Mobile charging maintenance method for three-dimensional heterogeneous traffic sensor network

Abstract

The invention provides a mobile charging maintenance method for a three-dimensional heterogeneous traffic sensor network. The mobile charging maintenance method comprises the following steps: S1, acquiring multi-source basic data; s2, constructing a road network driving graph model; s3, reconstructing a three-dimensional heterogeneous node and constructing an expansion graph; s4, generating a multi-algorithm joint optimization charging scheduling strategy; and S5, a charging instruction is issued and executed by the unmanned vehicle. According to the method, the graph model is constructed based on the real road network, so that the path planning of the mobile charging platform strictly follows the traffic road network topology constraint, the problem that a traditional model is disjointed with an actual scene is avoided, and the charging task can be executed on the ground.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of intelligent transportation, wireless sensor networks, and wireless power transfer, and in particular to a mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks. Background Technology

[0002] With the increasing demand for real-time road traffic condition perception in Intelligent Transportation Systems (ITS), Traffic Wireless Sensor Networks (TWSNs) are widely used in tasks such as traffic flow monitoring and road condition assessment, offering advantages such as low power consumption, low cost, and ease of deployment. Figure 2 As shown, the node deployment of a typical TWSN exhibits heterogeneous characteristics: sensor nodes are mostly deployed on the road surface or below ground, relying on battery power. Manually replacing batteries is time-consuming, costly, and disruptive to traffic. At the same time, uneven energy distribution among nodes can easily lead to premature failure. Relay nodes are erected at high locations such as streetlight poles, relying on existing roadside infrastructure for deployment. They are powered by solar energy and equipped with energy storage units, but they may still run out of energy in scenarios with insufficient sunlight, which in turn affects data aggregation and network connectivity.

[0003] To address the energy replenishment issue of TWSN nodes, wireless charging technology is gradually replacing manual battery swapping, and it is divided into two categories: static and dynamic. Static wireless charging has high deployment costs and limited coverage, making it difficult to meet the needs of large-scale TWSNs. Dynamic wireless charging, which uses mobile platforms such as unmanned vehicles (UGVs) and unmanned aerial vehicles (UAVs) for on-demand energy replenishment, has become a research hotspot. However, existing dynamic charging technologies have significant drawbacks: first, they ignore road network constraints and plan paths based on free-plane models, which cannot adapt to the requirement that mobile platforms need to travel along roads in real traffic scenarios; second, they lack three-dimensional deployment models and do not consider... Figure 2 The height difference between the sensor nodes and relay nodes makes it difficult to accurately model energy transmission and mobile energy consumption; thirdly, the heterogeneity of nodes is not considered, and the two types of nodes with significant differences in function and importance are homogenized, which makes it impossible to achieve differentiated power replenishment and affects the overall stability of the network. Summary of the Invention

[0004] This invention provides a mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks to address one or more shortcomings of existing TWSN mobile charging technologies.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks includes the following steps: S1. Acquire real road data, traffic wireless sensor network (TWSN) related information and unmanned vehicle parameters for the target area. The real road data includes road centerline, intersection nodes, and lane direction. The TWSN related information includes the geographic coordinates of sensor nodes, relay nodes and service stations, the real-time energy status and node type of sensor nodes and relay nodes, and the unmanned vehicle parameters include mobile energy consumption characteristics and charging power. S2. Based on the real road data and the geographic coordinates of the TWSN, construct an executable graph model driven by the road network. Specifically, the process involves parsing real road data to obtain an initial road network map, mapping the sensor nodes, relay nodes, and service stations of the TWSN to the nearest roadside or intersection on the initial road network map according to their geographical coordinates, and establishing topological connections. The process also involves labeling the length and driving cost attributes of each edge of the road network map based on the mobile energy consumption characteristics of the autonomous vehicle. S3. In the executable graph model Based on the geographic coordinates, real-time energy status of sensor nodes and relay nodes of TWSN, and node types, spatial reconstruction of the three-dimensional heterogeneous nodes is performed and necessary points are set to obtain an extended map. Specifically, with Based on the road network topology, a unified three-dimensional coordinate system is established to represent the three-dimensional positions of sensor nodes and relay nodes; energy thresholds are set according to the real-time energy status of sensor nodes and relay nodes, and nodes below the threshold are marked as nodes to be charged; the priority of each node to be charged is set according to the node type and importance; The nodes waiting to be charged along the roadside are processed into independent vertices, and these independent vertices are added to the system. In the middle, we obtain the extended graph. The newly added independent vertex is a necessary point for unmanned charging vehicles; S4. Based on extended graph Based on the topology and attribute parameters, a comprehensive optimization objective function is constructed. The objective function includes a mobile energy consumption model based on the mobile energy consumption characteristics of unmanned vehicles, a wireless energy transmission efficiency model based on the charging power and three-dimensional spatial relationship of unmanned vehicles, a node priority model, and path feasibility constraints. The solution is obtained by jointly solving the problem using reinforcement learning algorithm and intelligent optimization algorithm, thereby generating a charging scheduling strategy for unmanned vehicles. S5. The charging scheduling strategy is converted into a control command and sent to the unmanned vehicle. The unmanned vehicle then follows its own mobility energy consumption characteristics. It travels along the planned route, charging the nodes at the required points according to the charging power specified in the instructions, and returns to the service station when it has completed charging at all nodes or when its own energy is below the safety threshold.

[0006] In this specification, the specific process of mapping the TWSN sensor nodes, relay nodes, and service stations to the initial road network map in step S2 is as follows: Calculate the projected distance from each sensor node, relay node, and service station to all road edges, and select the road edge or intersection with the shortest distance as the mapping object; if a node is mapped to the midpoint of a road edge, establish a topological connection with the two endpoints of that road edge; if mapped to an intersection, directly add it to the road network topology to ensure the unmanned vehicle travels along the road. The road network allows you to reach the corresponding node simply by moving. The spatial reconstruction provides basic road network support.

[0007] In this specification, the specific method for establishing a unified three-dimensional coordinate system in step S3 is as follows: the height value in the three-dimensional coordinates of the sensor node is not greater than 0, and the height value in the three-dimensional coordinates of the relay node is 3-6 meters. The distance and height difference between the node and the autonomous vehicle are characterized through three-dimensional spatial relationships. The road network edge attributes are used to calculate the three-dimensional path cost, which is then used for subsequent calculation of wireless power transmission efficiency and estimation of mobile energy consumption.

[0008] In this specification, the specific steps of vertexization processing in step S3 are as follows: based on The roadside information is used to locate the nearest roadside to the charging node, and a projection point is inserted into the geometric representation of that roadside. The original roadside is divided into two segments, and independent vertices are created for the charging nodes. These independent vertices are connected to the projected points, and the connecting edges are labeled with attributes such as road distance, parking cost, and accessibility. This ensures that the charging locations are within the... It has clear path semantics and maintains consistency with Consistency of road network topology.

[0009] In this specification, the energy threshold mentioned in step S3 includes an energy warning threshold and an energy depletion threshold. When the real-time energy status of a sensor node or relay node is lower than the energy warning threshold, the independent vertex corresponding to that node is included in the set of necessary nodes. When it is lower than the energy depletion threshold, it is marked as an emergency high-priority node. The priority setting is based on the node's remaining energy ratio, node type (relay nodes have higher priority than sensor nodes), and node importance. Importance is based on the node's position in the network. and The weights of data transmission or sensing functions in the network topology are determined.

[0010] In this specification, the construction of the comprehensive optimization objective function in step S4 also takes into account the energy decay patterns of sensor nodes and relay nodes, the self-charging characteristics of autonomous vehicles, and the impact of different node failures on the network, while also incorporating... Road network operating costs and The three-dimensional spatial loss is considered to ensure that the objective function takes into account energy utilization efficiency, network stability, and mobility costs.

[0011] In this specification, the reinforcement learning algorithm described in step S4 uses the TD3 algorithm or the DDPG algorithm, and the intelligent optimization algorithm uses a genetic algorithm or a particle swarm optimization algorithm. During the solution process, the following methods are employed: Based on the topological structure, combined with The system addresses road network constraints while simultaneously handling discrete node access order decisions and continuous charging duration and movement speed adjustment decisions, generating a scheduling scheme that balances feasibility and efficiency.

[0012] In this specification, the charging scheduling strategy described in step S4 includes: the node access order of the autonomous vehicle, the stopping time and charging power configuration at each mandatory stop, the driving path and energy consumption estimation for each road segment, and the timing of the autonomous vehicle returning to the service station, wherein the driving path is based on... The connection relationship with The road network attributes are planned and generated to ensure that the path conforms to real traffic constraints.

[0013] In this specification, step S2 constructs the executable graph model. Then, if the target area has a dense road network, for Perform graph compression to reduce subsequent processing costs. The construction complexity; step S3 yields the extended graph. Subsequently, the energy status and road accessibility information of sensor nodes and relay nodes are synchronized in real time and updated synchronously. and The attribute parameters are used to ensure that the timeliness of the two graph models matches the actual scene.

