Model training method and device, computer equipment, storage medium and program product

Abstract

The invention relates to a model training method and device, computer equipment, a storage medium and a program product. The method comprises the steps that a vehicle end controller receives a task order issued by a cloud end algorithm model platform and used for training a target algorithm model; wherein the target algorithm model is deployed in the vehicle end controller, and the task order comprises an algorithm file and deployment position information of the target algorithm model; distributing the algorithm file to a virtual component corresponding to the target algorithm model according to the deployment position information; executing the algorithm file through the virtual component to obtain a training result of the target algorithm model; and feeding back a training result to the cloud algorithm model platform. By adopting the method, the training period of the target algorithm model can be shortened, the updating frequency of the target algorithm model can be improved, respective training of different algorithm models can be realized, vehicle end computing resources are fully utilized, model training and real vehicle work are decoupled, normal use of the vehicle is not affected by the training process in a real vehicle environment, the training authenticity is guaranteed, and the training effect is optimized.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a model training method, apparatus, computer equipment, storage medium and program product. Background Art

[0002] With the continuous development of intelligent driving in the automotive industry, people have higher and higher requirements for the accuracy of various algorithm models deployed on vehicles. At present, various algorithm models are generally trained with structured data in an off-vehicle environment, and then the trained algorithm models are deployed on the vehicle, or the algorithm models deployed on the vehicle are updated.

[0003] However, this conventional training method has a long cycle and it is difficult to adjust the algorithm model in a timely manner, resulting in poor real-time performance of the algorithm model and slow iteration speed. In addition, due to the large number of algorithm models and their different functions, the effect of unified training of various algorithm models is poor. Summary of the invention

[0004] Based on this, it is necessary to provide a model training method, device, computer equipment, storage medium and program product that can complete high-quality training of algorithm models on the vehicle side through interaction between the cloud and the vehicle side to address the above technical problems.

[0005] In a first aspect, the present application provides a model training method, which is applied to a vehicle-side controller, comprising:

[0006] Receive a task order for training a target algorithm model issued by a cloud algorithm model platform; wherein the target algorithm model is deployed on a vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model;

[0007] According to the deployment location information, the algorithm file is allocated to the virtual component corresponding to the target algorithm model;

[0008] Through the virtual component, the algorithm file is executed to obtain the training results of the target algorithm model;

[0009] Feedback the training results to the cloud algorithm model platform.

[0010] In one embodiment, according to the deployment location information, the algorithm file is allocated to the virtual component corresponding to the target algorithm model, including:

[0011] According to the deployment location information, determine the first sub-path corresponding to the target algorithm model from the sub-paths in each algorithm model;

[0012] The algorithm file is stored in the first sub-path for calling by the virtual component corresponding to the target algorithm model.

[0013] In one embodiment, before feeding back the training results to the cloud algorithm model platform, the method further includes:

[0014] Receive the recovery order corresponding to the task order issued by the cloud algorithm model platform.

[0015] In one embodiment, the training results are fed back to the cloud algorithm model platform, including:

[0016] Obtaining the training results stored in the second sub-path through the vehicle-side data acquisition system;

[0017] Through the vehicle-side data acquisition system, the corresponding data packet is generated according to the training results, and the data packet is uploaded to the cloud-based algorithm model platform; among them, the data packet includes the working status and dynamic event data of the target algorithm model.

[0018] In one embodiment, the method further comprises:

[0019] Obtain the algorithm model data required for training the target algorithm model through the virtual component;

[0020] The algorithm model data is uploaded to the cloud-based algorithm model platform through the vehicle-side data acquisition system; the algorithm model data is used for the cloud-based algorithm model platform to generate algorithm files.

[0021] In one embodiment, the algorithm model data is uploaded to the cloud algorithm model platform through the vehicle-side data acquisition system, including:

[0022] When the trigger conditions of the event trigger corresponding to the target algorithm model are detected, the algorithm model data is uploaded to the cloud algorithm model platform through the vehicle-side data acquisition system.

[0023] In the second aspect, the present application provides a model training method, which is applied to a cloud algorithm model platform, including:

[0024] Sending a task order for training a target algorithm model to at least one vehicle-side controller;

[0025] The target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; the deployment location information is used for the vehicle-side controller to allocate the algorithm file to the virtual component corresponding to the target algorithm model; the algorithm file is used for the vehicle-side controller to execute through the virtual component to obtain the training result of the target algorithm model;

[0026] Receive training results fed back by at least one vehicle-side controller.

[0027] In one of the embodiments, the deployment location information is used for the vehicle-side controller to determine the first sub-path corresponding to the target algorithm model from the sub-paths in each algorithm model, and store the algorithm file in the first sub-path.

[0028] In one embodiment, before receiving the training result fed back by at least one vehicle-side controller, the method further includes:

[0029] A recovery order corresponding to the task order is issued to at least one vehicle-side controller.

[0030] In one embodiment, receiving a training result fed back by at least one vehicle-side controller includes:

[0031] Receive a data packet uploaded by at least one vehicle-side controller through a vehicle-side data acquisition system; wherein the data packet is generated based on a training result, and the data packet includes a working status and dynamic event data of a target algorithm model.

