Training method and application of unmanned vehicle DSDA system resource planning model, unmanned vehicle and storage medium

Abstract

The invention discloses an unmanned vehicle DSDA system resource planning model training method and application, an unmanned vehicle and a storage medium, and the method comprises the steps: obtaining a sample data set which comprises vehicle state data, driving environment data and bandwidth demand data, and the driving environment data comprises base station position data; the vehicle state data, the driving environment data and the bandwidth demand data are input into corresponding branch networks in a DSDA system resource planning model, fusion is carried out through a cross attention mechanism, fusion data are obtained, and the offset of the cross attention mechanism is base station position data; and adjusting model parameters of the DSDA system resource planning model based on the fused data and a preset loss function. In this way, possibility is provided for landing of the DSDA system in an unmanned vehicle application scene.

Description

Technical Field

[0001] This application belongs to the field of autonomous driving technology, specifically relating to a training method and apparatus for an unmanned vehicle DSDA system resource planning model, an unmanned vehicle DSDA system resource planning method and apparatus using the method, an unmanned vehicle, and a storage medium. Background Technology

[0002] With the development of radio frequency technology, DSDA (Dual SIM Dual Active) technology has evolved, allowing terminal devices to use two SIM cards simultaneously for calls and data services, meaning both SIM cards are active. Currently, the resource planning strategies for DSDA systems are primarily based on call and data service scenarios, and have not yet been implemented in autonomous vehicle applications.

[0003] The information disclosed in this background section is intended only to enhance the understanding of the overall background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this application is to provide a training method for the resource planning model of the DSDA system for autonomous vehicles, which is used to solve the problem that the DSDA system has no implementation solution in autonomous vehicle application scenarios.

[0005] To achieve the above objectives, this application provides a training method for a resource planning model of an autonomous vehicle DSDA system, the method comprising: Obtain a sample dataset, wherein the sample dataset includes vehicle status data, driving environment data, and bandwidth requirement data, and the driving environment data includes base station location data; The vehicle status data, driving environment data, and bandwidth requirement data are respectively input into the corresponding branch networks in the DSDA system resource planning model, and then fused through a cross-attention mechanism to obtain fused data, wherein the bias of the cross-attention mechanism is the base station location data; Based on the fused data and the preset loss function, the model parameters of the DSDA system resource planning model are adjusted.

[0006] In one embodiment, the vehicle status data includes at least one of vehicle driving mode, vehicle speed, battery charge, battery temperature, and vehicle task priority, and the vehicle driving mode includes at least one of near-field driving mode, autonomous driving mode, remote driving mode, and standby mode.

[0007] In one embodiment, the driving environment data further includes at least one of signal quality parameters, network performance parameters, geographical location, base station spectrum resources, and coverage data; and / or, The bandwidth requirement data includes at least one of the bandwidth requirements and latency requirements for each type of task.

[0008] In one embodiment, adjusting the model parameters of the DSDA system resource planning model based on the fused data and a preset loss function specifically includes: Based on the matching degree of bandwidth requirements for each type of task, a first sub-loss function is constructed; and based on the matching degree of latency requirements for each type of task, a second sub-loss function is constructed. A comprehensive loss function is constructed based on the first and second sub-loss functions, and is determined as the preset loss function; and / or, The DSDA system resource planning model includes a Transformer-based multi-branch structure.

[0009] This application also provides a resource planning method for an autonomous vehicle DSDA system, the method comprising: The system acquires the current vehicle status data, driving environment data, and bandwidth requirement data of the target vehicle, and then calls the target model to process them for resource planning of the DSDA system. The target model is trained using the method described above.

[0010] In one embodiment, calling the target model for resource planning of the DSDA system specifically includes: The target model outputs a resource planning strategy, wherein the resource planning strategy includes at least one of the following: a bandwidth allocation strategy for each task, an antenna signal optimization strategy, and a dual-card task scheduling strategy. The antenna signal optimization strategy includes antenna adjustment angle and / or antenna signal mode.

[0011] This application also provides a training device for an autonomous vehicle DSDA system resource planning model, comprising: The first acquisition module is used to acquire a sample dataset, wherein the sample dataset includes vehicle status data, driving environment data, and bandwidth requirement data, and the driving environment data includes base station location data. The input module is used to input the vehicle status data, driving environment data and bandwidth requirement data into the corresponding branch networks in the DSDA system resource planning model, and fuse them through a cross-attention mechanism to obtain fused data, wherein the bias of the cross-attention mechanism is the base station location data; The parameter adjustment module is used to adjust the model parameters of the DSDA system resource planning model based on the fused data and the preset loss function.

[0012] This application also provides a resource planning device for an unmanned vehicle DSDA system, including a second acquisition module and a calling module. The second acquisition module is used to acquire the current vehicle status data, driving environment data and bandwidth requirement data of the target vehicle, and to call the target model through the calling module to process them for resource planning of the DSDA system. The target model is trained using the method described above.

