A cross-domain data security sharing method for vehicle-road cooperation

Abstract

The application relates to a cross-domain data security sharing method for vehicle-road synergy, and belongs to the technical field of mobile communication. In view of the problems of difficult sharing of multi-agent data, heterogeneous computing and cross-domain settlement in vehicle-road synergy, a cross-domain data security sharing method for vehicle-road synergy is provided. First, the vehicle end, the roadside unit and the charging station independently train local models based on vehicle-road data, upload the local model parameters to the DAG blockchain, and update the local model based on the cumulative weight asynchronous aggregation. Second, the power grid side calculates the upper limit of green power consumption in each region based on the instantaneous consumption boundary, adjusts the regional electricity price, combines the road network traffic flow prediction, the charging station queuing and the energy consumption state to solve the optimal charging path. Then, based on the DAG cumulative weight, the node priority is adaptively adjusted, and the weight recovery of the problem node is realized through random walk reanchoring. Finally, based on the DAG blockchain data, the independent settlement of the account books between the cross-domain subjects is completed, and the topological backtracking mechanism is adopted to complete the account book conflict resolution.

Description

Technical Field

[0001] This invention belongs to the field of mobile communication technology, specifically relating to a cross-domain data security sharing method for vehicle-road energy cooperation. Background Technology

[0002] With the rapid growth of electric vehicle ownership and the large-scale grid connection of distributed photovoltaic power, vehicle charging decisions are gradually shifting from simply choosing the nearest charging station to a collaborative decision-making problem that requires comprehensive consideration of multiple factors, including the vehicle's remaining battery power, current location, destination, road congestion status, travel time of candidate routes, charging station queues, station load, charging duration, and the grid's green energy consumption. In the actual road and grid environment, roadside units (RSUs) can obtain roadside information such as regional traffic flow, road speed, and congestion status; charging station operators can obtain station-side information such as station utilization, historical charging records, queue waiting time, and service status; the grid dispatch center has grid-side information such as transformer load, photovoltaic output forecasts, and energy storage status; and the vehicle itself has individual information such as state of charge (SOC), driving trajectory, destination, and user preferences. The above information sources are scattered, heterogeneous, and belong to different property owners. Directly uploading this information to a unified platform would lead to privacy leaks, exposure of commercial data, and risks of cross-domain data transmission.

[0003] Existing electric vehicle charging route planning methods typically rely on traffic control centers, vehicle-to-everything (V2X) platforms, or cloud-based dispatch systems to centrally acquire vehicle, road, and charging station statuses, then make recommendations based on the shortest path, waiting time, or overall cost. While these methods can improve charging station recommendations and route planning to some extent, they often require vehicles to upload user preferences or relatively complete vehicle status information, and charging station operators to provide raw operational data. Furthermore, they do not incorporate grid-side photovoltaic (PV) absorption constraints into the optimization objectives for charging guidance. In scenarios with multiple operators and strong privacy constraints, centralized route planning struggles to simultaneously meet the needs for data security, personalized decision-making, and real-time multi-domain status awareness.

[0004] Federated learning enables multi-party collaborative modeling without requiring the original data to remain within its domain, and has been applied to tasks such as vehicle-to-everything (V2X) perception, traffic condition prediction, and charging load prediction. Some studies combine DAG blockchain with federated learning to achieve asynchronous model aggregation, such as the proposed asynchronous federated learning methods based on DAG blockchain. However, these solutions do not involve cumulative weight-driven Markov Chain Monte Carlo (MCMC) random walk tip selection strategies, and they are not adapted to specific scenarios of vehicle-road-energy (V3X) collaboration. In particular, in V3X energy collaboration scenarios, effective solutions are still lacking for how to construct an instantaneous absorption boundary model on the grid side, guide vehicles to charge at green energy production sites through electricity price signals, and achieve optimal charging path mapping at the vehicle-road layer by jointly predicting the road network, charging station queuing, and vehicle energy consumption status.

[0005] Furthermore, at the cross-domain settlement level, power grid companies, charging pile operators, and photovoltaic power plants belong to different property owners. Due to commercial interests, each party refuses to share the underlying raw data, and there are instances of malicious nodes uploading falsified load data to fraudulently cancel subsidies. Existing DAG parallel execution schemes propose contract parallelization methods that achieve contract-level parallel processing, but they do not extend to the independent state trajectory mechanism and topology backtracking conflict resolution strategy at the property owner level. Traditional settlement systems use global locks to ensure data consistency; settlement transactions between different entities that do not inherently have data conflicts are forced to be executed serially due to sharing the same global lock. Therefore, existing technologies struggle to simultaneously achieve asynchronous federated model aggregation, dual-layer scheduling of grid macro-price guidance and vehicle-end path mapping, adaptive node weight recovery, and cross-property owner parallel settlement without global locks, under the condition that the data of heterogeneous vehicle-road-energy entities does not leave the domain.