[0014] In this specification, if in step S3 it is detected that some necessary points are inaccessible or the road is closed, based on Road network connectivity adjustment The topology is handled by strategies such as delaying charging, adjusting the node access order, or sending a repair notification to the backend system to ensure that the feasibility of the charging scheduling strategy is not affected.

[0015] In summary, the present invention has at least the following beneficial effects: The executability of scheduling paths is significantly improved: By constructing a graph model based on the real road network, the path planning of the mobile charging platform strictly follows the traffic network topology constraints, avoiding the problem of traditional models being out of touch with the actual scenario, and ensuring that charging tasks can be implemented.

[0016] Enhanced model adaptability to real-world environments: A three-dimensional heterogeneous node spatial model is established to accurately depict the spatial relationships and energy transmission characteristics of nodes at different heights. Combined with node vertexization processing, the accessibility of charging points is made clearer, effectively reducing the risk of charging failure caused by spatial modeling deviations.

[0017] Improved overall network stability and reliability: By modeling node heterogeneity and prioritizing mechanisms, nodes performing critical functions are provided with differentiated power replenishment, optimizing network energy distribution, reducing the probability of critical node failure, and enhancing the long-term continuous operation capability of TWSN.

[0018] Energy utilization efficiency and scheduling efficiency optimization: By incorporating path planning and charging decisions into a unified intelligent optimization framework, ineffective movement and redundant docking are reduced. Under the premise of ensuring that all low-energy nodes can be recharged, efficient coordination of energy conversion and scheduling execution is achieved, and the performance is better than traditional single algorithm solutions. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks involved in this invention.

[0020] Figure 2 This is a schematic diagram of the TWSN node deployment involved in this invention.

[0021] Figure 3 This is a schematic diagram of the method flow involved in this invention.

[0022] Figure 4 This is a schematic diagram of the distribution construction of the road network and TWSN involved in this invention.

[0023] Figure 5 This is a schematic diagram extending the figures involved in this invention.

[0024] Figure 6 This is a schematic diagram of the charging and maintenance process involved in this invention. Detailed Implementation

[0025] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] like Figure 1 As shown, this embodiment provides a mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks, including the following steps: S1. Acquire real road data, traffic wireless sensor network (TWSN) related information and unmanned vehicle parameters for the target area. The real road data includes road centerline, intersection nodes, and lane direction. The TWSN related information includes the geographic coordinates of sensor nodes, relay nodes and service stations, the real-time energy status and node type of sensor nodes and relay nodes, and the unmanned vehicle parameters include mobile energy consumption characteristics and charging power. S2. Based on the real road data and the geographic coordinates of the TWSN, construct an executable graph model driven by the road network. Specifically, the process involves parsing real road data to obtain an initial road network map, mapping the sensor nodes, relay nodes, and service stations of the TWSN to the nearest roadside or intersection on the initial road network map according to their geographical coordinates, and establishing topological connections. The process also involves labeling the length and driving cost attributes of each edge of the road network map based on the mobile energy consumption characteristics of the autonomous vehicle. S3. In the executable graph model Based on the geographic coordinates, real-time energy status of sensor nodes and relay nodes of TWSN, and node types, spatial reconstruction of the three-dimensional heterogeneous nodes is performed and necessary points are set to obtain an extended map. Specifically, with Based on the road network topology, a unified three-dimensional coordinate system is established to characterize the three-dimensional positions of sensor nodes and relay nodes. The nodes waiting to be charged along the roadside are processed into independent vertices and connected to the network. The road network topology is connected, and an energy threshold is set according to the real-time energy status of sensor nodes and relay nodes. Nodes below the threshold are marked as must-pass points, and the priority of each must-pass point is set according to the node type and importance. S4. Based on extended graph Based on the topology and attribute parameters, a comprehensive optimization objective function is constructed. The objective function includes a mobile energy consumption model based on the mobile energy consumption characteristics of unmanned vehicles, a wireless energy transmission efficiency model based on the charging power and three-dimensional spatial relationship of unmanned vehicles, a node priority model, and path feasibility constraints. The solution is obtained by jointly solving the problem using reinforcement learning algorithm and intelligent optimization algorithm, thereby generating a charging scheduling strategy for unmanned vehicles. S5. The charging scheduling strategy is converted into a control command and sent to the unmanned vehicle. The unmanned vehicle then follows its own mobility energy consumption characteristics. It travels along the planned route, charging the nodes at the required points according to the charging power specified in the instructions, and returns to the service station when it has completed charging at all nodes or when its own energy is below the safety threshold.

[0027] In some embodiments, the specific process of mapping the sensor nodes, relay nodes, and service stations of the TWSN to the initial road network map in step S2 is as follows: calculate the projected distance from each sensor node, relay node, and service station to all road edges, and select the road edge or intersection with the shortest distance as the mapping object; if the node is mapped to the midpoint of the road edge, establish a topological connection with the two endpoints of that road edge; if it is mapped to an intersection, directly add it to the road network topology to ensure that the unmanned vehicle travels along the road. The road network allows you to reach the corresponding node simply by moving. The spatial reconstruction provides basic road network support.

[0028] In some embodiments, the specific method for establishing a unified three-dimensional coordinate system in step S3 is as follows: the height value in the three-dimensional coordinates of the sensor node is not greater than 0, and the height value in the three-dimensional coordinates of the relay node is 3-6 meters. The distance and height difference between the node and the autonomous vehicle are characterized by three-dimensional spatial relationships. The road network edge attributes are used to calculate the three-dimensional path cost, which is then used for subsequent calculation of wireless power transmission efficiency and estimation of mobile energy consumption.

[0029] In some embodiments, the specific steps of the vertexization process in step S3 are as follows: based on The roadside information is used to locate the nearest roadside to the charging node, and a projection point is inserted into the geometric representation of that roadside. The original roadside is divided into two segments, and independent vertices are created for the charging nodes. These independent vertices are connected to the projected points, and the connecting edges are labeled with attributes such as road distance, parking cost, and accessibility. This ensures that the charging locations are within the... It has clear path semantics and maintains consistency with Consistency of road network topology.

[0030] In some embodiments, the energy threshold in step S3 includes an energy warning threshold and an energy depletion threshold. When the real-time energy status of a sensor node or relay node is lower than the energy warning threshold, the independent vertex corresponding to that node is included in the set of necessary nodes. When it is lower than the energy depletion threshold, it is marked as an emergency high-priority node. The priority is set based on the node's remaining energy ratio, node type (relay nodes have higher priority than sensor nodes), and node importance. Importance is based on the node's position in the network. and The weights of data transmission or sensing functions in the network topology are determined.

[0031] In some embodiments, the construction of the comprehensive optimization objective function in step S4 also refers to the energy decay patterns of sensor nodes and relay nodes, the self-charging characteristics of the autonomous vehicle, and the impact of different node failures on the network, while simultaneously fusing... Road network operating costs and The three-dimensional spatial loss is considered to ensure that the objective function takes into account energy utilization efficiency, network stability, and mobility costs.

[0032] In some embodiments, the reinforcement learning algorithm in step S4 employs the TD3 algorithm or the DDPG algorithm, and the intelligent optimization algorithm employs a genetic algorithm or a particle swarm optimization algorithm. During the solution process, [the algorithm is described in the original text]. Based on the topological structure, combined with The system addresses road network constraints while simultaneously handling discrete node access order decisions and continuous charging duration and movement speed adjustment decisions, generating a scheduling scheme that balances feasibility and efficiency.

[0033] In some embodiments, the charging scheduling strategy in step S4 includes: the node access order of the autonomous vehicle, the stopping time and charging power configuration at each mandatory point, the driving path and energy consumption estimation for each road segment, and the timing of the autonomous vehicle returning to the service station, wherein the driving path is based on... The connection relationship with The road network attributes are planned and generated to ensure that the path conforms to real traffic constraints.

[0034] In some embodiments, step S2 constructs an executable graph model. Then, if the target area has a dense road network, for Perform graph compression to reduce subsequent processing costs. The construction complexity; step S3 yields the extended graph. Subsequently, the energy status and road accessibility information of sensor nodes and relay nodes are synchronized in real time and updated synchronously. and The attribute parameters are used to ensure that the timeliness of the two graph models matches the actual scene.

[0035] In some embodiments, if in step S3 it is detected that the nodes corresponding to some necessary points are inaccessible or the road is closed, based on Road network connectivity adjustment The topology is handled by strategies such as delaying charging, adjusting the node access order, or sending a repair notification to the backend system to ensure that the feasibility of the charging scheduling strategy is not affected.

[0036] In some embodiments, the specific implementation of the road data parsing in step S2 is as follows: using the open-source data interface provided by OpenStreetMap (OSM), the road data of the target area is called through the osmnx library of Python, and the latitude and longitude coordinate sequence of the road centerline, the unique identifier ID of the intersection node and the Boolean value label of the lane direction are automatically extracted. The road centerline is stored in polyline format and the intersection node is represented by latitude and longitude coordinate pairs.

