A method, system, device, and storage medium for intelligent driving network topology reconstruction.

Abstract

This invention discloses a method, system, device, and storage medium for intelligent driving network topology reconstruction. The method includes: acquiring multi-source intelligent driving information, performing feature extraction and fusion processing on the multi-source information to obtain a spatiotemporal semantic map; constructing vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and simulating intelligent driving interaction behaviors between the agents based on the spatiotemporal semantic map; optimizing the network topology using a network optimization model based on the intelligent driving interaction behaviors to obtain predicted communication links; and adjusting the communication links between the vehicle, RSU, and the cloud based on the predicted communication links to reconstruct the network topology. This application processes multi-source intelligent driving information to obtain a spatiotemporal semantic map, providing data support; it optimizes the topology of the network optimization model using multiple intelligent agents, improving the real-time performance and reliability of topology reconstruction; and it can be widely applied in the field of Internet of Things (IoT) technology.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and in particular to a method, system, device, and storage medium for intelligent driving network topology reconstruction. Background Technology

[0002] In the field of L4 autonomous driving, achieving a reliable and efficient communication network is key to ensuring the safe and stable operation of vehicles.

[0003] Currently, relevant technologies mainly employ relatively fixed network topologies or dynamic adjustment methods based on simple rules. For example, some solutions connect vehicles to surrounding infrastructure (such as RSUs) through a pre-defined static network topology. In this approach, the communication link between the vehicle and the RSU is determined at the initial stage. While this ensures basic communication to a certain extent, it lacks the ability to adapt to dynamically changing environments. When encountering sudden traffic events (such as traffic accidents leading to temporary road closures) or significant changes in vehicle density, network performance deteriorates sharply, failing to meet the stringent real-time and reliability requirements of L4 autonomous driving. Summary of the Invention

[0004] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0005] Therefore, the purpose of this invention is to provide a reliable intelligent driving network topology reconstruction method, system, device, and storage medium.

[0006] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of the present invention include:

[0007] On one hand, embodiments of the present invention provide a method for intelligent driving network topology reconstruction, including the following steps: acquiring multi-source intelligent driving information, and performing feature extraction and fusion processing on the multi-source intelligent driving information to obtain a spatiotemporal semantic map; constructing vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and simulating intelligent driving interaction behaviors between intelligent agents based on the spatiotemporal semantic map; optimizing the network topology through a network optimization model based on the intelligent driving interaction behaviors to obtain predicted communication links; and adjusting the communication links between the vehicle, RSU, and the cloud based on the predicted communication links to reconstruct the network topology. This application processes multi-source intelligent driving information to obtain a spatiotemporal semantic map, providing data support; and optimizes the topology through a network optimization model using multiple intelligent agents, improving the real-time performance and reliability of the topology reconstruction.

[0008] In addition, the intelligent driving network topology reconstruction method according to the above embodiments of the present invention may also have the following additional technical features:

[0009] Furthermore, in the intelligent driving network topology reconstruction method of this embodiment of the invention, the network optimization model is established through the following steps:

[0010] Based on the spatiotemporal semantic map, a policy network is established;

[0011] Construct an experience replay buffer; wherein the experience replay buffer is used to store the agent's historical decisions;

[0012] Based on the decision-making objectives and scenarios of the intelligent agent, a reward function is constructed;

[0013] A network optimization model is constructed based on the policy network, the experience replay buffer, and the reward function.

[0014] Furthermore, the intelligent driving network topology reconstruction method in this application, wherein constructing a reward function based on the decision objective and scenario of the intelligent agent, includes:

[0015] The communication delay reward is determined based on the ratio of the current latency of the vehicle agent to the historical baseline latency.

[0016] The resource utilization reward is determined based on the target load rate and current load of the RSU agent or MEC agent.

[0017] The reliability reward is determined based on the packet loss rate of the agent;

[0018] A reward function is constructed by weighted summation of the communication delay reward, the resource utilization reward, and the reliability reward.

[0019] Furthermore, in the intelligent driving network topology reconstruction method of this application, the network optimization model is trained through the following steps:

[0020] Based on the policy network and a greedy algorithm, the current action, current state, current reward, and next state are determined.

[0021] The current action, the current state, the current reward, and the next state are stored in the experience replay buffer, and the priority of the current action is determined based on the error.

[0022] If the number of experience replay buffers is greater than or equal to a preset number, samples are taken from the experience replay buffers, and the target Q value is calculated based on the samples and the priority.

[0023] Based on the current Q-value corresponding to the action and the target Q-value, a loss function is constructed, and the policy network is updated based on the loss function;

[0024] If the number of training steps is greater than or equal to the preset number of steps, the target network is updated according to the policy network.

[0025] Furthermore, the intelligent driving network topology reconfiguration method in this application further includes:

[0026] Based on the urgency bid of the vehicle agent, the RSU agent allocates a first resource based on the bid, and the MEC agent allocates a second resource based on a global optimization algorithm; wherein, the range of the first resource is greater than the range of the second resource;

[0027] Alternatively, if multiple vehicle agents conflict, a first resource is allocated based on the priority of the agent or the network status.

[0028] Alternatively, if communication between the vehicle agent, the RSU agent, or the MEC agent is interrupted, a fast rerouting is initiated.

[0029] Furthermore, the intelligent driving network topology reconstruction method in this application, wherein acquiring intelligent driving multi-source information and performing feature extraction and fusion processing on the intelligent driving multi-source information to obtain a spatiotemporal semantic map, includes:

[0030] Obtain at least two of the following information sources: map information, traffic event information, vehicle sensor information, and communication status information, to determine multi-source information;

[0031] Spatial features of multi-source information are extracted using a CNN network.

[0032] The temporal features in multi-source information are processed by an LSTM network to predict temporal characteristics.

[0033] A spatiotemporal semantic map is determined by fusing the spatial features, temporal features, and traffic text features through a self-attention network.

[0034] Furthermore, the intelligent driving network topology reconstruction method in this application, wherein the construction of vehicle intelligent agent, RSU intelligent agent, and MEC intelligent agent includes:

[0035] The state of the vehicle agent is determined by its position, speed, available RSU list, and link quality. The decision objective of the vehicle agent includes communication latency.

[0036] The state of the RSU agent is determined by load rate, number of connected vehicles, and regional events. The decision objective of the RSU agent includes balanced bandwidth allocation.

[0037] The state of the MEC agent is determined by CPU utilization, task queue, and RSU state. The decision objective of the MEC agent includes optimizing computing power allocation.

