Autonomous planning of agent-driven power grid equipment patrol path optimization method

Abstract

The application relates to the technical field of inspection path optimization, in particular to a power grid equipment inspection path optimization method driven by autonomous planning intelligent agents. The application proposes the following scheme: based on global maps and demand information of power grid equipment, a to-be-inspected cluster containing shared equipment is generated through multi-objective clustering, and an overlapping cluster hypergraph is established with the shared equipment as a hyperedge. The initial value upper bound of the hyperedge is calculated to screen a candidate set, a cost game is performed on the to-be-inspected cluster corresponding to the shared equipment, and a joint inspection path maximizing the income is generated according to the auction pricing and VCG payment mechanism. The application can realize the collaborative optimization of tasks among multiple clusters under the conditions of geographical constraints, energy consumption and time efficiency, effectively avoid the problems of repeated coverage, inspection conflict and resource waste in traditional path planning, and improve the global income and execution efficiency of intelligent inspection.

Description

Technical Field

[0001] This application relates to the field of inspection path optimization technology, and in particular to a method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent. Background Technology

[0002] Existing power grid equipment inspection route planning methods typically rely on fixed rules or a single optimization objective for task allocation and route generation, making it difficult to achieve global coordination under conditions of multiple devices, multiple regions, and multiple constraints. Traditional route planning results often suffer from overlapping inspection routes, low resource utilization, and uneven distribution of inspection time, leading to repeated inspections of some devices while delaying or omitting some high-risk devices. For equipment clusters with spatial aggregation or geographical constraints, existing methods struggle to balance the risk priority and energy consumption limitations of inspection tasks, easily resulting in a decline in overall inspection efficiency. Even when some solutions introduce intelligent optimization algorithms, the lack of task collaboration and dynamic game mechanisms still fails to effectively balance the distribution of inspection benefits and costs across different regions, resulting in insufficient overall system benefits and limitations on the timeliness and coverage of inspection operations.

[0003] To address the above issues, this application proposes a method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent. Summary of the Invention

[0004] The technical problem this application aims to solve is to address the shortcomings of existing technologies by providing a method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent. Based on a global map and power grid equipment demand information, it generates clusters of equipment to be inspected that include shared devices through multi-objective clustering, and establishes an overlapping cluster hypergraph with shared devices as hyperedges. By calculating the initial value upper bound of the hyperedges, a candidate set is selected, and cost game theory is performed on the clusters of equipment to be inspected corresponding to the shared devices. Based on auction pricing and VCG payment mechanisms, a joint inspection path that maximizes revenue is generated, thereby improving overall inspection efficiency.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A method for optimizing power grid equipment inspection paths driven by an autonomous planning agent is applied to an inspection planning system. The inspection planning system includes an autonomous planning agent, multiple power grid devices, and inspection equipment. The autonomous planning agent is configured with a global map. The method includes:

[0007] Based on the global map and the demand information of the power grid equipment to be inspected, a multi-objective clustering is performed using a clustering algorithm to obtain multiple clusters to be inspected, wherein each cluster to be inspected includes at least one shared device.

[0008] Using the cluster to be inspected as the vertex, if there is a shared device between at least two clusters to be inspected, the shared device is used as a hyperedge to connect the corresponding clusters to be inspected, generating an overlapping cluster hypergraph, and the candidate set is obtained by calculating the upper bound of the initial value of the hyperedge.

[0009] Perform iterative operations, select the current shared device from the candidate set, and determine the joint inspection path that maximizes the benefits based on the cluster to be inspected connected by the hyperedge corresponding to the current shared device through cost game. Simultaneously delete the power grid devices in the joint inspection path from the cluster to be inspected, update the overlapping cluster hypergraph and the candidate set, until the candidate set is an empty set.

[0010] The final inspection path is generated by summarizing the joint inspection paths from each iteration and the power grid equipment to be inspected that has not been deleted from the cluster to be inspected.