[0032] In one embodiment, before issuing a task order for training the target algorithm model to at least one vehicle-side controller, the method further includes:

[0033] Compile the cleaned algorithm model data to obtain the algorithm file;

[0034] Store the algorithm file in the algorithm warehouse.

[0035] In one embodiment, before compiling the cleaned algorithm model data to obtain the algorithm file, the method further includes:

[0036] Receive algorithm model data uploaded by at least one vehicle-side controller through a vehicle-side data acquisition system when the triggering conditions of an event trigger corresponding to a target algorithm model are detected; wherein the algorithm model data is obtained by the vehicle-side controller through a virtual component corresponding to the target algorithm model.

[0037] In a third aspect, the present application also provides a model training device, which is configured in a vehicle-side controller and includes:

[0038] A task receiving module is used to receive a task order for training a target algorithm model issued by a cloud algorithm model platform; wherein the target algorithm model is deployed on a vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model;

[0039] A file allocation module, used to allocate the algorithm file to the virtual component corresponding to the target algorithm model according to the deployment location information;

[0040] The model training module is used to execute the algorithm file through the virtual component to obtain the training result of the target algorithm model;

[0041] The result feedback module is used to feed back the training results to the cloud algorithm model platform.

[0042] In a fourth aspect, the present application also provides a model training device, which is configured on a cloud algorithm model platform, including:

[0043] A task issuing module, used to issue a task order for training a target algorithm model to at least one vehicle-side controller;

[0044] The target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; the deployment location information is used for the vehicle-side controller to allocate the algorithm file to the virtual component corresponding to the target algorithm model; the algorithm file is used for the vehicle-side controller to execute through the virtual component to obtain the training result of the target algorithm model;

[0045] The result receiving module is used to receive the training results fed back by at least one vehicle-side controller.

[0046] In a fifth aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method of the first aspect or the second aspect when executing the computer program.

[0047] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method of the first aspect or the second aspect are implemented. In a fifth aspect, the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method of the first aspect or the second aspect are implemented.

[0048] The above-mentioned model training method, device, computer equipment, storage medium and program product, the vehicle-side controller receives the task order for training the target algorithm model issued by the cloud algorithm model platform; wherein the target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; according to the deployment location information, the algorithm file is assigned to the virtual component corresponding to the target algorithm model; through the virtual component, the algorithm file is executed to obtain the training result of the target algorithm model; the training result is fed back to the cloud algorithm model platform. This application completes the training of the target algorithm model in the real vehicle environment of the vehicle side through the interaction between the cloud and the vehicle side, which is conducive to shortening the training cycle of the target algorithm model and improving the update frequency of the target algorithm model; wherein the vehicle side can determine the virtual component corresponding to the target algorithm model according to the deployment location information, so as to achieve the purpose of distinguishing different algorithm models, and on this basis, different algorithm models can be trained separately to make full use of the vehicle-side computing resources; through the virtual component, the model training is decoupled from the real vehicle work, and the training process does not affect the normal use of the vehicle. At the same time, model training in the real vehicle environment can ensure the authenticity of the training and optimize the training effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments of the present application or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 An application environment diagram of a model training method in an embodiment;

[0051] Figure 2 A schematic diagram of a flow chart of a model training method in one embodiment;

[0052] Figure 3 A schematic diagram of a process of allocating algorithm files to virtual components in one embodiment;

[0053] Figure 4 A schematic diagram of a process for feeding back training results to a cloud algorithm model platform in one embodiment;

[0054] Figure 5 A schematic diagram of a process for uploading algorithm model data to a cloud algorithm model platform in one embodiment;

[0055] Figure 6 A schematic diagram of a flow chart of a model training method in another embodiment;

[0056] Figure 7A schematic diagram of a process for generating an algorithm file in one embodiment;

[0057] Figure 8 A system architecture diagram of a model training method in one embodiment;

[0058] Fig. 9 is a structural block diagram of a model training device in one embodiment;

[0059] Fig.10 is a structural block diagram of a model training device in another embodiment;

[0060] Fig.11 It is a structural block diagram of a model training device in yet another embodiment;

[0061] Fig.12 is an internal structure diagram of a computer device in one embodiment;

[0062] Fig.13 FIG. 4 is a diagram showing the internal structure of a computer device in another embodiment. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0064] The model training method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, any of the vehicle-side controllers 101-10n can communicate with the cloud algorithm model platform 100 through the network. The data storage system can store the data that the cloud algorithm model platform 100 needs to process. The data storage system can be integrated on the cloud algorithm model platform 100, or it can be placed on the cloud or other network servers. The vehicle-side controller receives a task order for training the target algorithm model issued by the cloud algorithm model platform; wherein the target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; according to the deployment location information, the algorithm file is assigned to the virtual component corresponding to the target algorithm model; through the virtual component, the algorithm file is executed to obtain the training result of the target algorithm model; the training result is fed back to the cloud algorithm model platform. The server deployed by the cloud algorithm model platform 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0065] In an exemplary embodiment, Figure 2 As shown, a model training method is provided, which is applied to Figure 1The vehicle-side controller in the example is used to illustrate, including:

[0066] S201, receiving a task order for training a target algorithm model issued by a cloud algorithm model platform.