[0013] This application also provides an unmanned vehicle, including: At least one processor; and The memory stores instructions that, when executed by the at least one processor, cause the at least one processor to execute the training method for the autonomous vehicle DSDA system resource planning model as described above, or the autonomous vehicle DSDA system resource planning method as described above.

[0014] This application also provides a machine-readable storage medium storing executable instructions, which, when executed, cause the machine to perform the training method for the unmanned vehicle DSDA system resource planning model as described above, or the unmanned vehicle DSDA system resource planning method as described above.

[0015] Compared with existing technologies, the training method of the resource planning model of the autonomous vehicle DSDA system according to this application trains the model using a sample dataset including vehicle state data, driving environment data, and bandwidth demand data. These different dimensions of information, cleverly configured in the autonomous vehicle application scenario, provide a foundation for the stability, diversity, and comprehensiveness of the model's output decisions. Furthermore, during training, various types of data are input into the corresponding branch networks of the model and fused through a cross-attention mechanism. The model parameters are adjusted based on the fused data and a preset loss function. In this process, the bias of the cross-attention mechanism is set to the relatively fixed factor of base station location data, making the model's learning direction more stable and improving the model's generalization ability, thus providing a possibility for the implementation of the DSDA system in autonomous vehicle application scenarios.

[0016] On another front, when using the model trained in this application for DSDA system resource planning, it is not limited to a simple allocation of a certain type of resource. Instead, it proposes multi-type / multi-angle resource planning strategies, including bandwidth allocation strategies for each task, antenna signal optimization strategies, and dual-card task scheduling strategies. These strategies can improve multiple performance indicators such as driving safety, data transmission latency, and bandwidth utilization, thereby enhancing the adaptability of the DSDA system in autonomous vehicle application scenarios. Furthermore, since it utilizes machine learning, it can better adapt to the dynamic resource allocation requirements of autonomous vehicle application scenarios. Attached Figure Description

[0017] Figure 1This is a training method for an unmanned vehicle DSDA system resource planning model according to an embodiment of this application, and an application scenario diagram of the unmanned vehicle DSDA system resource planning method. Figure 2 This is a flowchart of a training method for a resource planning model of an unmanned vehicle DSDA system according to an embodiment of this application; Figure 3 This is a schematic diagram of the architecture of a model in a training method for a resource planning model of an unmanned vehicle DSDA system according to an embodiment of this application. Figure 4 This is a physical model of a cellular network wireless environment exemplified in one embodiment of this application; Figure 5 This is a block diagram of a training device for a resource planning model of an unmanned vehicle DSDA system according to an embodiment of this application; Figure 6 A block diagram of a resource planning device for an unmanned vehicle DSDA system according to an embodiment of this application; Figure 7 This is a hardware structure diagram of an unmanned vehicle according to an embodiment of this application. Detailed Implementation

[0018] The present application will now be described in detail with reference to the embodiments shown in the accompanying drawings. However, these embodiments do not limit the present application, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the protection scope of the present application.

[0019] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] Before introducing the embodiments of this application, the basic technologies and some technical terms involved in the embodiments of this application will be explained illustratively: Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, and inductive learning.

[0021] Maneuvering Driving (MD) refers to a driving mode where the driver, inside the vehicle, performs low-speed, short-distance, and precise maneuvering. Its core characteristic is that the driver is present and responsible for control throughout the process, but the vehicle, through environmental perception and auxiliary control systems, performs precise trajectory tracking in complex local scenarios to achieve specific driving objectives. Typical applications include automatic / assisted parking (such as perpendicular parking and parallel parking), maneuvering in narrow parking spaces, low-speed following through other vehicles (such as passing through construction zones), and fixed-route summoning (such as driving from a parking space to the driver).

[0022] Autonomous Driving (AD) refers to the ability to guide and make decisions regarding a vehicle's driving tasks without requiring a driver to perform physical driving operations, thus enabling the vehicle to drive safely. Autonomous driving technologies typically include high-precision mapping, environmental perception, behavioral decision-making, path planning, and motion control. Autonomous driving systems are systems that achieve different levels of autonomous driving functions in vehicles, such as Level 2 driver assistance systems, Level 3 high-speed autonomous driving systems requiring human supervision, and Level 4 / 5 highly / fully autonomous driving systems.

[0023] Stereo-Driving (SD) refers to a driver operating the vehicle from a physically isolated location (e.g., a cockpit), acquiring real-time multimodal information about the vehicle's surroundings (such as video, radar, and vehicle status) via a communication network, and remotely controlling actuators such as the steering wheel and pedals to manage the vehicle. Typical applications include remote takeover of autonomous vehicles, operations in special areas (such as mines and ports), vehicle rescue, and testing and verification.