[0038] On the other hand, embodiments of the present invention propose an intelligent driving network topology reconfiguration system, comprising:

[0039] The acquisition module is used to acquire multi-source information for intelligent driving, and to perform feature extraction and fusion processing on the multi-source information for intelligent driving to obtain a spatiotemporal semantic map;

[0040] The intelligent driving module is used to construct vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and to simulate intelligent driving interaction behaviors between intelligent agents based on the spatiotemporal semantic map.

[0041] The prediction module is used to perform network topology optimization through a network optimization model based on the intelligent driving interaction behavior to obtain predicted communication links;

[0042] The reconstruction module is used to adjust the communication links between the vehicle, RSU and the cloud based on the predicted communication links, and to reconstruct the network topology.

[0043] On the other hand, embodiments of the present invention provide an intelligent driving network topology reconfiguration device, comprising:

[0044] At least one processor;

[0045] At least one memory for storing at least one program;

[0046] When the at least one program is executed by the at least one processor, the at least one processor implements the above-described intelligent driving network topology reconstruction method.

[0047] On the other hand, embodiments of the present invention provide a storage medium storing a processor-executable program, which, when executed by a processor, is used to implement the above-described intelligent driving network topology reconstruction method.

[0048] This invention provides a method, system, device, and storage medium for intelligent driving network topology reconstruction. The method includes: acquiring multi-source intelligent driving information, and performing feature extraction and fusion processing on the multi-source intelligent driving information to obtain a spatiotemporal semantic map; constructing vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and simulating intelligent driving interaction behaviors between the intelligent agents based on the spatiotemporal semantic map; optimizing the network topology according to the intelligent driving interaction behaviors through a network optimization model to obtain predicted communication links; and adjusting the communication links between the vehicle, RSU, and the cloud according to the predicted communication links to reconstruct the network topology. This application processes multi-source intelligent driving information to obtain a spatiotemporal semantic map, providing data support; and optimizes the topology through a network optimization model using multiple intelligent agents, improving the real-time performance and reliability of topology reconstruction. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0050] Figure 1 A flowchart illustrating an embodiment of the intelligent driving network topology reconstruction method provided by the present invention;

[0051] Figure 2 A flowchart illustrating one embodiment of the network optimization model establishment process provided by the present invention;

[0052] Figure 3 A flowchart illustrating one embodiment of the reward function determination process provided by the present invention;

[0053] Figure 4 A schematic flowchart illustrating one embodiment of the training process of the network optimization model provided by the present invention;

[0054] Figure 5 A flowchart illustrating one embodiment of the spatiotemporal semantic map construction process provided by the present invention;

[0055] Figure 6 A flowchart illustrating another embodiment of the spatiotemporal semantic map construction process provided by the present invention;

[0056] Figure 7 A schematic flowchart illustrating another embodiment of the training process of the network optimization model provided by the present invention;

[0057] Figure 8 A flowchart illustrating another embodiment of the intelligent driving network topology reconstruction method provided by the present invention;

[0058] Figure 9 A schematic diagram of the structure of an embodiment of the intelligent driving network topology reconfiguration system provided by the present invention;

[0059] Figure 10 This is a schematic diagram of one embodiment of the intelligent driving network topology reconfiguration device provided by the present invention. Detailed Implementation

[0060] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0061] First, the terms used in this application will be explained:

[0062] 1. Spatiotemporal Semantics: Integrating time-dimensional information (such as traffic flow changes over time and traffic light cycles), spatial-dimensional information (such as road topology and relative vehicle positions), and semantic information (such as traffic rules and the meaning of road signs) to comprehensively describe the integrated concept of the L4 intelligent driving environment.

[0063] 2. Level 4 Autonomous Driving: Highly automated driving, where the vehicle can complete driving tasks completely autonomously in specific scenarios (such as limited areas or specific weather conditions) without human driver intervention.

[0064] 3. Multi-Agent Reinforcement Learning (MARL): A machine learning technique in which multiple agents learn optimal behavioral strategies by interacting with the environment and based on reward mechanisms in a shared environment in order to achieve global goal optimization.

[0065] 4. Roadside Unit (RSU): Intelligent devices deployed in road infrastructure (such as intersections, the middle of the road, etc.) communicate with vehicles over short distances via the PC5 interface to transmit traffic information (such as traffic light status and road condition information). At the same time, they are connected to edge computing nodes (MEC) via wired or wireless means to play the role of data aggregation and forwarding.

[0066] 5. On-Board Unit (OBU): A communication terminal device installed inside the vehicle, supporting wireless communication protocols such as 5G-V2X. It is responsible for enabling data interaction between the vehicle and the outside world (such as Roadside Units (RSUs) and Edge Computing Nodes (MECs). It sends the status data collected by the vehicle's sensors (such as location, speed, and driving intention) to external devices and receives traffic information and communication commands from the outside. It is a key device for vehicle-road cooperative communication.

[0067] 6. Multi-Access Edge Computing (MEC): A computing technology deployed at the network edge (such as roadside base station equipment rooms), possessing localized computing, storage, and network resources. In L4 autonomous driving scenarios, MEC can process data uploaded by vehicles and roadside units in real time, run algorithms such as spatiotemporal semantic map construction and multi-agent reinforcement learning, and make dynamic decisions on network topology based on the computing results, reducing the latency caused by data back to the core network and improving system response speed.

[0068] 7. Convolutional Neural Network (CNN): A deep learning model that automatically extracts features from data through structures such as convolutional layers, pooling layers, and fully connected layers. CNNs can be used to extract features from image data (such as images captured by vehicle cameras or data from cameras deployed by RSUs), identify targets such as traffic signs, vehicles, and pedestrians, and provide basic feature information for the construction of spatiotemporal semantic maps.

[0069] 8. Long Short-Term Memory (LSTM): A special type of recurrent neural network (RNN) that uses gating mechanisms (input gate, forget gate, output gate) to solve the gradient vanishing and gradient exploding problems in traditional RNNs, effectively processing time-series data. In this application, LSTM is used to analyze data such as traffic flow and traffic light status that change over time, to mine the temporal dependencies in the data, predict future traffic trends, and assist in the dynamic updating of spatiotemporal semantic maps.

[0070] 9. Deep Q-Network (DQN): A reinforcement learning algorithm combining deep learning and Q-learning, it utilizes neural networks to approximate the Q-value function, addressing the low storage and computational efficiency issues of traditional Q-learning when dealing with high-dimensional state spaces. This application employs an improved DQN algorithm for multi-agent reinforcement learning training. Through experience replay and target network update mechanisms, it optimizes the agent's communication policy learning process, avoiding overfitting during training. It is commonly used in intelligent decision-making tasks.