[0067] Among them, the cloud-based algorithm model platform can be used for business functions such as algorithm model classification, management, compilation, recycling, training task distribution, training status management, and sample data element collection. The vehicle-side controller can refer to the vehicle-side autonomous driving domain controller, which has multiple software components deployed inside, including real control components and virtual components. The real control components and virtual components can run the same algorithm model. All algorithm models share the vehicle's operating environment, but the virtual components do not participate in vehicle control. The cloud-based algorithm model platform and the vehicle-side controller can transmit data through the vehicle-cloud interaction link.

[0068] The target algorithm model can be any algorithm model deployed on the vehicle-side controller, such as an obstacle detection model. Specifically, the target algorithm model is an algorithm model running on a virtual component. After the cloud algorithm model platform establishes a task order for training the target algorithm model, it sends the task order to the vehicle-side controller to instruct the vehicle-side controller to train the target algorithm model deployed on the vehicle-side controller in the actual vehicle operation environment.

[0069] The task order includes the algorithm file and deployment location information of the target algorithm model. The algorithm file, as an attachment in the task order, can refer to the compiled SWC (Software Component) algorithm file, which is used to execute the training steps of the target algorithm model after it takes effect. The deployment location information is used to indicate the deployment location of the target algorithm model on the vehicle side, including but not limited to the storage location of the algorithm file of the target algorithm model.

[0070] S202: Allocate the algorithm file to the virtual component corresponding to the target algorithm model according to the deployment location information.

[0071] According to the deployment location information of the task order, the target algorithm model to which the algorithm file in the task order belongs can be determined, and the deployment location of the target algorithm model on the vehicle side can be determined, including the storage location of the algorithm file. The algorithm file is stored in the storage location indicated by the deployment location information, so that the virtual component corresponding to the target algorithm model can read the algorithm file in the storage location, and the algorithm file is allocated to the virtual component corresponding to the target algorithm model.

[0072] It can be understood that one software component corresponds to one algorithm model, different algorithm models realize different vehicle intelligent functions, and different algorithm models correspond to different paths. That is, the storage locations of algorithm files of different algorithm models are divided based on certain rules, and the software components corresponding to each algorithm model can read the algorithm files at the corresponding storage location to train, test and update the algorithm model.

[0073] S203, executing the algorithm file through the virtual component to obtain the training result of the target algorithm model.

[0074] The vehicle-side controller controls the virtual component corresponding to the target algorithm model, reads the algorithm file in the corresponding storage location and executes it, thereby automatically training the running target algorithm model. The target algorithm model continuously generates data during the training process, including but not limited to the output data and working data of the target algorithm model. After the training process is completed, the training results of the target algorithm model are formed by combining the data generated during the training process.

[0075] Optionally, the algorithm file includes input data of the target algorithm model. After the virtual component executes the algorithm file and it becomes effective, the input data in the algorithm file is automatically input into the target algorithm model to obtain output data of the target algorithm model. Exemplarily, the target algorithm model is an obstacle detection model. The image data in the algorithm file is input into the target algorithm model through the virtual component. After the target algorithm model processes the image data, the output obstacle detection result is obtained. By executing the algorithm file through the virtual component, the target algorithm model can also be reversely iterated according to the training data in the algorithm file and the obstacle detection result, and finally generate the training result.

[0076] Optionally, the virtual component obtains the image data collected by the vehicle-side data acquisition system and uses it as the input data of the target algorithm model. After the target algorithm model processes the image data, the output obstacle detection result is obtained. The algorithm file is executed by the virtual component to perform reverse iterative training on the target algorithm model to finally generate the training result.

[0077] S204, feeding back the training results to the cloud algorithm model platform.

[0078] The vehicle-side controller uploads the training results of the target algorithm model to the cloud algorithm model platform. The training results include but are not limited to the output data, model accuracy and processing time of the target algorithm model.

[0079] Optionally, the vehicle-side controller can upload the training results to the cloud algorithm model platform through the vehicle-cloud interaction link.

[0080] Optionally, the vehicle-side controller can first transmit the training results to the vehicle-side data acquisition system, and then upload them to the cloud algorithm model platform through the vehicle-side data acquisition system.

[0081] In the above model training method, the vehicle-side controller receives a task order for training the target algorithm model issued by the cloud-based algorithm model platform; wherein the target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; according to the deployment location information, the algorithm file is assigned to the virtual component corresponding to the target algorithm model; through the virtual component, the algorithm file is executed to obtain the training result of the target algorithm model; and the training result is fed back to the cloud-based algorithm model platform. This embodiment completes the training of the target algorithm model in the real vehicle environment of the vehicle side through the interaction between the cloud and the vehicle side, which is conducive to shortening the training cycle of the target algorithm model and improving the update frequency of the target algorithm model; wherein the vehicle side can determine the virtual component corresponding to the target algorithm model according to the deployment location information, so as to achieve the purpose of distinguishing different algorithm models, and on this basis, different algorithm models can be trained separately to make full use of the vehicle-side computing resources; through the virtual component, the model training is decoupled from the real vehicle work, and the training process does not affect the normal use of the vehicle. At the same time, model training in the real vehicle environment can ensure the authenticity of the training and optimize the training effect.

