Internet of vehicles multi-domain intelligent optimization method and system under computing network fusion architecture

Abstract

The invention relates to an Internet of Vehicles multi-domain intelligent optimization method and system under a computing network convergence architecture, and the method comprises the steps: receiving a task request sent by an edge layer road side unit based on a pre-configured cloud edge end cooperation architecture, optimizing the user service time delay and network load balance through a differential evolution algorithm, and obtaining a network load balance optimization result. Generating an original intelligent model distribution path and distributing the pre-trained original intelligent model to an edge layer road side unit; receiving an original intelligent model and vehicle mobility data, optimizing vehicle mobility, model quality and alliance establishment cost through an alliance game algorithm, and constructing a vehicle fine-tuning alliance structure suitable for federal learning; on the basis of a vehicle fine tuning alliance structure, local model updating data and trust evaluation data of vehicles in an alliance are obtained, dynamic weighted federated learning is executed through multi-dimensional trust evaluation, security is enhanced through differential privacy and a block chain tracking mechanism, and a high-precision global intelligent model is generated and deployed to the vehicles. Reliable support is provided for automatic driving, V2X communication and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of computing-network fusion, and in particular to a multi-domain intelligent optimization method and system for an Internet of Vehicles (IoV) under a computing-network fusion architecture. Background Art

[0002] As a key branch of the Internet of Things (IoT), the Internet of Vehicles (IoV) promotes the development of intelligent transportation systems through the collaboration of onboard mobile networks, intra-vehicle networks, and inter-vehicle networks. With the widespread adoption of 5G technology and intelligent applications, the IoV plays a key role in tasks such as autonomous driving, high-precision map updates, real-time traffic optimization, and safety monitoring. However, these complex tasks place increasing demands on computing resources and real-time performance, leading to processing delays and resource bottlenecks on edge devices due to limited computing power. The Vehicular Edge Computing (VEC) architecture achieves vertical integration of cloud, edge, and devices by offloading data processing from the cloud to the edge, reducing transmission latency and improving user Quality of Service (QoS). Despite this, the VEC architecture has limitations in computational isolation, making horizontal collaboration between edge nodes difficult, resulting in "islands of computing power" that limit resource utilization and system performance.

[0003] Furthermore, the surge in compute-intensive and latency-sensitive applications in the Internet of Vehicles (IoV) has exacerbated computing bottlenecks, leading to increased service latency and a degraded user experience. Traditional approaches to address resource constraints often rely on hardware expansion, but this high cost is significantly in conflict with the benefits of improved QoS. Furthermore, the IoV system's heavy reliance on sensor data increases security and privacy risks, such as malicious attacks and data leaks, which threaten user privacy and traffic system stability. Existing edge computing research has proposed multi-objective optimization models that combine latency, energy consumption, load balancing, and privacy protection to optimize task allocation through improved algorithms. Furthermore, to address the network dynamics caused by high vehicle mobility, some solutions have introduced privacy-aware content distribution and incentive mechanisms. However, these solutions often focus solely on intelligent distribution, model fine-tuning, or security and privacy, lacking a comprehensive optimization framework, making it difficult to achieve a balance between resource coordination, performance improvement, and security assurance.

[0004] To address these challenges, computing networks, as the foundation of the digital economy, improve resource utilization and QoS, while reducing latency and costs through efficient collaboration between the cloud, network, and edge. An optimized IoV architecture based on computing networks is urgently needed to achieve comprehensive optimization for intelligent task distribution, model fine-tuning, and security and privacy, overcome computational isolation and resource bottlenecks, and improve system performance and user experience. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-domain intelligent optimization method and system for the Internet of Vehicles under a computing-network fusion architecture, so as to solve the technical problems of insufficient horizontal collaboration, low resource utilization, and separation of intelligent distribution, model fine-tuning, and security and privacy optimization caused by computing isolation in the existing vehicle edge computing technology.

[0006] To achieve one of the above-mentioned objectives, an embodiment of the present invention provides a multi-domain intelligent optimization method for an Internet of Vehicles (IoV) under a computing-network fusion architecture, the method comprising:

[0007] Receive task requests from edge layer roadside units, optimize user service latency and network load balancing through differential evolution algorithms, generate original intelligent model distribution paths, and distribute pre-trained original intelligent models to edge layer roadside units;

[0008] Receive the original intelligent model and vehicle mobility data, optimize vehicle mobility, model quality and alliance formation cost through the alliance game algorithm, and build a vehicle fine-tuning alliance structure suitable for federated learning;

[0009] Obtain local model update data and trust assessment data of vehicles in the vehicle fine-tuning alliance structure, perform dynamic weighted federated learning through multi-dimensional trust assessment, enhance security through differential privacy and blockchain tracking mechanisms, generate high-precision global intelligent models, and deploy them to vehicles.

[0010] As a further improvement of an embodiment of the present invention, the method further includes that the construction of the preconfigured cloud-edge-device collaborative architecture includes:

[0011] Obtaining the requirements of the Internet of Vehicles system, building a three-layer cloud-edge-end collaborative architecture, and having the edge layer roadside unit transmit task requests to the cloud layer macro base station to obtain defined architecture parameters;

[0012] The cloud-edge-device three-layer collaborative architecture includes:

[0013] A cloud-layer MBS configured to perform global task scheduling, model caching, and cache strategy optimization, and wherein the pre-trained original intelligent model is stored in the MBS;

[0014] Edge layer RSU, configured to perform signal relay, local data aggregation, and collaborative computing with vehicles;

[0015] The terminal-layer vehicle is configured to share computing tasks through on-board computing resources and collaboratively optimize the original intelligent model training.