[0037] In some embodiments, the conflict handling rule for node mapping in step S2 is as follows: when multiple TWSN nodes are mapped to the same roadside and the difference in projection distance is less than 5 meters, the projection points of these nodes are merged into a common projection point, a new common vertex is added and connected to the common projection point, and the three-dimensional coordinates of the common vertex are taken as the average of the coordinates of each node, so as to ensure that the unmanned vehicle can replenish power for multiple neighboring nodes in a single stop.

[0038] In some embodiments, the criteria for determining the density of the road network are: the density of road edges in the target area is greater than 0.8 per thousand square meters. The graph compression process adopts an edge merging strategy, which merges adjacent road edges with a length of less than 10 meters that connect the same type of road (such as all being branch roads) into one edge. The length of the merged edge is the sum of the lengths of the original edges, and the driving cost is the weighted average of the driving costs of the original edges (the weight is the proportion of the original edge length).

[0039] In some embodiments, the calibration rules for the unified three-dimensional coordinate system in step S3 are as follows: the lower left corner of the administrative boundary of the target area is taken as the origin of the coordinate system (x=0, y=0), the x-axis extends along the east-west direction, the y-axis extends along the north-south direction, and the coordinate unit is meters; the height value (z-axis) is based on the average altitude of the target area, the z-value of the sensor node is -0.5 meters (deployed under the road surface), the z-value of the relay node is 3-6 meters (streetlight pole height), and the z-value of the unmanned vehicle is 0.5 meters (vehicle charging device height).

[0040] In some embodiments, the position of the projection point insertion in the vertexization process in step S3 is restricted as follows: the distance between the projection point and the intersection nodes at both ends of the road edge is not less than 3 meters. If the distance between the projection point and an endpoint is less than 3 meters, the projection point will coincide with that endpoint, and the original road edge will only be split once to ensure that the road topology is not over-split.

[0041] In some embodiments, the energy threshold in step S3 is determined based on: the energy warning threshold ( The calculation is based on the average energy consumption rate of the node and the single charging task cycle of the autonomous vehicle. ,in The maximum energy storage capacity of the node; the energy depletion threshold ( The minimum startup energy consumption of the node is set to 1.2 times to ensure that the node can still maintain basic communication when an emergency power replenishment is triggered.

[0042] In some embodiments, the quantification method for the node importance in step S3 is as follows: the data transmission weight ranges from 0.3 to 0.6, the sensing function weight ranges from 0.2 to 0.4, and the node type weight (1 for relay nodes and 0.5 for sensor nodes) is fixed. The weighted sum of these three values ​​is the quantified value of the node importance, and the weighting coefficients are... The values ​​are 0.4, 0.3, and 0.3, respectively, and satisfy the following conditions: .

[0043] In some embodiments, the specific correlation model for the mobile energy consumption characteristics of the unmanned vehicle in step S1 is as follows: mobile energy consumption is linearly correlated with path length, and the correlation formula is: =k×L+b, where L is the length of the roadside (unit: meters), k is the energy consumption coefficient (range of 0.05-0.1Wh / m, determined by the battery capacity, load and motor efficiency of the autonomous vehicle), and b is the fixed start-stop energy consumption (value of 5-8Wh / time). This model is used to label the roadside driving cost in step S2.

[0044] In some embodiments, the joint solution mechanism of the reinforcement learning algorithm and the intelligent optimization algorithm in step S4 is as follows: First, an initial charging path sequence (discrete decision) is generated through a genetic algorithm (GA). This sequence must satisfy the coverage of all necessary points and the energy constraints of the unmanned vehicle. Then, the initial path is used as the initial state of reinforcement learning (TD3 algorithm). By adjusting the ratio of charging time (continuous decision) and moving speed, the comprehensive optimization objective function is minimized. The number of iterations is set to 500-1000. When the fluctuation of the reward function value in 20 consecutive iterations is less than 1%, the iteration is stopped and the optimal scheduling strategy is output.

[0045] In some embodiments, the specific dimension of the state space of the reinforcement learning in step S4 is: the dimension of the node feature matrix is ​​N×6 (N is... The total number of nodes in the MC (Multi-Node Array) consists of six features: three-dimensional coordinates (x / y / z), remaining energy percentage, node type identifier, and priority weight. The MC's own state parameters include five dimensions: remaining energy percentage, three-dimensional position coordinates, and current driving speed. The input data of the state space is aggregated through two convolutional layers of a graph neural network (GNN), and the output dimension is a 128-dimensional node embedding vector.

[0046] In some embodiments, the range of values ​​for each balance coefficient in the comprehensive optimization objective function described in step S4 is as follows: ∈[0.3,0.5], ∈[0.3,0.4], ∈[0.2,0.3], which can be adjusted according to the actual needs of the scenario. For example, in a scenario where network stability is the priority, increase it. The value of ; in scenarios where energy efficiency is prioritized, increase The value of .

[0047] In some embodiments, the docking cost in the vertex-processing connection edge attributes in step S3 is calculated as follows: the docking cost of sensor nodes is fixed at 2 minutes (including vehicle start-stop and charging device calibration time), the docking cost of relay nodes is 3 minutes (including high-altitude charging alignment time), and the accessibility flag is obtained through the real-time traffic data interface. If the road is in a temporary closed state, it is marked as "unreachable".

[0048] In some embodiments, the step S10 based on Road network connectivity adjustment The specific method for topology construction is as follows: when the road edge corresponding to a certain necessary point is marked as "unreachable", Dijkstra's algorithm is used to search for it. The nearest alternative road to this node, in An alternative connection edge is added, and the travel cost of the alternative connection edge is 1.5 times that of the original edge, ensuring that the scheduling strategy can still cover this essential point.

[0049] In some embodiments, the specific model for calculating wireless power transmission efficiency based on the three-dimensional spatial relationship in step S3 is as follows: ,in The ideal charging efficiency is 0.85 when the distance is 0. The distance attenuation coefficient (with a value of 0.002) ), The height difference attenuation coefficient (with a value of 0.1) ), Let denoted as MC and denoted as Δz, and let Δz be the height difference between them. This model is used to accurately calculate the charging energy loss of nodes at different heights.

[0050] The technical concept of this invention is as follows: This invention addresses the shortcomings of existing TWSN mobile charging technologies by proposing a comprehensive solution integrating road network constraints, 3D modeling, and intelligent scheduling. The core idea is as follows: First, an executable graph model is constructed based on real road data, mapping TWSN nodes to the road network and labeling their attributes to ensure path planning conforms to traffic constraints. Second, considering the heterogeneous 3D deployment characteristics of nodes, a unified 3D coordinate model is established, vertex-based processing is performed on nodes attached to roadside areas, and low-energy nodes are marked as necessary points, clarifying charging accessibility and coverage requirements. Finally, a comprehensive optimization objective is constructed, incorporating factors such as node priority, 3D energy transmission efficiency, and mobile energy consumption. Combining reinforcement learning and intelligent optimization algorithms, the access order, stopping time, and charging parameters of the mobile charging platform are jointly solved, generating a directly executable differentiated charging scheduling strategy to achieve efficient and sustainable energy replenishment for TWSN nodes.

[0051] This invention aims to address the problems of difficult energy replenishment for sensor nodes and relay nodes, unexecutable charging paths, and underutilization of node heterogeneity in three-dimensional heterogeneous traffic wireless sensor networks (TWSN) in traffic scenarios. It proposes a mobile charging maintenance method and system for three-dimensional heterogeneous traffic sensor networks to achieve efficient, executable, and sustainable remote charging scheduling for unmanned vehicles.

[0052] To overcome the shortcomings of existing technologies in terms of road network constraints, spatial modeling, and energy replenishment strategies, the main objectives of this invention include: 1) Construct an executable graph model to realize charging path planning based on road network.

[0053] This invention addresses the characteristic of autonomous vehicles in traffic scenarios that they must move along road networks. It proposes a method to construct a graph structure between service stations, sensor nodes, and relay nodes using real road data (such as OpenStreetMap). By establishing a path planning model that satisfies road topology constraints, the mobile charging task can be executed in real urban road networks, ensuring the accuracy of energy consumption estimation.

[0054] 2) Establish a three-dimensional deployment model to realize an energy replenishment mechanism for three-dimensional space.

[0055] In response to the heterogeneous deployment characteristics of sensor nodes located below the road surface and relay nodes located high on light poles in three dimensions, this invention constructs a three-dimensional wireless energy transmission model and a node proximity model, which can accurately describe the node height difference, three-dimensional positional relationship, and the resulting changes in energy transmission efficiency, thereby improving the model's adaptability to real traffic environments.

[0056] 3) Incorporate node heterogeneity to achieve differentiated charging and maintenance strategies.