[0071] In the field of L4 autonomous driving, achieving a reliable and efficient communication network is key to ensuring the safe and stable operation of vehicles.

[0072] Currently, existing technologies mainly employ relatively fixed network topologies or dynamic adjustment methods based on simple rules. For example, some solutions connect vehicles to surrounding infrastructure (such as RSUs) through a pre-set static network topology. In this approach, the communication link between the vehicle and the RSU is determined at the initial stage. While this ensures basic communication to a certain extent, it lacks the ability to adapt to dynamically changing environments. When encountering sudden traffic events (such as traffic accidents leading to temporary road closures) or significant changes in vehicle density, network performance deteriorates sharply, failing to meet the stringent real-time and reliability requirements of L4 autonomous driving.

[0073] Other existing technologies employ dynamic network adjustment strategies based on simple traffic monitoring. These schemes monitor data traffic in the network, and when the traffic on a particular link reaches a certain threshold, they attempt to switch some data traffic to other links. However, this approach only considers traffic volume and ignores other crucial information in the traffic scenario, such as changes in road topology, traffic rule constraints, and semantic information like vehicle intentions. For example, at complex intersections, even if a link has low traffic, vehicles may be unable to communicate through that link due to road construction or traffic control, and existing solutions cannot effectively identify and address such situations.

[0074] Furthermore, current vehicle-to-everything (V2X) communication lacks deep information fusion and collaboration mechanisms between different devices and systems. Vehicles, RSUs, MECs, and the cloud process data independently, failing to fully leverage each other's data advantages for global optimization. This results in unreasonable network resource allocation, high communication latency, and severely impacts the decision-making efficiency and safety of L4 autonomous driving systems.

[0075] In summary, current technologies still have the following key problems: they cannot achieve real-time and comprehensive perception of spatiotemporal semantic information in L4-level intelligent driving environments, and cannot make efficient and intelligent dynamic adjustments to the communication network topology based on this perception information. Consequently, they cannot meet the stringent requirements of low latency, high reliability, and high bandwidth for communication systems in complex and ever-changing driving scenarios.

[0076] This invention relates to the field of intelligent transportation and communication technology, specifically to a network topology dynamic reconstruction technology based on spatiotemporal semantic fusion and multi-agent reinforcement learning in L4 autonomous driving scenarios.

[0077] First refer to Figure 1This invention provides a method for intelligent driving network topology reconstruction. This method can be applied to terminals, servers, or software running on either terminal or server. Terminals can be tablets, laptops, desktop computers, etc., but are not limited to these. Servers can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. (See also...) Figure 1 The intelligent driving network topology reconstruction method in this embodiment of the invention mainly includes the following steps:

[0078] Step S100: Obtain multi-source information for intelligent driving, and perform feature extraction and fusion processing on the multi-source information for intelligent driving to obtain a spatiotemporal semantic map;

[0079] Step S200: Construct vehicle intelligent agent, RSU intelligent agent and MEC intelligent agent, and simulate intelligent driving interaction behavior between intelligent agents based on spatiotemporal semantic map;

[0080] Step S300: Based on the intelligent driving interaction behavior, network topology optimization is performed through a network optimization model to obtain the predicted communication link;

[0081] Step S400: Based on the predicted communication links, adjust the communication links between the vehicle, RSU and the cloud, and reconstruct the network topology.

[0082] In some possible implementations, the intelligent driving multi-source information in this application includes multiple signals, at least two different signals. This application provides data support for subsequent network topology reconstruction through a spatiotemporal semantic map. This application constructs vehicle agents, RSU agents, and MEC agents, and simulates the interaction behavior between agents based on the spatiotemporal semantic map. A network optimization model is then used to optimize the network topology based on the interaction behavior. The network optimization model can be any artificial intelligence model that implements data optimization. The predicted communication links are the optimal links between the vehicle agents, RSU agents, and MEC agents. Based on the obtained predicted communication links, the network topology is reconstructed to improve the accuracy of the network topology; simultaneously, an automated processing procedure enables real-time updates of the network topology.

[0083] For example, the method provided in this application includes:

[0084] Step 1: Spatiotemporal semantic map construction;

[0085] By integrating multi-source data, including high-precision maps, real-time traffic events (such as accident / control information), vehicle sensor data (such as camera / radar data), and communication status (such as base station load and signal strength), spatiotemporal semantic features are extracted through a CNN-LSTM-Transformer hybrid network to form a dynamic environment model (this is a basic step that provides data support for subsequent analysis).

[0086] Step 2: Multi-agent reinforcement learning training;

[0087] Vehicles, RSUs, and MECs are defined as intelligent agents. Interaction behaviors are simulated based on spatiotemporal semantic maps. The optimal communication strategy (such as link selection and bandwidth allocation) is learned through an improved Deep Q Network (DQN) algorithm (including experience replay and dynamic reward function), which breaks through the limitations of traditional single-agent optimization.

[0088] Step 3: Dynamic network topology reconstruction;

[0089] Based on the learned strategies, the vehicle-RSU-cloud communication links are dynamically adjusted to optimize resource allocation and routing paths, achieving efficient and stable communication in complex scenarios.

[0090] This technical solution mainly addresses the pain points of existing technologies, such as lack of spatiotemporal semantic perception (inability to understand traffic rules / temporary events) and insufficient intelligent reconstruction capabilities (reliance on fixed topology or simple traffic adjustment), and solves the problems of high communication latency, low reliability, and serious resource waste in L4 intelligent driving in complex scenarios.

[0091] The proposed solution enhances the safety and stability of L4 autonomous driving systems in various complex traffic scenarios, specifically in the following ways:

[0092] 1. Improved real-time performance: The transmission delay of key information is effectively reduced, meeting the real-time decision-making needs of L4 intelligent driving and ensuring timely interaction of decision commands;

[0093] 2. Enhanced reliability: Packet loss rate is significantly reduced in complex scenarios, communication success rate is greatly improved, and data transmission is ensured to be stable and reliable;

[0094] 3. Resource optimization: Improve network bandwidth utilization, reduce core network traffic consumption through semantic-driven dynamic allocation, and achieve efficient resource utilization.