[0006] To address the aforementioned issues, this invention proposes a cross-domain data security sharing method for vehicle-road energy collaboration. This method, under the premise that the raw data from RSUs, charging station operators, the power grid dispatch center, and vehicles remain locally, involves each node training its local prediction model in parallel and asynchronously. The model parameters are then woven into a DAG topology network via a cumulative weight-driven MCMC random walk to achieve asynchronous federated aggregation. Next, on the power grid side, the green energy consumption ceiling for each region is calculated based on the consumption boundary, using differentiated electricity prices to guide vehicles to charge at green energy production sites. At the vehicle-road layer, the optimal charging path is solved by combining road network traffic flow prediction, charging station queuing models, and vehicle-side energy consumption status. Furthermore, low-weight nodes are downgraded using DAG cumulative weights, and the weights of problematic nodes are adaptively restored through MCMC random walk re-anchoring. Finally, independent state trajectory lines are constructed for different property owners under the DAG architecture to achieve parallel execution of cross-domain settlement. A topology backtracking mechanism is used only to resolve read-write set conflicts, and the green energy subsidy settlement is confirmed by the PBFT consensus committee and automatically triggered by smart contracts. Therefore, while reducing the risks of cross-entity transmission of raw data and leakage of user preferences, it improves the adaptability of charging path decisions to changes in regional traffic, site services and grid absorption status, and enhances the traceability of federated learning and cross-domain settlement processes. Summary of the Invention

[0007] This invention provides a cross-domain data security sharing method for vehicle-road energy coordination, which addresses existing problems in vehicle-road energy coordination scheduling, such as the risk of cross-domain data sharing among multiple entities, heterogeneous computing power constraints, the disconnect between grid-side green energy consumption and charging guidance, the lack of adaptive weighting mechanisms for node task recovery, and global lock contention in cross-domain settlement.

[0008] To solve the above-mentioned technical problems, the implementation process of this invention is as follows:

[0009] Firstly, embodiments of the present invention provide a method for secure cross-domain data sharing in vehicle-road-energy coordination. This method, under the condition that the original data of roadside units, charging station operators, power grid dispatch centers, and vehicles are all retained locally, achieves secure data sharing and reliable scheduling for vehicle-road-energy coordination through asynchronous federated learning of DAGs, two-layer scheduling guided by macro-level electricity prices and vehicle-side path mapping, adaptive recovery of node weights, and cross-domain settlement of independent state trajectory lines. The method includes the following steps:

[0010] S1: An asynchronous federated learning method for vehicle-road energy based on DAG topology weaving;

[0011] S2: A grid-level dispatching method based on instantaneous absorption and electricity price guidance;

[0012] S3: A vehicle-road layer mapping method based on road network prediction and charging station queuing;

[0013] S4: Weight recovery method based on node cumulative weight and random walk re-anchoring;

[0014] S5: A cross-domain settlement method based on independent state trajectory lines and smart contracts.

[0015] Secondly, in step S1 of this embodiment of the invention, an asynchronous federated learning method for vehicle-road energy based on DAG topology weaving is proposed. Vehicle-end, roadside unit, and charging station independently train local models based on vehicle-road energy data. Specifically, the roadside unit trains a traffic flow prediction model based on a Temporal Graph Convolutional Network (T-GCN), the charging station trains a charging load prediction model based on historical load data, and the vehicle-end trains a path energy consumption prediction model based on maximum entropy estimation and dynamic energy consumption. After each node completes its local training, it encapsulates the model gradient or parameters into transaction nodes. These nodes then adaptively select and reference two unconfirmed transaction tips at the end of the DAG ledger using a Markov Chain Monte Carlo (MCMC) random walk algorithm, creating directed reference edges to weave themselves into the DAG topology. The transition probability of the random walk particles is jointly determined by the cumulative weight of the nodes and the sensitivity coefficient of the cumulative weight difference, causing the walk to favor high-weight paths. Model parameter transactions are asynchronously aggregated and updated with the parameters of the referenced predecessor nodes based on the DAG topology reference relationships. The aggregation weight is determined by the cumulative weight of the topology reference edges. Roadside units, charging stations, and vehicle terminals participate in federated aggregation on an equal footing, eliminating the need for a centralized parameter server and removing the bottleneck constraint of fast nodes waiting for slow nodes in traditional federated learning. The more frequently a model parameter transaction is referenced by subsequent nodes, the higher its cumulative weight in the DAG topology, naturally forming a global ranking of the contributions of nodes in each domain. When submitting model parameter transactions, nodes complete mutual witness verification by referencing predecessor transactions.

[0016] Thirdly, in step S2 of this embodiment, a grid-level scheduling method based on instantaneous absorption and electricity price guidance is proposed. A green electricity time-space scheduling model is constructed at the grid end, using the total power of currently charging electric vehicles and the remaining absorption capacity of the local energy storage system as inputs. The upper limit of green electricity absorption for each region at the current moment is calculated based on the min operator. Addressing the "duck curve" problem of distribution network net load caused by the strong random fluctuations of photovoltaic power generation—where high photovoltaic power generation at midday leads to power backflow and voltage rise, and the overlap of photovoltaic power generation at noon and the evening charging peak causes extremely high load ramp-up rates, resulting in long-term overload operation of distribution transformers and accelerated insulation thermal aging—a model for estimating the probability of maximum entropy spatial flow and coordinated transfer of green electricity is constructed. Based on the photovoltaic output forecast and load gap of each region, differentiated charging electricity price signals are generated. By guiding electric vehicles to spontaneously gather in the green electricity production areas in both time and space dimensions through electricity price differences, peak shaving and valley smoothing are achieved, reducing the amplitude of net load fluctuations, improving the local absorption rate of green electricity while delaying the aging of distribution transformers.

