicle torque control model is constructed using a residual network, a feature fusion network, an autoregressive recurrent network, and a deep neural network. Training data is acquired, including the vehicle's state information at the target time and the vehicle's first control torque at the previous time. The label corresponding to the training data is the vehicle's second control torque at the target time. The training data is input into the vehicle torque control model: multiple stage features corresponding to the state information are output through multiple stage networks of the residual network, and the first torque feature corresponding to the first control torque is output through the last stage network of the residual network. Multiple stage features are processed through a feature fusion network to obtain fused features. The fused features and the first torque feature are processed through an autoregressive recurrent network to obtain regression features. The regression features are processed through a deep neural network to obtain the processing result corresponding to the training data. The classification loss between the processing result and the label corresponding to the training data is calculated using a cross-entropy loss function, and the network parameters of the vehicle torque control model are updated based on the classification loss to complete the training of the vehicle torque control model. By employing the above technical means, the problem of unstable vehicle torque control in the prior art can be solved, thereby ensuring smooth vehicle driving and improving user comfort and satisfaction.

[0058] In some embodiments, the training module 307 is further configured as a first processing module 303, configured to input the labels corresponding to the training data into the vehicle torque control model, and output the second torque feature corresponding to the second control torque through the last stage network of the residual network; S202, calculate the supervision loss between the regression feature and the second torque feature through the mean squared error loss function; S203, update the network parameters of the vehicle torque control model according to the classification loss and the supervision loss to complete the training of the vehicle torque control model.

[0059] In some embodiments, the training module 307 is further configured to calculate the supervision loss between the regression features and the second torque feature using the following mean squared error loss function:

[0060]

[0061] Where MSE() is the mean squared error function (the mean squared error loss function includes the mean squared error function), W i Z represents the regression feature corresponding to the i-th training data point. i Let i be the second torque feature of the label (second control torque) corresponding to the i-th training data, where i is a natural number and its value is between 1 and N, and N is the number of training data.

[0062] In some embodiments, the training module 307 is further configured to perform multi-stage training on the vehicle torque control model: freezing the network parameters of the deep neural network, updating the network parameters of the residual network, feature fusion network, and autoregressive recurrent network based on the supervised loss, to complete the first stage training of the vehicle torque control model; after completing the first stage training, unfreezing the network parameters of the deep neural network, freezing the network parameters of the residual network, feature fusion network, and autoregressive recurrent network, updating the network parameters of the deep neural network based on the classification loss, to complete the second stage training of the vehicle torque control model.

[0063] In some embodiments, the first processing module 303 is further configured to input training data into a residual network: process the training data through a zero-stage network of the residual network to obtain zero-stage features; process the zero-stage features through a first-stage network of the residual network to obtain first-stage features; process the first-stage features through a second-stage network of the residual network to obtain second-stage features; process the second-stage features through a third-stage network of the residual network to obtain third-stage features; and process the third-stage features through a fourth-stage network of the residual network to obtain fourth-stage features; wherein the residual network includes a zero-stage network, a first-stage network, a second-stage network, a third-stage network, and a fourth-stage network; and wherein the multiple stage features include first-stage features, second-stage features, third-stage features, and fourth-stage features.

[0064] In some embodiments, the second processing module 304 is further configured to fuse the first-stage features, the second-stage features, the third-stage features, and the fourth-stage features through a feature fusion network to obtain fused features.

[0065] In some embodiments, the training module 307 is further configured to acquire the target state information of the target vehicle at the current moment and the first target control torque of the target vehicle at the previous moment; input the target state information and the first target control torque into the vehicle torque control model, output the second target control torque of the target vehicle at the current moment; and control the target vehicle at the current moment according to the second target control torque.

[0066] In some embodiments, the training module 307 is further configured to input target state information and a first target control torque into the vehicle torque control model: output multiple target stage features corresponding to the target state information through multiple stage networks of the residual network, output the first target torque feature corresponding to the first target control torque through the last stage network of the residual network; process the multiple target stage features through a feature fusion network to obtain target fusion features; process the target fusion features and the first target torque feature through an autoregressive recurrent network to obtain target regression features; and process the target regression features through a deep neural network to obtain the second target control torque.

[0067] In some embodiments, the training module 307 is further configured to upload target state information and a first target control torque to the cloud, determine the third target control torque of the target vehicle at the current moment through the autonomous driving service in the cloud; determine the reward for the vehicle torque control model to predict the second target control torque based on the target state information and the first target control torque according to the difference between the second target control torque and the third target control torque, wherein the smaller the difference between the second target control torque and the third target control torque, the greater the reward; update the network parameters of the vehicle torque control model according to the reward, so as to optimize the vehicle torque control model through reinforcement learning during the driving process of the target vehicle.

[0068] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0069] Figure 4 This is a schematic diagram of the electronic device 4 provided in an embodiment of this disclosure. Figure 4 As shown, the electronic device 4 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, it implements the steps in the various method embodiments described above. Alternatively, when the processor 401 executes the computer program 403, it implements the functions of each module / unit in the various device embodiments described above.

[0070] Electronic device 4 may include, but is not limited to, processor 401 and memory 402. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or different components.

[0071] Processor 401 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0072] The memory 402 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM of the electronic device 4. The memory 402 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 4. The memory 402 can also include both internal and external storage units of the electronic device 4. The memory 402 is used to store computer programs and other programs and data required by the electronic device.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0074] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0075] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions 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 this application, and should all be included within the protection scope of this application.