[0095] The method provided in this application includes: acquiring multi-source information for intelligent driving, and performing feature extraction and fusion processing on the multi-source information to obtain a spatiotemporal semantic map; constructing vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and simulating intelligent driving interaction behaviors between the intelligent agents based on the spatiotemporal semantic map; optimizing the network topology through a network optimization model based on the intelligent driving interaction behaviors to obtain predicted communication links; and adjusting the communication links between the vehicle, RSU, and the cloud based on the predicted communication links to reconstruct the network topology. This application processes multi-source information for intelligent driving to obtain a spatiotemporal semantic map, providing data support; and optimizes the topology of the network optimization model through multiple intelligent agents to improve the real-time performance and reliability of topology reconstruction.

[0096] Furthermore, referring to Figure 2 The intelligent driving network topology reconstruction method of this invention establishes the network optimization model through the following steps:

[0097] Step S310: Establish a policy network based on the spatiotemporal semantic map;

[0098] Step S320: Construct an experience replay buffer; wherein, the experience replay buffer is used to store the agent's historical decisions;

[0099] Step S330: Construct a reward function based on the agent's decision-making objectives and the scenario;

[0100] Step S340: Construct a network optimization model based on the policy network, experience replay buffer, and reward function.

[0101] The network optimization model constructed in this application determines state behaviors based on a policy network and stores multiple sets of state behaviors in an experience replay buffer. Furthermore, a reward function is constructed based on the decision goals and scenarios of each agent, thereby determining the reward corresponding to each behavior. Through state, reward, and behavior, the forward propagation of the network optimization model is constructed. The network optimization model enables the reconstruction of the network topology.

[0102] Furthermore, referring to Figure 3 The intelligent driving network topology reconstruction method in this application constructs a reward function based on the agent's decision-making objective and the scenario, including:

[0103] Step S331: Determine the communication delay reward based on the ratio of the current latency of the vehicle intelligent agent to the historical baseline latency;

[0104] Step S332: Determine the resource utilization reward based on the target load rate and current load of the RSU agent or MEC agent;

[0105] Step S333: Determine the reliability reward based on the packet loss rate of the agent;

[0106] Step S334: Perform a weighted summation of the communication delay reward, resource utilization reward, and reliability reward to construct a reward function.

[0107] In this application, the historical baseline latency is the historical average latency. This application determines the resource utilization reward for the RSU agent based on its target load rate and current load; and determines the resource utilization reward for the MEC agent based on its target load rate and current load. If packet loss exists, a reliability reward is determined based on the agent's packet loss rate. The reliability reward can be proportional to the packet loss rate or have other functional relationships.

[0108] On the other hand, this application also provides a weight adjustment scheme. Specifically, the scenario complexity is obtained, and the weights of the weighted sum are adjusted based on the scenario complexity and a corresponding threshold. The corresponding threshold is the scenario complexity for which the weights need to be adjusted.

[0109] Furthermore, referring to Figure 4 The intelligent driving network topology reconstruction method in this application obtains the network optimization model through the following steps:

[0110] Step S335: Based on the policy network and using a greedy algorithm, determine the current action, current state, current reward, and next state;

[0111] Step S336: Store the current action, current state, current reward, and next state into the experience replay buffer, and determine the priority of the current action based on the error.

[0112] Step S337: If the number of experience replay buffers is greater than or equal to the preset number, sample from the experience replay buffers and calculate the target Q value based on the sampling and priority.

[0113] Step S338: Construct a loss function based on the current Q-value and the target Q-value corresponding to the action, and update the policy network based on the loss function;

[0114] Step S339: If the number of training steps is greater than or equal to the preset number of steps, update the target network according to the policy network.

[0115] This application updates the parameters of the policy network based on a loss function. If the number of training steps is greater than or equal to a preset number of steps, the parameters of the target network are updated according to the parameters of the policy network. The method of error calculation can be set according to requirements, and this application does not impose specific limitations. The preset number is the threshold for sampling buffer data. The loss function can be set according to requirements, and the preset number of steps is the threshold for the number of training steps to update the target network. When the number of training steps is greater than or equal to the preset number of steps, the parameters of the policy network are used as the new parameters of the target network.

[0116] Furthermore, the intelligent driving network topology reconstruction method in this application further includes:

[0117] Based on the urgency of the vehicle agent's bid, the RSU agent allocates the first resource based on the bid, and the MEC agent allocates the second resource based on a global optimization algorithm; wherein, the range of the first resource is greater than the range of the second resource;

[0118] Alternatively, if multiple vehicle agents conflict, the first resource is allocated based on the agent's priority or network status.

[0119] Alternatively, if communication between the vehicle agent, RSU agent, or MEC agent is interrupted, a fast rerouting is initiated.

[0120] In this application, the scope represented by the first resource is larger than the scope represented by the second resource. If communication between any of the vehicle agent, RSU agent, or MEC agent is interrupted, fast rerouting is initiated.

[0121] Furthermore, referring to Figure 5 The intelligent driving network topology reconstruction method in this application acquires multi-source information for intelligent driving, and performs feature extraction and fusion processing on the multi-source information to obtain a spatiotemporal semantic map, including:

[0122] Step S110: Obtain at least two pieces of information from map information, traffic event information, vehicle sensor information, and communication status information to determine multi-source information;

[0123] Step S120: Extract spatial features of multi-source information through a CNN network;

[0124] Step S130: Process the temporal features in the multi-source information using an LSTM network to predict the temporal features;

[0125] Step S140: A spatiotemporal semantic map is determined by fusing spatial features, temporal features, and traffic text features through a self-attention network.

[0126] Furthermore, the intelligent driving network topology reconstruction method in this application constructs vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, including:

[0127] Determining the state of a vehicle agent includes its position, speed, available RSU list, and link quality; the decision objectives of a vehicle agent include communication latency.

[0128] Determining the state of the RSU agent includes load rate, number of connected vehicles, and regional events. The decision objectives of the RSU agent include balanced bandwidth allocation.

[0129] Determining the state of an MEC agent includes CPU utilization, task queue, and RSU state. The decision-making objectives of an MEC agent include optimizing computing power allocation.

[0130] The decision-making objectives of the vehicle's intelligent agents include minimizing communication latency. The construction of each intelligent agent in this application can be adjusted according to actual needs, and this application does not impose specific limitations.

[0131] In some embodiments, the method provided in this application further includes:

[0132] The RSU agent and vehicle agent will transmit the real-time status back to the MEC agent, and the MEC agent will verify the effect of network topology reconstruction.