[0082] In an exemplary embodiment, Figure 3 As shown, the above S202 includes:

[0083] S301, determining a first sub-path corresponding to a target algorithm model from among the sub-paths in each algorithm model according to the deployment location information.

[0084] Among them, the deployment location information includes the path corresponding to the target algorithm model, that is, the first sub-path, which is represented by a folder under a disk on the vehicle side.

[0085] The vehicle-side controller establishes the middle-stage path of the vehicle-side algorithm model, divides the paths corresponding to each algorithm model, and subdivides different sub-paths under the main path, so that each algorithm model corresponds to a unique sub-path.

[0086] The path corresponding to the target algorithm model in the deployment location information is matched one by one with the sub-paths in each algorithm model to obtain the matching first sub-path.

[0087] Optionally, the deployment location information includes a first subpath corresponding to the target algorithm model, and the deployment location information can be directly parsed to obtain the first subpath.

[0088] S302, storing the algorithm file in the first sub-path for calling by the virtual component corresponding to the target algorithm model.

[0089] By storing the algorithm file in the task order in the first sub-path, the algorithm file can be assigned to the virtual component corresponding to the target algorithm model. The vehicle-side controller can control the virtual component to call the algorithm file in the first sub-path, and the algorithm file takes effect on the vehicle side to train the target algorithm model.

[0090] In this embodiment, the vehicle-side controller establishes a middle-stage path for the algorithm model and sets a unique sub-path for each algorithm model. After receiving the task order for model training, the first sub-path corresponding to the target algorithm model to be trained can be found according to the deployment location information in the task order, so as to accurately allocate the algorithm file of the target algorithm model to the corresponding virtual component and complete the accurate issuance of the task order. On this basis, different algorithm models can be trained through different software components, and the training steps of each algorithm model do not conflict with each other, which is conducive to making full use of the computing power resources on the vehicle side and providing efficient and high-quality vehicle-side model training functions.

[0091] In an exemplary embodiment, before the above S204, the model training method further includes:

[0092] Receive the recovery order corresponding to the task order issued by the cloud algorithm model platform.

[0093] The step of the vehicle-side controller feeding back the training results to the cloud-based algorithm model platform may be triggered by the cloud-based algorithm model platform. That is, before the vehicle-side controller feeds back the training results to the cloud-based algorithm model platform, the cloud-based algorithm model platform sends a recycling order corresponding to the task order to the vehicle-side controller, instructing the vehicle-side controller to feed back the training results of the target algorithm model to the cloud-based algorithm model platform.

[0094] Optionally, the cloud-based algorithm model platform sends a recycling order to the vehicle-side controller through the vehicle-cloud interaction link.

[0095] In this embodiment, the cloud-based algorithm model platform triggers the recovery of the training results of the target algorithm model, and the vehicle-cloud interaction method improves the rationality and reliability of the vehicle-side model training.

[0096] In an exemplary embodiment, Figure 4 As shown, the above S204 includes:

[0097] S401, obtaining the training results stored in the second sub-path through the vehicle-side data acquisition system.

[0098] The vehicle-side data acquisition system can be used to automatically collect various vehicle data, such as the road ahead data during vehicle driving, driving data such as braking, steering or acceleration, and driver's health status data.

[0099] The vehicle-side data acquisition system can interact with various software components in the vehicle-side controller. For example, the vehicle-side controller uses the Linux system, and the vehicle-side controller can obtain the collected data from the vehicle-side data acquisition system through any software component with corresponding permissions. The corresponding permissions can refer to data processing permissions such as adding, deleting, reducing, modifying and copying.

[0100] The vehicle-side controller reports the training results of the target algorithm model to the cloud algorithm model platform through the vehicle-side data acquisition system. Specifically, the vehicle-side data acquisition system reads the data stored in the second sub-path to obtain the training results of the target algorithm model. The training results of the target algorithm model are stored in the second sub-path.

[0101] Optionally, various data during the operation of the target algorithm model, including data generated during the operation, are recorded through the vehicle-side event recording system and stored in the second sub-path.

[0102] S402, generate corresponding data packets according to the training results through the vehicle-side data collection system, and upload the data packets to the cloud algorithm model platform.

[0103] The obtained training results are packaged through the vehicle-side data acquisition system to obtain the corresponding data package, which is then uploaded to the cloud algorithm model platform. The data package includes the working status and dynamic event data of the target algorithm model. The working status can be expressed as a normal working status or an abnormal working status; the dynamic event data can be the data generated by the target algorithm model when the vehicle encounters a dynamic event, such as the image detection results output by the target algorithm model when there is an obstacle in front of the vehicle.

[0104] Optionally, the dynamic events encountered by the vehicle are recorded through the vehicle-side event recording system, and the data generated by the target algorithm model is provided to the vehicle-side data acquisition system for packaging and processing.

[0105] In this embodiment, the data generated during the training of the target algorithm model is packaged into a data packet through the vehicle-side data acquisition system, and then uploaded to the cloud-based algorithm model platform to complete a model training cycle, which can meet the high-frequency verification needs of vehicle intelligent driving.

