An edge-computing-based low-latency communication method and system for vehicle networking

Abstract

The application discloses a kind of based on edge computing's Internet of Vehicles low latency communication method and system, method includes: edge node obtains historical operation data and obtains real-time evaluation network communication link and the state parameters of edge node itself performance;Edge node is based on historical operation data and real-time state parameter prediction generates communication delay value, and real-time monitoring edge node resource to generate comprehensive load index;When communication delay value exceeds preset delay threshold or comprehensive load index shows that edge node is overloaded, triggers edge node to carry out task migration;Edge node selects another edge node as migration target node, and part of task is transferred to migration target node execution.The application identifies potential high latency state in advance by delay prediction, and then triggers migration, transfers task from high load node to low delay or resource sufficient target node, so as to significantly reduce average response latency at system level, improve service continuity.

Description

Technical Field

[0001] This invention relates to the field of vehicle-to-everything (V2X) communication technology, and in particular to a low-latency V2X communication method and system based on edge computing. Background Technology

[0002] With the rapid development of intelligent transportation and autonomous driving technologies, the Internet of Vehicles (IoV) has become a key technology for realizing vehicle intelligence and road cooperation. Traditional IoV systems mostly adopt a centralized cloud architecture, requiring vehicle data to be uploaded to a remote cloud server for processing and decision-making. This approach suffers from significant latency issues in high-density communication scenarios; the long data transmission path results in a large end-to-end response time, making it unsuitable for the millisecond-level control requirements of autonomous driving. To address these technical problems, the applicant proposes a low-latency communication method and system for IoV based on edge computing. Summary of the Invention

[0003] To address the issues of high latency, low reliability, and low bandwidth utilization in the background technology, this invention proposes a low-latency communication method and system for vehicle-to-everything (V2X) based on edge computing. By deploying edge computing modules at roadside nodes, lightweight artificial intelligence algorithms are used to achieve low-latency communication and distributed collaborative control. This method is applicable to scenarios such as autonomous driving, traffic signal coordination, platooning control, emergency vehicle priority passage, and remote driving assistance, and can achieve millisecond-level response and highly reliable communication in complex traffic environments.

[0004] In a first aspect, the present invention proposes a low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing, which is mainly executed by roadside edge nodes. The method includes:

[0005] Edge nodes acquire historical operational data and obtain real-time evaluation parameters of network communication links and their own performance.

[0006] Edge nodes predict and generate communication latency values ​​based on historical operating data and real-time status parameters, and monitor edge node resources in real time to generate comprehensive load indicators;

[0007] When the communication latency exceeds a preset latency threshold or the comprehensive load index shows that the edge node is overloaded, the edge node is triggered to perform task migration.

[0008] Based on the comprehensive load index, the current communication link channel status, and the task priority, the edge node selects another edge node as the migration target node and transfers some tasks to the migration target node for execution.

[0009] Preferably, the state parameters are obtained at least partially from vehicle operating status data, the edge node predicts the communication latency value based on a long short-term memory neural network model, the input of the long short-term memory neural network model includes a comprehensive load index, a historical communication latency sequence obtained from the historical operating data, and a vehicle operating status feature vector obtained from the vehicle operating status data; the output of the long short-term memory neural network model is the communication latency value.

[0010] Preferably, the long short-term memory neural network model adopts an adaptive time window mechanism to adjust the input time step according to changes in vehicle speed and fluctuations in the communication link.

[0011] Preferably, the output of the model also includes an uncertainty confidence value, which is used to evaluate the reliability of the communication delay value. When the uncertainty confidence value is lower than a preset value, the edge node extends the window time or increases the conditions for triggering task migration.

[0012] Preferably, "the edge node selects another edge node as the migration target node based on the comprehensive load index, the channel state of the current communication link, and the priority of the task" specifically includes:

[0013] Assign a priority coefficient P to each task i ;

[0014] Calculate the overall score S for each task. i S i = β1·P i + β2·(1 - L n ) + β3·(1 - B u ); where β1, β2, and β3 represent priority coefficients P respectively. i Comprehensive load index L n Channel state parameter B u Weighting coefficients;

[0015] Based on the comprehensive score S i Determine the scheduling order of tasks and select migration targets.

[0016] Preferably, "transferring some tasks to the target node for execution" specifically includes a preset warning interval. When the edge node detects that the communication latency value has fallen into the warning interval based on the growth trend of the communication latency value, the target node pre-creates a copy of the task. Furthermore, the migration cost and the reliability of the target node are assessed before migration. The assessment of migration cost includes:

[0017] Preset migration cost threshold;

[0018] Evaluate the cost of transferring the task to the target node, including network transmission latency, state synchronization overhead, and energy consumption.

[0019] By comparing the cost with a migration cost threshold, when the cost exceeds the migration cost threshold, the edge node either delays migration or uses some key information.

[0020] The reliability of the target node during migration includes:

[0021] Dynamically assess the availability and security level of each edge node and generate node reliability coefficients;

[0022] Edge nodes with high reliability coefficients should be prioritized as migration target nodes.

[0023] Preferably, the edge node selects the optimal communication path in real time based on the channel status, and uses clustered data transmission to achieve adaptive bandwidth allocation and traffic balancing.