[0011] The required information for the power grid equipment to be inspected includes at least one of the following:

[0012] Equipment operating status parameters, including voltage, current, temperature, partial discharge signal, and historical health index;

[0013] Equipment risk weight parameters are used to characterize the potential failure probability, environmental risk level, and load importance of equipment;

[0014] Inspection timeliness parameters are used to limit the latest response time for equipment inspection tasks;

[0015] Geographic constraint parameters include the geographical coordinates of the equipment, the boundaries of its region, and safety distance limits.

[0016] The multi-objective clustering method using a clustering algorithm yields multiple clusters to be inspected, including:

[0017] Calculate the geographical distance matrix between power grid equipment based on the geographical constraint parameters, and construct a multidimensional feature vector corresponding to the geographical distance matrix based on the equipment risk weight parameters;

[0018] The different feature dimensions of the multidimensional feature vector are normalized and weighted using the equipment operating status parameters and the inspection timeliness parameters to obtain a weighted feature vector.

[0019] Based on the weighted feature vector, the power grid equipment is clustered using a density clustering algorithm to obtain multiple initial clusters to be inspected. During the clustering process, a sharing threshold is preset. When the feature similarity of any power grid equipment relative to at least two initial clusters to be inspected is greater than or equal to the sharing threshold, the power grid equipment is simultaneously assigned to the corresponding initial cluster to be inspected and is used as a shared equipment.

[0020] Calculate the average risk index and geographical compactness index of each initial cluster to be inspected to update the initial clusters to be inspected, resulting in multiple clusters to be inspected.

[0021] The generation of the overlapping cluster hypergraph includes:

[0022] Multiple clusters to be inspected, obtained through clustering algorithms, are used as a set of vertices, and the elements in each vertex correspond to the power grid equipment to be inspected in the cluster.

[0023] Traverse the vertex elements and extract the set of shared devices in each vertex. If at least two vertices have the same shared device, establish a hyperedge between the corresponding vertices. The hyperedge is used to represent the shared inspection association between the clusters to be inspected.

[0024] The overlapping cluster hypergraph is formed by combining all hyperedges and their corresponding vertex sets, wherein each hyperedge connects at least two vertices, the hyperedge corresponds to at least one shared device, and the hyperedge is weighted according to the demand information of the shared device.

[0025] The process of obtaining the candidate set by calculating the initial upper bound of the superedge value includes:

[0026] For each hyperedge in the overlapping cluster hypergraph, extract the set of shared devices corresponding to the hyperedge and the cluster center coordinates of multiple clusters to be inspected to which each shared device belongs in the set of shared devices;

[0027] Calculate the overall risk-return value of the shared device set based on the weights corresponding to the hyperedges;

[0028] Based on the geographical distance between the cluster center coordinates, the average energy consumption within each cluster, and the preset resource parameters, the driving cost between the vertices connected by the hyperedge is calculated, wherein the resource parameters include the inspection equipment's range, energy capacity, and operating time window.

[0029] The difference between the comprehensive risk-return value and the driving cost value is used as the initial value upper bound of the hyperedge, wherein the initial value upper bound is used to characterize the maximum net return obtained by forming a joint inspection path by multiple vertices connected by the hyperedge under the constraint conditions.

[0030] Sort the values ​​of the initial value upper bound from high to low and generate a candidate set.

[0031] The order in which the current shared device is selected from the candidate set is top-down.

[0032] The joint inspection path that maximizes benefits through cost game theory includes:

[0033] Using the current shared device as the anchor point, multiple clusters to be inspected, including the anchor point, are identified as participants. Each participant generates multiple candidate sub-path packages based on the demand information of the power grid equipment within the cluster. Each candidate sub-path in the candidate sub-path package satisfies the following: the candidate sub-path includes the current shared device, and the covered power grid equipment comes from the power grid equipment within the corresponding participant's cluster; the candidate sub-path satisfies preset constraints and includes corresponding travel time, energy consumption, and waiting time estimates.

[0034] The candidate sub-paths in each candidate sub-path package are priced through an auction mechanism, and the priced candidate sub-paths are submitted to the autonomous planning agent as auction items. The price of each candidate sub-path is determined by the risk-reward value and execution cost of the candidate sub-path. The risk-reward value is determined based on the sum of the fault prevention benefits of the power grid equipment in the candidate sub-path, and the execution cost includes travel time, energy consumption and waiting time.

