V2V Interaction Method for Vehicular Ad Hoc Networks Based on Intelligent Connected Vehicle WiFi

Abstract

The present invention relates to the technical field of wireless communication networks, and particularly to a V2V interaction method for vehicle ad-hoc networks based on intelligent connected vehicle WiFi. It collects vehicle motion state and network load data through in-vehicle sensors, dynamically switches between AP / STA modes by combining the Deep Q-Network (DQN) model and the ε-greedy strategy, and constructs an adaptive topology structure. It extends the OLSR protocol packet to embed a path delay field, optimizes multi-hop routing selection by combining pheromone gradient calculation, and realizes message priority classification and differential privacy protection through a federated learning model. Based on the particle swarm algorithm, it dynamically adjusts the AP / STA ratio threshold to trigger node role reconfiguration to balance network load. The present invention reuses existing WiFi hardware to reduce deployment costs, combines cross-layer routing optimization and a distributed decision-making mechanism, and improves network real-time performance, scalability, and group collaboration ability in dynamic scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication networks, and particularly to a V2V interaction method for vehicle ad-hoc networks based on intelligent connected vehicle WiFi. Background Art

[0002] A vehicle ad-hoc network (VANET) of intelligent connected vehicle WiFi is a distributed communication network built based on wireless local area network technology. It autonomously forms a network with nodes such as in-vehicle communication units, neighboring vehicles, and roadside infrastructure to achieve real-time data interaction between vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I). This technology breaks through the dependence on a central base station in traditional cellular networks, uses a multi-hop relay transmission mechanism to expand the communication range, and adapts to network structure changes brought about by high-speed vehicle movement through dynamic topology awareness and self-organizing routing protocols. It can support low-latency communication requirements in scenarios such as collision warning, formation coordination, and road condition sharing. At the same time, it improves the utilization rate of wireless resources through spectrum reuse, and belongs to a key supporting technology for improving driving safety and efficiency in intelligent transportation systems.

[0003] Vehicle-to-vehicle (V2V) interaction of intelligent connected vehicles is a distributed cooperation mechanism based on wireless communication technology. Through in-vehicle terminals using dedicated short-range communication (DSRC) or cellular vehicle-to-everything (C-V2X) protocols, it realizes real-time interaction of status data, driving intentions, and environmental perception information between vehicles. This technology relies on a self-organizing network architecture, supports multi-hop relay transmission and dynamic routing optimization, can adapt to network topology fluctuations caused by high-speed vehicle movement without central node scheduling, and completes information synchronization in scenarios such as collision warning, cooperative obstacle avoidance, and platoon driving through low-latency and highly reliable communication links. At the same time, it combines channel anti-interference and data encryption technologies to improve transmission stability and security, and belongs to the core communication solution for improving the active safety and group cooperation efficiency of intelligent transportation systems.

[0004] Existing V2X technology solutions have high deployment costs due to dependence on infrastructure such as roadside units and communication base stations. Moreover, dedicated communication modules such as DSRC and LTE-V have problems of high hardware transformation costs and limited penetration rates. At the same time, traditional V2V communication has latency bottlenecks in terms of multi-hop routing efficiency and dynamic adaptability of network topology. The present invention constructs a dynamic ad-hoc network by reusing the existing WiFi modules of intelligent connected vehicles, realizes autonomous networking of vehicle nodes through the AP / STA / bridge mode switching mechanism, and combines multi-hop routing optimization and bridging interconnection technologies. Without the need to add dedicated hardware and infrastructure, it reduces the deployment cost of vehicle cooperative communication and improves the real-time performance and network scalability of message propagation in complex traffic scenarios. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a V2V interaction method for vehicle ad hoc networks based on the WiFi of intelligent connected vehicles, which solves the problems that the existing V2X technology has high deployment costs and poor scalability in vehicle cooperative communication due to dependence on dedicated hardware and infrastructure, and the traditional networking mode has low real-time performance caused by insufficient dynamic topology adaptation ability.

[0006] To solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0007] The present invention provides a V2V interaction method for vehicle ad hoc networks based on the WiFi of intelligent connected vehicles, including:

[0008] Collect vehicle motion state data and network state data through in-vehicle sensors, normalize the speed, position coordinates and network load rate, and generate a multi-dimensional feature vector;

[0009] Input the multi-dimensional feature vector into a pre-trained deep Q network model, output an operation instruction set containing the AP / STA mode switching probability, and select the optimal instruction according to the ε-greedy strategy to construct a dynamic ad hoc network topology;

[0010] Based on the node distribution information in the dynamic ad hoc network topology, embed a path delay field in the OLSR protocol packet to generate a pheromone data packet, and update the routing table entries of each node through periodic broadcasting;

[0011] Receive the V2V message of the in-vehicle application layer, input the message type and propagation hop count into the federated learning classification model, output a priority label and embed it in the message header, and trigger the MAC layer channel preemption mechanism;

[0012] Perform spatio-temporal alignment on the shared position data of neighboring vehicles and the local IMU sensor data, generate a vehicle pose estimate through the particle filter algorithm, calculate a cooperative control instruction based on the Nash equilibrium strategy matrix, and distribute it to the target vehicle through an optimized multi-hop path;

[0013] Collect the AP node load rate and the number of channel conflicts, iteratively calculate the AP / STA ratio threshold through the particle swarm optimization algorithm, trigger node role reconfiguration according to the calculation result, and feedback the load status to the ad hoc network topology construction module.

[0014] Further, in the V2V interaction method for vehicle ad hoc networks based on the WiFi of intelligent connected vehicles of the present invention, the construction of the dynamic ad hoc network topology includes:

[0015] Discretize the vehicle speed value into three states: low speed, medium speed, and high speed, and associate the current AP node density and channel occupancy rate to construct a three-dimensional state space;

[0016] Input the three-dimensional state space into the pre-trained deep Q-network model, output the value function evaluation values of each candidate operation, and generate an AP / STA mode switching instruction based on the ε-greedy strategy;

[0017] The switching instruction is written into the hardware register through the WIFI chipset driver interface, triggering the generation of a self-organizing network topology structure including bridging nodes, and transmitting the topology node distribution data to the routing layer.

[0018] Further, in the V2V interaction method of the vehicle self-organizing network based on the intelligent connected vehicle WiFi of the present invention, based on the node distribution information in the dynamic self-organizing network topology, embedding a path delay field in the OLSR protocol packet to generate a pheromone data packet, and updating the routing table entries of each node through periodic broadcast includes:

[0019] Expand the delay field in the OLSR protocol Hello packet header, and periodically generate a pheromone data packet including the hop count, RSSI value, and current queue length;

[0020] The intermediate node calculates the pheromone concentration gradient according to the delay field value and the RSSI attenuation coefficient, and updates the routing table entries in descending order of the gradient value;

[0021] The routing table is cross-layer transmitted to the priority queue manager in the message transmission layer through the Socket interface, driving high-priority messages to preempt low-delay paths. Further, in the V2V interaction method of the vehicle self-organizing network based on the intelligent connected vehicle WiFi of the present invention, receiving the V2V message of the in-vehicle application layer, inputting the message type and propagation hop count into the federated learning classification model, outputting a priority label and embedding it into the message header, and triggering the MAC layer channel preemption mechanism includes: extracting the message type, propagation hop count, and end-to-end delay in the vehicle locally to construct a three-element feature vector, and adding Laplace noise to implement differential privacy encryption;

[0022] Upload the encrypted feature vector to the federated learning classification model of the edge server, and aggregate multi-node data to update the message priority classification rule;

[0023] The classification rule is digitally signed and written into the in-vehicle communication protocol stack through OTA, adding the highest priority label to the emergency braking message and triggering the MAC layer CSMA / CA backoff mechanism to bypass. Further, in the V2V interaction method of the vehicle self-organizing network based on the intelligent connected vehicle WiFi of the present invention, aligning the shared location data of neighboring vehicles with the local IMU sensor data in time and space, generating a vehicle pose estimate through the particle filter algorithm, calculating a cooperative control instruction based on the Nash equilibrium strategy matrix, and distributing it to the target vehicle through the optimized multi-hop path includes:

[0024] Perform UTM projection transformation on the adjacent vehicle position data in the GPS coordinate system, align it with the local IMU angular velocity data through a sliding window, and generate vehicle pose estimation through the particle filter algorithm;

[0025] Calculate the safety distance threshold between the host vehicle and the adjacent vehicle based on the Nash equilibrium strategy matrix, and generate a control instruction including the collaborative deceleration gradient value;

[0026] The control instruction is distributed to the target vehicle group through the IPv6 multicast address, and the decision delay data is written into the routing layer path redundancy calculation module.