[0024] Preferably, the edge node divides adjacent vehicles into several dynamic clusters. When the density of the vehicle group is too high, each cluster selects the head vehicle for representative communication, thereby reducing redundant data transmission and reducing system bandwidth load.

[0025] Preferably, when a vehicle moves across the coverage area of ​​an adjacent edge node, the current edge node migrates the vehicle's task context, historical communication records, and priority information to the next edge node in real time.

[0026] Preferably, the edge node also acquires road environment parameters, including network channel strength, vehicle density, noise interference level, and weather conditions. The edge node dynamically adjusts the data transmission method and data transmission rate based on the environmental parameters.

[0027] Preferably, the method further includes:

[0028] The edge node makes local decisions based on vehicle operating status data, comprehensive load indicators, current communication link channel status, and task priority.

[0029] Local decisions are fed back to the vehicle terminal for real-time control.

[0030] Preferably, the edge node and the cloud server work together to execute the method;

[0031] The edge nodes upload model parameters or gradient information to the cloud server;

[0032] Receive model parameters from the cloud server, which have been globally aggregated and optimized through a federated learning mechanism.

[0033] Secondly, the present invention proposes a vehicle-to-everything (V2X) low-latency communication system based on edge computing, comprising a vehicle terminal, roadside edge nodes, and a cloud server;

[0034] The vehicle terminal is used to collect vehicle operating status data and send it to the roadside edge node;

[0035] The edge node is used to perform the method described in the first aspect;

[0036] The cloud server is used to aggregate model parameters from multiple roadside edge nodes through a federated learning mechanism, perform global model optimization, and distribute the updated model parameters to each edge node.

[0037] This invention identifies potential high-latency states in advance through latency prediction and then triggers migration to transfer tasks from high-load nodes to low-latency or resource-sufficient target nodes, achieving dynamic control and thus significantly reducing average response latency at the system level and improving service continuity. Attached Figure Description

[0038] Figure 1 This is a block diagram of a low-latency communication system for vehicle networking based on edge computing proposed in this invention.

[0039] Figure 2 This is a flowchart of a low-latency communication method for vehicle-to-everything (V2X) based on edge computing proposed in this invention. Detailed Implementation

[0040] Example 1: The present invention proposes a low-latency communication system for vehicle networking based on edge computing, which includes multiple vehicle terminals, multiple roadside edge nodes, and a cloud server. The vehicle terminals are wirelessly connected to the edge nodes, and the edge nodes are connected to the cloud server via wired or wireless networks.

[0041] The vehicle terminal is used to collect vehicle operating status data and send it to the roadside edge node.

[0042] Edge nodes are used to receive raw data sent by vehicle terminals, preprocess vehicle operation status data, extract features, dynamically schedule tasks and make localized decisions, and upload summary data or model parameters to the cloud server when necessary.

[0043] The cloud server is used to aggregate model parameters from multiple edge nodes through a federated learning mechanism, perform global model optimization, and periodically distribute global model parameters or optimization strategies to each edge node to achieve collaborative communication between the vehicle, edge, and cloud.

[0044] 1. Specifically, such as Figure 1 As shown, the vehicle terminal includes a data acquisition module, a communication module, an encryption module, and a control execution module.

[0045] The data acquisition module is used to collect dynamic information on vehicle speed, direction, acceleration, position, lane number, and surrounding vehicles.

[0046] The communication module is used for bidirectional data interaction with edge nodes via dual channels of C-V2X and 5G networks. During communication, when the quality of the primary channel degrades, the secondary channel automatically takes over data transmission to ensure communication continuity.

[0047] Specifically, the communication link between the edge nodes and the vehicle terminal supports a multi-path redundancy mechanism. The vehicle terminal can establish connections with multiple edge nodes simultaneously and dynamically switch the primary communication link based on channel quality; when the primary link is interrupted, the secondary link automatically takes over, ensuring the continuity and reliability of communication.

[0048] Specifically, the system supports cross-regional edge collaboration. When a vehicle moves across the coverage area of ​​an adjacent edge node, the current edge node will migrate the vehicle data (task context, priority information, historical communication records) to the next edge node to achieve seamless switching and business continuity.

[0049] Preferably, the communication data format between the edge node and the vehicle terminal conforms to the multi-protocol interoperability standard and can support communication methods such as C-V2X, 5G NR-V2X and WiFi 6 to adapt to the real-time vehicle-to-everything (V2X) communication needs under different network conditions.

[0050] The encryption module is used to encrypt the transmitted data using AES or ECC to ensure the security and integrity of the communication.

[0051] The control execution module is used to execute vehicle control behaviors based on the received edge decision instructions, including speed adjustment, emergency braking, or path planning.

[0052] 2. Specifically, the edge node includes a data preprocessing unit, a dynamic task scheduling unit, an intelligent decision-making unit, a caching unit, a network monitoring unit, an energy consumption management module, and a security authentication and anomaly detection module.

[0053] The data preprocessing unit is used to perform noise filtering, time alignment, outlier removal, and feature normalization on the received vehicle data to obtain standardized data features.

[0054] Among them, the intelligent decision-making unit uses a lightweight neural network model to predict the vehicle's operating status and assess risks, and calculates the priority of tasks and the urgency of communication tasks.

[0055] A latency prediction model is constructed based on a lightweight neural network model. This model can predict future latency trends based on historical communication data and channel state information.