[0106] In an exemplary embodiment, Figure 5 As shown, the model training method also includes:

[0107] S501, obtaining algorithm model data required for training the target algorithm model through the virtual component.

[0108] Before the cloud-based algorithm model platform issues a task order for the target algorithm model to the vehicle-side controller, the vehicle-side controller may upload the algorithm model data required to generate the task order to the cloud-based algorithm model platform.

[0109] Specifically, each software component in the vehicle-side controller has the ability to collect data elements, and can collect the data elements required by the corresponding algorithm model, such as perception image capture, vehicle trajectory prediction link interception, prediction node data recording, positioning deviation data and environmental data recording, etc. According to the function of the target algorithm model corresponding to each software component, the type of data element that each software component can capture is defined. Each software component can also interact with other software components through protocols such as SOME / IP (Scalable service-oriented middleware over IP), XCP (Universal Calibration Protocol) and Orin_consumer to obtain the data elements required by the corresponding algorithm model.

[0110] The vehicle-side controller controls the virtual components to capture the algorithm model data required by the target algorithm model, or interacts with other virtual components to obtain the algorithm model data required by the target algorithm model.

[0111] S502, upload the algorithm model data to the cloud algorithm model platform through the vehicle-side data acquisition system.

[0112] Optionally, when the triggering conditions of the event trigger corresponding to the target algorithm model are detected, the algorithm model data is uploaded to the cloud algorithm model platform through the vehicle-side data acquisition system.

[0113] The event trigger corresponding to the target algorithm model can be set through the vehicle-side event recording system. Different algorithm models can correspond to different event triggers. According to the event requirements of each algorithm model, the triggering conditions of the event trigger corresponding to each algorithm model are set.

[0114] For example, the triggering conditions of the target algorithm model are such as a vehicle collision, a vehicle function failure, etc. After detecting that the triggering conditions of the event trigger corresponding to the target algorithm model are currently met, the collected algorithm model data is uploaded through the vehicle-side data acquisition system.

[0115] Furthermore, the cloud algorithm model platform generates a corresponding algorithm file based on the algorithm model data. Specifically, the cloud algorithm model platform cleans the received algorithm model data, selects the algorithm model data required to generate the algorithm file, and then processes the cleaned algorithm model data to obtain the algorithm file of the target algorithm model, and stores it in the algorithm warehouse.

[0116] In this embodiment, real algorithm model data is collected in real time through virtual components, which can reduce data costs and improve the authenticity of model training. The algorithm model data collected from the real vehicle is then processed on the cloud to obtain an algorithm file, which is used for vehicle-side model training, thereby improving the model training effect.

[0117] In an exemplary embodiment, Figure 6 As shown, a model training method is provided, which is applied to Figure 1 The cloud-based algorithm model platform in is used as an example to illustrate, including:

[0118] S601, issuing a task order for training a target algorithm model to at least one vehicle-side controller.

[0119] The target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; the deployment location information is used for the vehicle-side controller to allocate the algorithm file to the virtual component corresponding to the target algorithm model; the algorithm file is used for the vehicle-side controller to execute through the virtual component to obtain the training result of the target algorithm model;

[0120] S602, receiving training results fed back by at least one vehicle-side controller.

[0121] In the above model training method, the cloud algorithm model platform issues a task order for training the target algorithm model to at least one vehicle-side controller; wherein the target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; the deployment location information is used for the vehicle-side controller to allocate the algorithm file to the virtual component corresponding to the target algorithm model; the algorithm file is used for the vehicle-side controller to execute through the virtual component to obtain the training result of the target algorithm model; and the training result fed back by at least one vehicle-side controller is received. This embodiment completes the training of the target algorithm model in the real vehicle environment of the vehicle side through the interaction between the cloud and the vehicle side, which is conducive to shortening the training cycle of the target algorithm model and improving the update frequency of the target algorithm model; wherein the vehicle side can determine the virtual component corresponding to the target algorithm model according to the deployment location information, so as to achieve the purpose of distinguishing different algorithm models, and on this basis, different algorithm models can be trained separately, making full use of the vehicle-side computing resources; the model training is decoupled from the real vehicle work through the virtual component, and the training process does not affect the normal use of the vehicle. At the same time, model training in the real vehicle environment can ensure the authenticity of the training and optimize the training effect.

[0122] As an optional implementation in this embodiment, the deployment location information is used to provide the vehicle-side controller to determine the first sub-path corresponding to the target algorithm model from the sub-paths in each algorithm model, and store the algorithm file in the first sub-path.

[0123] In this embodiment, the vehicle-side controller establishes a middle-stage path for the algorithm model and sets a unique sub-path for each algorithm model. After receiving the task order for model training, the first sub-path corresponding to the target algorithm model to be trained can be found according to the deployment location information in the task order, so as to accurately allocate the algorithm file of the target algorithm model to the corresponding virtual component and complete the accurate issuance of the task order. On this basis, different algorithm models can be trained through different software components, and the training steps of each algorithm model do not conflict with each other, which is conducive to making full use of the computing power resources on the vehicle side and providing efficient and high-quality vehicle-side model training functions.