[0016] As a further improvement of an embodiment of the present invention, the method further includes that optimizing user service delay and network load balancing by using a differential evolution algorithm includes:

[0017] Receive the task request of the RSU and the communication data generated based on the cloud-edge architecture, and define service delay as a service quality indicator; the service delay is composed of transmission delay and propagation delay; the RSU obtains the delay of the original intelligent model and iteratively calculates the delay of each intermediate node in the path set to generate service delay evaluation data;

[0018] Based on the communication data, defining the network load balance as a cost indicator, expressed as a standard deviation of link utilization, and generating a load balance evaluation result;

[0019] A differential evolution algorithm that integrates multiple mutation strategies is used, and the difference between the user service delay and the network load balance is used as the fitness function to optimize the original intelligent model distribution path, specifically including:

[0020] Initialize the population and randomly generate a real-number encoding vector representing the distribution strategy;

[0021] Perform mutation operations and generate mutation vectors. Mutation strategies include:

[0022] The first mutation strategy means randomly selecting three individuals to generate mutation vectors. The formula is:

[0023] ;

[0024] in, is the mutation vector generated based on the first mutation strategy, 、 、 are three different individuals randomly selected from the population, is the variation factor;

[0025] The second mutation strategy is to generate mutation vectors based on the optimal individual. The formula is:

[0026] ;

[0027] in, is the mutation vector generated based on the second mutation strategy, is the current optimal individual, 、 are two different individuals randomly selected from the population;

[0028] The third mutation strategy is to combine the current individual with the optimal individual to generate a mutation vector. The formula is:

[0029] ;

[0030] in, is the mutation vector generated based on the third mutation strategy, For the current individual;

[0031] Perform a crossover operation to mix the mutation vector with the target individual through the crossover rate to generate the test individual. The formula is:

[0032] ;

[0033] in, For the test individuals, is the crossover rate, is a randomly selected dimension index;

[0034] Perform a selection operation, compare the fitness of the test individual and the target individual, and retain the individual with better fitness; the fitness is the difference between the service delay and the load balance;

[0035] Repeating the mutation operation, the crossover operation, and the selection operation until a termination condition is satisfied, wherein the termination condition includes a maximum number of iterations or a convergence criterion;

[0036] Boundary checks are performed on the generated individuals to generate the optimal distribution path of the original intelligent model.

[0037] As a further improvement of an embodiment of the present invention, the method further includes that optimizing vehicle mobility, model quality and alliance formation cost through the alliance game algorithm includes:

[0038] The optimization of vehicle mobility, model quality and alliance formation cost through the alliance game algorithm includes:

[0039] Receiving vehicle mobility data and training data information, defining the model quality as a user service quality indicator to update the estimated accuracy representation of the global model; the estimated accuracy is a concave increasing function based on the training data quality, and generating a model quality evaluation result;

[0040] Based on the number of vehicles in the alliance, the alliance formation cost is defined as the cost expenditure indicator, expressed as a linear function of the number of vehicles, and the alliance cost evaluation result is generated;

[0041] Based on the vehicle mobility model and the alliance formation permission set, the alliance structure is optimized through the alliance game algorithm, including:

[0042] Configuring a set of participants and a value function that evaluates the utility of the alliance based on model quality;

[0043] Based on the interests of the vehicles and the value function, a mutually beneficial alliance structure is formed through negotiation;

[0044] Use the Shapley value method to distribute the benefits of alliance members;

[0045] Verify alliance stability and ensure that no vehicle has the motivation to leave the current alliance and join another alliance;

[0046] When the alliance is unstable, the alliance selection is reconfigured based on the vehicle mobility model, limiting the time that alliance vehicles stay within the coverage area of ​​the roadside unit to no less than the time required for training and inference. The alliance formation, benefit distribution and stability verification are repeated until a stable alliance structure is formed.

[0047] Output a stable vehicle fine-tuning alliance structure suitable for federated learning and its profit distribution results.

[0048] As a further improvement of an embodiment of the present invention, the method further includes that performing dynamic weighted federated learning through a multi-dimensional trust evaluation mechanism includes:

[0049] Collect vehicle identity authentication, network traffic, terminal security, service access, and threat notification data;

[0050] Evaluate the vehicle's communication reliability, behavior history, model update quality, and environmental dynamics through a data analysis model, and calculate the trust value and determine the trust level using information entropy theory;

[0051] Generate comprehensive trust assessment results based on a multi-dimensional trust assessment mechanism, and adjust historical trust data through a dynamic update mechanism;

[0052] The multi-dimensional trust evaluation mechanism includes:

[0053] Direct trust is based on vehicle identity authentication and communication reliability assessment;

[0054] Recommended trust is based on the trust recommendations of other vehicles or roadside units (RSUs) for the vehicle;

[0055] Feedback trust is based on the quality of the contribution of local model update data uploaded by the vehicle to the global model;

[0056] Historical trust is based on the vehicle’s past performance records in federated learning and is dynamically updated through a time decay function.

[0057] assigning a weighted coefficient in federated learning to the vehicle based on the comprehensive trust evaluation result;

[0058] Based on the weighted coefficients, the local model gradients uploaded by the vehicles are weighted averaged to generate a high-precision global intelligent model.

[0059] As a further improvement of an embodiment of the present invention, the method further includes: enhancing security through the differential privacy mechanism includes:

[0060] Receiving local model gradient data uploaded by the vehicle, adding random noise to the gradient data for protection through the differential privacy mechanism, generating privacy-protected gradient data to prevent reverse speculation attacks and reduce the impact on model performance; the local model gradient data is a parameter update vector generated by the vehicle after optimizing the original intelligent model based on local training data;

[0061] Based on the security authentication protocol, digital signatures, encrypted transmission and identity authentication are used to verify the legitimacy of both communicating parties, and end-to-end encryption and dynamic key updates are used to protect the integrity and security of data transmission, generating a secure communication channel.

[0062] As a further improvement of an embodiment of the present invention, the method further includes: enhancing security through the blockchain tracking mechanism includes:

[0063] After each round of federated learning, the comprehensive trust evaluation results, gradient contributions, abnormal behavior records, and task timestamps of participating nodes are stored in the alliance chain, which includes RSUs, regulators, and high-trust vehicles.

[0064] The gradient contribution is calculated based on the degree of influence of the local model gradient data uploaded by the vehicle on the global intelligent model optimization;

[0065] Optimize the block data structure, reduce storage requirements through data compression and sharding, use Merkle tree technology to verify data integrity, and perform hash encryption or homomorphic encryption on sensitive data to protect privacy;

[0066] Optimize trust assessment, perform abnormal behavior detection, and verify data integrity based on the blockchain records.