[0027] Furthermore, for the V2V interaction method of the vehicle ad-hoc network based on intelligent connected vehicle WiFi of the present invention, the steps of collecting the AP node load rate and the number of channel conflicts, iteratively calculating the AP / STA ratio threshold through the particle swarm optimization algorithm, triggering node role reconfiguration according to the calculation result, and feeding back the load status to the ad-hoc network topology construction module include:

[0028] Quantify the AP node density, the number of STA connections, and the channel conflict rate into a three-dimensional vector space, and iteratively calculate the optimal AP / STA ratio threshold through the particle swarm algorithm;

[0029] Trigger a downgrade instruction in the area where the AP density exceeds the threshold, switch the edge AP node to the STA mode, and establish an 802.11s bridging link with the central AP;

[0030] Write the role assignment result into the configuration parameter library of the ad-hoc network topology construction module, and trigger the reconstruction of the topology structure based on the current load status.

[0031] Furthermore, the V2V interaction method of the vehicle ad-hoc network based on intelligent connected vehicle WiFi of the present invention further includes:

[0032] Divide the vehicle speed into three intervals of [0, 20), [20, 60), and [60, ∞) km / h to define low-speed, medium-speed, and high-speed states, associate the current channel occupancy rate level with the AP node density interval, and construct a three-dimensional discrete state space;

[0033] Input the three-dimensional discrete state space into the depth Q network model, and output the value function evaluation values of three operations: maintaining the AP mode, switching to the STA mode, and starting the bridging mode;

[0034] Calculate the mode switching probability distribution according to the value function evaluation value, and drive the WIFI chipset to switch to the STA mode when the channel occupancy rate exceeds the threshold.

[0035] Furthermore, for the V2V interaction method of the vehicle ad-hoc network based on intelligent connected vehicle WiFi of the present invention, the calculation of the pheromone concentration gradient includes:

[0036] Record the path delay value in the OLSR protocol TCL field, calculate the initial pheromone concentration according to the RSSI signal strength, and set the delay weight factor as the reciprocal of the current network load rate;

[0037] When an intermediate node forwards a data packet, update the pheromone concentration value according to the formula concentration decay amount = number of hops × preset decay coefficient to generate gradient distribution data;

[0038] Write the gradient distribution data into the priority flag bit of the routing table entry, and preferentially select the path with the highest flag bit value to establish a TCP long connection. Further, in the V2V interaction method of the vehicle ad hoc network based on the intelligent connected vehicle WiFi of the present invention, the differential privacy encryption includes:

[0039] Add Laplace distribution noise data to the feature vector of the message type and propagation hop count to make the correlation entropy value between the single-hop transmission record and the message type exceed the preset threshold;

[0040] Aggregate the encrypted data through the edge server and update the weights of the Softmax output layer of the federated learning classification model;

[0041] After attaching the SHA-256 hash value to the classification rule, it is sent to the vehicle terminal for integrity verification in the communication protocol stack parsing module.

[0042] Further, in the V2V interaction method of the vehicle ad hoc network based on the intelligent connected vehicle WiFi of the present invention, the bridging link includes:

[0043] When detecting subnet isolation caused by the degradation of the AP node, select a dual-mode node that supports the 802.11s bridging protocol as a relay;

[0044] Maintain the mapping relationship of the source subnet ID, destination subnet ID, and next-hop MAC address in the routing table of the bridging node, and forward the control message through EtherType 0x88CC;

[0045] Write the throughput and delay data of the bridging link into the particle swarm algorithm iteration parameter constraint library to limit the calculation range of the AP / STA ratio.

[0046] Advantages of the present invention;

[0047] The present invention realizes dynamic self - organizing network by reusing the existing vehicle WiFi module, significantly reducing the cost of collaborative communication deployment. Based on the deep Q - network (DQN) model and the AP / STA mode switching mechanism in a three - dimensional discrete state space, combined with the ε - greedy strategy, it dynamically generates network topologies, enabling vehicle nodes to autonomously adapt to different moving speeds and network load scenarios. For example, in areas of high - speed movement or high - channel conflict, DQN triggers STA mode switching instructions and combines with the 802.11s bridging protocol to maintain network connectivity, avoiding the dependence of traditional V2X technologies on dedicated hardware and infrastructure. At the same time, the particle swarm optimization algorithm dynamically adjusts the node role allocation by iteratively calculating the AP / STA ratio threshold, reducing local channel congestion and enhancing network scalability and resource utilization.

[0048] The present invention improves real - time performance and reliability through cross - layer collaborative optimization and distributed decision - making mechanisms. The combination of the extended delay field of the OLSR protocol and pheromone gradient calculation realizes dynamic optimization of multi - hop paths, enabling high - priority messages to reduce end - to - end delay through the MAC - layer preemption mechanism. The federated learning model aggregates encrypted feature data to generate global classification rules, achieving fast identification and transmission of emergency messages while protecting privacy. In addition, the particle filter fuses multi - source sensor data to generate vehicle pose estimation, combines with the Nash equilibrium strategy matrix to output collaborative control instructions, and improves group collaboration efficiency through the IPv6 multicast distribution mechanism. The closed - loop load balancing and routing redundancy feedback mechanism further strengthens the network autonomy ability to adapt to the dynamic requirements of complex traffic scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the present invention, the following will briefly introduce the drawings required in the embodiments. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the drawings.

[0050] Figure 1 It is a timing diagram of the V2V interaction method of the vehicle self - organizing network based on intelligent - connected vehicle WiFi provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention in combination with the specific embodiments and corresponding drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. The following will, in combination with the drawings, detail the technical solutions provided by each embodiment of the present invention. To better understand the objectives of the present invention, the following will further describe the present invention in detail.

[0052] Please refer to Figure 1 , the present invention provides a V2V interaction method for vehicle ad-hoc network based on intelligent connected vehicle WiFi, including:

[0053] Collect vehicle motion state data and network state data through on-vehicle sensors, perform normalization processing on speed, position coordinates and network load rate, and generate a multi-dimensional feature vector;

[0054] Collect vehicle motion state data (including speed, position coordinates) and network state data (such as network load rate) through on-vehicle sensors in real time, and perform normalization processing on these heterogeneous data to eliminate the dimension difference and generate a multi-dimensional feature vector with a unified dimension. The normalization processing adopts the min-max scaling method to map each parameter to a preset numerical interval, providing standardized input for subsequent model processing. This step provides basic data support for subsequent dynamic networking decisions.

[0055] Input the multi-dimensional feature vector into a pre-trained deep Q-network model, output an operation instruction set including the AP / STA mode switching probability, and select the optimal instruction according to the ε-greedy strategy to construct a dynamic ad-hoc network topology;

[0056] Input the normalized multi-dimensional feature vector into a pre-trained deep Q-network (DQN) model, and output an operation instruction set including the switching probability of the AP (access point) mode and the STA (station) mode. The DQN model is trained through historical vehicle networking scenario data to learn the optimal mode switching strategy under different network states. Based on the ε-greedy strategy, the model selects the operation instruction with the highest current value function evaluation value with a certain probability, dynamically adjusts the AP / STA role of vehicle nodes, and generates an ad-hoc network topology structure including bridge nodes. The switching instruction is written into the hardware register through the WiFi chipset driver interface, the network topology configuration is updated in real time, and the topology distribution data is transmitted to the routing layer to realize the adaptive optimization of the network structure.

[0057] Based on the node distribution information in the dynamic ad-hoc network topology, embed a path delay field in the OLSR protocol packet to generate a pheromone data packet, and update the routing table entries of each node through periodic broadcasting;

[0058] On the basis of the dynamic ad-hoc network topology, extend the Hello message header of the OLSR (optimized link state routing) protocol, embed a path delay field to generate a pheromone data packet. The intermediate node calculates the path pheromone concentration gradient according to the received pheromone data packet in combination with the RSSI (received signal strength indication) attenuation coefficient, and updates the local routing table entries in descending order of the gradient value. The routing table is transmitted across layers through the Socket interface to the priority queue manager of the message transmission layer, so that high-priority messages can preempt low-delay paths first, reducing the end-to-end transmission delay.