[0024] Standby mode: This is a standby state in which some core electronic control units, network communication modules, and sensors continue to operate at low power even after the powertrain is turned off. In this mode, the vehicle can respond to remote commands in real time (such as unlocking / starting via a mobile app), monitor anti-theft alarms, receive OTA upgrade pushes, and provide instant wake-up support for intelligent functions such as keyless entry and remote air conditioning pre-start, achieving an intelligent experience of "uninterrupted power supply even when away from the vehicle." This mode strikes a balance between functional availability and energy consumption and is one of the fundamental capabilities of intelligent connected vehicles.

[0025] Typically, DSDA systems employ fixed-rule bandwidth and signal resource allocation strategies. Based on the type of service, the DSDA system allocates it to the corresponding SIM card link by default. A typical strategy separates voice (VoLTE) and data services. For example, if SIM card 1 is conducting a VoLTE HD call and needs to download a large file, the DSDA system will continue the voice call on SIM card 1 to ensure uninterrupted call quality. The file download task will be automatically assigned to the data network on SIM card 2, thus guaranteeing the absolute priority and stability of basic communication from a hardware control strategy perspective.

[0026] It can be seen that most of these technologies simply prioritize business operations and allocate corresponding SIM card resources for resource planning in DSDA systems. However, in autonomous vehicle applications, integrating the DSDA system with various business operations and needs of autonomous vehicles, and providing more multi-dimensional resource planning strategies from the perspective of improving the driving quality and safety of autonomous vehicles, is an issue that existing technologies have not yet considered. Therefore, one of the contributions of the technical solution in this application is reflected in: the technical framework for implementing the DSDA system in autonomous vehicle driving scenarios, which combines vehicle status data, driving environment data, and bandwidth requirement data with machine learning models to comprehensively plan bandwidth allocation, antenna signal optimization, and task scheduling.

[0027] Figure 1 This is an optional schematic diagram illustrating the system architecture for implementing the DSDA system in an autonomous vehicle, as described in an embodiment of this application. Figure 1As shown, the data acquisition module, either onboard or independently configured on the autonomous vehicle, collects multi-dimensional data, including vehicle status data, driving environment data, and bandwidth requirement data, and sends it to the server. The server then inputs this data into the DSDA system resource planning model for training, adjusting and optimizing the model parameters. The training process can be offline training based on collected historical data, or it can be further refined and optimized through incremental learning. When the autonomous vehicle sends the currently collected data to the server, the server can use the trained model to output the current DSDA system resource planning strategy and send it back to the autonomous vehicle for execution.

[0028] In the above architecture, the server can be a third-party server, such as an enterprise server, which is used to perform the steps of model training and DSDA system resource planning using the model; or, different servers can be configured to perform the steps of model training and DSDA system resource planning using the model respectively. For example, model training can be performed on an enterprise server, and another server can be integrated on the autonomous vehicle to perform DSDA system resource planning directly on the autonomous vehicle based on the data collected by the data acquisition module and the trained model.

[0029] Continue to participate Figure 1 In a specific scenario applying the training method and system resource planning method of the unmanned vehicle DSDA system resource planning model of this application, servers and terminal devices may also be included. The terminal devices may include in-vehicle terminals and user terminals. In-vehicle terminals may include vehicle computers or on-board units (OBUs), etc. In-vehicle terminals may also be applications (APPs) on the terminal, APPs on smart rearview mirrors, APPs or mini-programs on mobile phones, etc., without limitation. User terminals (UEs) may be wireless or wired terminal devices. Wireless terminal devices can refer to devices with wireless transceiver capabilities. User terminals may be mobile phones, tablets, computers with wireless transceiver capabilities, virtual reality (VR) user devices, augmented reality (AR) user devices, smart voice interaction devices, smart home appliances, in-vehicle terminals, aircraft, etc., without limitation.

[0030] Taking an in-vehicle device as an example, a DSDA system resource planning application is installed on the in-vehicle device. The server can send the trained model to the in-vehicle device. At the same time, the in-vehicle device, combined with the data collected by the data acquisition module, determines the current DSDA system resource planning scheme. Users can confirm the execution of all or part of the resource planning scheme through the operation interface of the in-vehicle device, or the resource planning scheme can be automatically issued and executed without the user's awareness.

[0031] It is understood that the above is only an example of a server or vehicle-mounted device to illustrate one possible execution subject of the various embodiments of this application. In more embodiments, all the above operations can also be directly executed by a dedicated terminal device with sufficient computing power. This application does not limit this.

[0032] Specific reference Figure 2 This paper introduces an embodiment of the training method for the resource planning model of the unmanned vehicle DSDA system of this application. In this embodiment, the method includes: S11. Obtain the sample dataset.

[0033] The sample dataset includes vehicle status data, driving environment data, and bandwidth requirement data. These three types of data will be explained in detail below.

[0034] Vehicle status data Vehicle status data includes at least one of the following: vehicle driving mode, vehicle speed, battery charge, battery temperature, and vehicle task priority.

[0035] ① The vehicle driving model includes at least one of the following: near-field driving mode, autonomous driving mode, remote driving mode, and standby mode. The following examples explain how the differences in these vehicle driving modes may affect the decision on resource planning for the DSDA system.