[0067] To achieve one of the above-mentioned objectives of the invention, an embodiment of the present invention further provides a multi-domain intelligent optimization system for an Internet of Vehicles (IoV) under a computing-network fusion architecture, the system comprising an optimization module, a management module, and an aggregation module;

[0068] The optimization module is used to receive task requests from edge layer roadside units based on a preconfigured cloud-edge-end collaborative architecture, optimize user service latency and network load balancing through a differential evolution algorithm, generate an original intelligent model distribution path, and distribute the pre-trained original intelligent model to the edge layer roadside units;

[0069] The management module is used to receive the original intelligent model and vehicle mobility data, optimize vehicle mobility, model quality and alliance formation cost through the alliance game algorithm, and build a vehicle fine-tuning alliance structure suitable for federated learning;

[0070] The aggregation module is used to fine-tune the alliance structure based on vehicles, obtain local model update data and trust assessment data of vehicles in the alliance, perform dynamic weighted federated learning through multi-dimensional trust assessment, enhance security through differential privacy and blockchain tracking mechanisms, generate high-precision global intelligent models, and deploy them to vehicles.

[0071] In order to achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention also provides an electronic device, including a memory and a processor, characterized in that the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the multi-domain intelligent optimization method of the Internet of Vehicles under the above-mentioned computing-network fusion architecture are implemented.

[0072] In order to achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention also provides a storage medium, which stores a computer program, and is characterized in that when the computer program is executed by a processor, it implements the steps of the multi-domain intelligent optimization method of the Internet of Vehicles under the above-mentioned computing-network fusion architecture.

[0073] Compared with the existing technology, the present invention provides a multi-domain intelligent optimization method and system for the Internet of Vehicles under a computing-network fusion architecture. Through the computing-network fusion architecture, efficient collaboration between the cloud, edge and end is achieved, computing isolation and resource island problems are solved, and resource utilization is improved; differential evolution and alliance game algorithms are used to optimize intelligent distribution and model fine-tuning, significantly reducing latency and improving model accuracy; security and privacy protection are enhanced through multi-dimensional trust evaluation, differential privacy and blockchain mechanisms, effectively responding to malicious attacks and data leakage risks; and ultimately achieving low-latency, high-precision, and load-balanced Internet of Vehicles service performance, providing reliable support for applications such as autonomous driving and V2X communication. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is an overall flow chart of the multi-domain intelligent optimization method of the Internet of Vehicles under the computing-network fusion architecture described in the present invention.

[0075] Figure 2 This is a schematic diagram of the cloud-edge-end three-layer collaborative architecture of the intelligent optimization architecture based on vehicle-computing fusion, which is the multi-domain intelligent optimization method of the Internet of Vehicles under the computing-network fusion architecture described in the present invention.

[0076] Figure 3 This is a flow chart of a differential evolution algorithm that integrates multiple mutation strategies for the multi-domain intelligent optimization method of the Internet of Vehicles under the computing-network fusion architecture described in the present invention.

[0077] Figure 4 This is a flow chart of the alliance game algorithm based on vehicle mobility in the multi-domain intelligent optimization method of the Internet of Vehicles under the computing-network fusion architecture described in the present invention.

[0078] Figure 5This is a schematic diagram of the architecture of the multi-domain intelligent optimization system of the Internet of Vehicles under the computing-network fusion architecture described in the present invention. DETAILED DESCRIPTION

[0079] The present invention will be described in detail below with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional changes made by those skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0080] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.

[0081] In the first embodiment of the present invention, the present invention provides a multi-domain intelligent optimization method for Internet of Vehicles under the computing-network fusion architecture, such as Figure 1 As shown, the method includes,

[0082] S1: Receives task requests from edge-layer roadside units, optimizes user service latency and network load balancing through differential evolution algorithms, generates original intelligent model distribution paths, and distributes pre-trained original intelligent models to edge-layer roadside units;

[0083] S2: Receives the original intelligent model and vehicle mobility data, optimizes vehicle mobility, model quality, and alliance formation costs through an alliance game algorithm, and constructs a fine-tuned vehicle alliance structure suitable for federated learning.

[0084] S3: Obtain local model update data and trust assessment data from vehicles in the vehicle fine-tuning alliance structure, perform dynamic weighted federated learning through multi-dimensional trust assessment, enhance security through differential privacy and blockchain tracking mechanisms, generate a high-precision global intelligent model, and deploy it to vehicles.

[0085] In a specific embodiment of the present invention, the construction of the preconfigured cloud-edge-device collaborative architecture is specifically as follows:

[0086] Obtaining the requirements of the Internet of Vehicles system, building a three-layer cloud-edge-end collaborative architecture, and having the edge layer roadside unit transmit task requests to the cloud layer macro base station to obtain defined architecture parameters;

[0087] The cloud-edge-device three-layer collaborative architecture includes:

[0088] A cloud-layer MBS configured to perform global task scheduling, model caching, and cache strategy optimization, and wherein the pre-trained original intelligent model is stored in the MBS;

[0089] Edge layer RSU, configured to perform signal relay, local data aggregation, and collaborative computing with vehicles;

[0090] The terminal-layer vehicle is configured to share computing tasks through on-board computing resources and collaboratively optimize the original intelligent model training.

[0091] It should be noted that if Figure 2 As shown, the intelligent optimization architecture based on vehicle-computing fusion is a three-layer collaborative architecture of cloud, edge and terminal, in which MBS is located in the top cloud layer, has strong computing power and caches all models; RSU nodes are located in the edge layer of the middle layer, have certain computing capabilities but still have problems such as insufficient resources and limited computing power due to hardware limitations; vehicles are located in the bottom terminal layer, which can interact with the middle edge layer to assist in the training of intelligent models. When the RSU sends a request for an intelligent task, the MBS plans an intelligent distribution path for the RSU. After receiving the intelligent model, the RSU incentivizes the terminal layer vehicles to form a fine-tuning alliance. The RSU acts as a convergence node to conduct federated learning within the alliance to achieve intelligent fine-tuning. A detailed description of each layer is as follows:

[0092] Cloud Layer: MBS, located at the top cloud layer, possesses strong computing power and caches all models. MBS plays a core role in the IoV. MBS also serves as the central controller, determining caching strategies and, in the case of collaborative caching, allowing users to retrieve requested content from the MBS or relay RSU nodes. Different relay RSU nodes can share content with each other.