[0059] Receive V2V messages from the in-vehicle application layer, input the message type and propagation hop count into the federated learning classification model, output the priority label and embed it into the message header, and trigger the MAC layer channel preemption mechanism;

[0060] Extract the type, propagation hop count, and end-to-end delay characteristics of V2V messages at the in-vehicle terminal, construct a ternary feature vector, and implement differential privacy encryption by adding Laplace noise to prevent the leakage of sensitive information. The encrypted feature vector is uploaded to the federated learning classification model of the edge server, and the message priority classification rule is updated by aggregating multi-node data. The classification rule is digitally signed and then sent to the vehicle terminal through OTA (Over-the-Air) technology and embedded in the communication protocol stack. Emergency braking messages are marked as the highest priority, triggering the MAC layer CSMA / CA backoff mechanism to bypass the regular channel competition process and achieve low-latency transmission.

[0061] Perform spatio-temporal alignment on the shared location data of neighboring vehicles and the local IMU sensor data, generate vehicle pose estimation through the particle filter algorithm, calculate the cooperative control instruction based on the Nash equilibrium strategy matrix, and distribute it to the target vehicle through the optimized multi-hop path;

[0062] Perform UTM (Universal Transverse Mercator) projection transformation on the GPS location data shared by neighboring vehicles, perform sliding window spatio-temporal alignment with the angular velocity data of the local IMU (Inertial Measurement Unit) to eliminate spatio-temporal deviation. Fusion of multi-source sensor data through the particle filter algorithm to generate high-precision vehicle pose estimation. Calculate the safety distance threshold between the vehicle and neighboring vehicles based on the Nash equilibrium strategy matrix, and generate a control instruction including the cooperative deceleration gradient value. The instruction is distributed to the target vehicle group along the optimized multi-hop path through the IPv6 multicast address, and at the same time, the decision delay data is recorded to optimize the routing redundancy calculation.

[0063] Collect the load rate of AP nodes and the number of channel conflicts, iteratively calculate the AP / STA ratio threshold through the particle swarm optimization algorithm, trigger node role reconfiguration according to the calculation result, and feedback the load status to the ad hoc network topology construction module.

[0064] Quantify the AP node density, STA connection number, and channel conflict rate into a three-dimensional vector space, and input the particle swarm optimization (PSO) algorithm to iteratively calculate the optimal AP / STA ratio threshold. When the AP node density exceeds the threshold, trigger the edge AP node downgrade instruction, switch it to the STA mode, and establish a bridging link based on the 802.11s protocol with the central AP. The role assignment result is fed back to the ad hoc network topology construction module to drive network topology reconstruction and achieve dynamic load balancing and efficient utilization of channel resources.

[0065] The above technical content constitutes a closed-loop collaborative system: data collection and feature processing provide input for the DQN model, driving the AP / STA mode switching and dynamic networking; the networking topology information supports the optimization of the routing protocol to ensure low-latency path selection; the message priority classification and collaborative decision-making module relies on the routing optimization result to achieve efficient instruction distribution; the load balancing mechanism monitors the network status in real time and feeds back to the topology construction module through node role reconfiguration, forming an adaptive adjustment loop. Through cross-layer data interaction and status sharing, each module jointly improves the real-time performance, reliability, and scalability of the vehicle ad hoc network.

[0066] Specifically, for the V2V interaction method of the vehicle ad hoc network based on the intelligent connected vehicle WiFi of the present invention, the construction of the dynamic ad hoc network topology includes:

[0067] Discretize the vehicle speed value into three states: low speed, medium speed, and high speed, and construct a three-dimensional state space by associating the current AP node density and channel occupancy rate;

[0068] Input the three-dimensional state space into a pre-trained deep Q-network model, output the value function evaluation values of each candidate operation, and generate an AP / STA mode switching instruction based on the ε-greedy strategy;

[0069] The switching instruction is written into the hardware register through the WIFI chipset driver interface, triggering the generation of an ad hoc network topology structure including bridging nodes, and transmitting the topology node distribution data to the routing layer.

[0070] Divide the real-time collected vehicle speed value into three discrete states: low speed, medium speed, and high speed according to a preset interval. The interval division is based on the urban road speed limit standard and vehicle movement characteristics, improving the model processing efficiency by reducing the state space dimension. For example, the low-speed state corresponds to congestion or parking scenarios, the medium speed corresponds to normal driving, and the high speed corresponds to highway or expressway scenarios. This discretization process provides a structured input for subsequent correlation analysis.

[0071] Associate the discretized speed state with the current AP (access point) node density level and channel occupancy rate level to form a three-dimensional discrete state space. Among them, the AP node density is obtained by scanning the number of surrounding SSIDs and signal strength statistics, and the channel occupancy rate is quantified by monitoring the proportion of channel idle time. The three-dimensional state space represents the comprehensive state of vehicle mobility, network coverage, and channel load, providing multi-dimensional environmental perception data for mode switching decisions.

[0072] The constructed three-dimensional state space is input into a pre-trained Deep Q-Network (DQN) model, and the value function evaluation values of three operations, namely AP mode maintenance, STA (Station) mode switching, and bridging mode activation, are output. The DQN model is trained with historical networking scenario data to learn the long-term benefits of each operation under different states for network throughput, latency, and other metrics. The model uses a convolutional neural network structure to extract state space features and outputs the value probability distribution of each operation.

[0073] Based on the value function evaluation values output by the DQN model, an ε-greedy strategy is used to generate mode switching instructions. This strategy selects the operation with the highest value evaluation value with a preset probability (exploitation), or selects other operations with a random probability (exploration), balancing network optimization and the ability to adapt to unknown states. For example, in a high-channel load state, it preferentially switches to the STA mode to reduce interference, while in a low-load scenario, it retains the bridging mode to expand network coverage.

[0074] The generated mode switching instructions are written into the hardware register through the WiFi chipset driver interface to trigger the working mode switching of the wireless communication module. The AP mode node creates an independent SSID and broadcasts beacon frames, the STA mode node scans and associates with neighboring APs, and the bridging mode node establishes a wireless distributed system (WDS) under the 802.11s protocol. Thus, an ad-hoc network topology containing multi-hop bridging nodes is generated to achieve dynamic interconnection between vehicle nodes.

[0075] The reconstructed topology node distribution data is transmitted to the routing layer through a shared memory or inter-process communication mechanism. The routing layer updates the neighbor node list based on the topology information, calculates the optimal path weight, and provides a topology change event trigger signal for the OLSR protocol. This data flow realizes the collaborative optimization between the network layer and the MAC layer and supports the dynamic update of routing table entries.

[0076] The above technical content forms a closed-loop control link: the discretization of vehicle states and the construction of a three-dimensional space provide structured input for the DQN model; model inference combines with the ε-greedy strategy to generate switching instructions that balance stability and exploration; the execution of hardware instructions directly drives the reconstruction of the network topology and affects subsequent state space parameters through a data feedback mechanism; the cross-layer transmission of topology data realizes the linkage between routing optimization and resource scheduling. Each step, through the progressive relationship of state perception, decision execution, and effect feedback, jointly realizes the adaptive adjustment of the vehicle ad-hoc network topology to dynamic traffic scenarios.

[0077] Specifically, for the V2V interaction method of the vehicle ad-hoc network based on intelligent connected vehicle WiFi of the present invention, based on the node distribution information in the dynamic ad-hoc network topology, a pheromone data packet is generated by embedding a path delay field in the OLSR protocol packet, and the routing table entries of each node are updated through periodic broadcasting, including:

[0078] Extend the delay field in the header of the OLSR protocol Hello message, and periodically generate pheromone data packets containing hop count, RSSI value, and the current queue length;

[0079] Intermediate nodes calculate the pheromone concentration gradient based on the delay field value and the RSSI attenuation coefficient, and update the routing table entries in descending order of the gradient value;

[0080] The routing table is cross-layer transmitted to the priority queue manager in the message transmission layer through the Socket interface, driving high-priority messages to preempt low-delay paths sorted in descending order based on the pheromone concentration gradient value. The pheromone concentration gradient is obtained by weighted calculation of path delay, RSSI attenuation coefficient, and network load rate. The multi-hop routing path generation method of the present invention has the following specific implementation steps and technical associations:

[0081] Extend a custom delay field in the header of the Hello message of the OLSR (Optimized Link State Routing) protocol to record the transmission delay data from the current node to the neighbor node. The Hello message is broadcast at a preset period, carrying hop count, RSSI (Received Signal Strength Indicator) value, and network layer queue length information to form a dynamic pheromone data packet. The queue length is obtained by counting the number of data packets to be sent at the network interface, reflecting the instantaneous load status of the node. This extension mechanism realizes the real-time collection and flooding transmission of routing information, providing a data basis for path optimization.