[0056] Edge nodes calculate vehicle communication latency predictions and task priorities. When the predicted communication latency exceeds a preset latency threshold, a task migration mechanism is triggered, enabling collaborative scheduling among multiple edge nodes to transfer some computing tasks to adjacent reliable edge nodes for execution. This avoids edge node overload and ensures that the overall system latency does not exceed the target upper limit.

[0057] The cache unit is used to store short-term communication data and decision results locally in order to cope with network interruptions or signal instability.

[0058] The energy management module is used to monitor the power consumption and operating status of edge nodes in real time, and dynamically adjust task allocation according to the current load and power supply status of edge nodes. When the energy status is lower than the threshold, the system automatically reduces the processing frequency of low-priority tasks or migrates some tasks to adjacent edge nodes to achieve energy-saving operation and extend the operating life of edge nodes.

[0059] The network monitoring unit is used to monitor the quality of the communication channel and the load of the edge nodes in real time, and dynamically adjust the data transmission rate and the inference frequency of the algorithm model.

[0060] The task scheduling unit allocates resources based on task priority, channel occupancy, and node computational load using an adaptive scheduling algorithm based on multi-objective optimization, thereby achieving low latency and high reliability communication.

[0061] During task scheduling, a priority-based dynamic task scheduling algorithm is used.

[0062] The algorithm comprehensively considers vehicle task type, latency requirements, communication channel status, computational load, and energy consumption constraints, and achieves optimal resource allocation through heuristic scheduling strategies.

[0063] When the system detects a high-priority task (such as a collision warning or emergency braking), the edge node will prioritize allocating computing resources and bandwidth to the task and provide instruction feedback through a low-latency channel.

[0064] Preferably, the edge node also includes a backend management platform connected to a cloud server for visualized monitoring and remote management of vehicle group status, node load, communication latency, and system logs. The backend management platform provides a centralized control interface for real-time display of the load, communication latency, model performance, and abnormal events of each edge node, and supports remote policy distribution and log traceability.

[0065] The back-end management platform can connect with the intelligent traffic control center to realize the linkage of road traffic lights, traffic flow prediction, and priority control of emergency vehicles.

[0066] The security authentication and anomaly detection module supports multi-layered security protection mechanisms, including a communication encryption layer (AES, ECC), an authentication layer (digital signature and device identity authentication), and a data integrity verification layer.

[0067] Specifically, after receiving vehicle operation status data, the edge nodes perform digital signature verification, device authentication, and anomaly detection to prevent forged instructions or malicious attacks, and to ensure the security and stability of the system operation.

[0068] The security authentication and anomaly detection module is also used to detect network congestion, node overload, data loss, and abnormal latency events in real time. When an anomaly is detected, the edge node immediately logs the event and reports it to the cloud server through the backend management platform. At the same time, the fault recovery mechanism is activated, and task resources are reallocated.

[0069] Preferably, redundant computing units are configured in the edge nodes. When the main processor fails or is overloaded, it automatically switches to the backup computing core to continue executing tasks, ensuring continuous system operation.

[0070] Preferably, the edge node also has a clock synchronization mechanism. The vehicle terminal, edge node, and cloud server maintain time consistency between the vehicle terminal and the edge node through Network Time Protocol (NTP) or GPS timing to ensure accurate data timing, avoid decision delays or task conflicts, and improve the response efficiency of collaborative control.

[0071] Preferably, the edge node also includes a self-learning module for automatically adjusting scheduling strategy parameters based on historical running records. This module uses a reinforcement learning algorithm to dynamically optimize scheduling weights based on feedback from task execution latency, energy consumption, and resource utilization, thereby improving long-term performance stability.

[0072] Preferably, the edge node also includes an environmental monitoring module, which is used to acquire road environmental parameters, including network channel strength, vehicle density, noise interference level and weather conditions. The edge node dynamically adjusts the signal modulation method and data transmission rate according to the environmental parameters to further reduce latency and improve communication reliability.

[0073] Preferably, edge nodes can combine vehicle positioning information with high-precision map data to optimize vehicle route selection and traffic light coordination control. Edge nodes predict future road congestion levels based on real-time traffic conditions and issue route adjustment instructions in advance, achieving vehicle-road cooperative optimization.

[0074] 3. Specifically, the cloud server employs a federated learning mechanism to achieve distributed model updates; edge nodes only upload model parameters or gradient information, not the original data, to protect user privacy. After aggregating updated parameters from multiple nodes, the cloud server performs global model aggregation and periodically distributes optimized models to each edge node, improving the overall prediction accuracy and adaptive capability of the system.

[0075] Example 2, as Figure 2 As shown, a low-latency communication method for vehicle-to-everything (V2X) communication implemented through the system of Embodiment 1 is mainly executed by roadside edge nodes, and the method includes:

[0076] Edge nodes acquire historical operational data, as well as real-time status parameters for evaluating network communication links and their own performance.

[0077] Edge nodes predict and generate communication latency values ​​based on historical operating data and real-time status parameters, and monitor edge node resources in real time to generate comprehensive load indicators;

[0078] When the communication latency reaches the preset latency threshold or the comprehensive load index shows that the edge node is overloaded, the edge node is triggered to perform task migration.

[0079] Based on comprehensive load metrics, the current communication link channel status, and task priorities, the edge node selects another edge node as the migration target node and transfers some tasks to the migration target node for execution.