[0093] Edge layer: RSU nodes, located in the middle layer, possess a certain level of computing power, but due to hardware limitations, they still face issues such as insufficient resources and limited computing power. As key communication nodes in the IoV, these nodes provide functions such as signal relay, traffic information collection, real-time road conditions, data exchange, and intelligent traffic management. However, their hardware architecture may not be able to match the powerful computing and storage capabilities of the cloud. The limited resources of RSUs can lead to performance bottlenecks when processing large amounts of data or complex computing tasks, especially in high-traffic or high-demand situations.

[0094] Terminal layer: As the bottom terminal layer, vehicles need to interact more closely with the edge layer to assist in the training and optimization of intelligent models. Due to the limited computing and storage capabilities of RSU nodes, they may not be able to handle all computing-intensive tasks independently, especially in scenarios with high traffic or huge amounts of data. Therefore, vehicles can share some of the computing tasks through on-board computing resources. This collaborative working mode can not only improve the computing efficiency of the entire system, but also reduce service delays caused by insufficient computing power, and enhance the performance and reliability of the Internet of Vehicles system. Through effective collaboration between vehicles and RSU nodes, the distribution of computing tasks can be adjusted more flexibly, making the training of intelligent models more accurate, while improving the real-time and response speed of Internet of Vehicles services.

[0095] In a specific embodiment of the present invention, the user service delay and network load balancing are optimized by using a differential evolution algorithm, specifically,

[0096] Receive the task request of the RSU and the communication data generated based on the cloud-edge architecture, and define service delay as a service quality indicator; the service delay is composed of transmission delay and propagation delay; the RSU obtains the delay of the original intelligent model and iteratively calculates the delay of each intermediate node in the path set to generate service delay evaluation data;

[0097] Based on the communication data, defining the network load balance as a cost indicator, expressed as a standard deviation of link utilization, and generating a load balance evaluation result;

[0098] A differential evolution algorithm that integrates multiple mutation strategies is used, and the difference between the user service delay and the network load balance is used as the fitness function to optimize the original intelligent model distribution path, specifically including:

[0099] Initialize the population and randomly generate a real-number encoding vector representing the distribution strategy;

[0100] Perform mutation operations and generate mutation vectors. Mutation strategies include:

[0101] The first mutation strategy means randomly selecting three individuals to generate mutation vectors. The formula is:

[0102] ;

[0103] in, is the mutation vector generated based on the first mutation strategy, 、 、 are three different individuals randomly selected from the population, is the variation factor;

[0104] The second mutation strategy is to generate mutation vectors based on the optimal individual. The formula is:

[0105] ;

[0106] in, is the mutation vector generated based on the second mutation strategy, is the current optimal individual, 、 are two different individuals randomly selected from the population;

[0107] The third mutation strategy is to combine the current individual with the optimal individual to generate a mutation vector. The formula is:

[0108] ;

[0109] in, is the mutation vector generated based on the third mutation strategy, For the current individual;

[0110] Perform a crossover operation to mix the mutation vector with the target individual through the crossover rate to generate the test individual. The formula is:

[0111] ;

[0112] in, For the test individuals, is the crossover rate, is a randomly selected dimension index;

[0113] Perform a selection operation, compare the fitness of the test individual and the target individual, and retain the individual with better fitness; the fitness is the difference between the service delay and the load balance;

[0114] Repeating the mutation operation, the crossover operation, and the selection operation until a termination condition is satisfied, wherein the termination condition includes a maximum number of iterations or a convergence criterion;

[0115] Boundary checks are performed on the generated individuals to generate the optimal distribution path of the original intelligent model.

[0116] It should be noted that many applications, such as autonomous driving and V2X communication, have extremely high real-time requirements, and latency directly determines the system's response speed and security. Low latency not only improves the user experience, ensuring smooth navigation and in-car entertainment, but also enhances service reliability. Furthermore, as a quantifiable and easily optimized metric, latency not only reflects network performance but also indirectly improves other key system performance. Therefore, latency is used as a metric to measure user QoS in the intelligent distribution phase. First, a communication model is constructed, assuming that latency is composed of both transmission and propagation delays. Furthermore, the latency acquired by the RSU is not the cumulative delay of a single link, but rather the iterative delay of each intermediate node in the RSU path set.

[0117] Furthermore, because traffic demands on different nodes in the IoV communication network are often uneven, it's easy for some nodes to experience bandwidth overload while others experience idle resources. Load balancing can dynamically allocate network resources, optimize bandwidth utilization, avoid communication congestion, and improve data transmission efficiency. Therefore, load balancing is used as a metric to measure bandwidth overhead. Link utilization is calculated using the established communication model, and its standard deviation is used to represent the degree of load balancing.

[0118] Furthermore, since latency, representing user QoS, and load balancing, representing cost, are both functions of the distribution strategy, a differential evolution algorithm that integrates multiple mutation strategies is used to optimize the distribution strategy, targeting the difference between latency and load balancing. This algorithm first divides the population into multiple subgroups, each of which employs a different mutation strategy to fully leverage the optimization capabilities of each strategy, thereby improving the balance between global and local search. Specifically, the difference between latency and load balancing is defined as the fitness function, comprehensively considering both minimizing communication latency and balancing network load. By guiding the subgroups with differentiated strategies, the algorithm can more efficiently explore the solution space. Ultimately, with fitness optimization as the goal, the structure of the path distribution matrix is ​​dynamically adjusted to achieve the global optimal solution.

[0119] Furthermore, if Figure 3 As shown in the figure, the differential evolution algorithm process of integrating multiple mutation strategies is as follows:

[0120] Step 1: Initialization: Randomly generate a population of initial candidate solutions. Each individual is a potential solution, that is, a potential distribution strategy, represented as a vector encoded by real numbers.

[0121] Step 2: Mutation operation: For each individual, select three different individuals as mutation vectors. Adjust the amplitude of the mutation vector according to the mutation factor to generate a new mutation vector.

[0122] Step 4: Select an action: Decide whether the test individual should replace the target individual. This is done by comparing the fitness of the test individual with the target individual. Fitness is the difference between latency and bandwidth cost. If the test individual has a better fitness, the target individual is replaced.

[0123] Step 5: Iterate until the conditions are met:

[0124] Step 6: Repeat steps 2 to 4 until the termination condition is met, such as reaching the maximum number of iterations or meeting the convergence criterion.

[0125] Step 7: Boundary condition processing: If necessary, perform boundary checks on the generated individuals to ensure that they are within the feasible solution space of the problem.