[0035] Based on the pricing of each candidate sub-path, multiple candidate sub-paths are selected to generate candidate joint inspection paths;

[0036] The candidate joint inspection path is settled and allocated through the VCG payment mechanism to obtain the joint inspection path.

[0037] The step of selecting multiple candidate sub-paths based on the pricing of each candidate sub-path to generate a candidate joint inspection path includes:

[0038] Prioritize each candidate sub-path based on its pricing, and select the candidate sub-path with the highest pricing as the main path;

[0039] Starting from the anchor point in the main path, each candidate power grid device in the main path is traversed sequentially, and a replacement device that can replace the candidate power grid device is selected from the remaining candidate sub-paths. The selection of the replacement device is based on the marginal benefit of the replacement device, which represents the difference in pricing on the main path after the replacement device replaces the candidate power grid device.

[0040] Candidate joint inspection paths are obtained by replacing the corresponding candidate power grid equipment with all alternative equipment.

[0041] The process of settling and allocating costs for the candidate joint inspection paths through the VCG payment mechanism to obtain the joint inspection paths includes:

[0042] For each participant, calculate the marginal contribution to the candidate joint inspection path, where the marginal contribution is the proportion of the risk-reward value and execution cost of the power grid equipment included in the participant in the candidate joint inspection path;

[0043] The total technical benefits of the candidate joint inspection path are calculated based on the marginal contributions of all participants.

[0044] Based on the total technical benefits, a payment is allocated to each participant through the VCG payment mechanism to adjust the payment cost for each participant. The final path is determined based on the payment allocation results, and a joint inspection path is output.

[0045] Updating the overlapping cluster hypergraph and the candidate set includes:

[0046] For the clusters to be inspected after deletion, the number of devices, average risk weight, geographic center coordinates and resource occupancy status of each cluster to be inspected are recalculated to update the characteristic information of the clusters to be inspected.

[0047] Based on the updated cluster feature information to be inspected, the overlapping cluster hypergraph is locally structurally adjusted, wherein the local structural adjustment includes: deleting the corresponding hyperedges; and recalculating the shared device set and the comprehensive risk-reward value corresponding to the shared device set based on the remaining hyperedges.

[0048] The initial value upper bound is updated, and the hyperedges are reordered to generate an updated candidate set. If the updated candidate set is not empty, the overlapping cluster hypergraph with local structure adjustment and the candidate set are input into the next iteration.

[0049] Compared with the prior art, the beneficial effects of this application are:

[0050] This application achieves joint path optimization among multiple inspection clusters by introducing autonomous planning agents and a cost-game mechanism, enabling the inspection planning of power grid equipment to dynamically balance risk-reward, energy consumption constraints, and timeliness. By constructing an overlapping cluster hypergraph containing shared equipment, it realizes the correlation modeling and resource collaborative allocation among inspection tasks, significantly reducing the probability of duplicate coverage and blind spot omissions. Combining auction pricing and VCG payment mechanisms, the path selection possesses interpretable economic rationality and global optimality, thereby improving the resource utilization rate and overall operational efficiency of power grid inspection. Attached Figure Description

[0051] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0052] Figure 1 An exemplary application scenario diagram provided for an embodiment of this application;

[0053] Figure 2 This is a schematic diagram of the inspection planning system provided in the embodiments of this application;

[0054] Figure 3 A flowchart illustrating the method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent, as provided in this application embodiment;

[0055] Figure 4 This is a schematic diagram of the structure of the cluster to be inspected provided in an embodiment of this application;

[0056] Figure 5 This is a schematic diagram of the structure of the overlapping cluster hypergraph provided in the embodiments of this application;

[0057] Figure 6 This is a schematic diagram illustrating the principle of main path replacement provided in an embodiment of this application.