Claims

1. A training method for a vehicle torque control model, characterized in that, include: A vehicle torque control model was constructed using residual networks, feature fusion networks, autoregressive recurrent networks, and deep neural networks. Acquire training data, wherein the training data includes the vehicle's state information at the target time and the vehicle's first control torque at the previous time before the target time, and the label corresponding to the training data is the vehicle's second control torque at the target time; the state information includes: vehicle speed, vehicle acceleration, steering wheel angle, steering wheel angular velocity and steering wheel angular acceleration, engine speed and yaw rate; Input the training data into the vehicle torque control model: The residual network outputs multiple stage features corresponding to the state information through multiple stage networks respectively, and outputs the first torque feature corresponding to the first control torque through the last stage network of the residual network. The feature fusion network processes features from multiple stages to obtain fused features. The fused features and the first torque feature are processed by the autoregressive recurrent network to obtain regression features; The regression features are processed by the deep neural network to obtain the processing results corresponding to the training data; The classification loss between the processing result and the label corresponding to the training data is calculated using the cross-entropy loss function, and the network parameters of the vehicle torque control model are updated based on the classification loss to complete the training of the vehicle torque control model.

2. The method according to claim 1, characterized in that, After calculating the classification loss between the processing result and the label corresponding to the training data using the cross-entropy loss function, the method further includes: The labels corresponding to the training data are input into the vehicle torque control model, and the second torque feature corresponding to the second control torque is output through the last stage network of the residual network. The supervision loss between the regression feature and the second torque feature is calculated using the mean squared error loss function; The network parameters of the vehicle torque control model are updated based on the classification loss and the supervision loss to complete the training of the vehicle torque control model.

3. The method according to claim 2, characterized in that, The network parameters of the vehicle torque control model are updated based on the classification loss and the supervision loss to complete the training of the vehicle torque control model, including: The vehicle torque control model is trained in multiple stages: The network parameters of the deep neural network are frozen, and the network parameters of the residual network, the feature fusion network, and the autoregressive recurrent network are updated according to the supervised loss to complete the first stage training of the vehicle torque control model. After completing the first stage of training, the network parameters of the deep neural network are unfrozen, while the network parameters of the residual network, the feature fusion network, and the autoregressive recurrent network are frozen. The network parameters of the deep neural network are then updated based on the classification loss to complete the second stage of training of the vehicle torque control model.

4. The method according to claim 1, characterized in that, Multiple stage features are output through multiple stage networks of the residual network, including: The training data is input into the residual network: The training data is processed through the zero-stage network of the residual network to obtain the zero-stage features; The zero-stage features are processed by the first-stage network of the residual network to obtain the first-stage features; The first-stage features are processed by the second-stage network of the residual network to obtain the second-stage features; The second-stage features are processed by the third-stage network of the residual network to obtain the third-stage features; The third-stage features are processed by the fourth-stage network of the residual network to obtain the fourth-stage features; The residual network includes a zero-stage network, a first-stage network, a second-stage network, a third-stage network, and a fourth-stage network. The multiple stage features include the first stage feature, the second stage feature, the third stage feature, and the fourth stage feature.

5. The method according to claim 4, characterized in that, The feature fusion network processes features from multiple stages to obtain fused features, including: The first-stage features, the second-stage features, the third-stage features, and the fourth-stage features are fused using the feature fusion network to obtain the fused features.

6. The method according to claim 1, characterized in that, After updating the network parameters of the vehicle torque control model based on the classification loss to complete the training of the vehicle torque control model, the method further includes: Obtain the target status information of the target vehicle at the current moment and the first target control torque of the target vehicle at the previous moment; The target state information and the first target control torque are input into the vehicle torque control model, and the second target control torque of the target vehicle at the current moment is output. At the current moment, the target vehicle is controlled according to the second target control torque.

7. The method according to claim 6, characterized in that, After inputting the target state information and the first target control torque into the vehicle torque control model, and outputting the second target control torque of the target vehicle at the current moment, the method further includes: The target state information and the first target control torque are uploaded to the cloud, and the third target control torque of the target vehicle at the current moment is determined through the autonomous driving service of the cloud. Based on the difference between the second target control torque and the third target control torque, the reward for the vehicle torque control model to predict the second target control torque based on the target state information and the first target control torque is determined, wherein the smaller the difference between the second target control torque and the third target control torque, the greater the reward; The network parameters of the vehicle torque control model are updated based on the reward, so as to optimize the vehicle torque control model through reinforcement learning during the driving process of the target vehicle.

8. A training device for a vehicle torque control model, characterized in that, include: The building module is configured to construct a vehicle torque control model using residual networks, feature fusion networks, autoregressive recurrent networks, and deep neural networks. The acquisition module is configured to acquire training data, wherein the training data includes the vehicle's state information at the target time and the vehicle's first control torque at the previous time before the target time, and the label corresponding to the training data is the vehicle's second control torque at the target time; the state information includes: vehicle speed, vehicle acceleration, steering wheel angle, steering wheel angular velocity and steering wheel angular acceleration, engine speed and yaw rate; The first processing module is configured to input the training data into the vehicle torque control model: output multiple stage features corresponding to the state information through multiple stage networks of the residual network, and output the first torque feature corresponding to the first control torque through the last stage network of the residual network. The second processing module is configured to process multiple stage features through the feature fusion network to obtain fused features; The third processing module is configured to process the fused features and the first torque features through the autoregressive recurrent network to obtain regression features; The fourth processing module is configured to process the regression features through the deep neural network to obtain the processing result corresponding to the training data; The training module is configured to calculate the classification loss between the processing result and the label corresponding to the training data using the cross-entropy loss function, and update the network parameters of the vehicle torque control model based on the classification loss to complete the training of the vehicle torque control model.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images