[0036] In near-field driving mode, the vehicle primarily relies on short-range communication technologies such as Wi-Fi and Bluetooth to transmit commands and exchange data with mobile terminals, roadside equipment, or other vehicles within a limited range, without needing to continuously interact with the cloud via cellular networks. Communication content mainly consists of concise control commands, status confirmations, and small-scale sensor data. Simultaneously, the system maintains low-frequency, low-volume cloud connections for purposes such as critical status synchronization, task distribution, and basic monitoring, with relatively low requirements for network bandwidth and latency.

[0037] In autonomous driving mode, high-precision positioning is a fundamental functional requirement. Typically, vehicles can achieve centimeter-level high-precision positioning using Real-Time Dynamic Carrier Phase Differential (RTK) technology. In this process, the vehicle receives multi-frequency GNSS satellite signals and fuses them with real-time differential correction data provided by a ground-based augmentation network or a self-built reference station to eliminate common errors such as satellite orbital errors, clock biases, and ionospheric interference. Simultaneously, it can optionally incorporate an inertial navigation system (INS) and onboard sensors such as wheel speedometers to maintain continuous and reliable positioning even during brief interruptions in satellite signals. It can be seen that while the reception of data such as differential correction data may place some demands on the network, a large amount of perception data is calculated on the vehicle side, and the requirements for network latency and bandwidth are generally considered moderate.

[0038] In remote driving mode, it is typically necessary to simultaneously achieve ultra-high bandwidth bidirectional data streaming and extremely low end-to-end latency. For example, in the downlink direction, multiple high-definition video streams from the vehicle need to be transmitted to the remote user's cockpit in real time and stably; otherwise, the remote user's view may be blurry or choppy. In the uplink direction, every control command from the user must be delivered to the vehicle and executed at near-zero latency. During this process, the latency of the entire "perception-decision-control" closed loop needs to be stabilized to the tens of milliseconds. Any insufficient bandwidth or latency jitter will directly lead to control failure and cause unnecessary safety risks. It is generally believed that the requirements for network latency and bandwidth are the highest.

[0039] In standby mode, the vehicle's network requirements are typically significantly reduced, but not completely disconnected. This mode primarily maintains a minimal network connection to support low-power, low-frequency periodic status reporting (such as vehicle location, battery level, and system health status) and real-time monitoring and rapid response to commands such as remote wake-up, OTA upgrades, and anti-theft. Therefore, this mode is generally considered to have the lowest requirements for network latency and bandwidth, provided that network connection reliability is guaranteed.

[0040] It can be seen that differences in vehicle status data can directly affect a vehicle's current network latency and bandwidth requirements, which in turn may affect the model's resource planning for the DSDA system. Therefore, the applicant proposed using this data as one of the sample datasets during the model training phase.

[0041] ② Vehicle speed is often positively correlated with network latency and bandwidth requirements. For example, when a vehicle is traveling at high speed, it traverses a larger area per unit time and switches between networks covered by different base stations more frequently, thus requiring more network resources for lower latency and more stable connections. As another example, in the aforementioned incremental learning scenario, increased speed means an increased amount of perceptual data, which may require greater bandwidth and a higher frequency to upload the perceptual data to the server for incremental model learning.

[0042] In this application embodiment, the vehicle speed can be collected as an actual data value, or it can be divided into speed ranges after the vehicle speed is collected, such as 0-10km / h, 10-25km / h, and above 25km / h. This application does not impose any restrictions on this.

[0043] ③ Battery level is a crucial constraint determining network connectivity strategies and power management. In low-battery conditions, priority is typically given to ensuring core vehicle operation and safety functions. Degraded communication modes can be adopted to save energy, such as switching from continuous high-bandwidth, low-latency connections to lower-bandwidth, higher-latency, or even periodic sleep / wake-up modes, reducing unnecessary data synchronization and high-definition video streaming.

[0044] ④ Battery temperature directly affects battery / vehicle safety, while high-power, high-bandwidth service demands increase the vehicle's electrical and thermal loads. For example, when battery temperature is high, reducing the applicable frame rate for video streaming may help improve safety. This feedback to DSDA's resource planning decisions may not only provide recommendations on the applicable frame rate for video streaming but may also simultaneously reduce the bandwidth allocation for video services.

[0045] Other vehicle status data, such as the impact of vehicle task priorities on the planning of DSDA system resources, are relatively easy to understand. Typically, the highest priority tasks will unconditionally preempt bandwidth and guarantee the lowest latency, while the opposite is true for lower priority tasks. In this embodiment, these vehicle status data are also included in the possible collection scope of the sample data.

[0046] Driving environment data The driving environment data includes base station location data, as well as at least one of the following: signal quality parameters, network performance parameters, geographical location, base station spectrum resources, and coverage data.