[0017] Fourthly, in step S3 of this embodiment, a vehicle-road layer mapping method based on road network prediction and charging station queuing is proposed. A charging path mapping model is constructed at the vehicle-road layer. Roadside units perform dynamic spatiotemporal flow prediction of the entire road network based on T-GCN to obtain the real-time travel time of each road segment. A multi-service queuing model is established for each charging station, using the number of charging piles as the number of parallel service stations and the traffic flow predicted by RSU as the vehicle arrival rate input, dynamically simulating the disorderly queuing behavior of vehicles within the charging station and solving for the queuing waiting time at the charging piles. At the vehicle end, based on maximum entropy estimation and dynamic energy consumption, combined with the vehicle's current state of charge (SOC), the achievable range and driving energy consumption of each candidate path are determined. On this basis, the average cumulative weight of a node referenced by the entire network in the DAG topology and the task execution success rate within the most recent time window are introduced to generate a weighted dynamic reliability score as a reliability penalty factor. Using candidate charging stations and their corresponding driving routes as the search space, a comprehensive cost is constructed—the comprehensive cost is amplified by a reliability penalty factor after considering driving time, charging station waiting time, and charging electricity price. The charging station and route with the minimum comprehensive cost are selected as the charging route mapping result. The lower the reliability of the charging station, the more its waiting time is amplified, thus naturally avoiding unreliable stations in the route decision.

[0018] Fifthly, in step S4 of this embodiment, a weight recovery method based on node weight and random walk re-anchoring is proposed. Based on the reliability score generated by the node's accumulated weight in the DAG topology and its historical execution success rate, nodes with sufficient computing power and stable execution gradually increase their weight and naturally increase their priority; nodes with insufficient computing power or frequent failures automatically decrease their weight and correspondingly decrease their priority. This mechanism is completely decentralized; any node can independently calculate the weight of any other node through its locally maintained DAG topology view, without the need for centralized monitoring nodes. When a node's accumulated weight decreases due to timeout or failure, the system does not use a first-in-first-out strategy to re-enqueue it. Instead, it extracts the accumulated weight distribution of all unconfirmed transactions in the current DAG topology and starts a new MCMC random walk for re-anchoring, starting with the transaction submitted by that node. The new topology position after re-anchoring is dynamically determined by the current total network accumulated weight—if the node has accumulated a higher topology weight during this period, the re-anchoring path naturally favors the higher-weight tip, giving the task a higher global priority than before the failure, thus accelerating weight recovery; conversely, the priority decreases, freeing up computing resources for tasks of high-weight nodes.

[0019] Sixthly, in step S5 of this embodiment, a cross-domain settlement method based on independent state trajectories and smart contracts is proposed. Under the DAG architecture, each ownership entity maintains an independent state trajectories (StateWire). Each state trajectories is a sub-DAG composed of all confirmed task transactions within that ownership entity in chronological order. Settlement transactions of different entities are completely parallel and do not interfere with each other. A new settlement transaction only needs to randomly sample and reference two previous healthy transactions in the ledger to be attached to the DAG graph. Local conflict detection is triggered only when the state dependency graphs of two settlement transactions intersect, i.e., when the read-write sets overlap. Conflict resolution adopts a topology backtracking strategy—transactions with higher cumulative weights are submitted and written to their corresponding state trajectories first, while transactions with lower cumulative weights backtrack along their DAG reference edges to find an ancestor node that does not conflict with the state of the high-weight transaction as a new anchor point. From the successor branch of this anchor point, a conflict-free path is re-selected to rebuild the execution context, avoiding global lock waiting. A Practical Byzantine Fault Tolerance (PBFT) consensus committee, composed of multiple maintenance nodes, provides final confirmation of the solidified DAG topology snapshot. Smart contracts automatically trigger cross-domain liquidation and value settlement based on the solidified task execution results, including electric vehicle charging fee settlement and green electricity consumption subsidy allocation. Settlement vouchers are hashed and stored on the blockchain, ensuring full lifecycle traceability and auditability. Attached Figure Description

[0020] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0021] Figure 1 This is a diagram of a cross-domain data security sharing system architecture for vehicle-road collaboration.

[0022] Figure 2 This is a diagram of a cross-domain data security sharing algorithm for vehicle-road-energy collaboration.

[0023] Figure 3 A flowchart for cross-domain data security sharing in vehicle-road collaboration. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In this article, the term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates an "or" relationship between the preceding and following related objects. In formulas, the character " / " indicates a "division" relationship between the preceding and following related objects. Unless otherwise specified, "multiple" in this article refers to two or more.

[0026] To facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish the same or similar items with essentially the same function or effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order.

[0027] In this embodiment of the invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this embodiment of the invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner. In this embodiment of the invention, unless otherwise stated, "a plurality of" means two or more.