[0080] It should be noted that "edge nodes acquiring historical operational data and real-time evaluating the status parameters of network communication links and the edge nodes' own performance" specifically includes:

[0081] On the one hand, the vehicle terminal collects vehicle operating status data at a fixed period Δt. :

[0082]

[0083] in, This represents the speed of the vehicle at time t. This represents the acceleration of the vehicle at time t. This represents the vehicle's heading angle at time t. , This represents the vehicle's position coordinates at time t (e.g., latitude and longitude or projected plane coordinates). This represents the timestamp used to record vehicle operating status information.

[0084] Vehicle operation status data is encrypted by the vehicle terminal using AES-256 or Elliptic Curve Cryptography (ECC) and then uploaded to the nearest roadside edge node via the communication module.

[0085] The edge nodes preprocess the received operational status data from multiple vehicles to obtain standardized data features. The preprocessing methods specifically include performing time-series alignment and data cleaning operations.

[0086] Preferably, time-series alignment employs a sliding time window method to unify the operational status data of different vehicles under the same time reference. Data cleaning includes outlier removal, noise filtering, and normalization. Preferably, noise filtering uses a combination of weighted moving average and Kalman filtering to effectively reduce the impact of abrupt changes and external interference. Since the preprocessing method is existing technology, those skilled in the art can implement it directly, and it will not be described in detail here.

[0087] Edge nodes extract features from the preprocessed vehicle operating status data to generate a vehicle operating status feature vector. Specifically, by calculating key indicators such as speed change rate, inter-vehicle distance density, channel signal strength (RSSI), and signal-to-noise ratio (SNR), the vehicle operating status feature vector is generated for subsequent calculations and task allocation. The formula for the vehicle operating status feature vector at time t is as follows:

[0088]

[0089] in, This represents the signal-to-noise ratio at time t; This represents the packet loss rate at time t; This represents the bandwidth utilization rate at time t; This represents the overall load index of the edge node at time t; This represents the CPU utilization rate at time t; This represents the vehicle speed at time t; This represents the vehicle's acceleration at time t; This represents the vehicle's heading angle at time t; This represents the vehicle's position coordinates at time t.

[0090] On the other hand, edge node detection obtains real-time status parameters of node performance to obtain comprehensive load indicators. .

[0091] The edge nodes utilize an internal resource monitoring unit for real-time detection and calculation. This unit, a fundamental system support module, periodically samples and evaluates the nodes' computational and communication resource usage. Specifically, the resource monitoring unit collects multiple operational parameters in real-time within preset time intervals (100ms–500ms), including CPU utilization, GPU or AI accelerator utilization, system memory and cache usage, current task queue length, and network bandwidth utilization. It then uses a weighted algorithm to comprehensively calculate the collected data, generating a comprehensive load index for the node. .

[0092]

[0093] in, , These represent the current CPU utilization and total number of cores at the edge node, respectively. , These represent the memory usage and total capacity of the edge nodes, respectively. , These represent the bandwidth usage and maximum bandwidth of the edge node, respectively. , These represent the current length and maximum task capacity of the task queue at the edge node, respectively. For adjustable weights (e.g., 0.4, 0.3, 0.2, 0.1), satisfying... .

[0094] Comprehensive load index The calculation period is (e.g., 100ms), updated in real time. The closer it is to 1, the higher the load.

[0095] The weighting coefficients of each indicator can be adaptively adjusted based on the node's computing power, task type, and energy consumption status to reflect the overall resource utilization of the node. The dynamic task scheduling module reads the calculated comprehensive load indicators in real time. It serves as an important input parameter for task migration and resource allocation decisions.

[0096] It should be noted that "predicting and generating communication latency values ​​based on historical operating data and real-time status parameters" specifically includes:

[0097] Edge nodes input historical communication delay sequences obtained from the historical operational data, and comprehensive load metrics. The system inputs real-time state parameters, such as vehicle operating status feature vectors, into a built-in Long Short-Term Memory (LSTM) neural network model, which then predicts the output communication latency value.

[0098] The Long Short-Term Memory (LSTM) neural network model takes the following form:

[0099]

[0100] in, This represents the predicted communication delay value for the next moment. This represents the prediction function of the LSTM model; The feature vector representing the vehicle's operating state at time t; This indicates an adaptive time window that dynamically adjusts according to changes in vehicle speed and communication link.

[0101] In the core formula of the LSTM model, this application adds a vehicle dynamic feature channel and a load channel to the weight matrices of the input gate, forget gate, and output gate. The formula is simplified and unified as follows:

[0102]

[0103] Finally, the predicted delay is output:

[0104]

[0105] in, Indicates the input gate; Indicates the Gate of Oblivion; Indicates the output gate; Indicates the cell state; Indicates the output of the hidden layer; , , , Indicates that it applies to the current input. The weights; , , , Indicates action on The weights; , , , For bias terms; , These represent the weights and biases of the output layer, respectively.

[0106] The LSTM model in this application has been extended in terms of input feature design. In addition to using historical communication delay sequences as input, it also introduces multi-dimensional real-time state parameters such as channel state, node load, vehicle dynamics, and CPU utilization. This enables the algorithm to better adapt to the high dynamics and real-time requirements of edge computing environments, thereby improving the sensitivity and generalization ability of the prediction results to complex network states.