[0133] If the network topology reconstruction effect does not meet the standard, update the intelligent driving interaction behavior and return to the step of optimizing the network topology through the network optimization model based on the intelligent driving interaction behavior to obtain the predicted communication link.

[0134] The intelligent driving network topology reconstruction method provided in this application will be described in detail below with reference to the accompanying drawings and specific embodiments:

[0135] System module architecture:

[0136] 1. Vehicle side: Equipped with sensors such as cameras and LiDAR, it communicates with the outside world through the On-Board Unit (OBU). The OBU supports the 5G-V2X protocol to achieve direct communication with the RSU, and also interacts with the cloud through edge computing nodes (MEC).

[0137] 2. Roadside Unit (RSU): Deployed at intersections and in the middle of roads, equipped with traffic sensing devices (cameras, millimeter-wave radar), responsible for collecting vehicle status data, broadcasting traffic information (such as traffic light status and congestion warnings), and uploading the data to MEC.

[0138] 3. Edge Computing Nodes (MECs): Possessing high-performance computing capabilities, they are deployed in base station equipment rooms or along roadsides. Their core modules include:

[0139] Spatiotemporal Semantic Map Construction Module: Employs Convolutional Neural Networks (CNNs) to extract features from multi-source data and analyzes time-series data using Long Short-Term Memory Networks (LSTMs) to construct a dynamic spatiotemporal semantic map. Multi-Agent Reinforcement Learning Module: Based on an improved Deep Q-Network (DQN) algorithm, it stores the agent's state-action-reward-next-state (SARS') quadruple through an experience replay mechanism. It updates parameters every 100 steps using the target network, avoiding oscillations in the value function during training.

[0140] Topology Reconfiguration Module: Based on the learned optimal strategy, it generates network topology adjustment instructions, including link switching rules, data transmission priority configuration, and other information.

[0141] 4. Cloud: Provides big data storage and analysis services, trains global parameters for multi-agent reinforcement learning models, and distributes the optimized model parameters to MEC.

[0142] The core of this solution lies in integrating spatiotemporal semantic information and achieving dynamic and intelligent reconstruction of network topology through multi-agent reinforcement learning. The overall solution encompasses three key modules: spatiotemporal semantic map construction, multi-agent reinforcement learning decision-making, and dynamic network topology reconstruction. These modules work collaboratively to form a closed-loop optimization system, ensuring low-latency and highly reliable communication services for L4 autonomous driving in various traffic scenarios. The core technology modules are as follows:

[0143] (I) Spatiotemporal semantic map construction module, refer to Figure 6 As shown, it specifically includes the following:

[0144] 1. Multi-source data fusion acquisition

[0145] High-precision map data: Basic geographic information such as road static topology, lane attributes, and traffic sign locations are obtained through vehicle-mounted LiDAR and UAV aerial surveys, providing a spatial reference for the map.

[0146] Real-time traffic events: By leveraging roadside equipment (such as cameras and millimeter-wave radar deployed by RSU), traffic management system interfaces, and crowdsourced data, dynamic events such as traffic accidents and construction control are collected. For example, by identifying the location of accident vehicles through cameras and combining it with the information recorded by the traffic management system, temporary restricted areas can be generated to support real-time obstacle avoidance decisions.

[0147] Vehicle sensor data: The vehicle-mounted LiDAR, camera and millimeter-wave radar are fused to perceive the distribution of obstacles and vehicle trajectory within 300 meters. For example, after the LiDAR point cloud is fused with the visual image, the obstacle position with spatiotemporal label is output.

[0148] Communication status data: Real-time monitoring of RSU load rate, signal strength, and available link quality of vehicle OBU, providing a basis for communication quality adjustments.

[0149] Spatiotemporal calibration: All devices are connected to the BeiDou clock synchronization system to achieve nanosecond-level time alignment. Joint calibration technology is employed to ensure spatial consistency of data.

[0150] 2. Fusion processing and feature extraction.

[0151] A hybrid neural network architecture of CNN-LSTM-Transformer is used for multimodal feature fusion and extraction.

[0152] CNN (ResNet-50+FPN) extracts spatial features from images and point clouds, such as obstacle outlines and semantic labels for traffic signs;

[0153] LSTM analyzes continuous frame features and sensor time series data to predict traffic light phase change cycles (such as the remaining time of a red light) or congestion spread trends.

[0154] Transformer uses a self-attention mechanism to fuse spatial features, temporal dynamics, and textual traffic rules (such as "speed limits in school zones") to generate a hierarchical semantic map that includes road topology, real-time events, and rule constraints.

[0155] During the fusion process, the spatiotemporal calibration of multimodal data is first completed. Then, spatial attention is used to focus on key areas and modal attention is used to dynamically adjust the weights of each sensor. Finally, a hierarchical spatiotemporal semantic map containing information such as road topology, real-time traffic conditions, and traffic rule constraints is generated. The bottom layer is the original sensor data, the middle layer is the fusion features, and the top layer is the semantic abstraction. The map area is dynamically updated through an incremental update mechanism.

[0156] (ii) Multi-agent reinforcement learning decision-making module

[0157] 1. Definition of intelligent agent and state space

[0158] Vehicle intelligent agent: The state includes location (latitude and longitude ±1m), speed, list of available RSUs and link quality (e.g., RSU_001 delay 60ms), and the goal is to minimize communication latency (<50ms).

[0159] RSU agent: The state includes load rate, number of connected vehicles and regional events (such as accidents / construction), and the goal is to balance bandwidth allocation (throughput ≥ 1Gbps).

[0160] MEC agents: The state includes CPU utilization, task queue and RSU state, with the goal of optimizing computing power allocation (task latency <20ms).

[0161] 2. Improved DQN algorithm and reward function

[0162] Network architecture optimization: An attention-based network structure is adopted as the policy network to focus on key features (such as the location of emergency vehicles) and enhance the ability to capture and process spatiotemporal semantic features. At the same time, a target network is set up, and a periodic soft update mechanism is used to stabilize the training process and reduce training oscillations.

[0163] Experience replay and priority experience replay: An experience replay buffer is introduced to store the agent's historical decision-making experience (state, action, reward, next state, and whether it has ended). A priority experience replay strategy is adopted, which samples experiences based on their importance (such as reward size and rarity of state transitions), prioritizing the learning of important experiences and accelerating the convergence speed.