[0057] This invention incorporates the functional differences, energy consumption differences, and importance of sensor nodes and relay nodes into a unified scheduling framework, and constructs a priority model based on node type, enabling relay nodes and key nodes to obtain higher power replenishment priority, thereby improving the stability and data reliability of the entire TWSN.

[0058] This invention constructs a graph model based on a real road network, establishes a node representation method in three-dimensional space, and combines intelligent optimization algorithms to jointly solve the charging path and charging strategy of unmanned vehicles, so as to achieve efficient energy replenishment for sensor nodes and relay nodes. Figure 3 A flowchart illustrating the method of the present invention is provided, wherein... For service stations, that is, the starting point of driverless vehicles, , , , The arrows indicate the paths taken by the autonomous vehicles, serving as charging nodes. The entire method comprises the following three main stages: (1) Road network-driven graph structure construction In the first stage of this invention, the system constructs an executable graph model of a Traffic Wireless Sensor Network (TWSN) based on real road data (such as OpenStreetMap). The specific process is as follows: Road data parsing: Obtain structured data such as road centerlines, intersection nodes, and lane directions in the target area, and parse it into an initial road network map composed of nodes and edges.

[0059] Node spatial mapping: Based on the geographical coordinates of sensor nodes, relay nodes and service stations, they are mapped to the nearest edge or intersection in the road network, and a connection relationship consistent with the road network topology is established so that the movement of autonomous vehicles meets the constraints of real traffic roads.

[0060] Attribute labeling: Record the length, travel cost (such as distance or time) and other attributes of each edge in the road network to provide basic information for subsequent route planning.

[0061] Through the above steps, the present invention constructs a TWSN graph under road topology constraints. It can accurately describe the feasible movement paths of autonomous vehicles in traffic scenarios.

[0062] (2) Spatial reconstruction and mandatory point setting of three-dimensional heterogeneous nodes Since sensor nodes in traffic scenarios are typically deployed on the road surface, while relay nodes are installed high on streetlight poles, this invention further reconstructs the spatial structure based on the initial graph to accurately reflect the characteristics of three-dimensional heterogeneous deployment. The main steps include: 3D location modeling: The actual heights of sensor nodes and relay nodes (e.g., sensor node 0m, street light pole relay node +3–6m) are incorporated into a unified coordinate system to construct a 3D coordinate (x,y,z), which is then used to calculate the 3D path cost and wireless power transmission efficiency model.

[0063] Vertexization of nodes to be charged: Since sensor nodes and relay nodes are usually attached to road edges, this invention separates the nodes to be charged from the road edges, creates them as independent vertices, and maintains their connection with the corresponding road edges. This allows the charging location to have clear reachability and path semantics in the graph.

[0064] Mandatory Vertex Labeling: All nodes to be charged are set as mandatory vertices in the graph. The autonomous vehicle must visit these vertices when solving the charging strategy to ensure that all low-energy nodes are recharged.

[0065] After this processing step, the original image Expanded into a graph containing all three-dimensional charging points This provides a high-precision spatial representation for subsequent scheduling strategies.

[0066] (3) Solving the mobile charging scheduling strategy based on optimization and reinforcement learning After the graph structure is constructed and the 3D position model is completed, this invention uses an intelligent optimization method to solve the charging task of the unmanned vehicle, so as to maximize the charging energy efficiency while ensuring that all nodes to be supplied are supplied. The solution steps are as follows: A cost function is established for the path planning problem: a comprehensive optimization objective function is constructed based on the autonomous vehicle's mobile energy consumption model, charging energy model, node priority model, and three-dimensional transmission efficiency model, with constraints including autonomous vehicle power limit, path feasibility, and mandatory node visits.

[0067] The invention employs a multi-source optimization method to solve the scheduling strategy. For complex three-dimensional charging decisions, the invention can combine the following intelligent optimization algorithms to achieve the solution: reinforcement learning methods (such as TD3, DDPG), genetic algorithm (GA), particle swarm optimization (PSO), etc.

[0068] By incorporating both path planning and charging decisions into the solution framework, this invention can obtain a comprehensive scheduling scheme that takes into account path cost, node importance, and energy efficiency.

[0069] The system ultimately generates an executable unmanned vehicle charging task, including: 1) the order of accessing nodes; 2) the stopping time and charging power configuration at each node; and 3) the timing and path for the unmanned vehicle to return to the service station.

[0070] Road Network-Driven Graph Structure Construction: The first stage of this invention aims to construct a road topology graph model based on real road data, which can be used for mobile charging scheduling of autonomous vehicles. This model forms the underlying graph structure of a Traffic Wireless Sensor Network (TWSN). The core objective of this stage is to ensure that the autonomous vehicle's operating path strictly meets the constraints of the real traffic network and to provide an accurate spatial basis for subsequent path planning and charging decisions. Figure 4 A schematic diagram of the road network-driven graph structure construction is provided. This stage includes the following steps: (1) Road data analysis This invention uses real road data sources (such as OpenStreetMap, OSM) to obtain traffic road structure information for the target area. The system automatically extracts the following structured elements through a data interface: Road centerline (polyline): describes the geometry of the road.

[0071] Junction / intersection nodes: These serve as the basic vertices of the graph model.

[0072] Road connectivity: Describes the drivable relationship between each road and its adjacent roads.

[0073] Lane direction, road type, road grade and other attributes (optional): used to set the movement constraints or movement costs of autonomous vehicles.

[0074] The parsed road structure is organized in the form of node sets and edge sets, forming a preliminary road topology skeleton.

[0075] (2) Construction of road network map model The parsed road data is mapped to an undirected graph G=(V,E), where each intersection is represented as a vertex. ,in This represents the coordinates of the intersection in a geographic coordinate system (such as WGS84 or UTM). Each road segment is represented as an undirected edge. ,in Indicates from the intersection At the intersection The passable road.

[0076] (3) Spatial mapping of TWSN nodes The traffic wireless sensor network of this invention includes sensor nodes, relay nodes, and service stations. These nodes are not necessarily located at intersections in real geographic space, thus requiring spatial mapping: Node location acquisition: Obtain the geographic coordinates of each sensor node Si, relay node Cj, and service station according to the deployment plan. Mapping to the nearest road element: Perform the following operations for each node: Calculate its projection points to all road edges, select the road edge or intersection with the shortest distance, establish the association between the node and the road network, and establish topological relationships (maintaining road network constraints): If the node falls at the midpoint of a road edge, first establish topological connections with the two endpoints of that road edge (further "vertexization" will be discussed in the next section). If mapped to an intersection, it is directly added to the road network topology. Through the above mapping, TWSN nodes and the road network form a unified topology, requiring autonomous vehicles to move along the road structure to reach the nodes.

[0077] (4) Road structure attribute labeling To support subsequent path planning and energy consumption calculation, this invention adds the following attributes to each road edge: edge length, mobile energy consumption coefficient, autonomous vehicle speed limit, road type (main road / secondary road / local road), and whether it allows two-way traffic (this invention defaults to an undirected graph, which can be extended to a directed graph).

[0078] Spatial Reconstruction and Mandatory Point Setting of 3D Heterogeneous Nodes: To accurately reflect the highly heterogeneous characteristics of nodes in traffic scenarios and ensure the feasibility of charging tasks in real road networks, such as... Figure 5 As shown, the road network map generated in the first stage of this invention Based on (V,E), further 3D spatial reconstruction (vertexization) is performed on sensor nodes attached to the roadside or under the road surface and relay nodes installed on elevated facilities, and all nodes requiring power replenishment are marked as mandatory vertices. This process includes three parts: 3D position modeling, vertexization of edge nodes, and labeling of mandatory vertices. Figure 5 The expansion of the diagram adds new vertices, which are the sensors to be charged and the routing nodes, i.e., the nodes that the wireless charging vehicle must pass through. , , , ).

[0079] 3D position modeling 1) Definition of a triplet for TWSN: TWSN = (S, A, BS), in: A set of sensor nodes; For the set of relay nodes; BS stands for Base Station / Service Station.

[0080] 2) Node 3D coordinate representation: Each sensor node: ,1≤i≤n, where This indicates the roadside ID where the node is located. Let z be its three-dimensional coordinates (for ground level, z ≤ 0).

[0081] Each relay node: ,1≤j≤m, where The ID of the roadside to which the relay node belongs. It represents three-dimensional coordinates (usually z>0).

[0082] 3) Three-dimensional Euclidean distance metric Given any two points and Its three-dimensional Euclidean distance is defined as: This three-dimensional distance is used for subsequent energy consumption estimation and wireless transmission loss modeling.

[0083] Vertexization of nodes to be charged along the road: To incorporate nodes to be charged along the road into the graph structure and ensure clear path semantics, the following steps are performed: (1) Node projection and nearest road segment location: For any node u∈S∪A, calculate its projection point to the road edge e∈E. And select the road edge e* corresponding to the minimum projection distance: ; Where dist(·) is the shortest Euclidean distance from the point to the line segment (considering the z-component to distinguish between underground and high poles).