[0124] In an exemplary embodiment, before the above S602, the model training method further includes:

[0125] A recovery order corresponding to the task order is issued to at least one vehicle-side controller.

[0126] In this embodiment, the cloud-based algorithm model platform triggers the recovery of the training results of the target algorithm model, and the vehicle-cloud interaction method improves the rationality and reliability of the vehicle-side model training.

[0127] In an exemplary embodiment, the above S602 includes:

[0128] Receive a data packet uploaded by at least one vehicle-side controller through a vehicle-side data acquisition system; wherein the data packet is generated based on a training result, and the data packet includes a working status and dynamic event data of a target algorithm model.

[0129] In this embodiment, the data generated during the training of the target algorithm model is packaged into a data packet through the vehicle-side data acquisition system, and then uploaded to the cloud-based algorithm model platform to complete a model training cycle, which can meet the high-frequency verification needs of vehicle intelligent driving.

[0130] In an exemplary embodiment, Figure 7 As shown, the model training method also includes:

[0131] S701, compile the cleaned algorithm model data to obtain an algorithm file.

[0132] S702, storing the algorithm file in the algorithm warehouse.

[0133] In this embodiment, the vehicle side collects real algorithm model data, which can reduce data costs and improve the authenticity of model training. The cloud side then processes the algorithm model data collected from the real vehicle to obtain an algorithm file, which is used for vehicle-side model training, which can improve the model training effect.

[0134] In an exemplary embodiment, before the above S701, the model training method further includes:

[0135] Receive algorithm model data uploaded by at least one vehicle-side controller through a vehicle-side data acquisition system when the triggering conditions of an event trigger corresponding to a target algorithm model are detected; wherein the algorithm model data is obtained by the vehicle-side controller through a virtual component corresponding to the target algorithm model.

[0136] In this embodiment, the vehicle-side controller can execute the reporting step of the algorithm model data based on the event trigger, reduce the data cost investment in model training, and reduce labor and time costs.

[0137] In an exemplary embodiment, a model training method is provided, comprising:

[0138] The vehicle-side controller obtains the algorithm model data required for training the target algorithm model through virtual components, and uploads the algorithm model data to the cloud algorithm model platform through the vehicle-side data acquisition system when the trigger conditions of the event trigger corresponding to the target algorithm model are detected.

[0139] After receiving the algorithm model data, the cloud-based algorithm model platform cleans and compiles the algorithm model data to obtain the algorithm file of the target algorithm model, and then stores it in the algorithm warehouse.

[0140] When the cloud-based algorithm model platform determines that the target algorithm model needs to be trained, it obtains the algorithm file from the algorithm warehouse, generates a task order including the algorithm file and deployment location information, and sends the task order to the vehicle-side controller.

[0141] After receiving the task order, the vehicle-side controller stores the algorithm file in the platform path of the algorithm model indicated by the deployment location information for the virtual component corresponding to the target algorithm model to call and execute, complete the training of the target algorithm model, and obtain the training results.

[0142] The cloud-based algorithm model sends a recycling order corresponding to the task order to the vehicle-side controller to instruct the vehicle-side controller to report the training results of the target algorithm model.

[0143] After receiving the recycling order, the vehicle-side controller packages the training results of the target algorithm model into a data packet through the vehicle-side data collection system and reports it to the cloud algorithm model. The data packet includes the data generated by the target algorithm model, not limited to the working status and dynamic event data of the target algorithm model.

[0144] In an exemplary embodiment, Figure 8As shown, the vehicle side and the cloud side together constitute the system architecture of the model training method, which complies with data compliance information security standards. Among them, the SAAS layer (Software as a Service, application service layer) cloud algorithm model training platform is the cloud algorithm model platform, which is the carrier of the cloud main function and can realize business functions such as task creation, algorithm configuration and PAAS layer (Platform as a Service, platform service layer) data management; the PAAS layer is the data storage layer, as a data warehouse function carrier for storing various types of vehicle-side data such as algorithm files, including but not limited to source data meta-warehouse, data algorithm model warehouse, training result feedback warehouse, perception, planning and control data storage warehouse, etc.; the edge computing layer corresponds to the vehicle-side controller equipped with the vehicle to realize the training of the algorithm model, including the vehicle-cloud interaction "end" system corresponding to the vehicle-cloud interaction link, which can be divided into the main control domain ADMC vehicle-side algorithm model middle platform and the auxiliary control domain ADSC vehicle-side algorithm model middle platform. The virtual application software SWC (i.e. virtual component) and the actual control application software SWC (i.e. actual control component) coexist in the vehicle-side controller, such as SWC1-SWCn, sharing Autosar (Automotive Open System Architecture, automotive open system architecture) / BSW (Basic Software) basic software layer / real-time runtime environment RTE (Runtime Environment, runtime environment), collects data elements through the shadow EDR (Endpoint Detection and Response) corresponding to the virtual application software SWC and the actual control EDR corresponding to the actual control application software SWC, and transmits them to the autonomous driving data collection system DC (Data Collection system for automated driving); edge hardware is the hardware foundation of the vehicle.

[0145] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0146] Based on the same inventive concept, the embodiment of the present application also provides a model training device for implementing the model training method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more model training device embodiments provided below can refer to the limitations on the model training method above, and will not be repeated here.