[0058] Figure 7 This is a schematic diagram of the main path replacement result provided in an embodiment of this application. Detailed Implementation

[0059] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0060] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0061] To facilitate understanding of the background and application scope of this application, the specific implementation methods are described below in conjunction with typical power grid operation and maintenance scenarios.

[0062] In daily inspections and emergency response tasks targeting mixed indoor and outdoor areas of substations, transmission corridors, and distribution transformer areas, the inspection targets exhibit a combination of high density, strong constraints, and high timeliness:

[0063] Within the same geographical unit, there is both close-range visual inspection of live equipment and cross-section infrared thermography, partial discharge listening, and local insulation observation. Outdoor routes are often restricted by restricted areas, terrain obstructions, and weather windows, while indoor routes are constrained by fire lanes and human-machine co-location rules. In practice, operating units typically deploy heterogeneous carriers such as wheeled ground robots and vehicle-mounted work platforms to perform collaborative tasks. Task requirements are usually dynamic (such as increased patrols triggered by protection action alarms, load surges, and weather warnings), while operation windows, take-off and landing points, and battery resources change constantly. The aforementioned complexity is amplified by a common phenomenon: the same equipment node may be candidate-assigned multiple times under different business scopes or different geographical clusters, resulting in implicit competition and redundant coverage across regions and carriers.

[0064] Existing methods mostly adopt two approaches: one is to take static geographic clustering as a guide, solve the path within each cluster, and then eliminate the conflict through cross-cluster mediation; the other is to directly solve the global journey on a large-scale graph using global vehicle path planning or reinforcement learning strategies.

[0065] The former often results in excessively high coordination costs when the same device is absorbed by multiple clusters at the same time, and generally leads to path pulling, repeated visits and response delays. The latter, although theoretically optimal, is difficult to meet the timeliness requirements of engineering deployment in terms of computational cost and interpretability when non-convex constraints such as time windows, no-fly zones, energized safety distances, and temporary construction barriers are superimposed, and it is even more difficult to replan stably when tasks are injected in a streaming manner.

[0066] The common dilemma is that once the shared equipment is delayed until the end of the solution process, the early allocation and path patterns have already been solidified, and no matter how much fine-tuning is done later, it is difficult to avoid resource waste and time loss.

[0067] The core processing logic of this application is based on the aforementioned common dilemma. It does not rely on a single work area, grid area, or pre-defined inspection subdomain division. Instead, it first recognizes the natural occurrence of cross-cluster sharing of power grid equipment under multi-objective clustering, and pre-emptively establishes this sharing relationship as the triggering factor for the solution. To this end, after obtaining the global map and task requirements, an overlapping cluster hypergraph is constructed, with the cluster to be inspected as the vertices and the set of shared equipment as the hyperedges, transforming the shared equipment from a source of conflict into an anchor point. A cluster-level alliance game is then triggered around the anchor point:

[0068] Instead of individual bidding, participating clusters submit candidate sub-path packets that must pass through anchor points, cover high-value subsets of their own cluster, and meet time windows and safety rules. These are then auctioned and spliced ​​using a combination of technical benefits and execution costs to form a joint inspection path spanning multiple clusters and passing through anchor points. Unlike conventional routing that involves first dividing and then adjusting, this embodiment places merging, pricing, splicing, clearing (deleting points), and re-clustering within a rolling outer loop. Each time a path is generated, covered nodes are simultaneously removed, dynamically shrinking the problem scale and avoiding large-scale conflict resolution later.

[0069] It should be noted that, due to the essential requirements of fairness and stability in the project, this application adopts a game-theoretic auction and settlement logic:

[0070] Different inspection vehicles and teams naturally differ in resources, flight time, and risk preferences. A purely greedy merging approach can easily lead to high-capability vehicles repeatedly absorbing high-value areas, causing uneven scheduling and execution bottlenecks. This embodiment introduces a settlement allocation system with incentive compatibility after the joint path is determined, ensuring that the optimal response of each participant no longer deviates from the global optimum, thereby mitigating the fluctuations in the plan caused by strategic reporting. Simultaneously, the combined potential energy of the remaining uncovered risk penalty and movement / waiting costs is used as the convergence characterization of the outer loop, ensuring that each round of clearing brings provable overall improvement and guarantees stable, engineering-executable results within minute-level rolling time slots.