[0126] In a specific embodiment of the present invention, the vehicle mobility, model quality and alliance formation cost are optimized through the alliance game algorithm, specifically,

[0127] Receiving vehicle mobility data and training data information, defining the model quality as a user service quality indicator to update the estimated accuracy representation of the global model; the estimated accuracy is a concave increasing function based on the training data quality, and generating a model quality evaluation result;

[0128] Based on the number of vehicles in the alliance, the alliance formation cost is defined as the cost expenditure indicator, expressed as a linear function of the number of vehicles, and the alliance cost evaluation result is generated;

[0129] Based on the vehicle mobility model and the alliance formation permission set, the alliance structure is optimized through the alliance game algorithm, including:

[0130] Configuring a set of participants and a value function that evaluates the utility of the alliance based on model quality;

[0131] Based on the interests of the vehicles and the value function, a mutually beneficial alliance structure is formed through negotiation;

[0132] Use the Shapley value method to distribute the benefits of alliance members;

[0133] Verify alliance stability and ensure that no vehicle has the motivation to leave the current alliance and join another alliance;

[0134] When the alliance is unstable, the alliance selection is reconfigured based on the vehicle mobility model, limiting the time that alliance vehicles stay within the coverage area of ​​the roadside unit to no less than the time required for training and inference. The alliance formation, benefit distribution and stability verification are repeated until a stable alliance structure is formed.

[0135] Output a stable vehicle fine-tuning alliance structure suitable for federated learning and its profit distribution results.

[0136] It should be noted that for client collaboration, model accuracy is generally assumed to be positively correlated with the size of the overall training data. Higher model accuracy leads to higher satisfaction, meaning greater benefits for the alliance. This assumes that each alliance in the system applies the same metric to reflect its benefits. This metric is defined as the estimated accuracy of the new global model, a concavely increasing function of the training data quality across all participating organizations. Therefore, a logarithmic function is used to characterize the relationship between model accuracy and training data, meaning that model accuracy is used as a metric to measure user QoS during intelligent fine-tuning.

[0137] Furthermore, regarding the cost metric, while cooperation can generate additional benefits for clients, these benefits are limited by the inherent costs that clients must incur when forming a coalition to conduct cooperative actions. As the number of vehicles in the coalition increases, the cost of forming a coalition increases linearly to coordinate with all vehicles. Therefore, a linear function of the number of vehicles in the coalition—the component coalition cost—is used as a cost metric for the intelligent fine-tuning process.

[0138] Furthermore, a coalition game algorithm based on vehicle mobility first constructs a vehicle mobility model. Due to the limited coverage of RSUs, it may be difficult for vehicles to fully participate in FL training within the limited time. To build a more accurate global model, high-quality vehicle model updates must be aggregated on the RSU server. Therefore, only vehicles that can complete training and inference within the RSU's coverage area are selected to form a coalition to participate in the RSU's federated learning training. Here, the vehicle coalition formation permission set is given, indicating whether vehicles can form a coalition to participate in the model fine-tuning process of RSU i.

[0139] Furthermore, if Figure 4 As shown in the figure, the coalition game algorithm process based on vehicle mobility is as follows:

[0140] Step 1: Initialization: Determine the set of game participants, each with their own resources or capabilities. Define the value function for each alliance, that is, how to evaluate the total value or utility of an alliance.

[0141] Step 2: Alliance formation: Vehicle v attempts to communicate and negotiate with other vehicles to form an alliance based on their respective interests and value functions. The formation of an alliance can be based on mutual benefit, that is, each member of the alliance can obtain more benefits from it than if they acted alone.

[0142] Step 3: Value distribution: Alliance is formed, and Shapely value is used to distribute the benefits of alliance members.

[0143] Step 4: Stability Detection: Check whether the formed alliance is stable, that is, no member has the motivation to leave the current alliance to join another alliance because it will bring them greater benefits.

[0144] Step 5: Iterative optimization: If the alliance is unstable, allow vehicle v to re-evaluate the alliance selection and join or leave the alliance. When joining the alliance, ensure that the time the vehicle stays within the RSU coverage is greater than or equal to the vehicle training and inference model time to ensure that they can fully participate in FL training and build a more accurate global model.

[0145] Step 6: Repeat steps 2 to 4 until a stable state is reached, i.e. all alliances are stable and no participants want to change their alliance affiliation.

[0146] Step 7: Result output: Output the final alliance structure and the value distribution results within each alliance.

[0147] In a specific embodiment of the present invention, dynamic weighted federated learning is performed through a multi-dimensional trust evaluation mechanism, specifically,

[0148] Collect vehicle identity authentication, network traffic, terminal security, service access, and threat notification data;

[0149] Evaluate the vehicle's communication reliability, behavior history, model update quality, and environmental dynamics through a data analysis model, and calculate the trust value and determine the trust level using information entropy theory;

[0150] Generate comprehensive trust assessment results based on a multi-dimensional trust assessment mechanism, and adjust historical trust data through a dynamic update mechanism;

[0151] The multi-dimensional trust evaluation mechanism includes:

[0152] Direct trust is based on vehicle identity authentication and communication reliability assessment;

[0153] Recommended trust is based on the trust recommendations of other vehicles or roadside units (RSUs) for the vehicle;

[0154] Feedback trust is based on the quality of the contribution of local model update data uploaded by the vehicle to the global model;

[0155] Historical trust is based on the vehicle’s past performance records in federated learning and is dynamically updated through a time decay function.

[0156] assigning a weighted coefficient in federated learning to the vehicle based on the comprehensive trust evaluation result;

[0157] Based on the weighted coefficients, the local model gradients uploaded by the vehicles are weighted averaged to generate a high-precision global intelligent model.

[0158] It should be noted that in the connected vehicle environment, to ensure system security and the efficient execution of federated learning tasks, the RSU (Roadside Unit) introduces a multi-dimensional dynamic trust assessment mechanism. This scientifically evaluates each vehicle's trust score through comprehensive calculations and real-time updates. This mechanism comprehensively analyzes a vehicle's communication reliability, behavioral history, model update quality, and environmental dynamics, combining multiple key dimensions.

[0159] Furthermore, the specific steps of the multi-dimensional trust evaluation mechanism are:

[0160] Step 1: Establishing a basic trust value: Collect the vehicle's identity information, basic information about the terminal device used, and data sensitivity information about the application that needs to be accessed, and calculate the vehicle's basic trust value through corresponding weighted and average operations.