[0147] In an exemplary embodiment, Fig. 9 As shown, a model training device 1 is provided, comprising:

[0148] The task receiving module 10 is used to receive a task order for training a target algorithm model issued by a cloud algorithm model platform; wherein the target algorithm model is deployed on a vehicle-side controller, and the task order includes an algorithm file and deployment location information of the target algorithm model;

[0149] The file allocation module 20 is used to allocate the algorithm file to the virtual component corresponding to the target algorithm model according to the deployment location information;

[0150] The model training module 30 is used to execute the algorithm file through the virtual component to obtain the training result of the target algorithm model;

[0151] The result feedback module 40 is used to feed back the training results to the cloud algorithm model platform.

[0152] In an exemplary embodiment, Fig. 9 On the basis of Fig.10 As shown, the above-mentioned file allocation module 20 includes:

[0153] A path determination unit 21, configured to determine a first sub-path corresponding to a target algorithm model from among the sub-paths in each algorithm model according to the deployment location information;

[0154] The file storage unit 22 is used to store the algorithm file in the first sub-path for calling by the virtual component corresponding to the target algorithm model.

[0155] In an exemplary embodiment, the above-mentioned model training device 1 further includes:

[0156] The instruction receiving module is used to receive the recovery order corresponding to the task order issued by the cloud algorithm model platform.

[0157] In an exemplary embodiment, the result feedback module 40 includes:

[0158] A result acquisition unit, used to acquire the training result stored in the second sub-path through the vehicle-side data acquisition system;

[0159] The data uploading unit is used to generate corresponding data packets according to the training results through the vehicle-side data acquisition system, and upload the data packets to the cloud-based algorithm model platform; wherein the data packets include the working status and dynamic event data of the target algorithm model.

[0160] In an exemplary embodiment, the above-mentioned model training device 1 further includes:

[0161] A model data acquisition module is used to acquire the algorithm model data required for training the target algorithm model through virtual components;

[0162] The model data reporting module is used to upload the algorithm model data to the cloud-based algorithm model platform through the vehicle-side data acquisition system; the algorithm model data is used for the cloud-based algorithm model platform to generate algorithm files.

[0163] In an exemplary embodiment, the model data reporting module includes:

[0164] The model data reporting unit is used to upload the algorithm model data to the cloud algorithm model platform through the vehicle-side data acquisition system when the triggering conditions of the event trigger corresponding to the target algorithm model are detected.

[0165] In an exemplary embodiment, Fig.11 As shown, a model training device 2 is provided, comprising:

[0166] A task issuing module 50, used to issue a task order for training a target algorithm model to at least one vehicle-side controller;

[0167] The target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; the deployment location information is used for the vehicle-side controller to allocate the algorithm file to the virtual component corresponding to the target algorithm model; the algorithm file is used for the vehicle-side controller to execute through the virtual component to obtain the training result of the target algorithm model;

[0168] The result receiving module 60 is used to receive the training result fed back by at least one vehicle-side controller.

[0169] In an exemplary embodiment, the deployment location information is used to provide a vehicle-side controller to determine a first sub-path corresponding to a target algorithm model from among the sub-paths in each algorithm model, and store the algorithm file in the first sub-path.

[0170] In an exemplary embodiment, the above-mentioned model training device 2 further includes:

[0171] The instruction sending module is used to send a recovery order corresponding to the task order to at least one vehicle-side controller.

[0172] In an exemplary embodiment, the result receiving module 60 includes:

[0173] The result receiving unit is used to receive a data packet uploaded by at least one vehicle-side controller through a vehicle-side data acquisition system; wherein the data packet is generated based on the training results, and the data packet includes the working status and dynamic event data of the target algorithm model.

[0174] In an exemplary embodiment, the above-mentioned model training device 2 further includes:

[0175] The data compilation module is used to compile the cleaned algorithm model data to obtain the algorithm file;

[0176] The file storage module is used to store algorithm files in the algorithm warehouse.

[0177] In an exemplary embodiment, the above-mentioned model training device 2 further includes:

[0178] The model data receiving module is used to receive the algorithm model data uploaded by at least one vehicle-side controller through the vehicle-side data acquisition system when the triggering conditions of the event trigger corresponding to the target algorithm model are detected; wherein the algorithm model data is obtained by the vehicle-side controller through the virtual component corresponding to the target algorithm model.

[0179] Each module in the above-mentioned model training device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0180] In an exemplary embodiment, a computer device is provided. The computer device may be a vehicle-mounted controller, and its internal structure diagram may be as shown in FIG. Fig.12As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (Near Field Communication, NFC) or other technologies. When the computer program is executed by the processor, a model training method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or may be a key, trackball or touchpad provided on the computer device housing.

[0181] In an exemplary embodiment, a computer device is provided. The computer device may be a cloud-based algorithm model platform, and its internal structure diagram may be as follows: Fig.13 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store algorithm file data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a model training method is implemented.

[0182] Those skilled in the art will understand that the structure shown in the above figure is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0183] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps of the above-mentioned model training method when executing the computer program.

[0184] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned model training method are implemented.