[0028] Figure 1This paper illustrates the cross-domain data security sharing system architecture for vehicle-road energy collaboration involved in an embodiment of the present invention. The system includes a data acquisition layer, a DAG federated learning layer, a power grid dispatching and vehicle-road mapping layer, and a DAG smart contract settlement layer. The data acquisition layer includes Roadside Units (RSUs), charging station operators, the power grid dispatching center, and vehicle terminals, responsible for collecting multi-source data such as traffic flow, charging load, photovoltaic output and energy storage status, vehicle SOC, and driving trajectory. The DAG federated learning layer is responsible for using the local model gradients of each node as transactions, weaving them into the DAG topology network via MCMC random walks to achieve asynchronous federated aggregation. The power grid dispatching and vehicle-road mapping layer is responsible for constructing instantaneous absorption boundary models and green electricity spatial flow models to generate differentiated electricity price signals, and jointly performing micro-level charging path mapping with road network prediction, charging station queuing models, and vehicle-side energy consumption status. The DAG smart contract settlement layer is responsible for maintaining independent state trajectory lines for different property owners to achieve parallel settlement, which is confirmed by the PBFT consensus committee, and the smart contract automatically executes cross-domain clearing and green electricity subsidy allocation.

[0029] 1. An asynchronous federated learning method for vehicle-road energy based on DAG topology weaving

[0030] In vehicle-road energy cooperative scenarios, the data of distributed nodes belong to different owners. Traditional centralized federated learning relies on a centralized parameter server for global model aggregation. After fast nodes complete local training, they must wait for slow nodes, resulting in "heterogeneous computing power" and wasting computing resources. Some solutions introduce linear blockchains as parameter ledgers for federated learning, but due to the single-chain serial structure, all model updates must be queued and packaged into a single block, leading to a sharp increase in confirmation latency under high concurrency scenarios. To address this, an asynchronous federated learning method for vehicle-road energy based on DAG topology weaving is constructed. This method replaces the linear blockchain with a DAG (Directed Acyclic Graph), and adaptive tip selection and asynchronous aggregation are achieved through MCMC (Multi-Channel Multiplication) random walks. This eliminates the centralized bottleneck while breaking the heterogeneous computing power constraint of federated learning.

[0031] Let the set of urban road areas be . The charging station collection is The set of nodes participating in federated training is For nodes Its local model parameters are denoted as Roadside Units (RSUs) utilize Temporal Graph Convolutional Networks (T-GCNs) to analyze regions. In the time window The system predicts the speed and density distribution of traffic flow within the station, and the charging station uses historical load data to analyze the station's performance. The charging load is predicted, and the vehicle-side predicts path energy consumption based on maximum entropy estimation and a dynamic energy consumption model. All training processes are completed locally, and raw data is not uploaded across domains. After each node completes local training, the model gradient is... Encapsulated as a transaction node, it uses its own decentralized digital identity (DID) private key for signing, within the DAG topology graph. Start An independent Markov Chain Monte Carlo (MCMC) random walk particle Each particle randomly selects several starting nodes from the DAG, and then moves to subsequent nodes according to the transition probability driven by the cumulative weight:

[0032]

[0033] In the formula, Let x be the set of all successor nodes. Accumulate weights for nodes. To accumulate the weight difference sensitivity coefficient, the walk's preference for high-weight paths is controlled. When a particle reaches an unconfirmed transaction... (i.e., there are currently no transactions referenced) Stop wandering when ). The particles eventually stop at most Each node creates two directed edges pointing to one of the selected tips, thus attaching itself to the DAG topology. This process combines task submission and verification into one—nodes submit their tasks while simultaneously performing spontaneous verification of preceding tasks, i.e., "mutual witnessing."

[0034] When trading Subsequent tasks After selecting and referencing using the above MCMC random walk... Cumulative weights Automatic increment. The more frequently a model parameter transaction is referenced by subsequent nodes, the higher its cumulative weight in the DAG topology, naturally forming a global ranking of the contributions of nodes in each domain. Model parameters are asynchronously updated with the parameters of the referenced preceding nodes through streaming aggregation based on the DAG topology reference relationships. The aggregation weight is naturally determined by the cumulative weight of the topology reference edges.

[0035]

[0036] In the formula In the first The set of nodes referenced in the round aggregation. For nodes The cumulative weights are calculated. Roadside units, charging stations, and vehicles participate in federated aggregation on an equal footing, eliminating the need for a centralized parameter server and removing the heterogeneous computing power constraint of fast nodes waiting for slow nodes in traditional federated learning. Nodes verify each other's transactions by referencing previous transactions while submitting model parameter transactions, avoiding the exponential growth in confirmation delays caused by transaction congestion in traditional blockchains.

[0037] 2. A grid-level dispatching method based on instantaneous absorption and electricity price guidance

[0038] Grid-side photovoltaic (PV) power generation exhibits strong random fluctuations. Midday PV surges and load gaps lead to power backflow and voltage spikes. The evening PV shutdown coinciding with the evening charging peak results in extremely high load ramp-up rates, forming a "duck curve" of net load in the distribution network. This causes distribution transformers to operate under overload for extended periods, accelerating insulation thermal aging. Traditional charging guidance methods primarily focus on queuing times at charging stations, rarely incorporating the grid-side PV instantaneous absorption boundary as a spatial guidance constraint into charging decisions. Therefore, this paper constructs a grid-level scheduling method based on instantaneous absorption and price guidance. The instantaneous absorption boundary determines the upper limit of green energy absorption in each region, and a differentiated price signal is generated through a maximum entropy spatial flow model to guide the spatiotemporal migration of electric vehicles to green energy production areas.