[0084] (2) Create independent vertices in the graph: For each node u with energy to be replenished, add a new graph vertex. Its coordinates are .

[0085] (3) Establish connection with road topology (insert connection edge): It is recommended to use the "insert split point" method: insert projection points on the geometric representation of road edge e*. , to the original edge Divided into and Then and Connected edges. The weight of the connecting edge is the arc length along the road or the travel cost.

[0086] (4) Connector edge attributes: Add attributes to the newly added connector edge: connector_length: the distance along the road from the road to the node; docking_cost: the time / energy cost of the unmanned vehicle stopping and starting at the node; accessibility_flag: accessibility flag (such as one-way / closed / temporary construction, etc.).

[0087] After processing, the extended graph is obtained. .

[0088] Mandatory Vertices Labeling and Set Definition: (1) Threshold triggering rule: Let the remaining energy of the node be... .when (For example When (=40%), node u is added to the set to be charged; when (For example When the percentage is 10%, the node is marked as an urgent high-priority node.

[0089] (2) Definition of the set of necessary vertices: .

[0090] (3) Priority weight definition: for each Assign priority weights to ∈M ,For example: ; in The indicator function is 1 if u is a relay node, and 0 otherwise. criticality(u) represents the importance of the node (such as the criticality of an intersection). These are the weighting coefficients.

[0091] Graph extension: From arrive The mathematical relationships and final graph definitions of the initial road map: .

[0092] Extended graph after vertexization: , , in For the sensor vertex set, For relay vertex set, For service sites.

[0093] Edge set ,in This is the newly added set of connecting edges.

[0094] Each edge e∈ in the graph Each vertex has a weight function c(e) to measure the cost (energy consumption, time, or a linear combination of both) for the autonomous vehicle to traverse that edge; each charging vertex With docking costs (Time / Energy Consumption) and Charging Efficiency Parameters (Related to the height z of u).

[0095] Implementation notes and optional optimizations: Vertex processing can be performed offline or inserted online as needed; graph compression can be used to accelerate the solution when the road network is dense; strategies (delay / reordering / repair notification) are required for unreachable or closed nodes; node energy, road reachability, and other information need to be synchronized in real time to ensure... Timeliness.

[0096] For example: roadside Sensors on Its projection point is p. Insert a vertex in method A. Split e* into and New and Connected. If ≤ ,but ∈M. The scheduler in Search for the minimum cost path covering M and output the charging docking order and charging time.

[0097] Solving the mobile charging scheduling strategy based on reinforcement learning: Problem formalization and optimization objective: in the extended graph Within this framework, the scheduling problem is formalized as a sequential decision-making process.

[0098] 1. Node Energy Dynamic Model: Energy of each node u∈M(t) to be charged It decays over time and is replenished when accessed by MC. Its discrete-time update formula is modified as follows: ; For node energy consumption rate; This is an indicator function; MC is 1 when node u is being charged at time t, and 0 otherwise. The efficiency function for three-dimensional wireless charging is the three-dimensional Euclidean distance between the MC and the node. and height difference This function explicitly characterizes the significantly different energy transfer losses when charging underground or high-altitude nodes, representing a significant upgrade to the two-dimensional model. This represents the transmit power of the MC.

[0099] 2. MC Energy Dynamic Model: MC Energy The core constraints on its mobility, charging, and self-charging behavior are: c(e) represents the energy consumption of traversing road edges or connecting edges e (based on length, slope, and traffic conditions); U(t) represents the set of nodes that are charging at time t. An indicator function for MC to recharge itself at a service station; Power to charge service stations.

[0100] 3. Comprehensive Optimization Objective: The objective of this invention is not simply cost minimization, but rather a balanced optimization of multiple objectives. The composite reward function R(t) at time t is defined as the immediate manifestation of the optimization objective: ; EUE(t) (Energy Usage Efficiency): The ratio of the energy effectively charged into the node within period t to the total energy consumed by the MC (movement, charging loss, self-charging), reflecting the energy conversion efficiency; : The weighted loss of new dead nodes within period t. Inherited from the previous definition, ensure that the death of high-priority nodes (such as critical relays and intersection sensors) is subject to greater penalties; The movement cost of a MC can be characterized by path length or time. A balancing factor is used to adjust the relative importance of energy efficiency, network survivability, and mobility costs; the ultimate goal of reinforcement learning is to maximize the long-term cumulative discounted reward. This indirectly optimizes the long-term overall performance of the system.

[0101] Hybrid Action Space Solving Based on Deep Reinforcement Learning: To handle the hybrid decision-making process of "discrete path selection" and "continuous charging duration," this invention employs an improved framework based on the TD3 algorithm, whose design fully considers the three-dimensional road constraint map. Structural information.

[0102] 1. State-space design state The dynamic environment and constraints need to be fully coded, including: Graph structure information: represented in the form of a node feature matrix. The attributes of all nodes, such as 3D coordinates (x, y, z), type (sensor / relay), and current energy. Priority weight Whether it belongs to M(t), etc.; the state of MC itself: including remaining energy. 3D position and current velocity; global and local relationships: aggregated through a graph neural network (GNN) encoder to obtain the topological and spatial relationships between the current MC position and each node to be charged. 2. Definition of Mixed Action Space: Action Defined as: discrete action Starting from the current MC position, in Select the next vertex to visit. The action space consists of all necessary vertices M(t) and service stations. This is essentially selecting the next target on the road constraint map; continuous actions. : is a two-dimensional vector, representing ① the planned charging time for the target node. , and ② the percentage of movement speed that MC expects to adjust before reaching the target or after charging (to cope with changes in traffic flow). The charging time decision is jointly influenced by the node's missing energy and the three-dimensional charging efficiency η.

[0103] 3. Policy Networks and Value Networks: Policy Network (Actor): Input State Output mixed motion The network first processes the graph state using a GNN to generate node embeddings; then, through an attention mechanism, it integrates the MC's own state to calculate the probability distribution (discrete) of selecting each candidate node as the next target, and generates the corresponding charging time and speed parameters (continuous) in parallel; the value network (Critic) uses a double Q-network structure to prevent overestimation. Input state and actions Assess its long-term value The network also utilizes GNNs to understand graph structures, ensuring that value judgments are based on global topological relationships. 4. Training Mechanism: The TD3 algorithm is used for offline or online training. This is achieved through MC and a simulation environment (based on...). Interacting with dynamic models to collect experience trajectories During training: The Critic network learns accurate value estimation by minimizing temporal difference error; the Actor network optimizes the policy by maximizing the Q-value of the Critic evaluation, thereby learning how to make optimal decisions on "where to charge" and "how long to charge" under the complex three-dimensional constraints of terrain, altitude, and road network; the reward function R(t) includes a weighted death penalty. The direct guidance strategy prioritizes high-value nodes, enabling intelligent scheduling based on business needs.

[0104] Scheduling strategy generation and output: After training, the policy network can generate the optimal scheduling instructions based on the real-time input status (node ​​energy, MC location, road conditions, etc. synchronized by the monitoring system).

[0105] 1. Online decision-making and path generation: Given the current state The policy network outputs the next target vertex. and charging parameters. The path planning module then... Call the shortest path algorithm (such as A*) to calculate the distance from the current position of MC to the destination. The specific driving path (consisting of a series of edges); 2. Complete Task Plan: The system connects a series of discrete decisions with paths to form a complete charging task plan, including: 1) an ordered vertex visit sequence. ;2) Planned charging time for each target point;3) Segment travel time and energy consumption estimated based on real-time edge weight c(e); 3. Command Issuance and Execution. The plan is translated into control commands and issued to the MC. The MC moves along the planned path, performs a charging operation for a specified duration upon reaching each vertex, and returns to the service station when its energy falls below a threshold or after completing its task. .

[0106] Key technical points of this invention: (1) Executable graph construction method based on real road network: This invention uses real road data (such as OpenStreetMap) to parse the road centerline, intersection and lane direction, maps sensor nodes, relay nodes and service stations to the road network and establishes a topologically consistent connection relationship to form an executable graph model that satisfies traffic constraints, so as to realize the feasible path planning of unmanned vehicles on real urban roads.

[0107] (2) Spatial reconstruction and mandatory point setting mechanism of three-dimensional heterogeneous nodes: In view of the three-dimensional heterogeneous characteristics such as sensor nodes located on the road surface and relay nodes installed at the height of the pole, this invention establishes a unified three-dimensional coordinate model and performs "vertexization" on the nodes to be replenished attached to the roadside, making them independent reachable points. At the same time, all nodes that need to be replenished are marked as mandatory points to ensure that the scheduling path can cover all low-energy nodes.

[0108] (3) Importance modeling and differentiated charging strategy for heterogeneous nodes: This invention comprehensively considers node function, energy consumption and network importance, and constructs a priority model based on node type, so that key nodes (such as relay nodes) can obtain higher charging priority in scheduling, realize differentiated maintenance of three-dimensional heterogeneous TWSN, and improve the overall stability of the network and data reliability.