[0047] ① Base station location can affect signal quality, available bandwidth, environmental interference, signal coverage, and overlap. For example, when a vehicle senses it is approaching the edge of a base station's coverage area or a weak signal area, it can request more redundant resources (such as bandwidth and a shorter reporting cycle) in advance to cache critical data. In densely populated base station areas, vehicles can combine location information to select the cell with the strongest signal and lightest load to access, thereby obtaining better bandwidth and latency. Furthermore, since base station location is relatively fixed data, including it in the sample dataset essentially transforms it into one of the stable bases for planning DSDA system resources.

[0048] ② Signal quality parameters can be, for example, signal strength, signal-to-noise ratio, and attenuation. In specific 4G / 5G cellular network application scenarios, RSPR (Reference Signal Receiving Power) and SINR (Signal to Interference plus Noise Ratio) can be collected, and similarly divided into intervals, such as rsrp -85~-95dBm\-95dBm~-105dBm\<-105dBm, sinr 0>\0~10\>10, etc. This application does not impose any restrictions on this.

[0049] ③ Network performance parameters can include, for example, network latency, jitter, and bandwidth. Different types of tasks may have different requirements for various network performance parameters. For example, the transfer of large files and video streaming may have the highest bandwidth requirements and be less sensitive to latency and jitter; however, data transmission in remote driving mode, as mentioned above, has the highest requirements for latency and low jitter, while the bandwidth requirements may be relatively secondary.

[0050] ④ Geographic location data can be, for example, latitude and longitude data and road type data. Latitude and longitude can be combined with base station locations to determine the vehicle's position relative to the base station. Road types may include open urban roads and closed structured roads (such as ports, enclosed express delivery transfer parks, parking lots), etc. Typically, on open urban roads, vehicles may need to handle multiple tasks simultaneously, such as high-precision positioning and cloud monitoring; while on closed structured roads, vehicle operation is more predictable and the tasks are relatively simple, thus creating differences in the demand for DSDA system resources.

[0051] ⑤ Base station spectrum resources and coverage data can include, for example, the base station's frequency band, frequency point, bandwidth, and cell type. These factors collectively influence the coverage area, network capacity, and theoretical speed limit provided by the base station, thereby affecting the resource planning of the DSDA system.

[0052] Bandwidth demand data The bandwidth requirement data includes at least one of the bandwidth and latency requirements for each type of task. For example, the bandwidth requirements for near-field driving mode, autonomous driving mode, and remote driving mode can be 3Mbps, 5Mbps, and 10Mbps, respectively, and the latency requirements can be <1000ms, <500ms, and <300ms, respectively.

[0053] In this embodiment, normalization and denoising processes can be performed on various types of data in the sample dataset to generate standardized input features. For example, driving modes can correspond to one-hot encoding, signal quality to numerical features, and bandwidth requirements to time series data, etc. This embodiment will not elaborate on these aspects.

[0054] S12. Input the vehicle status data, driving environment data, and bandwidth requirement data into the corresponding branch networks in the DSDA system resource planning model, and fuse them through the cross-attention mechanism to obtain fused data.

[0055] S13. Based on the fused data and the preset loss function, adjust the model parameters of the DSDA system resource planning model.

[0056] Coordination Figure 3 In one exemplary embodiment, the DSDA system resource planning model may include a Transformer-based multi-branch structure to adapt to heterogeneous data from multiple modalities and sources.

[0057] Specifically, the model structure can establish an independent dedicated branch for each data modality or data type in the sample dataset. Each branch can include a feature embedding layer tailored to the characteristics of that modality or data type, and a Transformer encoder to extract deep features. The model structure can also include a cross-modal fusion module (based on a cross-attention mechanism) to project and align the feature sequences output from each branch, achieving information fusion between different modalities or data types. Subsequently, the obtained fused data (i.e., joint representation) is input into the downstream task head to output planning decisions.

[0058] In this embodiment, during model training, the bias of the cross-attention mechanism is set to base station location data. In the Transformer's attention mechanism, the bias can adjust or emphasize the correlation strength between different information elements. Setting the relatively fixed data of base station location as the bias allows the model to pay special attention to and weight the correlation between different features based on the base station location, ensuring the stability of the model's learning direction and improving the model's generalization ability.

[0059] Coordination Figure 4This paper exemplifies a classic physical model of a cellular network wireless environment using the "terminal location" and two "base station locations," along with their derived distances, path losses, signal overlap areas, and handover boundaries. Understandably, the base station itself, and related factors between base stations such as signal coverage, load capacity, interference patterns, signal overlap areas, handover boundaries, and cooperation opportunities, can be considered "causes" that may cause path loss, signal strength attenuation, and multipath effects. By using the base station location as a strong guide, the model can more easily learn the related "results" (signal attenuation, interference changes, etc.) from the collected sample data, thereby internalizing these physical laws. Based on this, the resource planning scheme output by the model will be more physically interpretable and forward-looking. For example, it can reserve resources in advance when approaching the handover boundary to achieve seamless handover; predict signal strength trends based on the relative location and coverage of base stations, thereby dynamically adjusting resource planning strategies.