[0082] After receiving the pheromone data packet, the intermediate node parses the delay field value and associates it with the RSSI attenuation coefficient to calculate the path pheromone concentration gradient. The attenuation coefficient is dynamically adjusted according to the wireless channel fading model, characterizing the attenuation degree of the signal strength with distance and obstacles. The node sorts the adjacent paths in descending order of the gradient value and preferentially selects the path with a high concentration gradient to update the local routing table entries. This calculation process integrates link quality and network load metrics, avoiding path congestion or signal attenuation problems caused by a single metric.

[0083] The updated routing table is cross-layer transmitted to the priority queue manager in the message transmission layer through the Socket interface. The Socket interface is bound to a specific port number and uses the UDP protocol to achieve low-overhead data transmission. The priority queue manager parses the delay and hop count data in the routing table entries, dynamically allocates low-delay path resources for high-priority messages, and triggers the MAC layer channel preemption mechanism. For example, emergency security messages are directly mapped to a preset high-priority queue to bypass the conventional scheduling algorithm for fast forwarding.

[0084] Based on the path delay information provided by the routing table, the priority queue manager implements a differentiated scheduling strategy. High-priority messages preempt low-latency paths by reserving time slots or increasing retransmission priorities, while low-priority messages are distributed to multiple paths using a load balancing strategy. The preemption mechanism is synchronized with the routing maintenance cycle of the OLSR protocol to avoid message disorder or packet loss caused by frequent path switching.

[0085] The above technical contents constitute the adaptive routing optimization link: protocol extension realizes the periodic collection of topology and load data; pheromone gradient calculation associates the signal strength of the physical layer with the status of the network layer to generate a comprehensive path evaluation result; the cross-layer transmission mechanism opens up the data interaction between the routing layer and the transmission layer to support priority scheduling decisions; the path preemption strategy dynamically adjusts resource allocation based on real-time routing information to form a closed-loop control of "data collection-path evaluation-resource scheduling". Each step improves the stability and real-time performance of multi-hop routing in high-speed vehicle movement scenarios through protocol extension, cross-layer collaboration and dynamic scheduling linkage.

[0086] Specifically, the V2V interaction method of the vehicle ad hoc network based on the WiFi of the intelligent connected vehicle described in the present invention, the V2V message of the vehicle application layer is received, the message type and the number of propagation hops are input into the federated learning classification model, the priority label is output and embedded in the message header, and the MAC layer channel preemption mechanism is triggered, including: extracting the message type, the number of propagation hops and the end-to-end delay in the vehicle to construct a ternary feature vector, adding Laplace noise to implement differential privacy encryption;

[0087] Upload the encrypted feature vector to the federated learning classification model of the edge server, aggregate multi-node data and update the message priority classification rules;

[0088] The classification rules are digitally signed and written into the vehicle communication protocol stack via OTA, adding the highest priority tag to the emergency braking message and triggering the MAC layer CSMA / CA backoff mechanism to bypass.

[0089] In the vehicle local communication protocol stack, the type identifier, number of transmitted hops and end-to-end delay measurement of the V2V message are extracted in real time to construct a three-dimensional feature vector. The message types are defined according to the SAE J2735 standard and are divided into categories such as safety warnings, collaborative control, and infotainment. To protect data privacy, Laplace noise that complies with the differential privacy mechanism is added to the feature vector so that the transmission record of a single vehicle cannot be reversed through feature association. The noise injection process is completed in the on-board processing unit, and the original sensitive information is deleted immediately after the encrypted feature data is generated.

[0090] The encrypted feature vectors are uploaded to the federated learning classification model of the edge server through the vehicle networking edge computing nodes. The server uses a secure multi-party computation protocol to aggregate the data of multiple nodes and update the decision boundary of the message priority classification rules. The federated learning model adopts a convolutional neural network structure and completes the update of model parameters in the encrypted data domain through a distributed gradient descent algorithm, avoiding the original data leaving the vehicle terminal and achieving a balance between data privacy and model accuracy.

[0091] The updated classification rules are digitally signed by the edge server using the RSA algorithm to generate a rule file containing the version number and the effective timestamp. The signed file is distributed to the vehicle terminal through an OTA (Over-the-Air) encrypted channel, and the signature verification and integrity check are completed in the rule parsing module of the communication protocol stack. The verified rule file is loaded into the QoS (Quality of Service) policy engine of the protocol stack to form a message priority mapping table.

[0092] When an emergency braking message is recognized, the classification engine embeds the highest priority label in the message header. This label triggers the optimization process of the CSMA / CA backoff mechanism at the MAC layer: the priority message directly enters the zero backoff window transmission stage, skipping the regular random backoff counting process. At the same time, the physical layer modulation and coding scheme is automatically switched to the more robust QPSK modulation to improve the transmission reliability of high-priority messages in a complex channel environment.

[0093] The above technical content forms a collaborative optimization link: the construction and encryption of feature vectors achieve the unity of local data validity and privacy; the federated learning mechanism completes the iteration of the classification model while protecting data privacy; the secure distribution process ensures the integrity and legality of the rule file; the priority label triggers cross-layer transmission optimization to achieve low-latency transmission of emergency messages. Each link builds a V2V message classification system that takes into account both privacy protection and transmission efficiency through the progressive relationship of data encryption, distributed learning, secure transmission, and protocol optimization.

[0094] Specifically, for the V2V interaction method of the vehicle ad-hoc network based on the WiFi of intelligent connected vehicles of the present invention, the spatio-temporal alignment of the shared location data of neighboring vehicles and the local IMU sensor data, the generation of vehicle pose estimation through the particle filter algorithm, and the calculation of collaborative control instructions based on the Nash equilibrium strategy matrix and the distribution to the target vehicle through the optimized multi-hop path include:

[0095] Perform UTM projection transformation on the location data of neighboring vehicles in the GPS coordinate system, perform sliding window alignment with the local IMU angular velocity data, and generate vehicle pose estimation through the particle filter algorithm;

[0096] Calculate the safety distance threshold between the vehicle and neighboring vehicles based on the Nash equilibrium strategy matrix, and generate a control instruction containing the collaborative deceleration gradient value;

[0097] The control instructions are distributed to the target vehicle group through the IPv6 multicast address, and the decision delay data is written into the routing layer path redundancy calculation module.

[0098] The GPS position data shared by neighboring vehicles is converted into the UTM (Universal Transverse Mercator) plane coordinate system to eliminate the influence of the earth's curvature on the calculation of the relative positions of nearby vehicles. At the same time, the angular velocity data of the local IMU (Inertial Measurement Unit) is collected, and the sliding window mechanism is used to align the timestamps to compensate for the difference in the sensor sampling frequencies. The length of the sliding window is dynamically adjusted according to the vehicle motion state, shortening the window in high-speed scenarios to improve real-time performance and lengthening the window in low-speed scenarios to enhance the data fusion accuracy. This step realizes the spatio-temporal reference unification of multi-source heterogeneous data.

[0099] The spatio-temporally aligned UTM coordinate data and IMU angular velocity data are input into the particle filter algorithm, and the particle weights are iteratively updated through the importance sampling and resampling mechanisms. The particle state space includes vehicle position, heading angle, and velocity components, and through the kinematic model prediction and multi-sensor observation value matching, a vehicle pose estimation result with high confidence is generated. This fusion process effectively suppresses the positioning drift problem caused by GPS signal loss or IMU cumulative error.

[0100] A multi-vehicle cooperative game model is established based on the Nash equilibrium theory, and the vehicle pose estimation results, road friction coefficient, and braking response time are quantified as the policy matrix parameters. The Pareto optimal solutions of the host vehicle and neighboring vehicles in the longitudinal following and lateral lane-changing scenarios are calculated to generate a dynamic safety distance threshold. The threshold varies non-linearly with the square value of the relative speed to adapt to the safety margin requirements under different traffic flow densities.

[0101] According to the deviation of the safety distance threshold, a PID control algorithm is used to generate a cooperative control instruction including the target deceleration gradient value and the action time window. The gradient value is distributed according to an exponential law, triggering a steep deceleration instruction in the emergency collision avoidance scenario and generating a gentle speed adjustment instruction in the conventional following scenario. The instruction data is encapsulated in the ASN.1 encoding format, and a timestamp and vehicle group identifier are added.

[0102] The control instructions are sent to the target vehicle group through the IPv6 multicast address, and the MLD (Multicast Listener Discovery) protocol is used to maintain the dynamic group member relationship. The routing layer preferentially selects a multicast path with a delay jitter lower than the threshold and the fewest hops according to the decision delay data stored in the path redundancy calculation module. The delay data is obtained by calculating the timestamp difference and is used to evaluate the path stability and trigger the routing table update.