[0071] It should be emphasized that the above description is not an exhaustive list of algorithmic details, but rather clarifies the technical starting point and reproducible implementation boundaries of this embodiment:

[0072] Given a known global map and multi-source task flow, and with constraints uniformly coded, this approach achieves joint path generation and stable convergence across clusters, carriers, and time windows by constructing an overlapping cluster hypergraph and anchor-triggered alliance game, coupled with a rolling mechanism of synchronous point deletion and re-clustering. This transforms the conflict caused by sharing into shared-driven collaboration, fundamentally resolving the triple challenges of redundant coverage, replanning overhead, and fair allocation in existing technologies. It provides a unified solution framework for the large-scale, routine, and emergency inspection of power grid equipment.

[0073] refer to Figure 1 , Figure 1 This is an exemplary application scenario diagram provided for an embodiment of this application.

[0074] like Figure 1 As shown, the autonomous planning intelligent agent-driven power grid equipment inspection path optimization method of this application can be deployed in an inspection planning system, which interacts with multiple power grid devices and inspection equipment through a communication link.

[0075] Figure 1 Various types of power grid equipment and multiple inspection devices are shown, such as substations or high-voltage distribution switchgear, which represent power grid equipment with high-voltage switching, protection relay and measurement and control functions; distribution boxes or control terminal equipment, which represent power grid equipment distributed at various branch nodes and undertaking local monitoring and segmented control tasks; and instrument panels or detection panels, which represent power grid equipment with signal acquisition, temperature measurement or partial discharge monitoring functions.

[0076] Figure 1 The middle arrows represent information flow and command flow, used to illustrate the data interaction path between power grid equipment and the inspection planning system.

[0077] Understandably, the inspection equipment can be ground-based inspection robots, drones, or vehicle-mounted inspection terminals, such as... Figure 1 Inspection equipment 1 to inspection equipment N; power grid equipment may include substation secondary equipment, distribution ring network cabinets, line switch boxes, metering devices, etc. Figure 1 This is merely an exemplary application scenario to illustrate the overall deployment logic and technical principles of this application. In actual implementation, adjustments and expansions can be made based on site conditions and power grid structure.

[0078] refer to Figure 2 , Figure 2 This is a schematic diagram of the inspection planning system provided in an embodiment of this application.

[0079] Figure 2 The inspection planning system is shown to include a communication module, a controller, and an autonomous planning agent, wherein:

[0080] The communication module is used to enable data exchange between the inspection planning system and external devices, including uploading the operating status information of power grid equipment, issuing inspection tasks, and providing feedback on execution status. The communication module supports multiple communication protocols, such as industrial Ethernet, 4G / 5G cellular communication, private network wireless links, or satellite communication, to ensure reliable data transmission under different power scenarios.

[0081] The controller is the core unit that connects the data flow and the decision flow. The controller can also interact with the actuators of the inspection equipment to complete task initiation, path import and progress monitoring.

[0082] The autonomous planning agent is the core decision-making module of the system. Based on the global map and the demand information of power grid equipment, it performs cluster analysis, construction of overlapping cluster hypergraphs, calculation of hyperedge value, and cost game solving. After receiving the inspection requirements, the autonomous planning agent can generate joint inspection paths through game-theoretic path planning algorithms and return the results to the controller for task scheduling. It can be deployed on a central server, cloud computing platform, or edge computing node to support multi-task parallel optimization and real-time replanning.

[0083] In some optional implementations, the communication module, controller, and autonomous planning agent can be integrated into the same computing node or deployed in a distributed manner across different network layers. For example, the controller and autonomous planning agent can be deployed at a regional control center, while a lightweight communication module can be configured on the field inspection terminal to achieve coordinated operation of remote intelligent scheduling and local path optimization.

[0084] Figure 2 Solid arrows indicate the data flow between systems, while dashed arrows indicate the interaction between the communication module and the outside world.