[0060] In this embodiment, when constructing the preset loss function, a first sub-loss function can be constructed based on the matching degree of bandwidth requirements of each type of task; and a second sub-loss function can be constructed based on the matching degree of latency requirements of each type of task; then a comprehensive loss function can be constructed based on the first sub-loss function and the second sub-loss function, which is determined as the preset loss function.

[0061] For the first sub-loss function, this means that for each task, the gap between its requested bandwidth and the actual allocated bandwidth can be calculated (e.g., using the mean squared error method) to penalize strategies that fail to allocate sufficient bandwidth. For example, Task A (remote control): requests 20 Mbps bandwidth; Task B (high-precision map update): requests 5 Mbps bandwidth; Task C (status reporting): requests 0.2 Mbps bandwidth; the model allocates 18 Mbps, 6 Mbps, and 0.05 Mbps bandwidth to the above tasks respectively. In this case, the first sub-loss function will mainly penalize the 2 Mbps shortfall in Task A and the 0.15 Mbps shortfall in Task C.

[0062] For the second sub-loss function, its calculation logic can be a latency-based "upper limit constraint," that is, calculating the difference between the actual latency of the task and its maximum tolerable latency, and including the loss when the actual latency exceeds the tolerable latency. For example, the maximum tolerable latency for three tasks are: Task A: 100ms, Task B: 500ms, and Task C: 1000ms. If, after resource planning, the actual latency of the corresponding tasks is 80ms, 800ms, and 200ms, then the second sub-loss function will only penalize Task B (whose actual latency exceeds the limit by 300ms).

[0063] The first sub-loss function and the second sub-loss function can be directly added together, or they can be assigned a corresponding weight coefficient and then added together to determine the comprehensive loss function. This application does not impose any restrictions on this.

[0064] This application also provides a method for resource planning of an autonomous vehicle DSDA system using a target model obtained by the above training method.

[0065] Specifically, the system can acquire the target vehicle's current vehicle status data, driving environment data, and bandwidth requirement data, and call the target model to process them for resource planning of the DSDA system, outputting a resource planning strategy. The output resource planning strategy may include at least one of the following: bandwidth allocation strategy for each task, antenna signal optimization strategy, and dual-card task scheduling strategy. The antenna signal optimization strategy includes antenna adjustment angle and / or antenna signal mode.

[0066] As can be seen, the embodiments of this application provide a DSDA system resource comprehensive planning method that matches the application scenario of autonomous vehicles. The bandwidth allocation strategy for each task can be to allocate a specific data rate to each task when the total bandwidth is limited; the antenna signal optimization strategy can actively improve signal quality and cope with signal fluctuations by adjusting the antenna radiation direction, 5G / 4G signal mode, etc., which is a physical layer optimization strategy; the dual-SIM task scheduling strategy determines which task is carried by which SIM card and how the two SIM cards coordinate. This architecture of multi-modal / multi-type data selection and multi-type / multi-angle resource planning strategy output is one of the core creative manifestations of the technical solution of this application.

[0067] For example, in the bandwidth allocation strategy for each task, Task 1: ensures its basic requirement of 4Mbps and allocates it to the stable channel of the secondary card (connected to high-quality base station B); Task 2: splits it into two parts. The latency-sensitive small differential map data is allocated to the primary card, utilizing its remaining bandwidth; the latency-insensitive large map packets are scheduled to be downloaded on the secondary card's 20Mbps peak bandwidth.

[0068] For example, in antenna signal optimization strategies, one aspect is adjusting the antenna angle: the model determines that the vehicle is heading towards base station B and the main SIM card signal is degraded, therefore it is recommended to dynamically adjust the main lobe direction of the main antenna beam from pointing towards base station A to pointing towards the optimal handover direction between base stations A and B, in order to optimize signal continuity during handover. Another aspect is switching the signal mode: the model determines that the vehicle is located in a densely built-up urban area, therefore it is recommended to switch the antenna from wide-coverage mode to directional mode to penetrate obstacles.

[0069] For example, in the dual-SIM task scheduling strategy, the model suggests that in remote driving mode, the primary SIM card with better network connectivity should be responsible for the video stream data from three cameras, while the secondary SIM card should be responsible for the video stream data from one camera.

[0070] Reference Figure 5 , an embodiment of a training device for the resource planning model of the unmanned vehicle DSDA system of the present application is introduced. In this embodiment, the training device for the resource planning model of the unmanned vehicle DSDA system includes a first acquisition module 211, an input module 212, and a parameter adjustment module 213.

[0071] The first acquisition module 211 is used to acquire a sample data set. Among them, the sample data set includes vehicle state data, driving environment data, and bandwidth demand data. The driving environment data includes base station location data; the input module 212 is used to input the vehicle state data, driving environment data, and bandwidth demand data into the corresponding branch networks in the DSDA system resource planning model respectively, and fuse them through a cross-attention mechanism to obtain fused data. Among them, the bias of the cross-attention mechanism is the base station location data; the parameter adjustment module 213 is used to adjust the model parameters of the DSDA system resource planning model based on the fused data and a preset loss function.