[0103] The above technical content forms a closed-loop decision-making link: spatio-temporal alignment and pose estimation provide high-precision environmental perception data for collaborative decision-making; the Nash equilibrium model converts the motion relationship between vehicles into a mathematical optimization problem and generates safety strategies; control instruction encoding and multicast distribution achieve reliable transmission of decision-making results; delay data is fed back to the routing layer to form a basis for path optimization. Through the two-way interaction of data flow and control flow in each link, the real-time performance and safety of vehicle collaborative decision-making are unified. Among them, the pose estimation accuracy directly affects the convergence speed of the policy matrix, and the routing optimization effect determines the end-to-end transmission reliability of control instructions.

[0104] Specifically, for the V2V interaction method of the vehicle ad-hoc network based on intelligent connected vehicle WiFi of the present invention, collecting the load rate of AP nodes and the number of channel conflicts, iteratively calculating the AP / STA ratio threshold through the particle swarm optimization algorithm, triggering node role reconfiguration according to the calculation result, and feeding back the load status to the ad-hoc network topology construction module includes:

[0105] Quantify the AP node density, STA connection number, and channel conflict rate into a three-dimensional vector space, and iteratively calculate the optimal AP / STA ratio threshold through the particle swarm algorithm;

[0106] Trigger a downgrade instruction in the area where the AP density exceeds the threshold, switch the edge AP node to the STA mode, and establish an 802.11s bridging link with the central AP;

[0107] Write the role assignment result into the configuration parameter library of the ad-hoc network topology construction module, and trigger the reconstruction of the topology structure based on the current load status.

[0108] Real-time collect data on the AP (access point) node density, STA (station) connection number, and channel conflict rate through the vehicle-mounted communication module, and map them to the preset normalization interval respectively. The AP node density is statistically obtained by scanning the surrounding SSID broadcast frequency, the STA connection number is obtained based on the association request frame count, and the channel conflict rate is calculated by the physical layer carrier sense failure rate. Construct the three types of parameters into a three-dimensional vector space to characterize the distribution characteristics of the regional network load. This quantization model provides a standardized input for dynamic threshold calculation.

[0109] Initialize the particle swarm parameters, and set the AP / STA ratio threshold as the optimization objective function. The position of each particle corresponds to a candidate threshold value, and the fitness function is defined as the weighted reciprocal of network throughput and transmission delay. The algorithm adjusts the particle velocity through the inertia weight and iteratively searches for the Pareto optimal solution in the solution space. When the fitness fluctuation of three consecutive iterations is less than the preset tolerance, it is determined that the algorithm converges and outputs the current optimal threshold.

[0110] In the area where the AP density exceeds the threshold, the edge AP nodes are screened according to the node position entropy value. After the downgrade instruction is triggered, the target node stops broadcasting the SSID and releases the IP address pool, and establishes a Mesh bridging link with the central AP through the IEEE 802.11s protocol. The bridging link enables the Airtime fair scheduling algorithm to allocate the wireless channel occupancy time according to the node load ratio, eliminating the problem of channel overload in the local area.

[0111] Write the node role assignment result into the configuration parameter library of the ad hoc network topology construction module, and update the weight of the AP / STA mode switching rule. After receiving the update instruction, the topology construction module rescans the network status and calls the deep Q network model to generate a topology reconstruction plan. The reconstructed topology data is synchronized to the routing layer, driving the OLSR protocol to update the multi-hop routing path priority list, forming a closed-loop control of load status perception - threshold calculation - topology adjustment.

[0112] The above technical content constitutes an adaptive load balancing loop: the network status is quantitatively modeled to provide input features for the particle swarm algorithm; the optimization algorithm outputs a threshold to drive the node role switching decision; the bridging link is established to relieve local congestion and maintain network connectivity; the role assignment result is fed back to the topology construction module to trigger reconstruction, and the reconstructed topology status affects the network parameter acquisition again. Each link is linked through the parameter transfer and event triggering mechanism to achieve load balancing and resource optimal allocation of the vehicle ad hoc network in the dynamic traffic scenario, where the particle swarm algorithm solves the multi-objective optimization problem, the bridging protocol ensures the smooth transition of the topology reconstruction process, and the deep Q network model ensures the matching degree of the reconstruction strategy with the real-time scenario.

[0113] Specifically, the V2V interaction method of the vehicle ad hoc network based on the intelligent connected vehicle WiFi of the present invention further includes:

[0114] Divide the vehicle speed into three intervals of [0, 20), [20, 60), and [60, ∞) km / h to define the low-speed, medium-speed, and high-speed states, associate the current channel occupancy rate level with the AP node density interval, and construct a three-dimensional discrete state space;

[0115] Input the three-dimensional discrete state space into the above-mentioned deep Q network model, and output the value function evaluation values of three operations: AP mode retention, STA mode switching, and bridging mode startup;

[0116] Calculate the mode switching probability distribution according to the value function evaluation value, and drive the WIFI chipset to switch to the STA mode when the channel occupancy rate exceeds the threshold.

[0117] According to the typical working conditions of urban roads, the vehicle speed is divided into three discrete intervals: [0, 20) km / h, [20, 60) km / h, and [60, ∞) km / h, corresponding to low speed (congestion or parking), medium speed (urban normal driving), and high speed (expressway or high-speed driving) states respectively. The interval thresholds refer to the association definition of vehicle speed and road type in the SAE J3016 autonomous driving classification standard, and the model processing efficiency is improved by reducing the dimension of continuous speed data.

[0118] The channel occupancy rate level and the AP node density level are collected in real time and quantified into three discrete values of low, medium, and high respectively. The channel occupancy rate is calculated by monitoring the busy-idle ratio of the target channel CCA (Clear Channel Assessment), and the AP node density is based on the statistics of the number of SSID broadcasts obtained by WiFi spectrum scanning. The speed state, the channel occupancy rate level, and the AP node density level are used to construct a three-dimensional discrete state space. Each dimension contains 3 discrete values, forming 27 combined states, which fully characterize the coupling relationship between vehicle mobility and network load.

[0119] The three-dimensional discrete state space is input into a pre-trained Deep Q-Network (DQN) model, and the value function evaluation values of three operations, namely AP mode retention, STA mode switching, and bridging mode activation, are output. The DQN model adopts a dual-network architecture, including an evaluation network and a target network, and obtains the state-action mapping relationship through training with historical networking data. The output layer of the model uses a linear activation function to directly map the expected network benefits of each operation in a specific state.

[0120] Based on the value function evaluation values output by the DQN model, the mode switching probability distribution is calculated using the Boltzmann distribution. High-value operations obtain higher selection probabilities, and at the same time, a low-probability random selection mechanism is retained to avoid local optimal solutions. When it is detected that the channel occupancy rate exceeds the preset threshold, the probability calculation module sends a mode switching instruction to the WiFi chipset to trigger the STA mode switching process.

[0121] The mode switching instruction is written into the hardware control register through the PCIe interface of the WiFi chipset. When switching from AP mode to STA mode, stop the SSID broadcast and release the DHCP server resources, and start the passive scanning process to associate with neighboring AP nodes; when the bridging mode is activated, enable the Mesh networking function of the 802.11s protocol to establish a multi-hop wireless distributed system. After the switching is completed, the new network state data is transmitted back to the state space construction module to form a closed-loop feedback.

[0122] The discrete state space constructs structured input for deep reinforcement learning, converting continuous environmental parameters into processable discrete states; the DQN model generates action value evaluations based on state space reasoning to drive the probabilistic decision-making mechanism; the mode switching instruction directly changes the network topology configuration through the hardware interface, and the topology change in turn affects the channel occupancy rate and AP density parameters, forming a closed-loop control link of "state perception - decision execution - environmental feedback". Each step realizes the real-time adaptive adjustment of the vehicle ad hoc network topology to the dynamic traffic scenario through the progressive relationship of parameter discretization, model reasoning, probabilistic decision-making, and hardware control.

[0123] Specifically, for the V2V interaction method of the vehicle ad hoc network based on the WiFi of intelligent connected vehicles described in the present invention, the calculation of the pheromone concentration gradient includes:

[0124] Expand the TCL (delay control list) field in the TC (topology control) message of the OLSR protocol to record the path delay value, calculate the initial pheromone concentration according to the RSSI signal strength, and set the delay weight factor as the reciprocal of the current network load rate;

[0125] When the intermediate node forwards the data packet, update the pheromone concentration value according to the formula concentration decay amount = number of hops × preset decay coefficient, and the decay coefficient is dynamically adjusted according to the network topology stability. Specifically, when the average vehicle moving speed exceeds 60 km / h, the decay coefficient increases to 0.8; when the network load rate is higher than 70%, the decay coefficient decreases to 0.5. Generate gradient distribution data;

[0126] Write the gradient distribution data into the priority flag bit of the routing table entry, and preferentially select the path with the highest flag bit value to establish a TCP long connection.