[0185] In one embodiment, a computer program product is provided, comprising a computer program, which implements the steps of the above-mentioned model training method when executed by a processor.

[0186] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0187] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0188] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A model training method, characterized in that: Applied to a vehicle-side controller, the method comprises: Receiving a task order for training a target algorithm model issued by a cloud algorithm model platform; wherein the target algorithm model is deployed on the vehicle-side controller, and the task order includes an algorithm file and deployment location information of the target algorithm model; Allocating the algorithm file to the virtual component corresponding to the target algorithm model according to the deployment location information; Executing the algorithm file through the virtual component to obtain the training result of the target algorithm model; The training results are fed back to the cloud algorithm model platform.

2. The method according to claim 1, characterized in that The allocating the algorithm file to the virtual component corresponding to the target algorithm model according to the deployment location information includes: According to the deployment location information, determining a first sub-path corresponding to the target algorithm model from among the sub-paths in each algorithm model; The algorithm file is stored in the first sub-path for calling by the virtual component corresponding to the target algorithm model.

3. The method according to claim 1, characterized in that Before feeding back the training results to the cloud algorithm model platform, the method further includes: Receive a recovery order corresponding to the task order issued by the cloud algorithm model platform.

4. The method according to claim 3, characterized in that Feeding back the training results to the cloud algorithm model platform includes: Acquire the training result stored in the second sub-path through the vehicle-side data acquisition system; Through the vehicle-side data acquisition system, a corresponding data packet is generated according to the training results, and the data packet is uploaded to the cloud-based algorithm model platform; wherein the data packet includes the working status and dynamic event data of the target algorithm model.

5. The method according to claim 1, characterized in that The method further comprises: Obtaining algorithm model data required for training the target algorithm model through the virtual component; The algorithm model data is uploaded to the cloud-based algorithm model platform through the vehicle-side data acquisition system; wherein the algorithm model data is used for the cloud-based algorithm model platform to generate an algorithm file.

6. The method according to claim 5, characterized in that The method of uploading the algorithm model data to the cloud algorithm model platform through the vehicle-side data acquisition system includes: When it is detected that the triggering condition of the event trigger corresponding to the target algorithm model is met, the algorithm model data is uploaded to the cloud algorithm model platform through the vehicle-side data acquisition system.

7. A model training method, characterized in that: Applied to a cloud algorithm model platform, the method includes: Sending a task order for training a target algorithm model to at least one vehicle-side controller; The target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; the deployment location information is used for the vehicle-side controller to allocate the algorithm file to the virtual component corresponding to the target algorithm model; the algorithm file is used for the vehicle-side controller to execute through the virtual component to obtain the training result of the target algorithm model; Receive the training result fed back by the at least one vehicle-side controller.

8. The method according to claim 7, characterized in that The deployment location information is used by the vehicle-side controller to determine the first sub-path corresponding to the target algorithm model from the sub-paths in each algorithm model, and store the algorithm file in the first sub-path.

9. The method according to claim 7, characterized in that: Before receiving the training result fed back by the at least one vehicle-side controller, the method further includes: A recovery order corresponding to the task order is issued to the at least one vehicle-side controller.

10. The method according to claim 9, characterized in that The receiving the training result fed back by the at least one vehicle-side controller comprises: Receive a data packet uploaded by at least one vehicle-side controller through a vehicle-side data acquisition system; wherein the data packet is generated based on the training result, and the data packet includes the working status and dynamic event data of the target algorithm model.

11. The method according to claim 7, characterized in that Before issuing a task order for training the target algorithm model to at least one vehicle-side controller, the method further includes: Compile the cleaned algorithm model data to obtain the algorithm file; The algorithm file is stored in the algorithm warehouse.

12. The method according to claim 11, characterized in that Before compiling the cleaned algorithm model data to obtain the algorithm file, the method further includes: Receive algorithm model data uploaded by at least one vehicle-side controller through a vehicle-side data acquisition system when the triggering condition of an event trigger corresponding to the target algorithm model is detected; wherein the algorithm model data is obtained by the vehicle-side controller through a virtual component corresponding to the target algorithm model.

13. A model training device, characterized in that: Configured in a vehicle-side controller, the device includes: A task receiving module, used to receive a task order for training a target algorithm model issued by a cloud algorithm model platform; wherein the target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; A file allocation module, used to allocate the algorithm file to the virtual component corresponding to the target algorithm model according to the deployment location information; A model training module, used to execute the algorithm file through the virtual component to obtain the training result of the target algorithm model; The result feedback module is used to feed back the training results to the cloud algorithm model platform.

14. A model training device, characterized in that: Configured on a cloud algorithm model platform, the device includes: A task issuing module, used to issue a task order for training a target algorithm model to at least one vehicle-side controller; The target algorithm model is deployed on the vehicle-side controller, and the task order includes the algorithm file and deployment location information of the target algorithm model; the deployment location information is used for the vehicle-side controller to allocate the algorithm file to the virtual component corresponding to the target algorithm model; the algorithm file is used for the vehicle-side controller to execute through the virtual component to obtain the training result of the target algorithm model; The result receiving module is used to receive the training result fed back by the at least one vehicle-side controller.

15. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 12 are implemented.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

17. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images