[0107] In this application, the Long Short-Term Memory Neural Network model adopts an adaptive time window mechanism, which only saves data that has arrived in the most recent period and continuously replaces old data with new data. That is, the model adjusts the input time step according to the changes in vehicle speed and the fluctuations in the communication link, so as to avoid the prediction lag problem of traditional fixed window models in high-speed moving scenarios.

[0108] Preferably, the model output also includes an uncertainty confidence value to evaluate the reliability of the communication latency value. When the uncertainty confidence value is lower than a preset value, the edge node expands the adaptive time window. Alternatively, the conditions for triggering task migration can be improved to reduce the probability of false triggering. Improving these conditions includes raising the communication latency threshold, increasing the overall load index threshold, adding a multi-condition joint triggering mechanism, extending the duration of anomalies, and smoothing out short-term anomalies. These measures can suppress false triggering behavior and improve the stability of task migration when the model's prediction confidence is insufficient.

[0109] It should be noted that the edge nodes use a dynamic task scheduling module to schedule tasks based on comprehensive load metrics. The system comprehensively determines whether to perform task migration and selects the target node based on the current communication link channel status and task priority, thereby achieving dynamic balancing and global optimization of computing resources.

[0110] Specifically, this application classifies tasks based on their functional attributes and real-time requirements in the vehicle-to-everything (V2X) communication and computing process. Different types of tasks have different latency tolerances and resource requirements. The main task types include the following: 1. Safety control tasks, such as collision warning, emergency braking, and lane keeping, requiring millisecond-level response and having the highest priority; 2. Cooperative control tasks, such as platooning control and vehicle-road cooperative scheduling, with high real-time requirements; 3. Perception and computing tasks, such as target recognition, environmental perception, and path planning, tolerating sub-hundred-millisecond latency; 4. Information interaction and uploading tasks, such as sensor data uploading and log recording, with relatively low real-time requirements; 5. Infotainment tasks, such as map updates and media services, with the lowest priority.

[0111] Specifically, in each scheduling cycle, the dynamic task scheduling module assigns a priority coefficient to each task. Task priority is determined by the dynamic task scheduling module based on a comprehensive calculation of multiple dimensions, including task attributes, real-time requirements, energy consumption characteristics, and urgency. Priority coefficient. The calculation formula is as follows:

[0112]

[0113] in, , , These are adjustable weighting coefficients; The urgency level of a task is indicated by system-preset weights or calculated from event trigger levels, and is limited by a logical normalization function. The value ranges from 0 to 1, with real-time tasks automatically receiving a higher urgency weight. The calculation formula is as follows:

[0114]

[0115] in, This indicates the priority of task categories (defined by the task type table, for example, security control tasks = 1.0, collaborative control tasks = 0.8, perception and computing tasks = 0.6, information interaction and uploading tasks = 0.4, infotainment tasks = 0.2). Indicates the urgency level of the event that triggers the task (e.g., collision warning = 1, normal communication = 0.2). , To adjust the parameters (can be 0.6 or 0.4).

[0116] in, This indicates the task's sensitivity to communication latency; the smaller the communication latency value, the lower the sensitivity. The larger it is. The calculation formula is as follows:

[0117]

[0118] in, This indicates the maximum allowed communication latency for the task (in milliseconds), defined by the task type or service level. This represents the upper limit of the global maximum communication latency set by the system, for example... To avoid excessively large values, the calculation was adjusted accordingly. Normalization is performed:

[0119]

[0120] Make For example, collision warning tasks. , ; Infotainment tasks , .

[0121] in, This represents energy sensitivity, reflecting the impact of task execution on node energy consumption, and its value ranges from 0 to 1. Specifically, it can be calculated based on the average power required for task execution and the node's remaining energy.

[0122]

[0123] in, This represents the average power consumption during task execution, collected by the energy management module of the edge node; Indicates the maximum available power of the edge node; This represents the current remaining energy of the edge node; This represents the total power consumption of the edge nodes.

[0124] When the remaining energy of the edge node is low or the power consumption of the task is high, A larger value indicates that the task is more sensitive to energy consumption. After calculation, normalization is performed: To ensure .

[0125] The dynamic task scheduling module then calculates the overall score for each task. , in, , , These represent priority coefficients. Comprehensive load index Channel state parameters The weighting coefficients are determined by introducing an adaptive weight optimization mechanism at the scheduling algorithm level. This mechanism can adjust the weighting coefficients of the priority coefficient, comprehensive load index, and channel state parameter online based on the current network load distribution and task type ratio. , , This is to achieve a dynamic balance in scheduling decisions.

[0126] Channel state parameters The calculation formula is as follows:

[0127]

[0128] Wherein, SNR represents the signal-to-noise ratio; PLR represents the packet loss rate; and RTT represents the round-trip time. , , This represents the normalized weight. A higher value indicates a worse communication link condition. The vehicle terminal's communication module (such as C-V2X or DSRC) collects communication parameters in real time, updates them once per second, and uploads them to the edge node.

[0129] Edge nodes are based on a comprehensive score Determine the scheduling order of tasks and the selection of migration targets, and select tasks with high comprehensive scores as priority processing objects.

[0130] In a preferred embodiment, transferring some tasks to the target node for execution specifically includes:

[0131] Preset migration cost threshold;

[0132] Evaluate the cost of transferring the task to the target node, including network transmission latency, state synchronization overhead, and energy consumption.