[0127] Expand the TCL (delay control list) field in the header of the TC (topology control) message of the OLSR (optimized link state routing) protocol to record the end-to-end transmission delay data of the path. The delay value is calculated by the timestamp difference method, and the cumulative transmission delay of the data packet from the source node to the current node is recorded. At the same time, calculate the initial pheromone concentration based on the RSSI (received signal strength indication) value of the received data packet. The RSSI strength is associated with the transmission power and antenna gain parameters and is mapped to the initial concentration value by the look-up table method. The TCL field adopts the TLV (type - length - value) encoding format, supporting dynamic field extension and parsing.

[0128] Take the current network load rate as the basis for calculating the delay weight factor, and set the weight factor as the reciprocal of the network load rate. The network load rate is obtained by statistically calculating the proportion of the busy duration of the channel within a unit time. When the load rate increases, the weight proportion of the delay parameter is reduced to avoid over-selection of high-load paths. The weight factor configuration module periodically collects MAC layer statistical information, generates dynamic adjustment parameters, and writes them into the configuration register of the routing protocol stack.

[0129] When an intermediate node forwards a data packet, it parses the path delay value in the TCL field and calculates the pheromone concentration attenuation amount in combination with a preset attenuation coefficient. The attenuation coefficient is dynamically adjusted according to the network topology stability. In the scenario of high-speed vehicle movement, the attenuation coefficient is increased to accelerate the elimination of old paths, and in a stable scenario, the coefficient is reduced to maintain routing consistency. The attenuation calculation result is updated to the local pheromone concentration table to form path quality gradient distribution data.

[0130] Write the updated pheromone gradient data into the priority flag bit of the routing table entry. The flag bit value is mapped to the 0 - 255 range through normalization processing. The routing selection engine preferentially selects the path with the highest flag bit value to establish a TCP long connection and reserves bandwidth resources for this path. The TCP connection enables the fast retransmission mechanism and dynamically adjusts the congestion window growth factor according to the pheromone gradient value to achieve the matching optimization of the transmission rate and path quality.

[0131] The above technical content constitutes an adaptive routing optimization mechanism: protocol field extension realizes the standardized carrying of path quality parameters; dynamic weight factor configuration balances the decision-making weights of delay and load indicators; concentration attenuation calculation reflects the spatio-temporal variation characteristics of path quality; the priority mapping mechanism converts abstract gradient data into an operable routing strategy. Each link forms a closed-loop control link of "data collection - quality evaluation - path optimization" through cross-layer interaction and parameter linkage of the protocol stack. Among them, the pheromone gradient data not only guides single-path selection but also realizes transport layer optimization through TCP connection parameter adjustment, improving the routing stability and transmission reliability of vehicle ad hoc networks in a dynamic environment.

[0132] Specifically, in the V2V interaction method of the vehicle ad hoc network based on intelligent connected vehicle WiFi of the present invention, the differential privacy encryption includes:

[0133] Add Laplace distribution noise data to the feature vectors of the message type and propagation hop count, so that the correlation entropy value between the single-hop transmission record and the message type exceeds a preset threshold;

[0134] Aggregate the encrypted data through an edge server and update the weights of the Softmax output layer of the federated learning classification model;

[0135] The classified rules are sent to the vehicle terminal after appending the SHA-256 hash value, and integrity verification is implemented in the communication protocol stack parsing module.

[0136] In the feature extraction module of the in-vehicle communication protocol stack, a binary feature vector is constructed for the message type encoding (such as emergency braking, cooperative lane change, etc.) and the propagation hop count value. A Laplace noise generator injects noise data conforming to the differential privacy mechanism into each dimension of the feature vector, and the noise magnitude is dynamically adjusted according to the preset privacy budget parameter. After injecting the noise, the mutual information between the single-hop transmission record and the message type is suppressed below the correlation entropy threshold, making it impossible for attackers to reverse-engineer the original message source and propagation path through reverse engineering. The noise injection process is completed in the vehicle's local security coprocessor, and the original data is encrypted and temporarily stored in the memory and then immediately erased.

[0137] The encrypted feature vector is uploaded to the edge server through the vehicle networking edge computing node, and the server uses the secure multi-party computing protocol to implement the update of the federated learning model. The encrypted data uploaded by each node performs gradient calculation in the ciphertext domain, and the weights of the Softmax output layer of the federated learning classification model are updated through the homomorphic encryption algorithm. During the model update process, the local data of each vehicle terminal always remains on the original device, and the server only obtains the encrypted gradient parameters, achieving a balance between privacy protection and model generalization ability.

[0138] The updated message priority classification rules are digitally signed using the RSA algorithm, and the timestamp and version number information are appended. The rule file generates a digest value through the SHA-256 hash algorithm, and the rule content and the digest value are bound and encapsulated into a secure data packet. The encapsulation process is completed in the trusted execution environment (TEE) of the edge server to prevent rule tampering and man-in-the-middle attacks.

[0139] The secure data packet is sent to the vehicle terminal through the cellular network or the DSRC channel. The parsing module of the communication protocol stack separates the rule content and the hash value, recalculates the digest value using the same SHA-256 algorithm, and verifies the consistency. The verified rule file is loaded into the QoS policy engine of the protocol stack, and the message type is mapped to the MAC layer priority label. After the emergency braking message triggers the priority label, it bypasses the conventional CSMA / CA backoff counter and directly enters the channel preemption process.

[0140] The above steps form a collaborative mechanism for privacy protection and transmission optimization: noise injection confuses sensitive features and blocks the risk of privacy leakage; federated learning realizes model iteration based on encrypted data to ensure classification accuracy; the security encapsulation and verification mechanism maintains the integrity of the rules; the priority label drives cross-layer transmission optimization. Each link constructs a V2V message processing system that takes into account both data privacy and communication efficiency through a closed-loop process of "local encryption - cloud aggregation - secure distribution - terminal execution". Among them, the combination of differential privacy mechanism and federated learning solves the contradiction between data sharing and privacy protection, and hash verification and digital signature establish an end-to-end secure trust chain.

[0141] Specifically, for the V2V interaction method of the vehicle ad-hoc network based on the intelligent connected vehicle WiFi of the present invention, the bridging link includes:

[0142] When it is detected that subnet isolation is caused by the degradation of the AP node, a dual-mode node that supports the 802.11s bridging protocol is selected as the relay;

[0143] Maintain the mapping relationship between the source subnet ID, the destination subnet ID and the next-hop MAC address in the routing table of the bridging node, and forward the control message through EtherType 0x88CC;

[0144] Write the throughput and delay data of the bridging link into the particle swarm algorithm iteration parameter constraint library to limit the calculation range of the AP / STA ratio.

[0145] Detect the subnet connectivity by periodically sending ICMPv6 neighbor discovery messages. When the response from the AP node of the target subnet is not received three times in a row, it is determined that a subnet isolation event has occurred. Start the candidate node screening process, and preferentially select a node that supports the IEEE802.11s protocol and has the dual-mode (AP / STA) working ability as the relay. The screening criteria include the signal strength threshold, the node load status and the remaining battery level, and the node with the highest comprehensive score is selected through the weighted scoring algorithm to activate the bridging mode.

[0146] Establish the mapping relationship between the source subnet ID, the destination subnet ID and the next-hop MAC address in the routing table of the bridging node, store it in a hash table structure and establish a fast query index. The control message is encapsulated with the EtherType 0x88CC identifier and transparently transmitted across subnets at the data link layer. After the bridging node parses the destination subnet ID of the message, it queries the mapping table to obtain the next-hop MAC address, repackages the frame header and forwards it to the target subnet to maintain the communication continuity of the control plane between subnets.

[0147] Real-time monitor the throughput, latency, and packet loss rate metrics of the bridging link. After formatting the data into a JSON structure, write it into the constraint library for the iterative parameters of the particle swarm algorithm. The constraint library defines the upper and lower boundaries of the AP / STA ratio threshold. When the latency of the bridging link exceeds the preset threshold, dynamically shrink the algorithm search space and limit the calculation range of the AP node density. The constraint parameters are synchronized to the topology construction module through the shared memory mechanism to avoid large-scale topology oscillations caused by the degradation of the bridging link performance.