[0161] Step 2: Trust indicator data collection: Trust indicator data is collected by collecting key data that affect trust values, such as vehicle identity authentication, network traffic, terminal security, business access, and threat notification, to support subsequent trust value calculation and trust evaluation.

[0162] Step 3: Trust indicator data processing: Use data processing rules to process the collected data and provide pre-processed data for the data analysis model.

[0163] Step 4: Trust indicator data analysis: Analyze the processed data through the data analysis model to evaluate multiple dimensions such as vehicle communication reliability, behavior history, model update quality, and environmental dynamics.

[0164] Step 5: Trust value measurement: Use information entropy theory to smooth the multi-dimensional trust decision attribute measurement value and calculate the trust value.

[0165] Step 6: Trust Level Assessment: Based on the calculated trust value and trust level, the vehicle is dynamically authorized and controlled. The trust level determines the vehicle's permissions and functions in the connected vehicle network.

[0166] Step 7: Comprehensive trust evaluation: Comprehensively consider the four measurement dimensions of direct trust, recommendation trust, feedback trust and historical trust to implement a hybrid trust evaluation scheme that can measure the reliability of both vehicle nodes and vehicle messages.

[0167] Step 8: Dynamic trust update: In multi-dimensional trust evaluation, historical trust is based on the vehicle’s past performance records in federated learning and is dynamically updated through a time decay function to reflect the timeliness of the trust relationship. It is defined as the time weight coefficient of historical trust records, which is used to adjust the change of trust value over time. Specifically, the time decay function can be expressed as an exponential decay function:

[0168]

[0169] in, Indicates the time interval of the i-th historical trust record (the difference between the current time and the record time), is the decay coefficient, which controls the decay speed of the trust value. The historical trust record vector is stored locally or in RSU. The updated historical trust value is:

[0170]

[0171] in, is the original trust value of the i-th historical trust record.

[0172] To ensure the real-time nature of trust evaluation, a time window is defined. If the time interval of the records If the time window is exceeded, the record is discarded.

[0173] Step 9: Decision and interaction: The service requester can use trust evaluation to decide whether to choose the node to interact with in order to obtain the service; the service provider can also use trust evaluation to decide the service level to provide to the requester.

[0174] Step 10: Feedback and adjustment: Based on the interaction and feedback between vehicles, the trust value and trust level are updated to adapt to changes and threats in the Internet of Vehicles environment.

[0175] In a specific embodiment of the present invention, security is enhanced by the differential privacy mechanism, specifically,

[0176] Receiving local model gradient data uploaded by the vehicle, adding random noise to the gradient data for protection through the differential privacy mechanism, generating privacy-protected gradient data to prevent reverse speculation attacks and reduce the impact on model performance; the local model gradient data is a parameter update vector generated by the vehicle after optimizing the original intelligent model based on local training data;

[0177] Based on the security authentication protocol, digital signatures, encrypted transmission and identity authentication are used to verify the legitimacy of both communicating parties, and end-to-end encryption and dynamic key updates are used to protect the integrity and security of data transmission, generating a secure communication channel.

[0178] It should be noted that the specific implementation steps of the federated learning architecture with differential privacy and dynamic weighting are as follows:

[0179] Step 1: Differential Privacy Enhancement: To prevent malicious nodes or servers from obtaining this information through reverse inference attacks, differential privacy technology is applied to gradient protection. Before the vehicle uploads the gradient, the system adds random noise to the gradient, making it difficult for an attacker to recover the original data even if the gradient is obtained. This differential privacy enhancement strategy protects privacy while minimizing the impact of noise on model performance, providing higher privacy and security for federated learning.

[0180] Step 2: Secure Communication Protocol: A secure authentication protocol based on the V2X standard ensures the integrity and immutability of uploaded and downloaded data. This protocol uses digital signatures, encrypted transmission, and identity authentication to verify the legitimate identities of both communicating parties, preventing malicious nodes from posing as spoofed entities. Furthermore, end-to-end encryption protects data from being eavesdropped during transmission, and a dynamic key update mechanism enhances the system's resilience to attacks. These measures establish a highly secure communication system, ensuring reliable data transmission for federated learning.

[0181] Step 3: Dynamically Weighted Federated Aggregation: In federated learning tasks, the trustworthiness and data quality of different vehicles can vary significantly. To ensure the quality and robustness of the global model, RSU introduces a dynamically weighted federated aggregation mechanism. RSU assigns weights based on each vehicle's trust score, which is calculated using a multi-dimensional evaluation mechanism that includes communication reliability, behavioral history, and model contribution. During the model update process, RSU performs a weighted average of uploaded gradients based on the trust weights, giving more influence to highly trusted nodes and weakening the influence of low-quality or malicious nodes.

[0182] In one embodiment of the present invention, security is enhanced by the blockchain tracking mechanism, specifically,

[0183] After each round of federated learning, the comprehensive trust evaluation results, gradient contributions, abnormal behavior records, and task timestamps of participating nodes are stored in the alliance chain, which includes RSUs, regulators, and high-trust vehicles.

[0184] The gradient contribution is calculated based on the degree of influence of the local model gradient data uploaded by the vehicle on the global intelligent model optimization;

[0185] Optimize the block data structure, reduce storage requirements through data compression and sharding, use Merkle tree technology to verify data integrity, and perform hash encryption or homomorphic encryption on sensitive data to protect privacy;

[0186] Optimize trust assessment, perform abnormal behavior detection, and verify data integrity based on the blockchain records.

[0187] It should be noted that in federated learning tasks, to ensure transparency of node behavior and system security, the RSU uses blockchain technology to record and manage the behavior of each node. After each round of federated learning, the RSU stores key behavioral information of participating nodes (such as trust scores, gradient update contributions, and anomaly detection records) on the blockchain.

[0188] Furthermore, in blockchain behavior tracking, a consortium chain architecture is used as an implementation method, with participating nodes including RSUs, regulators, and high-trust vehicle nodes. Consortium chains offer high throughput, low latency, and limited access rights, making them suitable for handling the frequent data recording needs of federated learning. At the same time, the block data structure is optimized, with each block storing information such as trust scores, gradient contributions, abnormal behavior records, and task timestamps. Data compression and sharding reduce storage requirements and improve efficiency in the connected vehicle environment. To ensure data security, RSUs upload collected behavior data to the chain, using Merkle Tree technology to quickly verify data integrity and hash-encrypt storage for sensitive behavior data. Using homomorphic encryption or zero-knowledge proof technology, the legitimacy of node data can be verified while protecting privacy, ensuring the security and privacy of the system.