[0164] Dynamic reward function:

[0165] Based on the decision-making objectives and scenario requirements of different intelligent agents, a hierarchical, multi-dimensional reward function design is adopted, specifically including:

[0166] 21. Communication Delay Reward: Emergency tasks are given a higher penalty weight when comparing current delay to historical baseline delay. In one embodiment, the ratio of current delay to historical baseline delay is determined, and the communication delay reward is negatively correlated with this ratio.

[0167] r_delay = -α*(current_delay / baseline_delay),

[0168] Dynamic parameter α: based on task priority (α=0.6 for urgent tasks, α=0.4 for normal tasks);

[0169] Baseline latency: The average latency value based on historical data statistics;

[0170] For example, when the emergency task α=0.6, R_delay=-0.6×(80ms / 50ms).

[0171] 22. Resource utilization bonus: Deviation penalty calculated based on the target load rate of RSU / MEC;

[0172] r_resource = β*(1-abs(utilization-target) / target);

[0173] Dynamic parameter β: based on resource scarcity (β=0.3 under high load, β=0.2 under low load).

[0174] Target load rate / target: RSU target load 70%, MEC target 60%; utilization is the current load rate;

[0175] For example, when β=0.3, R_resource=0.3×(1-|75%-70%| / 70%).

[0176] 23. Reliability Bonus: Positive rewards are given when there is no packet loss, and penalties are applied proportionally when there is packet loss.

[0177] r_reliability=if loss_rate==0 return γ else return -γ*loss_rate*100;

[0178] Dynamic parameter γ: based on the importance of transmitted data (security-critical data γ=0.4, ordinary data γ=0.2).

[0179] 24. Total Reward: The total reward is composed of weighted factors such as communication latency, resource utilization, and reliability. The dynamic weights are adjusted according to the complexity of the scenario or the priority of the task.

[0180] r_total=w1*r_delay + w2*r_resource + w3*r_reliability;

[0181] Dynamic weight adjustment mechanism:

[0182] The reward dimension weights are automatically switched based on the complexity of the scenario (such as vehicle density and the urgency of the event), for example, in the case of an accident scenario:

[0183] def adjust_weights(complexity):

[0184] if complexity > 0.8: # High complexity threshold

[0185] return {"delay": 0.6, "reliability": 0.4, "stability": 0.1}

[0186] else:

[0187] return {"delay": 0.4, "resource": 0.3, "reliability": 0.3}.

[0188] 3. Improve the DQN algorithm training process, referring to... Figure 7 As shown.

[0189] Core steps of the training process:

[0190] i. Initialization: Synchronize the parameters of the policy network and the target network, prioritize the experience replay buffer (capacity 100,000), and use the Adam optimizer (lr=0.0001).

[0191] ii. Sampling and Exploration: ε-greedy strategy (initial ε=0.9, exponentially decaying to 0.01) selects actions to balance exploration and exploitation;

[0192] iii. Experience storage: Store the state, action, reward, next state, and termination flag in the buffer, and calculate the priority according to the TD error;

[0193] iv. Batch training: When the buffer data size is ≥5000, sample 64 samples to calculate the target Q value.

[0194] with torch.no_grad():

[0195] next_q = target_net(next_states).max(1)[0].unsqueeze(1)

[0196] target_q = rewards + (1 - dones) * 0.99 * next_q # γ=0.99 discount factor

[0197] v. Parameter update: Calculate the smoothed L1 loss between the current Q value and the target Q value, perform gradient clipping (norm ≤ 1.0), and then backpropagate;

[0198] vi. Target network update: Softly update parameters every 1000 steps (τ=0.005) to prevent training oscillations.

[0199] for t_param, p_param in zip(target_net.parameters(), policy_net.parameters()):

[0200] t_param.data.copy_(τ * p_param.data + (1-τ) * t_param.data).

[0201] 4. Multi-agent collaborative mechanism.

[0202] a) Competition and cooperation: Vehicles and RSUs allocate bandwidth in an auction mode, with vehicles bidding according to urgency and RSUs allocating according to resource efficiency; MEC uses global optimization algorithms (such as Lyapunov optimization) to coordinate regional resources and improve overall performance.

[0203] b) Knowledge sharing: Based on federated learning, the agent uploads the model gradient after local training, and combines differential privacy protection of sensitive information (such as vehicle location) to achieve secure knowledge aggregation.

[0204] c) Abnormal response: When communication is interrupted (such as RSU failure), trigger fast rerouting and enable satellite link to maintain communication; when there is multi-agent conflict (such as multiple vehicles competing for bandwidth), use game theory to allocate resources according to priority and network status.

[0205] (III) Network Topology Dynamic Reconstruction Module

[0206] 1. Decision generation: The MEC multi-agent reinforcement learning module generates topology adjustment schemes (such as "vehicle V_007 switches to RSU_003, bandwidth 50Mbps") based on the spatiotemporal semantic map and the state information of each agent through policy optimization calculation.

[0207] 2. Command Issuance: The MEC topology reconfiguration decision module translates the strategy into a set of commands executable by the devices. For multi-device collaborative scenarios (such as joint adjustment of multiple RSUs in a congested area), it generates RSU cluster control commands; for single-device adjustment scenarios (such as single vehicle link switching), it generates single-device adjustment commands and issues commands (such as RSU cluster load balancing) through the 5G-V2X PC5 interface (between vehicle and RSU), wired link, or 5G Uu interface (between MEC and RSU).

[0208] 3. Equipment execution:

[0209] RSU performs the following actions: updates the wireless channel configuration according to instructions, such as adjusting the frequency band and power; establishes a new communication link; allocates dedicated resource blocks to the target vehicle; and ensures communication quality.

[0210] Vehicle execution action: After receiving the instruction, the vehicle OBU switches the communication link (e.g., from RSU_001 to RSU_003) and adjusts the data transmission parameters, such as the transmission rate and encoding method, to ensure compatibility with the new link (e.g., power 2W).

[0211] 4. Feedback Optimization: The RSU and vehicles transmit execution status (such as link quality and latency metrics) back to the MEC in real time. The MEC verifies the effectiveness of the strategy based on the feedback results. If the expected results are not achieved, a new round of reinforcement learning training is triggered to adjust the strategy, forming a closed loop of "decision-execution-optimization" to continuously improve network performance.

[0212] Reference Figure 8 As shown, the specific implementation is as follows:

[0213] This invention can be applied in various scenarios, such as urban intersections, multi-vehicle platoons on highways, and detours around construction zones. Its application is currently focused on a traffic accident scenario at an urban intersection.

[0214] Scenario: A traffic accident occurs on a road in the city, resulting in the closure of half of the road and affecting the passage of 30 Robotaxi vehicles.