[0148] The above steps form a closed-loop maintenance mechanism: subnet isolation detection triggers the selection of relay nodes to ensure network connectivity; routing table maintenance realizes the efficient forwarding of cross-subnet control messages; performance data is fed back to the particle swarm algorithm constraint library to reverse-optimize the network parameter decision. Each link is linked by event triggering and data flow. Among them, the performance data of the bridging link is used not only for real-time routing optimization but also as historical experience data to guide the subsequent calculation of the AP / STA ratio threshold, realizing the dynamic balance between network load balancing and topology stability.

[0149] The technical solution of the present invention is specifically implemented as follows:

[0150] Embodiment 1: Dynamic self-organizing network topology construction and dynamic switching of node roles;

[0151] This embodiment realizes the autonomous switching of the vehicle node's AP / STA mode and network topology reconstruction. First, on-vehicle sensors collect data on vehicle speed, position, and network load rate, and generate a multi-dimensional feature vector through normalization processing; discretize the speed into three states: low speed (0 - 20 km / h), medium speed (20 - 60 km / h), and high speed (>60 km / h), and construct a three-dimensional discrete state space by associating the channel occupancy rate with the AP node density. The Deep Q-Network (DQN) model outputs evaluation values for AP mode retention, STA switching, and bridging mode activation based on this state space, and uses the ε-greedy strategy to generate switching instructions. For example, when a high-speed moving vehicle enters an area with a high density of APs, the DQN triggers an STA mode switching instruction, which is written into the register through the WiFi chipset driver interface, forming a dynamic topology containing 802.11s bridging nodes, and passing the topology data to the routing layer. This mechanism reduces the dependence on infrastructure through distributed decision-making and improves network connectivity in dynamic scenarios.

[0152] Embodiment 2: Multi-hop routing optimization and message priority classification;

[0153] Expand the Hello message delay field in the OLSR protocol, and periodically generate pheromone data packets containing hop count, RSSI value, and queue length. Intermediate nodes calculate the pheromone gradient based on the delay weight factor (the reciprocal of the network load rate) and the RSSI attenuation coefficient, and update the routing table entries in descending order of the gradient value. At the same time, the on-vehicle terminal extracts the message type and propagation hop count to construct a three-element feature vector. After adding Laplace noise for differential privacy encryption, it is uploaded to the federated learning model of the edge server. The model aggregates multi-node data to update the classification rules, and issues them to the vehicle terminal through digital signatures. For example, the emergency braking message is marked with the highest priority, triggering the MAC layer CSMA / CA backoff mechanism to bypass and preferentially preempt low-delay paths. This solution realizes end-to-end low-delay transmission of high-priority messages through cross-layer routing optimization and distributed learning.

[0154] Example 3: Closed-loop load balancing and collaborative decision-making control;

[0155] Iteratively calculate the AP / STA ratio threshold through the particle swarm optimization algorithm. When the AP density exceeds the threshold, trigger the edge node degradation instruction to establish a bridging link to relieve local congestion. At the same time, perform UTM projection transformation on the GPS data of neighboring vehicles. After aligning with the local IMU angular velocity data, use the particle filter algorithm to generate vehicle pose estimation. Calculate the collaborative deceleration gradient instruction based on the Nash equilibrium strategy matrix and distribute it to the target vehicle group through IPv6 multicast. For example, in the vehicle platoon scenario, the bridging node dynamically shunts the communication load, and the collaborative control instruction is distributed along the optimized path to reduce the multi-hop transmission delay. This example improves the resource utilization rate and group collaboration efficiency in complex scenarios through the closed-loop linkage of network state feedback and collaborative decision-making.

[0156] Explanation of the technical features of the present invention:

[0157] Dynamic ad hoc network topology construction and AP / STA mode switching;

[0158] AP / STA mode: The AP (access point) mode means that the vehicle acts as the central node of the wireless network to broadcast the SSID and manage the connections; the STA (station) mode means that the vehicle acts as a terminal node and associates with the neighboring AP. The mode switching realizes the hardware layer configuration change through the WiFi chipset driver interface.

[0159] Deep Q-Network (DQN) model: A reinforcement learning model that, based on state data such as vehicle speed and network load, evaluates the long-term benefits of different operations (such as maintaining the AP mode and switching the STA) through a value function to drive dynamic decision-making.

[0160] ε-greedy strategy: A decision-making mechanism that balances exploration and exploitation. It randomly selects an operation with probability ε (exploring a new state) and selects the current optimal operation with probability 1 - ε (exploiting the known benefits) to avoid local optimal solutions.

[0161] Multi-hop Routing Optimization and Pheromone Mechanism

[0162] OLSR Protocol Extension: Add a delay field to the Hello message header of the Optimized Link State Routing (OLSR) protocol to record the path transmission delay for dynamically calculating the routing quality.

[0163] Pheromone Data Packet: A broadcast message containing the hop count, RSSI value, and queue length. Intermediate nodes update the pheromone concentration gradient according to the delay weight factor (the reciprocal of the network load rate) to guide the priority sorting of the routing table.

[0164] Federated Learning Classification Model: A distributed machine learning framework where vehicle local encrypted feature data (message type, hop count) is uploaded to the edge server for aggregation, generating global classification rules and distributing them to achieve dynamic annotation of message priorities under privacy protection.

[0165] Load Balancing and Cooperative Decision-making

[0166] Particle Swarm Optimization Algorithm: By iteratively calculating the three-dimensional vector space of the AP node density, STA connection count, and channel conflict rate, output the optimal AP / STA ratio threshold to trigger node role reconfiguration for load balancing.

[0167] Nash Equilibrium Strategy Matrix: A multi-vehicle cooperation model based on game theory that maps vehicle pose estimation and safety distance thresholds to a strategy matrix to generate Pareto optimal cooperative deceleration instructions.

[0168] 802.11s Bridging Link: A Mesh network protocol compliant with the IEEE 802.11s standard that establishes a multi-hop wireless distributed system during subnet isolation and forwards control messages through the EtherType 0x88CC identifier to maintain network connectivity.

[0169] The above features form a closed-loop technology system: The DQN model and ε-greedy strategy achieve topological dynamic reconstruction, providing a basis for routing optimization; OLSR extension and pheromone mechanism improve path selection efficiency, supporting the transmission of high-priority messages in the federated learning model; the particle swarm algorithm and bridging protocol dynamically adjust network load, and cooperate with the Nash equilibrium strategy to achieve low-latency distribution of group control instructions. Each module is linked through state perception, decision execution, and feedback optimization to solve the deficiencies of traditional V2X technologies that rely on infrastructure and have insufficient real-time performance.

[0170] The present invention effectively solves the problems of high deployment cost and poor scalability caused by the dependence on dedicated hardware and infrastructure in existing V2X technologies, and the low real-time performance caused by the insufficient dynamic topology adaptation ability of traditional networking modes through the following technical solutions:

[0171] First, the vehicle motion state and network load data are collected in real time through in-vehicle sensors. Combining the Deep Q-Network (DQN) model with the ε-greedy strategy, the dynamic switching of the vehicle node's AP / STA role is realized. The DQN model infers and generates mode switching instructions based on multi-dimensional feature vectors, driving the WiFi chipset to reconstruct the network topology at the hardware layer. For example, in high-channel-load areas, edge AP nodes are downgraded to STA mode, while maintaining network connectivity through the 802.11s bridging protocol. This mechanism does not rely on dedicated communication modules or roadside infrastructure and can achieve distributed networking by only reusing existing WiFi hardware, significantly reducing deployment costs and hardware transformation costs. In addition, the three-dimensional discrete state space modeling and dynamic decision-making mechanism enable the network topology to adapt to the high-speed movement scenario of vehicles, reducing communication interruptions caused by frequent topology changes.

[0172] Second, by expanding the OLSR protocol packet to embed the path delay field and combining pheromone gradient calculation, the dynamic optimization of multi-hop routing is achieved. Intermediate nodes update the routing table entries according to the delay and RSSI attenuation coefficient, and the priority queue manager schedules high-priority messages across layers to preempt low-delay paths. At the same time, the federated learning classification model aggregates multi-node encrypted feature data on the edge server, generates message priority classification rules and distributes them to in-vehicle terminals. For example, emergency braking messages bypass the conventional channel competition process through the MAC layer CSMA / CA backoff mechanism and directly trigger low-delay transmission. This solution avoids the dependence on a centralized base station in traditional V2V communication through distributed learning and cross-layer cooperation, and at the same time improves the real-time performance and reliability of multi-hop routing, meeting the dynamic expansion requirements of a large number of vehicle nodes.