[0189] In the second embodiment of the present invention, the present invention provides a multi-domain intelligent optimization system for an Internet of Vehicles (IoV) under a computing-network fusion architecture, the system comprising an optimization module 1, a management module 2, and an aggregation module 3;

[0190] The optimization module 1 is used to receive task requests from edge layer roadside units based on a preconfigured cloud-edge-end collaborative architecture, optimize user service latency and network load balancing through a differential evolution algorithm, generate an original intelligent model distribution path, and distribute the pre-trained original intelligent model to the edge layer roadside units;

[0191] The management module 2 is used to receive the original intelligent model and vehicle mobility data, optimize vehicle mobility, model quality and alliance formation cost through the alliance game algorithm, and build a vehicle fine-tuning alliance structure suitable for federated learning;

[0192] The aggregation module 3 is used to fine-tune the alliance structure based on vehicles, obtain local model update data and trust assessment data of vehicles in the alliance, perform dynamic weighted federated learning through multi-dimensional trust assessment, enhance security through differential privacy and blockchain tracking mechanisms, generate a high-precision global intelligent model, and deploy it to vehicles.

[0193] In embodiment three of the present invention, the present invention provides an electronic device, including a memory and a processor, characterized in that the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps of the multi-domain intelligent optimization method of the Internet of Vehicles under the above-mentioned computing-network fusion architecture are implemented.

[0194] In embodiment 4 of the present invention, the present invention provides a storage medium storing a computer program, characterized in that when the computer program is executed by a processor, the steps of the multi-domain intelligent optimization method of the Internet of Vehicles under the computing-network fusion architecture as described above are implemented.

[0195] In summary, the present invention provides a multi-domain intelligent optimization method and system for the Internet of Vehicles under a computing-network fusion architecture. Through the computing-network fusion architecture, efficient cloud-edge-end collaboration is achieved, computing isolation and resource island problems are solved, and resource utilization is improved. Differential evolution and alliance game algorithms are used to optimize intelligent distribution and model fine-tuning, significantly reducing latency and improving model accuracy. Multi-dimensional trust evaluation, differential privacy, and blockchain mechanisms are used to enhance security and privacy protection, effectively responding to malicious attacks and data leakage risks. Ultimately, low-latency, high-precision, and load-balanced Internet of Vehicles service performance is achieved, providing reliable support for applications such as autonomous driving and V2X communication.

[0196] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the modules described above can refer to the corresponding process in the aforementioned method implementation, and will not be repeated here.

[0197] Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network modules. Some or all of these modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0198] In addition, the functional modules in each embodiment of the present application may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0199] The above-mentioned integrated modules implemented in the form of software functional modules can be stored in a computer-readable storage medium. The above-mentioned software functional modules are stored in a storage medium and include a number of instructions for causing a computer system (which may be a personal computer, server, or network system, etc.) or a processor to execute some of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A multi-domain intelligent optimization method for the Internet of Vehicles (IoV) based on a computing-network fusion architecture, characterized by: include, Based on the pre-configured cloud-edge-device collaborative architecture, it receives task requests from edge-layer roadside units, optimizes user service latency and network load balancing through differential evolution algorithms, generates original intelligent model distribution paths, and distributes the pre-trained original intelligent models to edge-layer roadside units. Receive the original intelligent model and vehicle mobility data, optimize vehicle mobility, model quality and alliance formation cost through the alliance game algorithm, and build a vehicle fine-tuning alliance structure suitable for federated learning; Based on the vehicle-based fine-tuning alliance structure, local model update data and trust assessment data of vehicles in the alliance are obtained, dynamic weighted federated learning is performed through multi-dimensional trust assessment, and security is enhanced through differential privacy and blockchain tracking mechanisms. A high-precision global intelligent model is generated and deployed to vehicles.

2. The multi-domain intelligent optimization method for the Internet of Vehicles under the computing-network fusion architecture according to claim 1 is characterized by: The construction of the pre-configured cloud-edge-end collaborative architecture includes: Obtaining the requirements of the Internet of Vehicles system, building a three-layer cloud-edge-end collaborative architecture, and having the edge layer roadside unit transmit task requests to the cloud layer macro base station to obtain defined architecture parameters; The cloud-edge-device three-layer collaborative architecture includes: A cloud-layer MBS configured to perform global task scheduling, model caching, and cache strategy optimization, and wherein the pre-trained original intelligent model is stored in the MBS; Edge layer RSU, configured to perform signal relay, local data aggregation, and collaborative computing with vehicles; The terminal-layer vehicle is configured to share computing tasks through on-board computing resources and collaboratively optimize the original intelligent model training.

3. The multi-domain intelligent optimization method for the Internet of Vehicles under the computing-network fusion architecture according to claim 2 is characterized by: The optimization of user service delay and network load balancing by differential evolution algorithm includes: Receive the task request of the RSU and the communication data generated based on the cloud-edge architecture, and define service delay as a service quality indicator; the service delay is composed of transmission delay and propagation delay; the RSU obtains the delay of the original intelligent model and iteratively calculates the delay of each intermediate node in the path set to generate service delay evaluation data; Based on the communication data, defining the network load balance as a cost indicator, expressed as a standard deviation of link utilization, and generating a load balance evaluation result; A differential evolution algorithm that integrates multiple mutation strategies is used, and the difference between the user service delay and the network load balance is used as the fitness function to optimize the original intelligent model distribution path, specifically including: Initialize the population and randomly generate a real-number encoding vector representing the distribution strategy; Perform mutation operations and generate mutation vectors. Mutation strategies include: The first mutation strategy means randomly selecting three individuals to generate mutation vectors. The formula is: ； in, is the mutation vector generated based on the first mutation strategy, 、 、 are three different individuals randomly selected from the population, is the variation factor; The second mutation strategy is to generate mutation vectors based on the optimal individual. The formula is: ； in, is the mutation vector generated based on the second mutation strategy, is the current optimal individual, 、 are two different individuals randomly selected from the population; The third mutation strategy is to combine the current individual with the optimal individual to generate a mutation vector. The formula is: ； in, is the mutation vector generated based on the third mutation strategy, For the current individual; Perform a crossover operation to mix the mutation vector with the target individual through the crossover rate to generate the test individual. The formula is: ； in, For the test individuals, is the crossover rate, is a randomly selected dimension index; Perform a selection operation, compare the fitness of the test individual and the target individual, and retain the individual with better fitness; the fitness is the difference between the service delay and the load balance; Repeating the mutation operation, the crossover operation, and the selection operation until a termination condition is satisfied, wherein the termination condition includes a maximum number of iterations or a convergence criterion; Boundary checks are performed on the generated individuals to generate the optimal distribution path of the original intelligent model.