[0133] By comparing the cost with the migration cost threshold, when the cost exceeds the migration cost threshold, the edge node will either delay migration or use partial state incremental transmission.

[0134] The edge nodes utilize a power management module to monitor their power consumption and operational status in real time. This module collects CPU power consumption, temperature, and power supply voltage data in real time and calculates the state of energy. :

[0135]

[0136] in, This represents the remaining available energy of the edge node at time t. This indicates the maximum available energy of the edge node under full power or rated power conditions; Indicates the edge node in time The actual power consumption function, including the energy consumption of CPU, AI accelerator and communication module; This represents the cumulative energy consumption from system startup to the current moment.

[0137] When performing task migration, this method comprehensively considers network transmission delay, state synchronization overhead and energy consumption. When the migration cost exceeds the set threshold, the scheduling module will delay the migration or use partial state incremental transmission to avoid system oscillation.

[0138] The scheduling module adopts a lightweight migration strategy of partial state increment transmission. In order to avoid system oscillation or performance degradation, it does not immediately perform a complete task migration, but only transmits some key information (i.e., "state increment") to rebuild or partially synchronize the task environment on the target node.

[0139] Specifically, when using partial state incremental transmission, the migration content includes task context, task priority information, and historical communication records, and the migration time does not exceed 500ms.

[0140] After the migration is complete, the original edge node enters a listening state to ensure task continuity and execution integrity.

[0141] Existing technologies also suffer from low system reliability; once cloud communication is interrupted or delayed, the inter-vehicle collaborative function is limited. In a preferred embodiment, this method addresses this issue:

[0142] When a vehicle moves across the coverage area of ​​an adjacent edge node, the current edge node uses the Fast Context Transfer Protocol (FCP) to migrate the vehicle's historical communication records, task context, and task priority information to the next edge node in real time, achieving seamless switching. Furthermore, task execution remains uninterrupted during the migration process, ensuring the continuity of vehicle control commands.

[0143] To address the issue of low system reliability, in a preferred embodiment, a pre-defined warning interval is established. When an edge node detects that the communication latency value has fallen into the warning interval based on the increasing trend of the communication latency value, a copy of the task is pre-created on the target node. That is, when the predicted communication latency value repeatedly falls into the warning interval and shows a rapid upward trend, the system triggers the pre-replication mechanism. Through this incremental and pre-replication migration mechanism, when the latency prediction module detects that the latency is about to exceed the threshold, the system pre-creates a task copy on the target node, achieving smooth task migration. This significantly reduces the interruption time caused by task migration and ensures that the task remains uninterrupted even if the main edge node processing the task experiences an anomaly.

[0144] In a preferred embodiment, to further improve communication reliability, the method further includes:

[0145] The availability and security level of each edge node are dynamically evaluated to generate a node reliability coefficient. When a node experiences performance fluctuations or security risks, the scheduling module will automatically reduce its task allocation weight.

[0146] Prioritize edge nodes with high reliability as migration target nodes to improve the overall system stability and security.

[0147] Specifically, by comprehensively evaluating multi-dimensional operating parameters such as edge node performance stability, communication link stability, energy consumption status, security level, and historical task completion rate, these indicators are weighted and fused according to system policies to form a comprehensive score reflecting node availability and security level, which is the node reliability coefficient. When the node reliability coefficient decreases, the dynamic task scheduling module automatically reduces the task allocation weight of that node and prioritizes edge nodes with high reliability coefficients as migration target nodes during task migration, thereby improving the overall communication reliability and task continuity of the system.

[0148] Preferably, the edge node also acquires road environment parameters, including network channel strength, vehicle density, noise interference level and weather conditions. The edge node dynamically adjusts the data transmission method and data transmission rate based on the environmental parameters.

[0149] Preferably, existing technologies suffer from severe bandwidth consumption: multiple vehicles simultaneously uploading large amounts of data can easily cause communication congestion. To address this issue, edge nodes select the optimal communication path in real time based on channel conditions, employing clustered data transmission to achieve adaptive bandwidth allocation and traffic balancing, thereby avoiding link congestion.

[0150] For example, in high-interference environments, the system automatically enables robust modulation methods (such as QPSK) and reduces bandwidth usage; when channel conditions are good, it switches to high-throughput mode (such as 16QAM) to improve the overall communication rate.

[0151] Edge nodes divide adjacent vehicles into several dynamic clusters. When the density of the vehicle group is too high, each cluster selects the head vehicle for representative communication, thereby reducing redundant data transmission and reducing system bandwidth load.

[0152] In the application of vehicle-to-everything (V2X) communication, this method extends the input features of the LSTM latency prediction model and introduces an adaptive time window mechanism and an uncertainty confidence interval evaluation mechanism. In terms of scheduling algorithms, this method introduces an adaptive weight optimization mechanism and an incremental and pre-copy migration mechanism. Through multiple improvements and optimizations, this method outperforms traditional algorithms in terms of latency prediction accuracy, task scheduling stability, and migration continuity, realizing a highly reliable, low-latency edge intelligent computing architecture for V2X communication.

[0153] In a preferred embodiment, to address the problem of severe bandwidth consumption in the prior art, the edge node and cloud server collaboratively execute the method;

[0154] Each edge node uploads model parameters or gradient information and sends it to the cloud server;

[0155] The cloud server uses a federated learning mechanism to globally aggregate and optimize model parameters, and then distributes the optimized model parameters to edge nodes.