[0072] In one embodiment, the vehicle state data includes at least one of a vehicle driving mode, a vehicle speed, a battery power, a battery temperature, and a vehicle task priority. The vehicle driving mode includes at least one of a near-field driving mode, an autonomous driving mode, a remote driving mode, and a standby mode.

[0073] In one embodiment, the driving environment data further includes at least one of a signal quality parameter, a network performance parameter, a geographical location, a base station spectrum resource, and a coverage range data.

[0074] In one embodiment, the bandwidth demand data includes at least one of the bandwidth demand and the latency demand of each type of task.

[0075] In one embodiment, adjusting the model parameters of the DSDA system resource planning model based on the fused data and a preset loss function specifically includes: constructing a first sub-loss function based on the matching degree of the bandwidth demand of each type of task; and constructing a second sub-loss function based on the matching degree of the latency demand of each type of task; constructing a comprehensive loss function based on the first sub-loss function and the second sub-loss function, and determining it as the preset loss function.

[0076] In one embodiment, the DSDA system resource planning model includes a multi-branch structure based on Transformer.

[0077] Reference Figure 6This paper introduces an embodiment of the unmanned vehicle DSDA system resource planning device of this application. In this embodiment, the unmanned vehicle DSDA system resource planning device includes a second acquisition module 221 and an invocation module 222. The second acquisition module 221 is used to acquire the current vehicle state data, driving environment data and bandwidth requirement data of the target vehicle, and to call the target model trained by the above method through the invocation module 222 to process it for resource planning of the DSDA system.

[0078] As per the above reference Figures 1 to 4 This specification describes a training method for an autonomous vehicle DSDA system resource planning model and an autonomous vehicle DSDA system resource planning method according to embodiments thereof. The details mentioned in the above description of the method embodiments also apply to the training device for the autonomous vehicle DSDA system resource planning model and the autonomous vehicle DSDA system resource planning device according to embodiments thereof. These devices can be implemented in hardware, software, or a combination of hardware and software.

[0079] Figure 7 A hardware structure diagram of an unmanned vehicle according to an embodiment of this specification is shown. Figure 7 As shown, the unmanned vehicle 30 may include at least one processor 31, a memory 32 (e.g., non-volatile memory), a RAM 33, and a communication interface 34, and the at least one processor 31, memory 32, RAM 33, and communication interface 34 are connected together via an internal bus 35. At least one processor 31 executes at least one computer-readable instruction stored or encoded in the memory 32.

[0080] It should be understood that the computer-executable instructions stored in memory 32, when executed, cause at least one processor 31 to perform the above-described combinations in the various embodiments of this specification. Figures 1 to 4 The description includes various operations and functions.

[0081] In the embodiments of this specification, the unmanned vehicle 30 can be configured with a functional terminal to carry the above-mentioned hardware structure. The terminal may include, but is not limited to: personal computer, server computer, workstation, desktop computer, laptop computer, notebook computer, mobile electronic device, smartphone, tablet computer, cellular phone, personal digital assistant (PDA), handheld device, messaging device, wearable electronic device, consumer electronic device, etc.

[0082] According to one embodiment, a program product, such as a machine-readable medium, is provided. The machine-readable medium may have instructions (i.e., the elements implemented in software as described above), which, when executed by a machine, cause the machine to perform the above-described combinations of the various embodiments of this specification. Figures 1-4The various operations and functions described. Specifically, a system or apparatus equipped with a readable storage medium storing software program code that implements the functions of any of the embodiments described above, and enabling the computer or processor of the system or apparatus to read and execute the instructions stored in the readable storage medium.

[0083] In this case, the program code itself, which can be read from the readable medium, can perform the functions of any of the above embodiments, and therefore the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of this specification.

[0084] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer or the cloud via a communication network.

[0085] Those skilled in the art will understand that the various embodiments disclosed above can be modified and varied without departing from the spirit of the invention. Therefore, the scope of protection of this specification should be defined by the appended claims.

[0086] It should be noted that not all steps and units in the above process and system structure diagrams are mandatory; some steps or units can be omitted according to actual needs. The execution order of each step is not fixed and can be determined as needed. The device structure described in the above embodiments can be a physical structure or a logical structure. That is, some units may be implemented by the same physical client, or some units may be implemented by multiple physical clients, or they may be jointly implemented by certain components in multiple independent devices.

[0087] In the above embodiments, the hardware units or modules can be implemented mechanically or electrically. For example, a hardware unit, module, or processor may include permanent dedicated circuitry or logic (such as a dedicated processor, FPGA, or ASIC) to perform the corresponding operation. The hardware unit or processor may also include programmable logic or circuitry (such as a general-purpose processor or other programmable processor), which can be temporarily configured by software to perform the corresponding operation. The specific implementation method (mechanical, dedicated permanent circuitry, or temporarily configured circuitry) can be determined based on cost and time considerations.