[0085] Next, with reference to the accompanying drawings, the method for optimizing power grid equipment inspection paths driven by autonomous planning intelligent agents, as provided in the embodiments of this application, will be further described. Figure 3 The method shown is applied to an inspection planning system, which includes an autonomous planning agent, multiple power grid devices, and inspection equipment. The autonomous planning agent is configured with a global map. The method includes:

[0086] S1: Based on the global map and the demand information of the power grid equipment to be inspected, multi-objective clustering is performed through clustering algorithm to obtain multiple clusters to be inspected;

[0087] The cluster to be inspected includes at least one shared device;

[0088] In this embodiment, the global map includes the spatial coordinates, connectivity, restricted areas, and feasible passage information of the power grid equipment. The demand information for the power grid equipment to be inspected is dynamically generated by the operation scheduling terminal and may include equipment operating status parameters, risk weight parameters, inspection timeliness parameters, and geographical constraint parameters. Based on this information, a density clustering algorithm is used to perform multi-objective clustering of the equipment. A sharing threshold is set during the clustering process. When the feature similarity of a certain device simultaneously meets the clustering conditions of multiple clusters, the device is simultaneously assigned as a shared device to different clusters to be inspected.

[0089] Those skilled in the art will understand that the specific method for collecting demand information can be flexibly adjusted according to the system configuration, as long as it can accurately determine the operating status of the equipment based on the data information; this application does not impose any specific limitations.

[0090] S2: Using the cluster to be inspected as a vertex, if there is a shared device between at least two clusters to be inspected, connect the corresponding clusters to be inspected with the shared device as a hyperedge to generate an overlapping cluster hypergraph.

[0091] In this embodiment, each cluster to be inspected is abstracted as a vertex, and connections are established using a set of shared devices as hyperedges, thus forming an overlapping cluster hypergraph. Each hyperedge represents a sharing relationship between multiple clusters, and each hyperedge corresponds to at least one shared device. The traditional point-to-point clustering results are transformed into a hypergraph model with higher-order associations, enabling the potential linkage relationships between inspection tasks to be expressed in a graph structure.

[0092] S3: By calculating the initial value upper bound of the hyperedges of the overlapping cluster hypergraph, a candidate set is obtained through screening;

[0093] In this embodiment, an initial upper bound for the value of each hyperedge is calculated. This initial upper bound is obtained based on the difference between the overall risk-reward value of the shared equipment set and the corresponding geographical travel cost value. The overall risk-reward value is determined by the superposition of the risk weights and fault prevention benefits of the shared equipment, while the travel cost value is adjusted based on the energy consumption, range, and time window constraints of the inspection equipment. After considering the upper bound of the hyperedge's value as the maximum obtainable net benefit, high-value hyperedges are selected by numerical ranking to form a candidate set.

[0094] S4: Perform iterative operation, select the current shared device from the candidate set, determine the joint inspection path that maximizes the benefit based on the cluster to be inspected connected by the hyperedge corresponding to the current shared device through cost game, and synchronously delete the power grid devices in the joint inspection path from the cluster to be inspected, update the overlapping cluster hypergraph and the candidate set until the candidate set is an empty set;

[0095] In this embodiment, the current shared device is selected from the candidate set in descending order of its initial value upper bound. Using this shared device as the anchor point, a coalition game is triggered among multiple clusters to be inspected that include this anchor point. Each participating cluster generates several candidate sub-path packages based on its own device requirements. Each sub-path package uses the shared device as a necessary node and, under the condition of satisfying geographical and time constraints, submits the risk-reward and execution cost of the path as a bid.

[0096] Furthermore, the autonomous planning agent selects the optimal combination of several sub-paths through an auction-style game to form a joint inspection path, and adopts a revenue-sharing mechanism to achieve fair settlement. Subsequently, the devices covered by the joint path are synchronously removed from the cluster to be inspected, the overlapping cluster hypergraph is locally updated, and the upper bound of the value of the remaining hyperedges is recalculated.