[0039] A green energy time-space scheduling model is constructed at the grid end. Let the upper limit of instantaneous green energy absorption in region r at time t be... The total power of the electric vehicles currently charging is The remaining absorption capacity of the local energy storage system is The rated capacity of the distribution transformer is The base load is Then the instantaneous absorption boundary model is:

[0040]

[0041] This formula indicates that the upper limit of instantaneous green electricity absorption is determined by the smaller of "current charging power plus remaining energy storage absorption capacity" and "remaining transformer capacity." When a region has sufficient green electricity absorption capacity, it has the potential to attract new charging loads; when the absorption capacity approaches saturation, excess charging demand needs to be directed to adjacent regions. To quantitatively characterize the collaborative guidance relationship of green electricity between regions, a model for estimating the maximum entropy spatial flow direction and the collaborative transfer probability of green electricity is constructed. Let the regions be... To the area The probability of green electricity space transfer is :

[0042]

[0043] In the formula For the region To the area Spatial distance, This represents the distance attenuation coefficient. Regions with larger absorption capacity and closer distances achieve a higher transfer probability. Differential electricity price signals are generated based on regional photovoltaic output forecasts and load gaps. When there is ample space for green electricity consumption in a certain area, the electricity price is lowered to attract vehicles to charge there; when the space for consumption is tight, the electricity price is raised to guide vehicles to other areas. Through this differentiated electricity price signal, electric vehicles are guided to spontaneously gather in green electricity production areas in both time and space, achieving peak shaving and valley filling, reducing the amplitude of net load fluctuations, improving the local consumption rate of green electricity, and delaying the aging of distribution transformers.

[0044] 3. A vehicle-road layer mapping method based on road network prediction and charging station queuing

[0045] After the grid side determines differentiated electricity prices for each region, vehicles still need to choose between specific charging stations and driving routes. While the Roadside Unit (RSU) can sense regional traffic flow and road conditions, charging station operators have information on station queuing, and vehicles possess State of Charge (SOC) and energy consumption characteristics, this information belongs to different entities. Existing charging guidance methods typically consider either road network travel time or charging station waiting time separately, lacking a complete solution that jointly models road network prediction, charging station queuing, and vehicle energy consumption. Therefore, this paper constructs a vehicle-road layer mapping method based on road network prediction and charging station queuing. This method combines RSU traffic flow prediction, a multi-service queuing model for charging stations, and a vehicle-side maximum entropy energy consumption model, and introduces a DAG cumulative weight reliability score as a penalty factor to solve for the charging path with the minimum overall cost.

[0046] RSU uses T-GCN to perform dynamic spatiotemporal flow prediction on the entire road network, obtaining real-time travel times for each road segment. It also analyzes the data for each charging station. Establish a multi-service counter queuing model for charging stations. Number of charging stations The number of parallel service stations is used as the vehicle arrival rate, with the traffic flow predicted by the RSU as the benchmark. Input. Assume the vehicle arrival parameter is... The Poisson process, the service time of a single pile follows the parameter... If the exponential distribution is followed, then the charging station steady-state probability of queuing system Given by the process of birth and death:

[0047]

[0048] In the formula The idle probability is determined by the normalization condition. Confirmed. Charging station. Average queue length and average vehicle waiting time They are respectively and ,in Let this be the system's full congestion probability. The vehicle-side system, based on maximum entropy estimation and a dynamic energy consumption model, combined with the vehicle's current state of charge (SOC), determines the achievable range and the energy consumption of each candidate path. Assume the vehicle's battery capacity is... The current state of charge is Safely retain power level Road section Energy consumption per unit distance is The length of the road segment is Then the energy reachability constraint is:

[0049]

[0050] Sites and paths satisfying this constraint are retained as candidate combinations. Based on this, the average cumulative weight of a node's total network references within the DAG topology is introduced. Success rate of task execution within the most recent time window Weighted dynamic reliability score generation:

[0051]

[0052] In the formula For nodes The average cumulative weight normalized value of all task transactions in the DAG reflects the historical frequency of the node being verified and referenced by other nodes. This represents the success rate of task execution within the most recent time window. The weighting coefficients balance historical reputation and recent performance. A comprehensive cost function is constructed using candidate charging stations and their corresponding driving routes as the search space:

[0053]

[0054] In the formula Candidate paths Predicted travel time The predicted waiting time for charging station s. For charging stations Differential electricity pricing, For the estimated charging time, This is the reliability penalty strength coefficient. The cost is the cross-domain data transmission cost from the vehicle's current domain to the domain of the candidate charging station, where η is the transmission cost coefficient. The first term of this cost function... The second item combines three main cost categories: travel time, waiting time, and charging costs. As a reliability penalty factor, the waiting time is amplified when the charging station's reliability is low; the third item This incurs costs associated with cross-domain data transfer. The option to make... The minimized charging station and route are used as the charging route mapping result. The lower the reliability of the charging station, the greater the waiting time is amplified, thus naturally avoiding unreliable stations in the route decision.

[0055] 4. Weight recovery method based on node weights and random walk reanchoring

[0056] In the vehicle-road-energy cooperative scheduling system, there are significant differences in the local computing power, network stability, and historical behavior reliability of each distributed node. The cumulative weight of a node naturally fluctuates with its computing power sufficiency, task execution success rate, and frequency of reference across the entire network. Nodes with sufficient computing power and stable execution gradually increase in weight and naturally improve in priority; nodes with insufficient computing power or frequent failures automatically decrease in weight and correspondingly decrease in priority. Traditional solutions use a first-in-first-out (FIFO) strategy to re-add tasks to the tail of the queue when tasks time out or fail, without considering the cumulative weight differences of the nodes to which the task belongs in the topology graph, resulting in high-value tasks being blocked by low-value tasks. To address this, a weight adaptive recovery method based on DAG cumulative weight driving and random walk re-anchoring is constructed. The cumulative weight represents the node priority, and failed tasks are rewritten into the DAG topology through MCMC random walk re-anchoring. The new writing position is determined by the current cumulative weight distribution across the entire network, achieving accelerated recovery of high-weight nodes and natural demotion of low-weight nodes.