[0109] (4) The charging scheduling joint solution framework integrating optimization algorithms and reinforcement learning: This invention constructs a comprehensive cost function that includes factors such as mobile energy consumption, three-dimensional transmission efficiency, node priority and path feasibility, and combines reinforcement learning (such as TD3 / DDPG) and optimization algorithms (GA, PSO) to jointly solve the access order, stopping time and charging power of unmanned vehicles, and generate an optimal mobile charging scheduling strategy that can be directly executed.

[0110] This invention constructs an executable graph model based on a real road network, enabling mobile charging vehicles to plan their routes strictly according to the road network topology. This avoids the difficulty of implementing traditional two-dimensional modeling and significantly improves the executability and accuracy of scheduling routes. The invention establishes a three-dimensional heterogeneous node model, unifying ground sensor nodes and high-pole relay nodes into a three-dimensional coordinate system. It also innovatively performs vertexization processing on nodes to be charged, ensuring clear reachability for all charging points. This improves the model's adaptability to real traffic environments and effectively reduces the risk of network failure due to insufficient energy at critical nodes. Through node heterogeneity modeling and a priority mechanism, this invention allows the charging strategy to automatically tilt towards relay nodes and important nodes, resulting in a more reasonable overall network energy distribution, lower node mortality, and more stable system operation, significantly improving the continuous working capability of TWSNs. This invention unifies route planning and charging decisions within an optimization and reinforcement learning framework. It can improve energy utilization efficiency while satisfying the charging needs of all nodes, reducing ineffective movement and redundant stops, and quickly generating executable scheduling schemes. Overall performance is superior to traditional heuristic or single optimization algorithms.

[0111] In some embodiments, reinforcement learning (TD3), genetic algorithm (GA), and particle swarm optimization (PSO) are used to form a three-dimensional fusion interaction mechanism, as detailed below: I. Core Logic of Algorithm Integration This embodiment achieves deep integration of TD3, GA, and PSO through a closed-loop architecture of "path generation - weight optimization - parameter fine-tuning - feedback iteration". 1. The genetic algorithm (GA) is responsible for generating initial charging paths (discrete decision) that satisfy the road network constraints and the coverage of necessary points, providing a feasible solution space for subsequent optimization; 2. Particle Swarm Optimization (PSO) optimizes the energy consumption weight coefficient of the path based on the path energy consumption data generated by GA (through bidirectional interaction with GA), thereby improving the energy consumption adaptability of the path. 3. Reinforcement learning (TD3) incorporates the optimal path of GA and the optimization weights of PSO into the state space, optimizes the ratio of charging time to movement speed (continuous decision-making), and feeds back the optimized energy consumption to GA and PSO to adjust the direction of subsequent iterations (bidirectional interaction with the former two). 4. Through multiple rounds of iteration, the three systems ultimately output a scheduling strategy that balances path feasibility, energy efficiency, and precise energy replenishment.

[0112] II. Construction, Training and Application of Algorithm Fusion Models (a) Step 1: Genetic Algorithm (GA) constructs the initial set of charging paths (discrete decision layer) The core function of GA is to break through the local optima of path search, generate an initial charging path that covers all necessary points and conforms to the road network constraints (G0 / G1 topology), and provide a high-quality feasible solution for subsequent algorithms.

[0113] 1. Model Building: Population definition: Let the population of GA be... ,in Population size (value is 50), each individual This represents a charging path. For service sites, The order of visits to essential points (satisfying) ).

[0114] Fitness Function: The fitness function of a GA must simultaneously constrain path feasibility and energy consumption, and is defined as: ; For path Total mobile energy consumption, based on Calculation of road network edge attributes ( roadside (driving costs); For path feasibility indicator function, if Covering all essential points and ensuring the autonomous vehicles have sufficient energy. ( For nodes (the docking cost), then ,otherwise ; For path The number of times a node is repeatedly visited (constraining redundant paths); , , The fitness weight coefficients of GA (satisfying) This is used to balance energy consumption, feasibility, and path simplicity.

[0115] Genetic operations: Selection: Roulette wheel selection is used, fitness value The higher the individual's height, the greater the probability of being selected for the next generation; Crossover: Partial Mapping Crossover (PMX) is used to cross over the two paths. and The intermediate node segments are swapped to ensure that the path still conforms to the road network topology after the crossover; Mutation: the node order of two non-service stations in the path is randomly swapped to avoid premature convergence of the population.

[0116] 2. Model Training: Initializing the Population : Randomly generate 50 paths that meet the feasibility constraints (by...) Topological traversal ensures coverage of essential points); iterative training: set the number of iterations. Each iteration performs selection, crossover, and mutation operations to generate a new generation of population. Convergence condition: When the optimal fitness value is reached after 10 consecutive iterations. When the fluctuation is less than 1%, stop training and output the initial path set. (Take the top 10 paths in terms of fitness).

[0117] 3. First interaction output with PSO The optimal path of GA and its corresponding energy consumption data Node docking cost set As input to PSO, it is used for PSO weight optimization.

[0118] (II) Second step: Particle Swarm Optimization (PSO) to optimize energy consumption weight coefficients (weight optimization layer) The core function of PSO is to dynamically optimize the weight ratio of "road network driving cost" and "node stopping cost" in the path based on the actual energy consumption data of the path provided by GA, so that the path energy consumption calculation is more in line with the real scenario. At the same time, the optimized weights are fed back to GA to adjust the actual effect of its fitness function.

[0119] 1. Model Building Particle definition: Let the particle swarm of PSO be... ,in The number of particles (value is 30), each particle Represents a set of weighting coefficients. As a weight for road network operating costs, As a weight for node docking costs, satisfying ( ).

[0120] Fitness Function: The fitness function of PSO aims to minimize the "weighted energy consumption error" of the optimal path in GA, and is defined as follows: ; Weighted energy consumption (integrating PSO weights and GA path data); The actual test energy consumption of the GA optimal path (obtained through actual measurement on an autonomous vehicle simulation platform, for example) Wh); fitness value The larger the value, the smaller the error between the weighted energy consumption and the actual energy consumption. The better. Particle update formula: PSO finds the optimal weights by updating particle position and velocity. The update formula is: Speed ​​updates: ; Location update: ; For the first Sub-particles velocity vector ( The initial velocity range is ; Inertial weights (to balance global and local search); , These are learning factors (which guide particles toward their own optimal and global optimal directions, respectively). for Random numbers within a range; For particles The historical best position (the weight with the highest fitness). The position is the globally optimal position for the entire particle swarm (the weight with the highest population fitness); position updates are ensured through normalization. .

[0121] 2. Model Training: Initializing the Particle Swarm Optimization : Randomly generate 30 sets that satisfy Weight coefficients; Iterative training: setting the number of iterations The fitness of each particle is calculated in each iteration. ,renew and And update the particle velocity and position using the above formula; convergence condition: when the global optimal fitness is achieved... When the error is greater than 0.95 (i.e., the weighted energy consumption error is less than 5%), stop training and output the optimized weights. .

[0122] 3. Two-way interaction and feedback Positive interaction: Optimization weights of PSO As a dynamic adjustment factor for the GA fitness function, the updated GA fitness function is: This makes the paths subsequently generated by GA more closely match the actual energy consumption characteristics; Reverse feedback: GA path energy consumption data It provides a "real energy consumption benchmark" for PSO, preventing PSO optimization from deviating from actual path scenarios and ensuring the effectiveness of weight optimization.

[0123] (III) Step 3: Reinforcement Learning (TD3) Optimizes Continuous Decision Parameters (Precise Scheduling Layer) The core function of TD3 is to integrate the feasible paths of GA and the optimization weights of PSO to solve the continuous decision optimization problem of "charging time" and "movement speed", and feed the optimization results back to GA and PSO to form a closed loop iteration, ultimately achieving the optimal balance between energy consumption efficiency and energy replenishment effect.

[0124] 1. Model Building: state space The state space needs to fully encode the GA path, PSO weights, and dynamic environment information, defined as: ; The optimal path for GA output (discrete decision input); The optimized weights (weight inputs) are output by PSO. For path The real-time energy state vectors of all nodes in the system; The real-time remaining energy of the driverless car; The three-dimensional position coordinates of the autonomous vehicle ( ); This represents the real-time driving speed of the driverless car.

[0125] Action space The action space is a continuous decision vector, defined as: ; Charging time for the current node (range of values) ); The ratio of movement speed (relative to the road speed limit, with a range of values). (This is used to adjust the speed of the driverless car on the current road segment.)

[0126] reward function The reward function, which integrates the feasibility of the GA path, the energy consumption characteristics of the PSO weights, and the energy replenishment effect, is defined as follows: ; Energy efficiency (integrated three-dimensional charging efficiency) ); For nodes Priority weights (inherited from the definition of S3); For the node death indicator function, if but Otherwise, it is 0; for Energy consumption of autonomous vehicles at all times. Energy consumption for docking; , , The reward weighting coefficient (balancing energy efficiency, network stability, and cost); , The weights optimized for PSO (reflecting the interaction with PSO).