[0215] (I) Spatiotemporal semantic map construction (t=0s~2s);

[0216] 1. Multi-source data acquisition:

[0217] High-precision map: Lane distribution at intersections is obtained from the cloud (accuracy ±10cm);

[0218] Traffic incident: RSU camera identifies the accident location (119.481°E, 37.921°N), millimeter-wave radar detects the congestion area;

[0219] Vehicle sensing: Robotaxi's LiDAR scans a vehicle ahead that has suddenly stopped, and the camera identifies the type of vehicle involved in the accident;

[0220] Communication status: MEC detected that RSU_001 load rate is 75% and signal strength is -75dBm.

[0221] 2. Feature Extraction and Map Generation:

[0222] ResNet-50 identifies accident vehicles and traffic cones, LSTM predicts the congestion spread speed (5m / s), and Transformer generates temporary restricted areas.

[0223] (ii) Multi-agent decision-making process (t=3s~5s).

[0224] 1. Agent state update:

[0225] Vehicle V_007: Location (119.481°E, 37.921°N), speed 40km / h, available RSUs are RSU_001 (delay 60ms) and RSU_003 (delay 80ms).

[0226] RSU_001: Load rate 90%, 25 connected vehicles, cache queue 100ms;

[0227] MEC: CPU utilization is 80%, and the task queue includes path planning requirements.

[0228] 2. Reward Calculation and Action Decision:

[0229] Vehicle V_007: R_delay=-0.72, R_reliability=0.4, total reward -0.272: Decision switch to RSU_003;

[0230] RSU_001: R_resource=0.278, R_fairness=0.285, total reward 0.353: decision load balancing;

[0231] MEC: R_delay=-0.8, R_resource=0.3, total reward -0.31: Decision to allocate 4 CPU cores.

[0232] (III) Application of DQN training process in accident scenarios.

[0233] 1. Offline pre-training phase (before the incident);

[0234] Training environment: Simulates 10+ scenarios (urban intersections, highways, construction areas);

[0235] Key parameters: 1 million training steps, batch size 64, priority experience replay α=0.6;

[0236] Training results: The communication latency optimization strategy achieved a success rate of 92%, reducing latency by 58% compared to the random strategy.

[0237] 2. Online fine-tuning phase (t=0s~10s when the accident occurs);

[0238] t=3s: Vehicle V_007 loads the pre-trained model and executes the exploration strategy at ε=0.7;

[0239] t=4s: Execute "Switch RSU_003", obtain reward R=-0.272, and store it in the buffer;

[0240] t=5s: The buffer data volume reaches 500, triggering the first batch training;

[0241] t=10s: After 5 iterations of training, the model's Q-value estimation bias decreased from 25% to 8%.

[0242] (iv) Topology reconstruction execution (t=6s~8s)

[0243] 1. MEC command issuance:

[0244] Vehicle command: {"vehicleId":"V_007","action":"switch_link","targetRsu":"RSU_003","power":2W};

[0245] RSU command: {"rsuId":"RSU_001","action":"load_balance","targetRsu":"RSU_003","bandwidth":50Mbps}.

[0246] 2. Execution result:

[0247] The latency of V_007 decreased from 80ms to 45ms, the load rate of RSU_001 decreased from 90% to 65%, and the MEC path calculation latency decreased from 150ms to 30ms.

[0248] As shown in Table 1, the implementation effect comparison shows that the intelligent driving network topology reconstruction method provided in this application reduces communication latency, improves accident response time, reduces packet loss rate, and improves the real-time performance and reliability of network topology reconstruction overall.

[0249]

[0250] Table 1

[0251] This application constructs a spatiotemporal semantic map based on multi-source data fusion, providing comprehensive and accurate environmental information for dynamic network topology reconstruction. Compared to existing technologies that rely only on partial environmental information for network adjustments, the spatiotemporal semantic map of this invention integrates multi-source information such as high-precision maps, real-time traffic events, vehicle sensor data, and communication status. This allows for a more comprehensive and accurate reflection of the intelligent driving environment, making network topology reconstruction decisions more scientific, significantly improving the adaptability of the communication network to complex scenarios, reducing communication latency, and enhancing network reliability. This application employs a multi-agent reinforcement learning algorithm to enable autonomous learning and strategy optimization by agents such as vehicles, RSUs, and MECs within the spatiotemporal semantic map environment, thereby dynamically reconstructing the network topology. Unlike traditional network topology management methods based on fixed rules or simple dynamic adjustments, multi-agent reinforcement learning enables each node in the network to autonomously learn the optimal communication strategy according to the real-time environment, achieving intelligent dynamic optimization of the network topology. This not only improves network resource utilization but also enhances the network's adaptability to dynamically changing environments, effectively improving network performance and ensuring efficient communication of the L4 intelligent driving system in different scenarios. In the process of dynamic network topology reconstruction, this application utilizes the Multi-Agent Reinforcement Learning (MEC) to coordinate the adjustment of communication links and parameters by vehicles, RSUs, and other devices. Through the unified coordination of the MEC, this application avoids communication conflicts and resource waste caused by vehicles and RSUs operating independently, achieving rational allocation and efficient utilization of network resources. In complex traffic scenarios, it can quickly and effectively optimize network topology, ensure communication stability and real-time performance, and improve the overall operational efficiency of the L4 autonomous driving system.

[0252] Potential future applications of this application:

[0253] 1. Commercial operation: Robotaxi operates without safety drivers in designated areas;

[0254] 2. Logistics and Transportation: High-speed autonomous driving truck platooning communication improves freight efficiency (e.g., port / industrial park scenarios);

[0255] 3. Smart City: Multi-vehicle collaboration optimizes traffic flow and reduces urban congestion.

[0256] This application is beneficial for cost reduction: reducing the deployment density of roadside equipment by 30% and reducing operators' infrastructure investment; this application is beneficial for safety improvement: through real-time communication, the accident rate of L4 intelligent driving system is expected to decrease by 70%; this application is beneficial for ecosystem collaboration: promoting standardized cooperation among vehicle, road, and cloud manufacturers and accelerating the implementation of the intelligent transportation industry.

[0257] Secondly, refer to the appendix Figure 9 A smart driving network topology reconfiguration system according to an embodiment of the present invention is described, the system specifically comprising:

[0258] The acquisition module 810 is used to acquire multi-source information for intelligent driving, and to perform feature extraction and fusion processing on the multi-source information for intelligent driving to obtain a spatiotemporal semantic map;

[0259] The Intelligent Driving Module 820 is used to construct vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and simulates intelligent driving interaction behaviors between intelligent agents based on spatiotemporal semantic maps.