[0057] During the adaptive weight adjustment process, the system generates a reliability score based on the cumulative weight of each node in the DAG topology, the historical execution success rate, and the reliability scoring formula in step S3. The weight is derived from adaptive adjustment. This score reflects both the node's historical reliability and recent performance. A higher cumulative weight indicates that the node is verified and referenced more frequently across the network, while a higher historical success rate indicates better execution stability. A node's task priority is positively correlated with its reliability score. The priority ranking of nodes in the system is entirely driven by the cumulative weight of the DAG, forming a decentralized priority self-adjustment mechanism.

[0058]

[0059] In the formula For nodes The average cumulative weight normalized value of all task transactions in the DAG. This represents the task execution success rate within the most recent time window. Nodes with sufficient computing power and stable execution performance are considered. and Nodes with high computing power or frequent failures will have their priorities decrease as both metrics continue to decline. Any node can independently calculate the priority of any other node through its locally maintained DAG topology view, without relying on a centralized monitoring node.

[0060] During the re-anchoring recovery process, when a node fails to execute a task due to insufficient computing power, network instability, or hardware failure, its accumulated weight... With recent success rate The task automatically moves lower as it declines. Instead of using the traditional FIFO strategy to push failed tasks back to the end of the queue, the system extracts the cumulative weight distribution of all unconfirmed transaction tips in the current DAG topology. Starting with the node's most recently successfully submitted transaction, a new MCMC random walk is initiated for re-anchoring. The walk context contains the failed node's latest topological position information. After re-anchoring, the task's new position in the DAG topology is determined by the current cumulative weight distribution across the entire network, and its recovery priority is expressed as follows:

[0061]

[0062] In the formula To re-anchor the selected set of tips, For nodes The cumulative weight of the most recently successfully submitted transaction. This is the cumulative weight difference sensitivity coefficient. If the node... Before failure, due to sufficient computing power and stable execution, a high topology weight was accumulated. The reanchoring path naturally favored tips with higher weights, giving the recovered task a higher global priority than before the failure, thus accelerating the recovery of high-weight nodes. Conversely, if a node... If a node continues to perform poorly and its cumulative weight continues to decline, its new, re-anchored position will be shifted backward compared to its pre-failure position, and its global priority will be reduced accordingly, freeing up computing resources for urgent tasks of high-weight nodes. This mechanism achieves automatic adaptation of node priority to its computing power and behavior at the system level, without the need for a third-party arbitration body.

[0063] 5. Cross-domain settlement method based on independent state trajectory lines and smart contracts

[0064] Power grid companies, third-party charging pile operators, and photovoltaic power plants belong to different property owners. Driven by commercial interests, each party refuses to share underlying raw data, and there are even instances of malicious nodes uploading falsified load data to fraudulently cancel subsidies. Traditional settlement systems use global locks to ensure data consistency. Settlement transactions between different property owners that have no data conflicts are forced to be executed serially due to sharing the same global lock, severely limiting settlement efficiency. Current proposed contract parallelization methods based on DAG models achieve contract-level parallel processing, but they do not extend to the independent state trajectory mechanism and topology backtracking conflict resolution strategy at the property owner level. Therefore, this paper constructs a cross-domain settlement method based on independent state trajectories and smart contracts. Under the DAG architecture, each property owner maintains an independent state trajectory to achieve parallel settlement. Topology backtracking resolution is triggered only when there is a conflict in the read / write set, and the final confirmation is handled by the PBFT consensus committee, with the smart contract automatically executing the settlement.

[0065] In a DAG architecture, each property owner (such as State Grid distribution areas, China Southern Power Grid distribution areas, and urban investment charging station operators) has its own independent state wire on the blockchain. Each state wire... It is a sub-DAG composed of all confirmed task transactions within this property rights entity in chronological order:

[0066]

[0067] different The tasks between them are executed in complete parallelism and do not interfere with each other. The settlement contracts of charging station A and charging station B are two independent state trajectory lines, which can be processed in parallel by different execution nodes. A new settlement transaction only needs to be randomly sampled on the topology and referenced from the previous two healthy transactions in the ledger to be attached to the DAG graph, without going through the serial process of "block packaging - network broadcast - multiple rounds of block confirmation" in traditional blockchains. Local conflict detection is triggered only when the state dependency graphs of two settlement transactions intersect, that is, when the read and write sets overlap. The judgment condition for conflict detection is:

[0068]

[0069] Conflict resolution employs a topological backtracking strategy. Transactions with higher cumulative weights are committed first and written to their corresponding state trajectory lines. Transactions with lower cumulative weights backtrack along their DAG reference edges to find an ancestor node that does not conflict with the state of the higher-weighted transaction as a new anchor point. Formally, let... And if the two transactions conflict, then Searching for:

[0070]

[0071] by As a new logical starting point, a conflict-free path is re-selected from the subsequent branches of this anchor point to rebuild the execution context, avoiding global lock waits. The Practical Byzantine Fault Tolerance (PBFT) consensus committee, composed of multiple maintenance nodes, performs final confirmation on the solidified DAG topology snapshot.