4. The multi-domain intelligent optimization method for the Internet of Vehicles under the computing-network fusion architecture according to claim 3 is characterized by: The optimization of vehicle mobility, model quality and alliance formation cost through the alliance game algorithm includes: receiving vehicle mobility data and training data information, defining the model quality as a user service quality indicator to update an estimated accuracy representation of a global model; The estimated accuracy is a concave increasing function based on the quality of the training data, generating a model quality assessment result; Based on the number of vehicles in the alliance, the alliance formation cost is defined as the cost expenditure indicator, expressed as a linear function of the number of vehicles, and the alliance cost evaluation result is generated; Based on the vehicle mobility model and the alliance formation permission set, the alliance structure is optimized through the alliance game algorithm, including: Configuring a set of participants and a value function that evaluates the utility of the alliance based on model quality; Based on the interests of the vehicles and the value function, a mutually beneficial alliance structure is formed through negotiation; Use the Shapley value method to distribute the benefits of alliance members; Verify alliance stability and ensure that no vehicle has the motivation to leave the current alliance and join another alliance; When the alliance is unstable, the alliance selection is reconfigured based on the vehicle mobility model, limiting the time that alliance vehicles stay within the coverage area of ​​the roadside unit to no less than the time required for training and inference. The alliance formation, benefit distribution and stability verification are repeated until a stable alliance structure is formed. Output a stable vehicle fine-tuning alliance structure suitable for federated learning and its profit distribution results.

5. The multi-domain intelligent optimization method for the Internet of Vehicles under the computing-network fusion architecture according to claim 4 is characterized by: The method of performing dynamic weighted federated learning through a multi-dimensional trust evaluation mechanism includes: Collect vehicle identity authentication, network traffic, terminal security, service access, and threat notification data; Evaluate the vehicle's communication reliability, behavior history, model update quality, and environmental dynamics through a data analysis model, and calculate the trust value and determine the trust level using information entropy theory; Generate comprehensive trust assessment results based on a multi-dimensional trust assessment mechanism, and adjust historical trust data through a dynamic update mechanism; The multi-dimensional trust evaluation mechanism includes: Direct trust is based on vehicle identity authentication and communication reliability assessment; Recommended trust is based on the trust recommendations of other vehicles or roadside units (RSUs) for the vehicle; Feedback trust is based on the quality of the contribution of local model update data uploaded by the vehicle to the global model; Historical trust is based on the vehicle’s past performance records in federated learning and is dynamically updated through a time decay function. assigning a weighted coefficient in federated learning to the vehicle based on the comprehensive trust evaluation result; Based on the weighted coefficients, the local model gradients uploaded by the vehicles are weighted averaged to generate a high-precision global intelligent model.

6. The multi-domain intelligent optimization method for the Internet of Vehicles under the computing-network fusion architecture according to claim 5 is characterized by: Enhanced security through the described differential privacy mechanism includes: Receiving local model gradient data uploaded by the vehicle, adding random noise to the gradient data for protection through the differential privacy mechanism, generating privacy-protected gradient data to prevent reverse speculation attacks and reduce the impact on model performance; the local model gradient data is a parameter update vector generated by the vehicle after optimizing the original intelligent model based on local training data; Based on the security authentication protocol, digital signatures, encrypted transmission and identity authentication are used to verify the legitimacy of both communicating parties, and end-to-end encryption and dynamic key updates are used to protect the integrity and security of data transmission, generating a secure communication channel.

7. The multi-domain intelligent optimization method for the Internet of Vehicles under the computing-network fusion architecture according to claim 6 is characterized by: Enhanced security through the blockchain tracking mechanism includes: After each round of federated learning, the comprehensive trust evaluation results, gradient contributions, abnormal behavior records, and task timestamps of participating nodes are stored in the alliance chain, which includes RSUs, regulators, and high-trust vehicles. The gradient contribution is calculated based on the degree of influence of the local model gradient data uploaded by the vehicle on the global intelligent model optimization; Optimize the block data structure, reduce storage requirements through data compression and sharding, use Merkle tree technology to verify data integrity, and perform hash encryption or homomorphic encryption on sensitive data to protect privacy; Optimize trust assessment, perform abnormal behavior detection, and verify data integrity based on the blockchain records.

8. A multi-domain intelligent optimization system for the Internet of Vehicles (IoV) based on a computing-network fusion architecture, characterized by: Includes optimization module, management module and aggregation module; The optimization module is used to receive task requests from edge layer roadside units based on a preconfigured cloud-edge-end collaborative architecture, optimize user service latency and network load balancing through a differential evolution algorithm, generate an original intelligent model distribution path, and distribute the pre-trained original intelligent model to the edge layer roadside units; The management module is used to receive the original intelligent model and vehicle mobility data, optimize vehicle mobility, model quality and alliance formation cost through the alliance game algorithm, and build a vehicle fine-tuning alliance structure suitable for federated learning; The aggregation module is used to fine-tune the alliance structure based on vehicles, obtain local model update data and trust assessment data of vehicles in the alliance, perform dynamic weighted federated learning through multi-dimensional trust assessment, enhance security through differential privacy and blockchain tracking mechanisms, generate high-precision global intelligent models, and deploy them to vehicles.

9. An electronic device comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the multi-domain intelligent optimization method of the Internet of Vehicles under the computing-network fusion architecture as described in any one of claims 1-7 are implemented.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, the steps of the multi-domain intelligent optimization method of the Internet of Vehicles under the computing-network fusion architecture as described in any one of claims 1 to 7 are implemented.

Images