[0260] The prediction module 830 is used to perform network topology optimization through a network optimization model based on intelligent driving interaction behavior to obtain predicted communication links.

[0261] The reconfiguration module 840 is used to reconfigure the network topology by adjusting the communication links between the vehicle, RSU and the cloud based on the predicted communication links.

[0262] It is evident that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0263] Reference Figure 10 This invention provides an intelligent driving network topology reconfiguration device, comprising:

[0264] At least one processor 910;

[0265] At least one memory 920 is used to store at least one program;

[0266] When at least one program is executed by at least one processor 910, the intelligent driving network topology reconfiguration method is implemented by at least one processor 910.

[0267] Similarly, the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0268] This invention also provides a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to perform the above-described intelligent driving network topology reconstruction method.

[0269] Similarly, the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0270] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0271] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0272] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several programs to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0273] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable programs for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, a program execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can retrieve and execute a program from or in conjunction with such a program execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with a program execution system, apparatus, or device.

[0274] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0275] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable program execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0276] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0277] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0278] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for intelligent driving network topology reconstruction, characterized in that, The method includes the following steps: Acquire multi-source information for intelligent driving, and perform feature extraction and fusion processing on the multi-source information for intelligent driving to obtain a spatiotemporal semantic map; Construct vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and simulate intelligent driving interaction behaviors between intelligent agents based on the spatiotemporal semantic map; Based on the intelligent driving interaction behavior, network topology optimization is performed through a network optimization model to obtain predicted communication links; Based on the predicted communication links, the communication links between vehicles, RSUs, and the cloud are adjusted to reconstruct the network topology; The process of acquiring multi-source information for intelligent driving and performing feature extraction and fusion processing on the multi-source information to obtain a spatiotemporal semantic map includes: Obtain at least two of the following information sources: map information, traffic event information, vehicle sensor information, and communication status information, to determine multi-source information; Spatial features of multi-source information are extracted using a CNN network. The temporal features in multi-source information are processed by an LSTM network to predict temporal characteristics. A spatiotemporal semantic map is determined by fusing the spatial features, temporal features, and traffic text features through a self-attention network.

2. The intelligent driving network topology reconstruction method according to claim 1, characterized in that, The network optimization model is established through the following steps: Based on the spatiotemporal semantic map, a policy network is established; Construct an experience replay buffer; wherein the experience replay buffer is used to store the agent's historical decisions; Based on the decision-making objectives and scenarios of the intelligent agent, a reward function is constructed; A network optimization model is constructed based on the policy network, the experience replay buffer, and the reward function.

3. The intelligent driving network topology reconstruction method according to claim 2, characterized in that, The step of constructing a reward function based on the decision-making objective and scenario of the intelligent agent includes: The communication delay reward is determined based on the ratio of the current latency of the vehicle agent to the historical baseline latency. The resource utilization reward is determined based on the target load rate and current load of the RSU agent or MEC agent. The reliability reward is determined based on the packet loss rate of the agent; A reward function is constructed by weighted summation of the communication delay reward, the resource utilization reward, and the reliability reward.

4. The intelligent driving network topology reconstruction method according to claim 1, characterized in that, The network optimization model is trained through the following steps: Based on the policy network and a greedy algorithm, the current action, current state, current reward, and next state are determined. The current action, the current state, the current reward, and the next state are stored in the experience replay buffer, and the priority of the current action is determined based on the error. If the number of experience replay buffers is greater than or equal to a preset number, samples are taken from the experience replay buffers, and the target Q value is calculated based on the samples and the priority. Based on the current Q-value corresponding to the action and the target Q-value, a loss function is constructed, and the policy network is updated based on the loss function; If the number of training steps is greater than or equal to the preset number of steps, the target network is updated according to the policy network.

5. The intelligent driving network topology reconstruction method according to claim 1, characterized in that, The method further includes: Based on the urgency bid of the vehicle agent, the RSU agent allocates a first resource based on the bid, and the MEC agent allocates a second resource based on a global optimization algorithm; wherein, the range of the first resource is greater than the range of the second resource; Alternatively, if multiple vehicle agents conflict, a first resource is allocated based on the priority of the agent or the network status. Alternatively, if communication between the vehicle agent, the RSU agent, or the MEC agent is interrupted, a fast rerouting is initiated.

6. The intelligent driving network topology reconstruction method according to claim 1, characterized in that, The construction of the vehicle intelligent agent, RSU intelligent agent, and MEC intelligent agent includes: The state of the vehicle agent is determined by its position, speed, available RSU list, and link quality. The decision objective of the vehicle agent includes communication latency. The state of the RSU agent is determined by load rate, number of connected vehicles, and regional events. The decision objective of the RSU agent includes balanced bandwidth allocation. The state of the MEC agent is determined by CPU utilization, task queue, and RSU state. The decision objective of the MEC agent includes optimizing computing power allocation.

7. A smart driving network topology reconfiguration system, characterized in that, include: The acquisition module is used to acquire multi-source information for intelligent driving, and to perform feature extraction and fusion processing on the multi-source information for intelligent driving to obtain a spatiotemporal semantic map; The intelligent driving module is used to construct vehicle intelligent agents, RSU intelligent agents, and MEC intelligent agents, and to simulate intelligent driving interaction behaviors between intelligent agents based on the spatiotemporal semantic map. The prediction module is used to perform network topology optimization through a network optimization model based on the intelligent driving interaction behavior to obtain predicted communication links; The reconstruction module is used to adjust the communication links between the vehicle, RSU and the cloud based on the predicted communication links, and to reconstruct the network topology. The process of acquiring multi-source information for intelligent driving and performing feature extraction and fusion processing on the multi-source information to obtain a spatiotemporal semantic map includes: Obtain at least two of the following information sources: map information, traffic event information, vehicle sensor information, and communication status information, to determine multi-source information; Spatial features of multi-source information are extracted using a CNN network. The temporal features in multi-source information are processed by an LSTM network to predict temporal characteristics. A spatiotemporal semantic map is determined by fusing the spatial features, temporal features, and traffic text features through a self-attention network.

8. A smart driving network topology reconfiguration device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the intelligent driving network topology reconfiguration method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to implement the intelligent driving network topology reconstruction method as described in any one of claims 1 to 6.

Images