[0072] The DAG ledger does not serve as a data relay channel for models. Uploaded model parameter updates, user profile summaries, and local training metric summaries are sent directly to the coordinating RSU without passing through the DAG ledger. The ledger only records payload summaries, hashes, signatures, and confirmation results for each stage; it does not store raw traffic data, raw charging data, raw user preference values, raw user profile data, or complete model parameters. Each ledger record includes a record identifier, record type, federation round, region identifier, RSU cluster identifier, creation timestamp, parent record hash set, payload summary, digital signature, and a summary of the verification node signature set.

[0073] DAG record confirmation requires the following conditions to be met:

[0074]

[0075] In the formula To record the number of valid verification signatures obtained by x, The set of RSUs participating in the verification of DAG records. For indicator functions, To verify RSU Records The verification result is 1 when the signature, parent record hash, payload digest, participating entity status, and record type all pass the verification; otherwise, it is 0. The majority confirmation ratio of DAG. To confirm the threshold, Indicates rounding up. When At that time, record The record is confirmed as valid. The smart contract automatically triggers cross-domain clearing and value settlement based on the solidified task execution results, including electric vehicle charging fee settlement and green electricity consumption subsidy allocation. Settlement vouchers are hashed and stored on the blockchain, ensuring full lifecycle traceability and auditability.

[0076] 5. Method Execution Flow

[0077] Figure 3 The diagram shown illustrates the cross-domain data security sharing process for vehicle-road energy cooperation according to the present invention. The specific steps are as follows:

[0078] S601: Initialize the vehicle-road-energy cooperative network, and each node collects raw local data.

[0079] S602: The roadside unit trains a traffic flow prediction model based on a time-graph convolutional network, the charging station trains a charging load prediction model based on historical data, and the vehicle-side unit trains a path energy consumption prediction model based on the maximum entropy and energy consumption model. The model parameters are used as transactions and referenced by a Markov chain Monte Carlo random walk to write unconfirmed transactions at the end of the DAG into the DAG blockchain.

[0080] S603: Each node selects and references two unconfirmed transaction tips at the end of the DAG ledger using an MCMC random walk to create directed reference edges and writes them into the DAG topology. Model parameter transactions are asynchronously aggregated and updated based on topological reference relationships. The aggregation weight is determined by the accumulated weight of the reference edges. Each node participates in federated aggregation on an equal footing. The accumulated weight drives the contribution ranking. Upon submission, the referenced preceding transactions are mutually witnessed and verified.

[0081] S604: Construct an instantaneous absorption boundary model based on the min operator on the grid side, and determine the upper limit of green electricity absorption for each region at the current moment by taking the current total charging power, the remaining absorption space of local energy storage and the rated capacity of the distribution transformer as inputs.

[0082] S605: Construct a model for estimating the probability of green energy co-transfer based on the spatial flow direction of maximum entropy and the co-transfer probability of green energy between regions, and calculate the probability of green energy co-transfer between regions based on the spatial distance and absorption space differences between regions.

[0083] S606: Based on the photovoltaic output forecast and load gap of each region, combined with the upper limit of green electricity consumption and the probability of spatial transfer, a differentiated charging price signal is generated. When there is sufficient consumption space, the price is reduced to attract vehicles. When the consumption space is tight, the price is increased to guide vehicles to other regions, realizing the spatial and temporal migration of vehicles to green electricity production areas.

[0084] S607: RSU performs dynamic spatiotemporal flow prediction of the entire road network based on T-GCN to obtain the real-time travel time of each road segment; it establishes a multi-service queuing model for each charging station, using the number of charging piles as the number of service stations and the predicted traffic flow as the arrival rate, to dynamically simulate the disorderly queuing behavior of vehicles and solve the queuing waiting time at the charging pile.

[0085] S608: The vehicle-side system uses maximum entropy estimation and dynamic energy consumption model, combined with the vehicle's current state of charge, to determine the achievable range and driving energy consumption for each candidate path.

[0086] S609: Introducing the average cumulative weight of a node's references across the entire network in the DAG topology and its recent task execution success rate, a weighted dynamic reliability score is generated as a penalty factor. A comprehensive cost function is constructed, and the charging station and path with the minimum comprehensive cost are selected as the charging path mapping result.

[0087] S610: Generates a reliability score based on the cumulative weight of the node's DAG and its historical execution success rate. Nodes with sufficient computing power and stable execution will gradually increase their weight and naturally improve their priority; nodes with insufficient computing power or frequent failures will automatically decrease their weight and correspondingly decrease their priority.

[0088] S611: When a node fails to complete a task due to insufficient computing power or network failure, extract the cumulative weight distribution of all unconfirmed transaction tips in the current DAG topology, and start a new MCMC random walk for re-anchoring, starting from the most recently successfully submitted transaction of the node.

[0089] S612: The new position of a task in the DAG topology after re-anchoring is determined by the current cumulative weight distribution across the entire network. After re-anchoring, nodes with higher weights gain higher global priority; conversely, nodes with lower weights have lower priority, making way for tasks on higher-weight nodes to free up computing resources.