[0156] Federated learning enables each edge node to complete model updates locally, uploading only model parameters or gradient information, thus fundamentally reducing the amount of raw data transmission.

[0157] Specifically, each edge node trains the model independently, uploading only the parameter increments. Instead of uploading the original data.

[0158] The cloud server uses a weighted average algorithm to aggregate global model parameters:

[0159]

[0160] in, This represents the aggregated global model parameters; This indicates the number of edge nodes participating in federated learning; This represents the model parameters after local training at the i-th edge node; This represents the parameter increment of the i-th edge node in the most recent local update relative to the previous training round; This represents the number of data samples used in training by the i-th edge node; , representing the total number of data samples participating in all edge nodes; This is a weighting term used to ensure that edge nodes with larger sample sizes contribute more to the global model, thereby improving the generalization performance of the aggregated model.

[0161] Preferably, the edge node also stores short-term communication data and decision results locally through a caching unit. When the network is temporarily interrupted, the node can continue to provide local decision-making services based on the cached data, and then synchronize it to the cloud after the network is restored, so as to ensure the continuity of communication.

[0162] Preferably, edge nodes make local decisions and provide real-time feedback through an intelligent decision-making unit. The intelligent decision-making unit deploys a lightweight neural network model (EdgeNet), where the input to EdgeNet is a feature vector. The output is a control command. :

[0163]

[0164] in, This represents the local control command output vector of the i-th vehicle; This represents a lightweight neural network model (EdgeNet) deployed on edge nodes.

[0165] , Let represent the state vector of the i-th vehicle at time t. , Let represent the speed of the i-th vehicle at time t. Let represent the acceleration of the i-th vehicle at time t. This represents the direction angle of the i-th vehicle at time t. This represents the position coordinates of the i-th vehicle at time t. Let be the channel state parameters of the edge node at time t. This represents the overall load index of the edge node at time t. This is the priority coefficient.

[0166] This represents the vehicle speed adjustment amount (recommended acceleration / deceleration magnitude) calculated using EdgeNet. This represents the adjustment amount of the vehicle steering angle calculated using EdgeNet.

[0167] EdgeNet employs a structure combining convolutional neural networks (CNNs) and recurrent neural networks (LSTMs): CNN layers extract spatial features, LSTM layers capture temporal dependencies, and finally, a softmax classification layer generates the optimal control decision. The decision result is fed back to the vehicle terminal via a wireless link to achieve closed-loop control.

[0168] Preferably, the edge node also implements identity verification and data integrity checks between the vehicle and the edge node through a security authentication and anomaly detection module. The authentication process adopts a certificate-based two-way verification mechanism, combined with the HMAC-SHA256 algorithm to verify the integrity of the communication content. When unauthorized access or forged data packets are detected, the system immediately triggers isolation and alarm mechanisms to prevent the spread of network attacks.

[0169] In one specific embodiment, the vehicle-to-everything (V2X) communication implementation process includes:

[0170] The vehicle terminal first initiates a self-test process to ensure that all internal modules are functioning properly.

[0171] The vehicle terminal then collects the vehicle's operating status and uploads the data to the edge node;

[0172] After receiving the data, the edge nodes perform data preprocessing, feature extraction, and delay prediction.

[0173] The computation order is determined based on priority, and the EdgeNet model is called for local decision-making.

[0174] The decision results are fed back to the vehicle via a wireless channel, enabling real-time control.

[0175] The cloud server periodically collects model parameters from edge nodes, performs federated learning aggregation and policy optimization, and forms a global intelligent closed loop.

[0176] This method achieves distributed intelligent computing and communication optimization through a three-layer collaborative architecture of "vehicle-edge-cloud," effectively reducing vehicle-to-everything (V2X) communication latency and improving decision-making response speed and system reliability. Simultaneously, it employs latency prediction and task migration mechanisms to maintain stable communication even in complex and dynamic traffic environments; combined with energy consumption sensing and security authentication mechanisms, it ensures continuous and efficient system operation under various working conditions.

[0177] In terms of system innovation, this invention is the first to modularly integrate a latency prediction module, a dynamic task scheduling module, an intelligent decision-making module, and an energy management module at the roadside edge node, forming a closed-loop system of "prediction-decision-execution-feedback". Unlike traditional passive task redistribution mechanisms, this invention drives task migration through latency prediction, achieving proactive and perceptive resource scheduling. Tasks can be migrated in advance when communication latency is about to exceed a threshold, greatly reducing latency spikes. Simultaneously, the system establishes a federated learning and policy feedback channel between the cloud, edge, and endpoint layers, enabling the edge model to achieve self-learning and optimization without uploading raw data, ensuring data security while improving global intelligence.

[0178] In terms of algorithmic innovation, this invention proposes a multi-factor adaptive scheduling algorithm that incorporates multi-dimensional parameters such as task priority, node load, and channel state into a unified comprehensive scoring system. The algorithm can dynamically adjust weights based on the current network state, achieving self-optimization of the scheduling strategy. Furthermore, the migration mechanism incorporates incremental synchronization and replica pre-copying techniques. Before task migration, the system pre-creates lightweight replicas of the tasks and performs differential state updates, achieving "zero-interruption switching." Simultaneously, migration cost constraints are considered in the migration decision; when the migration cost exceeds a threshold, delayed migration or partial synchronization is automatically selected to balance latency and stability.