[0097] Understandably, this cyclical game and synchronous point deletion mechanism can achieve dynamic task convergence and balanced resource allocation, ensuring stable optimality and real-time performance even in scenarios with high-frequency task injection.

[0098] S5: Summarize the joint inspection paths of each iteration and the power grid equipment to be inspected that has not been deleted in the cluster to be inspected, and generate the final inspection path;

[0099] In this embodiment, after all iterations are completed, the joint inspection paths generated in each round are integrated, while supplementary task paths are formed by retaining devices that have not yet been covered, ultimately resulting in a global inspection path planning result. This result can be directly invoked by the inspection devices upon output and can be further allocated to different execution carriers based on task type, priority, and resource availability.

[0100] Before delving into the specific technical details of the steps, the embodiments of this application need to be emphasized again.

[0101] The core logic employed in this application is not simply path shortestification or task load balancing, but rather a planning concept based on shared-induced collaborative approaches. Traditional path optimization algorithms often treat task points as static nodes, focusing on minimizing path cost or maximizing overall coverage, while ignoring the coupling relationships between tasks caused by business dependencies, equipment constraints, and execution timing. In fact, in power grid scenarios, the spatial topology and business coupling between devices highly overlap. If static point sets are still used as the optimization object, it is easy to create a contradiction between local priority coverage and global response delay.

[0102] In this embodiment, the logical associations of inspection objects are mapped to computable higher-order connections, forming an overlapping cluster hypergraph structure. Unlike traditional node graphs, the overlapping cluster hypergraph structure uses shared devices as links to connect multi-source inspection tasks in parallel via hyperedges, allowing multiple membership relationships of any device to be explicitly represented in the same graph space. On the one hand, by raising the structural hierarchy, path dependencies between multiple clusters can be modeled; on the other hand, it provides an information channel for subsequent dynamic game theory, making strategic path merging and payoff redistribution possible. Thus, this embodiment eliminates the task independence assumption at the model level, making path planning no longer a static solution but a dynamic negotiation process.

[0103] In this specific implementation, the alliance auction logic introduced through game theory is not a transplantation of the economic model, but rather a solution to the asymmetry problem of shared objectives among multiple stakeholders in inspection tasks. Each participating cluster has different risk weights, resource constraints, and time window constraints. If a centralized scheduling or a single greedy strategy is adopted, it will inevitably lead to an imbalance where some high-risk areas are frequently allocated, while some peripheral areas are subject to long-term delays in inspection. Through game-theoretic auctions, each cluster independently bids under the anchor point constraint of the shared equipment, and the rationality of the path formation is judged by comparing the difference between the risk-reward and the execution cost.

[0104] It is easy to understand that each participant's optimal strategy aligns with the overall optimal direction, thus avoiding strategic misreporting and resource contention, achieving local rational convergence under the maximization of global benefits. In other words, this application transforms resource-constrained collaborative tasks into benefit-driven self-organizing optimization, thereby maintaining the stability and fairness of path decisions under dynamic scheduling conditions.

[0105] Furthermore, by synchronously deleting and reconstructing nodes in each iteration, the hypergraph structure can be updated at minimal cost. Unlike conventional global replanning, this embodiment only adjusts the graph structure in local regions where the associated clusters of shared devices change, and updates the hyperedge weights in real time. This differential update strategy enables the method to have near real-time computing capabilities, achieving minute-level path convergence even in large-scale power grid scenarios. It effectively reduces the number of repeatedly accessed nodes, shortens the overall execution time, and significantly improves task response speed.

[0106] Next, we will further elaborate on the technical aspects of the method in this application regarding the generation of clusters to be inspected.

[0107] It is understood that the requirements information of this application includes at least one of the following:

[0108] Equipment operating status parameters, including voltage, current, temperature, partial discharge signal, and historical health index;

[0109] Equipment risk weight parameters are used to characterize the potential failure probability, environmental risk level, and load importance of equipment;

[0110] Inspection timeliness parameters are used to limit the latest response time for equipment inspection tasks;

[0111] Geographic constraint parameters include the geographical coordinates of the equipment, the boundaries of its region, and safety distance limits.