[0090] S613: Under the DAG architecture, each property owner maintains an independent state trajectory line. Settlement transactions of different owners are completely parallel and do not interfere with each other. New settlement transactions only need to reference the previous two healthy transactions to be attached to the DAG graph.

[0091] S614: When conflicts exist in the read / write sets of settlement transactions, local conflict detection is triggered. Conflict resolution employs a topology backtracking strategy to resolve conflicts and avoid global lock waiting.

[0092] S615: The PBFT consensus committee, composed of multiple maintenance nodes, provides final confirmation. The smart contract automatically triggers cross-domain clearing and value settlement based on the solidified task execution results, including electric vehicle charging fee settlement and green electricity consumption subsidy allocation. Settlement vouchers are hashed and stored on the blockchain, ensuring full lifecycle traceability and auditability.

[0093] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network-side device, etc.) to execute the cell handover method described in the various embodiments of the present invention.

[0095] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A method for secure cross-domain data sharing in a vehicle-road cooperative manner, characterized in that, To address the challenges of data sharing difficulties among multiple stakeholders, heterogeneous computing power, and cross-domain settlement in the coordination of vehicles, charging piles, and power grids, as well as in the consumption of green energy, this method includes the following steps: S1: An asynchronous federated learning method for vehicle-road energy based on DAG topology weaving; S2: A grid-level dispatching method based on instantaneous absorption and electricity price guidance; S3: A vehicle-road layer mapping method based on road network prediction and charging station queuing; S4: Weight recovery method based on node cumulative weight and random walk re-anchoring; S5: A cross-domain settlement method based on independent state trajectory lines and smart contracts.

2. The method according to claim 1, characterized in that, In step S1, each node trains its local model in parallel and asynchronously. The roadside unit trains a traffic flow prediction model based on a time-graph convolutional network, the charging station trains a charging load prediction model based on historical data, and the vehicle-side unit trains a path energy consumption prediction model based on a maximum entropy and energy consumption model. The model parameters are written into the DAG blockchain as transactions via a Markov chain Monte Carlo random walk referencing unconfirmed transactions at the end of the DAG. During the random walk, when a particle starts from the current node and chooses its next path, the probability of each candidate successor node being selected is determined by the cumulative weight of that successor node in the DAG topology, so that the walk path naturally leans towards the direction with higher overall network trust. The parameters are asynchronously aggregated according to the topological relationship, and the weight is determined by the cumulative weight of the referenced edges. Each node participates equally without a centralized server. The higher the frequency of reference, the higher the cumulative weight, forming a contribution ranking. When submitting, the preceding transactions are mutually witnessed.

3. The method according to claim 1, characterized in that, In step S2, a green electricity time-shaving scheduling model is constructed at the grid end: taking the total charging power and energy storage absorption capacity as inputs, the upper limit of green electricity consumption in each region is determined based on the min operator; in response to the problems caused by strong random fluctuations in photovoltaic power generation, such as the midday surge in photovoltaic power generation leading to power backflow and voltage rise, and the superposition of photovoltaic power generation and charging peak in the evening causing extremely high load ramp-up rates, resulting in transformer overload and accelerated insulation aging, a maximum entropy space flow direction and collaborative transfer probability model is constructed. Differentiated electricity prices are generated based on photovoltaic power output and load gap to guide vehicles to consume green electricity, thereby achieving peak shaving and valley filling and reducing net load fluctuations.

4. The method according to claim 1, characterized in that, In step S3, a charging path mapping model is constructed at the vehicle-road layer: the roadside unit performs dynamic spatiotemporal flow prediction of the road network based on the time-graph convolutional network to obtain the road segment travel time; A multi-service queuing model is established for charging stations, using the number of charging piles as the number of service stations and predicted traffic flow as the arrival rate to solve for the waiting time at the charging piles. For the vehicle, a maximum entropy and energy consumption model is used, combined with the state of charge, to determine the driving range and path energy consumption. A reliability score is generated by introducing the average cumulative weight of the node's DAG and the recent success rate. The comprehensive cost is calculated by adding the cross-domain transmission cost to the reliability-amplified values ​​of travel time, waiting time, and electricity price, and selecting the minimum cost as the charging path.

5. The method according to claim 1, characterized in that, In step S4, a reliability score is generated based on the cumulative weight of the node DAG and the historical success rate, and the priority of low-weight nodes is reduced, while high-weight nodes are given higher priority. When a node's weight decreases due to timeout or failure, the weight distribution of currently unconfirmed transactions is extracted, and a random walk is performed to re-anchor the node, starting from the transaction of that node. The new position is dynamically determined by the weight of the entire network, thus realizing the recovery of high-weight nodes and the demotion of low-weight nodes.

6. The method according to claim 1, characterized in that, In step S5, based on DAG blockchain data, independent settlement of ledgers between cross-domain entities is achieved. A new transaction can be attached to the DAG by referencing two healthy transactions. When read and write sets overlap, high-weight transactions are submitted first, while low-weight transactions backtrack along the reference edges of the DAG to find an ancestor node that has no read / write conflict with the high-weight transaction to resolve the conflict. Finally, the practical Byzantine fault-tolerant consensus committee confirms that the smart contract automatically triggers cross-domain clearing and green electricity subsidy settlement, and the entire lifecycle of the certificate on the chain is traceable.

Images