[0127] Strategy Networks and Value Networks: Policy Network (Actor): Input Output action The network structure is "GNN feature extraction layer + fully connected layer". The GNN layer is used to aggregate path topology information, and the fully connected layer outputs continuous actions. Value Network (Critic): Employs a dual-Q network structure ( ),enter The long-term value of output actions To avoid overestimating its value.

[0128] 2. Model Training: Experience Replay Pool Initialization: Set the capacity of the experience replay pool. Storing experience samples Pre-training: Utilizing the GA path Weights of PSO Generate 1000 simulation trajectories and store them in the experience replay pool to complete pre-training; Iterative training: Set the number of iterations. In each iteration, a batch of samples (batch_size=256) is randomly sampled from the replay pool to update the Critic network and the Actor network: Critic update: minimizes the temporal difference error. ,in As a discount factor, Actions output by the target policy network; Actor update: by maximizing Optimize policy network parameters; convergence condition: when the loss function of the dual-Q network... When the value is less than 0.01, stop training and output the optimal policy parameters. Corresponding to optimal continuous decision .

[0129] 3. Multi-directional interaction and closed-loop feedback Interaction with GA: Optimal charging time for TD3 output This will affect the energy replenishment effect of the nodes, and thus update the energy state of the nodes. This serves as a constraint for the next round of population initialization in GA (to avoid generating paths that require excessively long charging times). Interaction with PSO: Energy Efficiency of TD3 As a supplementary benchmark to the PSO fitness function, update the actual energy consumption of the PSO. ( (Energy savings after TD3 optimization) make PSO weight optimization more aligned with the complete scheduling process; Final output: The optimal path for merging GA Optimization weights of PSO Continuous decision-making in TD3 It generates a complete charging scheduling strategy, including "access order + charging duration + driving speed + energy consumption estimation".

[0130] III. Core Contributions and Interaction Descriptions of Algorithm Fusion (I) Core Contributions 1. Overcoming the limitations of a single algorithm: GA solves the path feasibility problem, PSO optimizes the energy consumption weight adaptability, and TD3 solves the problem of accurate optimization of continuous parameters. The integration of the three covers the entire dimension of decision-making of "path-weight-parameter", avoiding the local optimum trap of a single algorithm in complex scenarios. 2. Dynamic closed-loop adaptation: Through a two-way feedback mechanism, the algorithm optimization always fits the real scenario (such as road network constraints, node energy dynamics, and energy consumption characteristics of autonomous vehicles), improving the executability and energy efficiency of the scheduling strategy; 3. Multi-objective optimization: It simultaneously achieves four objectives: "path feasibility, minimum energy consumption, accurate energy replenishment, and network stability", which is superior to the single-objective optimization effect of traditional single algorithms.

[0131] (ii) Verification of pairwise interaction relationships 1. GA↔PSO: GA provides path energy consumption data → PSO optimizes weights → PSO weight feedback adjusts the GA fitness function, forming a two-way closed loop of "path generation - weight optimization - path iteration"; 2. GA↔TD3: GA provides feasible paths → TD3 optimizes continuous parameters based on the paths → TD3 adjusts GA path constraints through nodal energy feedback to ensure that the paths match the energy replenishment requirements; 3. PSO↔TD3: PSO provides energy consumption weights → TD3 calculates weighted energy consumption based on the weights → TD3's energy efficiency feedback adjusts the actual energy consumption benchmark of PSO to ensure that the weight optimization fits the energy replenishment scenario.

[0132] IV. Model Application Examples Assuming there is one service site in the target area. 3 essential points (relay nodes) Sensor nodes , The specific application process is as follows: 1. GA generates an initial path set. Optimal path ,That Wh, Wh; 2. PSO optimizes the weights based on the above energy consumption data to obtain... Weighted energy consumption Wh, relative to actual energy consumption The error of Wh is less than 2%; 3. TD3 Input and Optimization yielded ; 4. Feedback and Adjustments: TD3 Optimization (Increase by 15%), Node (Relay nodes) did not run out of energy, and all paths generated in the next round of GA were retained. Priority access order, PSO update Wh, further optimize the weights to ; 5. Final scheduling strategy: (Speed ​​limited to 1.0x, charging time 180s) → (Speed ​​limited to 0.8x, charging time 120s) → (Speed ​​limited to 0.9x, charging time 90s) → The total energy consumption is 450Wh, the node survival rate is 100%, and the energy utilization efficiency is 0.85.

[0133] V. The new process of the 3D fusion interaction mechanism using three algorithms is as follows: S1. Acquire real road data, traffic wireless sensor network (TWSN) information, and autonomous vehicle parameters for the target area. The real road data includes road centerlines, intersection nodes, and lane directions (for subsequent graph model topology construction). The TWSN information includes the geographic coordinates of sensor nodes, relay nodes, and service stations (for 3D modeling and road network mapping), the real-time energy status of sensor nodes and relay nodes (for mandatory point selection and priority calculation), and node types (for differentiated energy replenishment strategies). The autonomous vehicle parameters include mobile energy consumption characteristics (including a correlation model between mobile energy consumption and path length). ), charging power (used for 3D wireless charging efficiency calculation), and all data serve as the basic input for subsequent graph model construction and algorithm solution.

[0134] S2. Based on the real road data and the geographic coordinates of the TWSN, construct an executable graph model driven by the road network. Specifically, the process involves parsing real road data (such as OpenStreetMap) using Python's osmnx library to extract structured elements such as road centerlines and intersection nodes to obtain an initial road network map. The projected distances from each sensor node, relay node, and service station to all road edges are calculated. The road edge or intersection with the shortest distance is selected as the mapping object. If a node is mapped to the midpoint of a road edge, a topological connection is established between it and the two endpoints of that road edge; if it is mapped to an intersection, it is directly added to the road network topology. A common projection point merging strategy is used to resolve multi-node mapping conflicts. This is combined with the mobile energy consumption characteristics of autonomous vehicles (…). The concentration is 0.05-0.1 Wh / m. (5-8Wh / trip), representing the length of each edge of the road network map and the travel cost attribute; if the target area has a dense road network (density greater than 0.8 roads / thousand square meters), for Perform edge merging and compression processing. Ultimately, as a follow-up The road network infrastructure and the path constraints used in the algorithm solution are based on this.

[0135] S3. In the executable graph model Based on the geographic coordinates, real-time energy status of sensor nodes and relay nodes of TWSN, and node types, spatial reconstruction of the three-dimensional heterogeneous nodes is performed and necessary points are set to obtain an extended map. Specifically: Based on the road network topology, a unified three-dimensional coordinate system is established (the origin is the lower left corner of the administrative boundary of the target area, and the sensor nodes are...). Relay Node Rice, driverless car (meters), using three-dimensional Euclidean distance to characterize the spatial relationship between nodes and autonomous vehicles; for The nodes to be charged attached to the roadside are processed into vertices. After locating the nearest roadside, a projection point is inserted (at a distance of no less than 3 meters from the road endpoint). The original roadside is split and used to create independent vertices for the nodes to be charged. Connecting edges are added to indicate the distance along the road, the stopping cost (2 minutes for sensor nodes, 3 minutes for relay nodes), and the accessibility marker. Thresholds (early warning thresholds) are set based on the real-time energy status of the nodes. = (Average charging interval of autonomous vehicles × Average energy consumption rate of nodes) + 20% Depletion threshold =Node minimum startup energy consumption × 1.2), nodes below the threshold are marked as mandatory nodes; combining the remaining energy consumption ratio of nodes, node type (relay node weight 1, sensor node weight 0.5) and importance (data transmission weight 0.3-0.6, sensing function weight 0.2-0.4), the formula is used to... ( , , Set the priority of each necessary point. Ultimately, it provides complete spatial topology and node attribute support for the algorithm solution.

[0136] S4. Based on extended graph Based on the topology and attribute parameters, a comprehensive optimization objective function is constructed. This objective function includes a mobile energy consumption model based on the mobile energy consumption characteristics of the autonomous vehicle, and a wireless energy transmission efficiency model based on the charging power and three-dimensional spatial relationship of the autonomous vehicle. , , , The node priority model and path feasibility constraints are solved using a combination of reinforcement learning algorithms (TD3 / DDPG) and intelligent optimization algorithms (Genetic Algorithm GA / Particle Swarm Optimization (PSO)). First, an initial set of charging paths is generated using GA (population size 50, 100 iterations). The top 10 paths with the highest fitness are output as the feasible solution space. The optimal path is then... Its energy consumption data is input into PSO; PSO aims to minimize the weighted energy consumption error, using the velocity-position update formula ( , Optimize the weighting of road network driving costs Weighted by docking costs (satisfy ), and feed the optimized weights back to GA to adjust the fitness function; TD3 will and Included in the state space (dimension is the node feature matrix) +5-dimensional autonomous vehicle status), with action space Optimize continuous decision-making through reward functions ( , , To maximize long-term cumulative rewards, the algorithm iterates 2000 times until the double-Q network loss is less than 0.01 and then converges. The three algorithms work together through a closed loop of "path generation - weight optimization - parameter fine-tuning - feedback iteration" to finally generate a charging scheduling strategy that includes node access order, docking time, charging power configuration, driving path and energy consumption estimation.