[0173] Finally, based on the particle swarm optimization algorithm, the AP / STA ratio threshold is iteratively calculated to dynamically adjust the node role to balance the network load. When the density of AP nodes exceeds the threshold, an edge AP downgrade instruction is triggered and a bridging link is established to relieve local channel congestion. In addition, the cooperative decision-making module fuses multi-source sensor data through the particle filter algorithm to generate vehicle pose estimation, calculates cooperative control instructions in combination with the Nash equilibrium strategy matrix, and distributes them to the target vehicle group through IPv6 multicast. This mechanism closes the loop linkage of network state awareness, path optimization, and cooperative control, improving network scalability and group cooperation efficiency in complex traffic scenarios while reducing the channel conflict rate. In summary, through dynamic resource allocation and multi-protocol cooperation, this solution realizes the efficient autonomy and low-latency interaction of vehicle ad hoc networks.

Claims

1. A V2V interaction method for a vehicle ad hoc network based on the WiFi of an intelligent connected vehicle, characterized in that, Including: Collect vehicle motion state data and network state data through in-vehicle sensors, normalize the speed, position coordinates, and network load rate, and generate a multi-dimensional feature vector; Input the multi-dimensional feature vector into a pre-trained deep Q-network model, output an operation instruction set including the AP / STA mode switching probability, and select the optimal instruction according to the ε-greedy strategy to construct a dynamic ad-hoc network topology; Based on the node distribution information in the dynamic ad-hoc network topology, embed a path delay field in the OLSR protocol packet to generate a pheromone packet, and update the routing table entries of each node through periodic broadcasting; Receive V2V messages from the in-vehicle application layer, input the message type and propagation hop count into the federated learning classification model, output a priority label and embed it in the message header, and trigger the MAC layer channel preemption mechanism; Perform spatio-temporal alignment on the shared location data of neighboring vehicles and the local IMU sensor data, generate vehicle pose estimation through the particle filter algorithm, calculate the cooperative control instruction based on the Nash equilibrium strategy matrix, and distribute it to the target vehicle through the optimized multi-hop path; Collect the AP node load rate and the number of channel conflicts, iteratively calculate the AP / STA ratio threshold through the particle swarm optimization algorithm, trigger node role reconfiguration according to the calculation result, and feedback the load status to the ad-hoc network topology construction module.

2. The V2V interaction method of the vehicle ad hoc network based on the WiFi of the intelligent connected vehicle according to claim 1, characterized in that, The constructing the dynamic ad-hoc network topology includes: Discretize the vehicle speed value into three states: low speed, medium speed, and high speed, and associate the current AP node density and channel occupancy rate to construct a three-dimensional state space; Input the three-dimensional state space into a pre-trained deep Q-network model, output the value function evaluation values of each candidate operation, and generate an AP / STA mode switching instruction based on the ε-greedy strategy; Write the switching instruction into the hardware register through the WIFI chipset driver interface, trigger the generation of an ad-hoc network topology structure including bridging nodes, and transfer the topology node distribution data to the routing layer.

3. The V2V interaction method of the vehicle ad hoc network based on the intelligent connected vehicle WiFi according to claim 1, characterized in that, The embedding a path delay field in the OLSR protocol packet based on the node distribution information in the dynamic ad-hoc network topology to generate a pheromone packet and updating the routing table entries of each node through periodic broadcasting includes: Expand the delay field in the OLSR protocol Hello packet header, and periodically generate a pheromone packet including the hop count, RSSI value, and current queue length; The intermediate node calculates the pheromone concentration gradient according to the delay field value and the RSSI attenuation coefficient, and updates the routing table entries in descending order of the gradient value; The routing table is cross-layer transmitted to the priority queue manager in the message transmission layer through the Socket interface, and drives the high-priority message to preempt the low-delay path.

4. The V2V interaction method of the vehicle ad hoc network based on the WiFi of the intelligent connected vehicle according to claim 1, characterized in that The receiving V2V messages from the in-vehicle application layer, inputting the message type and propagation hop count into the federated learning classification model, outputting a priority label and embedding it in the message header, and triggering the MAC layer channel preemption mechanism includes: Extract the message type, propagation hop count, and end-to-end delay in the vehicle local to construct a three-element feature vector, and add Laplace noise to implement differential privacy encryption; Upload the encrypted feature vector to the federated learning classification model on the edge server, and aggregate multi-node data to update the message priority classification rule; The classification rules are written into the vehicle communication protocol stack via OTA after being digitally signed, adding the highest priority tag to the emergency braking message and triggering the MAC layer CSMA / CA backoff mechanism to bypass it.

5. The V2V interaction method of the vehicle ad-hoc network based on the WiFi of the intelligent connected vehicle according to claim 1, characterized in that, The spatio-temporal alignment of the neighboring vehicle's shared position data and the local IMU sensor data, generating the vehicle pose estimation through the particle filter algorithm, calculating the cooperative control command based on the Nash equilibrium strategy matrix, and distributing it to the target vehicle through the optimized multi-hop path includes: Performing UTM projection transformation on the neighboring vehicle's position data in the GPS coordinate system, performing sliding window alignment with the local IMU angular velocity data, and generating the vehicle pose estimation through the particle filter algorithm; Calculating the safety distance threshold between the vehicle and the neighboring vehicle based on the Nash equilibrium strategy matrix, and generating a control command including the cooperative deceleration gradient value; The control command is distributed to the target vehicle group through the IPv6 multicast address, and the decision delay data is written into the routing layer path redundancy calculation module.

6. The V2V interaction method of the vehicle ad-hoc network based on the intelligent connected vehicle WiFi according to claim 1, characterized in that Collecting the AP node load rate and the number of channel collisions, iteratively calculating the AP / STA ratio threshold through the particle swarm optimization algorithm, triggering node role reconfiguration according to the calculation result, and feeding back the load status to the ad hoc network topology construction module includes: Quantifying the AP node density, the number of STA connections, and the channel collision rate into a three-dimensional vector space, and iteratively calculating the optimal AP / STA ratio threshold through the particle swarm algorithm; Triggering a downgrade command in the area where the AP density exceeds the threshold, switching the edge AP node to the STA mode and establishing an 802.11s bridging link with the central AP; The role assignment result is written into the configuration parameter library of the ad hoc network topology construction module, triggering the reconstruction of the topology structure based on the current load status.

7. The V2V interaction method of the vehicle ad hoc network based on the WiFi of the intelligent connected vehicle according to claim 2, wherein It also includes: Dividing the vehicle speed into three intervals of [0, 20), [20, 60), and [60, ∞) km / h to define the low-speed, medium-speed, and high-speed states, associating the current channel occupancy rate level with the AP node density interval, and constructing a three-dimensional discrete state space; Inputting the three-dimensional discrete state space into the above-mentioned deep Q network model, and outputting the value function evaluation values of three operations: AP mode maintenance, STA mode switching, and bridging mode startup; Calculating the mode switching probability distribution according to the value function evaluation value, and driving the WIFI chipset to switch to the STA mode when the channel occupancy rate exceeds the threshold.

8. The V2V interaction method of the vehicle ad-hoc network based on the intelligent connected vehicle WiFi according to claim 3, wherein, The calculation of the pheromone concentration gradient includes: Recording the path delay value in the OLSR protocol TCL field, calculating the initial pheromone concentration according to the RSSI signal strength, and setting the delay weight factor as the reciprocal of the current network load rate; When the intermediate node forwards the data packet, updating the pheromone concentration value according to the formula: concentration attenuation amount = number of hops × preset attenuation coefficient, and generating gradient distribution data; The gradient distribution data is written into the priority flag bit of the routing table entry, and the path with the highest flag bit value is preferentially selected to establish a TCP long connection.

9. The V2V interaction method of the vehicle ad-hoc network based on the intelligent connected vehicle WiFi according to claim 4, characterized in that, The differential privacy encryption includes: Adding Laplace distribution noise data to the feature vector of the message type and the propagation hop count, so that the correlation entropy value between the single-hop transmission record and the message type exceeds the preset threshold; Aggregate encrypted data through an edge server to update the weights of the Softmax output layer of the federated learning classification model; After appending the SHA-256 hash value to the classification rule, it is sent to the vehicle terminal for integrity verification in the communication protocol stack parsing module.

10. The V2V interaction method of the vehicle ad-hoc network based on the intelligent connected vehicle WiFi according to claim 6, wherein The bridging link includes: When subnet isolation caused by the degradation of the AP node is detected, select a dual-mode node that supports the 802.11s bridging protocol as a relay; Maintain the mapping relationship between the source subnet ID, destination subnet ID, and the next-hop MAC address in the routing table of the bridging node, and forward control packets through EtherType 0x88CC; Write the throughput and delay data of the bridging link into the particle swarm algorithm iteration parameter constraint library to limit the calculation range of the AP / STA ratio.

Images