[0088] The specific embodiments described above with reference to the accompanying drawings are exemplary embodiments, but do not represent all embodiments that can be implemented or fall within the scope of the claims. The term "exemplary" as used throughout this specification means "serving as an example, instance, or illustration" and does not imply that it is "preferred" or "advantageous" compared to other embodiments. Specific details are included to provide an understanding of the described techniques. However, these techniques can be practiced without these specific details. In some instances, well-known structures and apparatuses are shown in block diagram form to avoid obscuring the concepts of the described embodiments.

[0089] The foregoing description of this disclosure is provided to enable any person skilled in the art to implement or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles applicable herein can be applied to other variations without departing from the scope of this disclosure. Therefore, this disclosure is not limited to the examples and designs described herein, but is consistent with the widest scope of the principles and novel features disclosed herein.

Claims

1. A training method for a resource planning model of an unmanned vehicle DSDA system, characterized in that, The method includes: Obtain a sample dataset, wherein the sample dataset includes vehicle status data, driving environment data, and bandwidth requirement data, and the driving environment data includes base station location data; The vehicle status data, driving environment data, and bandwidth requirement data are respectively input into the corresponding branch networks in the DSDA system resource planning model, and then fused through a cross-attention mechanism to obtain fused data, wherein the bias of the cross-attention mechanism is the base station location data; Based on the fused data and the preset loss function, the model parameters of the DSDA system resource planning model are adjusted.

2. The training method for the resource planning model of the unmanned vehicle DSDA system according to claim 1, characterized in that, The vehicle status data includes at least one of the following: vehicle driving mode, vehicle speed, battery charge, battery temperature, and vehicle task priority. The vehicle driving mode includes at least one of the following: near-field driving mode, autonomous driving mode, remote driving mode, and standby mode.

3. The training method for the resource planning model of the unmanned vehicle DSDA system according to claim 1, characterized in that, The driving environment data also includes at least one of the following: signal quality parameters, network performance parameters, geographical location, base station spectrum resources, and coverage data; and / or, The bandwidth requirement data includes at least one of the bandwidth requirements and latency requirements for each type of task.

4. The training method for the resource planning model of the unmanned vehicle DSDA system according to claim 1, characterized in that, Based on the fused data and the preset loss function, the model parameters of the DSDA system resource planning model are adjusted, specifically including: Based on the matching degree of bandwidth requirements for each type of task, a first sub-loss function is constructed; and based on the matching degree of latency requirements for each type of task, a second sub-loss function is constructed. A comprehensive loss function is constructed based on the first and second sub-loss functions, and is determined as the preset loss function; and / or, The DSDA system resource planning model includes a Transformer-based multi-branch structure.

5. A resource planning method for an unmanned vehicle DSDA system, characterized in that, The method includes: The system acquires the current vehicle status data, driving environment data, and bandwidth requirement data of the target vehicle, and calls the target model to process them for resource planning of the DSDA system. The target model is trained using the method described in any one of claims 1-4.

6. The resource planning method for an unmanned vehicle DSDA system according to claim 5, characterized in that, The resource planning of a DSDA system using the target model specifically includes: The target model outputs a resource planning strategy, wherein the resource planning strategy includes at least one of the following: a bandwidth allocation strategy for each task, an antenna signal optimization strategy, and a dual-card task scheduling strategy. The antenna signal optimization strategy includes antenna adjustment angle and / or antenna signal mode.

7. A training device for a resource planning model of an unmanned vehicle DSDA system, characterized in that, include: The first acquisition module is used to acquire a sample dataset, wherein the sample dataset includes vehicle status data, driving environment data, and bandwidth requirement data, and the driving environment data includes base station location data. The input module is used to input the vehicle status data, driving environment data and bandwidth requirement data into the corresponding branch networks in the DSDA system resource planning model, and fuse them through a cross-attention mechanism to obtain fused data, wherein the bias of the cross-attention mechanism is the base station location data; The parameter adjustment module is used to adjust the model parameters of the DSDA system resource planning model based on the fused data and the preset loss function.

8. A resource planning device for an unmanned vehicle DSDA system, characterized in that, This includes a second acquisition module and a calling module. The second acquisition module is used to acquire the current vehicle status data, driving environment data and bandwidth requirement data of the target vehicle, and to call the target model through the calling module to process them for resource planning of the DSDA system. The target model is trained by the method described in any one of claims 1-4.

9. An unmanned vehicle, characterized in that, include: At least one processor; as well as The memory stores instructions that, when executed by the at least one processor, cause the at least one processor to perform the training method for the autonomous vehicle DSDA system resource planning model as described in any one of claims 1 to 4, or the autonomous vehicle DSDA system resource planning method as described in any one of claims 5 to 6.

10. A machine-readable storage medium storing executable instructions, characterized in that, When the instruction is executed, it causes the machine to perform the training method for the unmanned vehicle DSDA system resource planning model as described in any one of claims 1 to 4, or the unmanned vehicle DSDA system resource planning method as described in any one of claims 5 to 6.

Images