[0137] S5. The charging scheduling strategy is converted into a control command and sent to the autonomous vehicle. The autonomous vehicle, based on its own mobile energy consumption characteristics, and The planned path (based on) Connection relationship and (Road network attributes generated) Driving, charging at nodes that are mandatory stops according to the charging power specified in the instructions; during the charging process, the node energy status and road accessibility information are synchronized and updated in real time. and The attribute parameters; when a node is detected as unreachable, based on Road network connectivity adjustment The topology is optimized and the scheduling strategy is improved. When all nodes are charged or the autonomous vehicle's own energy is lower than the safety threshold, it returns to the service station along the planned path. At the same time, the actual energy consumption and energy replenishment effect are fed back to the algorithm model in step S4 to provide an iterative basis for subsequent scheduling.

[0138] like Figure 6 As shown, the charging maintenance described in this invention is not a one-time static process, but a periodic maintenance process covering the entire lifecycle of the Traffic Wireless Sensor Network (TWSN). TWSNs typically require long-term continuous operation, with a continuous working time of up to L years (usually L>5 years). To ensure long-term stable network operation, this invention sets a fixed maintenance time period M, for example, M=24 hours.

[0139] At the start of each maintenance cycle, the system checks the energy status of all sensor nodes and relay nodes in the current TWSN, filters out nodes with energy below a preset threshold, and marks them as nodes to be charged in this maintenance cycle. These nodes are designated as mandatory vertices during the maintenance cycle, and an extended graph is dynamically constructed based on this, scheduling unmanned vehicles to perform charging and maintenance tasks.

[0140] Once the current maintenance cycle is complete and all nodes awaiting charging have finished charging, the required vertices are deleted, and the expanded graph is restored to the executable graph model driven by the initial road network. The system then enters a waiting state until the next maintenance cycle begins, repeating the above detection, modeling, and scheduling process. Through this periodic and dynamically updated charging maintenance mechanism, this invention achieves continuous and efficient maintenance of the TWSN throughout its entire lifecycle.

Claims

1. A mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks, characterized in that, include: S1. Acquire real road data, traffic wireless sensor network (TWSN) related information, and unmanned vehicle parameters for the target area. The real road data includes road centerline, intersection nodes, and lane direction. The TWSN related information includes the geographic coordinates of sensor nodes, relay nodes, and service stations, the real-time energy status and node type of sensor nodes and relay nodes, and the unmanned vehicle parameters include mobile energy consumption characteristics and charging power. S2. Based on the real road data and the geographic coordinates of the TWSN, construct an executable graph model driven by the road network. Specifically, the process involves parsing real road data to obtain an initial road network map, mapping the sensor nodes, relay nodes, and service stations of the TWSN to the nearest roadside or intersection on the initial road network map according to their geographical coordinates, and establishing topological connections. The process also involves labeling the length and driving cost attributes of each edge of the road network map based on the mobile energy consumption characteristics of the autonomous vehicle. S3. In the executable graph model Based on the geographic coordinates, real-time energy status of sensor nodes and relay nodes of TWSN, and node types, spatial reconstruction of the three-dimensional heterogeneous nodes is performed and necessary points are set to obtain an extended map. Specifically, with Based on the road network topology, a unified three-dimensional coordinate system is established to represent the three-dimensional positions of sensor nodes and relay nodes; energy thresholds are set according to the real-time energy status of sensor nodes and relay nodes, and nodes below the threshold are marked as nodes to be charged; the priority of each node to be charged is set according to the node type and importance; The nodes waiting to be charged along the roadside are processed into independent vertices, and these independent vertices are added to the system. In the middle, we obtain the extended graph. The newly added independent vertex is a necessary point for unmanned charging vehicles; S4. Based on extended graph Based on the topology and attribute parameters, a comprehensive optimization objective function is constructed. The objective function includes a mobile energy consumption model based on the mobile energy consumption characteristics of unmanned vehicles, a wireless energy transmission efficiency model based on the charging power and three-dimensional spatial relationship of unmanned vehicles, a node priority model, and path feasibility constraints. The solution is obtained by jointly solving the problem using reinforcement learning algorithm and intelligent optimization algorithm, thereby generating a charging scheduling strategy for unmanned vehicles. S5. The charging scheduling strategy is converted into a control command and sent to the unmanned vehicle. The unmanned vehicle then follows its own mobility energy consumption characteristics. It travels along the planned route, charging the nodes at the required points according to the charging power specified in the instructions, and returns to the service station when it has completed charging at all nodes or when its own energy is below the safety threshold.

2. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, The specific process of mapping the TWSN sensor nodes, relay nodes, and service stations to the initial road network map in step S2 is as follows: Calculate the projected distance from each sensor node, relay node, and service station to all road edges, and select the road edge or intersection with the shortest distance as the mapping object; if a node is mapped to the midpoint of a road edge, establish a topological connection with the two endpoints of that road edge; if mapped to an intersection, directly add it to the road network topology to ensure the unmanned vehicle travels along the road. The road network allows you to reach the corresponding node simply by moving. The spatial reconstruction provides basic road network support.

3. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, The specific method for establishing a unified three-dimensional coordinate system in step S3 is as follows: the height value in the three-dimensional coordinates of the sensor node is not greater than 0, and the height value in the three-dimensional coordinates of the relay node is 3-6 meters. The distance and height difference between the node and the autonomous vehicle are characterized by the three-dimensional spatial relationship. The road network edge attributes are used to calculate the three-dimensional path cost, which is then used for subsequent calculation of wireless power transmission efficiency and estimation of mobile energy consumption.

4. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, The specific steps of vertexization processing in step S3 are as follows: Based on The roadside information is used to locate the nearest roadside to the charging node, and a projection point is inserted into the geometric representation of that roadside. The original roadside is divided into two segments, and independent vertices are created for the charging nodes. These independent vertices are connected to the projected points, and the connecting edges are labeled with attributes such as road distance, parking cost, and accessibility. This ensures that the charging locations are within the... It has clear path semantics and maintains consistency with Consistency of road network topology.

5. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, The energy threshold mentioned in step S3 includes an energy warning threshold and an energy depletion threshold. When the real-time energy status of a sensor node or relay node is lower than the energy warning threshold, the independent vertex corresponding to that node is included in the set of necessary points. When the energy level falls below the energy depletion threshold, the node is marked as an emergency high-priority node. Priority is determined based on the node's remaining energy percentage, node type, and node importance. Importance is based on the node's position in the energy hierarchy. and The weights of data transmission or sensing functions in the network topology are determined.

6. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, The construction of the comprehensive optimization objective function in step S4 also takes into account the energy decay patterns of sensor nodes and relay nodes, the self-charging characteristics of autonomous vehicles, and the impact of different node failures on the network, while simultaneously integrating... Road network operating costs and The three-dimensional spatial loss is considered to ensure that the objective function takes into account energy utilization efficiency, network stability, and mobility costs.

7. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, The reinforcement learning algorithm mentioned in step S4 uses the TD3 algorithm or the DDPG algorithm, and the intelligent optimization algorithm uses the genetic algorithm or the particle swarm optimization algorithm. During the solution process, the algorithm employs... Based on the topological structure, combined with The system addresses road network constraints while simultaneously handling discrete node access order decisions and continuous charging duration and movement speed adjustment decisions, generating a scheduling scheme that balances feasibility and efficiency.

8. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, The charging scheduling strategy in step S4 includes: the node access order of the autonomous vehicle, the stopping time and charging power configuration at each necessary point, the driving path and energy consumption estimation for each road segment, and the timing of the autonomous vehicle returning to the service station, wherein the driving path is based on... The connection relationship with The road network attributes are planned and generated to ensure that the path conforms to real traffic constraints.

9. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, Step S2: Construct the executable graph model Then, if the target area has a dense road network, for Perform graph compression to reduce subsequent processing costs. The construction complexity; step S3 yields the extended graph. Subsequently, the energy status and road accessibility information of sensor nodes and relay nodes are synchronized in real time and updated synchronously. and The attribute parameters are used to ensure that the timeliness of the two graph models matches the actual scene.

10. The mobile charging maintenance method for three-dimensional heterogeneous traffic sensor networks according to claim 1, characterized in that, In step S3, if it is detected that some necessary points are inaccessible or the road is closed, based on Road network connectivity adjustment The topology is handled by strategies such as delaying charging, adjusting the node access order, or sending a repair notification to the backend system to ensure that the feasibility of the charging scheduling strategy is not affected.

Images