[0179] Finally, regarding collaborative optimization and security mechanism innovation, the collaborative work of the energy consumption awareness module and the security authentication module enables the system to consider not only latency and load during resource scheduling, but also energy consumption and node trustworthiness, thereby achieving a unified balance between computing performance, communication efficiency, and security. Furthermore, this invention introduces a lightweight model compression and dynamic quantization mechanism, enabling edge nodes to significantly reduce computational and energy consumption overhead while maintaining inference accuracy.

[0180] In this invention, the task migration mechanism primarily addresses latency issues; federated learning and multipath communication mechanisms reduce bandwidth consumption; and redundant computation and security authentication mechanisms enhance system reliability. These three mechanisms work together to form the complete optimization system of this invention, enabling the system to simultaneously achieve low latency, low bandwidth consumption, and high reliability communication goals in complex vehicular network environments.

[0181] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing, characterized in that, Primarily executed by the edge nodes on the roadside, the method includes: Edge nodes acquire real-time and historical vehicle operation data, as well as real-time evaluation parameters of network communication links and edge node performance, and generate vehicle operation status feature vectors and comprehensive load indicators of edge nodes themselves. The edge node is based on a long short-term memory neural network model and predicts and generates communication latency values ​​by inputting the vehicle operating state feature vector; When the communication latency exceeds a preset latency threshold or the comprehensive load index shows that the edge node is overloaded, the edge node is triggered to perform task migration. Based on the comprehensive load index, the channel status of the current communication link, and the priority of the task, the edge node selects another edge node as the migration target node and transfers some tasks to the migration target node for execution. The long short-term memory neural network model adopts an adaptive time window mechanism to adjust the input time step according to changes in vehicle speed and fluctuations in the communication link. The output of the model also includes an uncertainty confidence value, which is used to evaluate the reliability of the communication delay value. When the uncertainty confidence value is lower than a preset value, the edge node extends the window time or increases the conditions for triggering task migration. The transfer of some tasks to the target node for execution specifically includes a preset warning interval. When the edge node detects that the communication latency value has fallen into the warning interval based on the growth trend of the communication latency value, the target node pre-creates a copy of the task.

2. The low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing according to claim 1, characterized in that, The formula for the vehicle operating state feature vector is: in, express The vehicle's operational state feature vector at any given time. express Signal-to-noise ratio at any given moment express Packet loss rate at any given time express Bandwidth utilization at any given time Represents edge nodes Comprehensive load indicators at any given time express CPU utilization at any given moment express Vehicle speed at any given moment Indicates vehicle acceleration at any moment Indicates vehicle The direction angle at that moment, Indicates vehicle The position coordinates at that moment.

3. The low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing according to claim 1, characterized in that, The selection of another edge node as the migration target node based on the comprehensive load index, the current channel state of the communication link, and the priority of the task specifically includes: Assign a priority coefficient to each task ; Calculate the overall score for each task. ;in, These represent priority coefficients. Comprehensive load index Channel state parameters Weighting coefficients; Based on the comprehensive score Determine the scheduling order of tasks and select migration targets.

4. The low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing according to claim 1, characterized in that, This also includes assessing migration costs and the reliability of the target node before migration, whereby the assessment of migration costs includes: Preset migration cost threshold; Evaluate the cost of transferring the task to the target node, including network transmission latency, state synchronization overhead, and energy consumption. By comparing the cost with the migration cost threshold, when the cost exceeds the migration cost threshold, the edge node delays migration or transmits some critical information; The reliability of the target node during migration includes: Dynamically assess the availability and security level of each edge node and generate node reliability coefficients; Edge nodes with high reliability coefficients should be prioritized as migration target nodes.

5. The low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing according to claim 1, characterized in that, When a vehicle moves across the coverage area of ​​an adjacent edge node, the current edge node migrates the vehicle's task context, historical communication records, and priority information to the next edge node in real time.

6. The low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing according to claim 1, characterized in that, The edge node divides adjacent vehicles into several dynamic clusters. When the density of the vehicle group is too high, the head vehicle of each cluster is selected for representative communication.

7. The low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing according to claim 1, characterized in that, Also includes: The edge node makes local decisions based on vehicle operating status data, comprehensive load indicators, current communication link channel status, and task priority. Local decisions are fed back to the vehicle terminal for real-time control.

8. A low-latency communication method for vehicle-to-everything (V2X) networks based on edge computing according to claim 7, characterized in that, The edge node and the cloud server work together to execute the method; The edge nodes upload model parameters or gradient information to the cloud server; Receive model parameters from the cloud server, which have been globally aggregated and optimized through a federated learning mechanism.

9. A low-latency communication system for vehicle-to-everything (V2X) based on edge computing, characterized in that, This includes vehicle terminals, roadside edge nodes, and cloud servers; The vehicle terminal is used to collect vehicle operating status data and send it to the roadside edge node; The edge node is used to perform the method as described in any one of claims 1 to 8; The cloud server is used to aggregate model parameters from multiple roadside edge nodes through a federated learning mechanism, perform global model optimization, and distribute the updated model parameters to each edge node.

Images