[0112] In one example, the multi-objective clustering obtained by the clustering algorithm yields multiple clusters to be inspected, including:

[0113] S1.1: Calculate the geographical distance matrix between power grid equipment based on the geographical constraint parameters, and construct a multi-dimensional feature vector corresponding to the geographical distance matrix based on the equipment risk weight parameters;

[0114] Specifically, to avoid neglecting restricted areas, terrain undulations, and road accessibility by simply using straight-line distance, the geographic distance matrix is ​​generated based on a traversable grid and a traversable weight map (both of which can be preset in advance, and will not be elaborated here):

[0115] The passable grid consists of passageways, corridors, maintenance access roads, landing points, bridges, and culverts. Restricted areas and high-voltage safety boundaries are represented by inaccessible units. The passability weight map overlays slope, road surface material, wind corridor impact, temporary construction, and traffic restrictions onto the passable units, ensuring that the elements between two devices represent the minimum cost passability length rather than a geometric straight-line length. For devices distributed across domains, the cost distance within their respective domains is calculated first, while the cross-domain distance is obtained by concatenating gateway points between domains, avoiding unreachable or distorted values ​​in the matrix. After distance calculation, risk weight parameters composed of historical faults, load criticality, environmental exposure, and maintenance history are retrieved and combined with distance information to construct a feature vector. The feature vector corresponding to each device includes at least the passability cost, directional component, reachability time to key nodes, environmental sensitivity level, and historical defect density, ensuring that subsequent clustering can simultaneously perceive the constraints of spatial proximity and business importance.

[0116] S1.2: Normalize and weight the different feature dimensions of the multidimensional feature vector using the equipment operating status parameters and the inspection timeliness parameters to obtain a weighted feature vector;

[0117] Specifically, a robust normalization process is adopted for the characteristic dimensions of different dimensions and distributions:

[0118] First, point locations are limited and intervals are pruned. Then, the scale is aligned by position or quantile intervals to avoid the impact of extreme values ​​on the cluster radius. Next, the relative weights of each dimension are determined based on the equipment operating status parameters and the inspection timeliness parameters. The weights are set based on two sources: one is empirical rules and maintenance strategies (e.g., increasing the weight of tasks approaching their deadlines and increasing the weight of equipment exceeding temperature limits), and the other is adaptive inference from historical task completion data (e.g., which dimension changes are more effective in improving task completion rates under the same resource conditions). The two types of information are combined to obtain a stable weight vector. After weighting, the feature vector maintains spatial separability while highlighting tasks nearing their expiration date and equipment in abnormal states to a closer feature space.

[0119] S1.3: Based on the weighted feature vector, the power grid equipment is clustered using a density clustering algorithm to obtain multiple initial clusters to be inspected. A shared threshold is preset during the clustering process. When the feature similarity of any power grid equipment relative to at least two initial clusters to be inspected is greater than or equal to the shared threshold, the power grid equipment is simultaneously assigned to the corresponding initial cluster to be inspected as a shared equipment.

[0120] Specifically, a density-based adaptive clustering method is employed: core and boundary points are determined by local sample density and nearest-neighbor reachability. The cluster radius is not a fixed constant but is adaptively determined locally based on the nearest-neighbor distance distribution of each device, allowing high-density urban areas and low-density suburbs to achieve reasonable clustering within the same parameter framework. To reduce the impact of map errors, adjacency relationships are filtered based on global map reachability, retaining only adjacency pairs with feasible paths.

[0121] The shared threshold is determined synchronously during the clustering process:

[0122] When the similarity between a device and two or more cluster centers exceeds the threshold and there is a passable connection between the two centers, it is directly recorded as a shared node during clustering and written into the member list of two or more clusters, rather than being detected afterward. This ensures that the subsequently constructed overlapping structure remains consistent with the actual business conflict.

[0123] In this embodiment, the size and shape of the cluster are subject to two types of boundary constraints:

[0124] One type is the upper limit of geographical span, to prevent clusters from having excessively large spans;

[0125] Another type is the time window reachable limit, which requires that any two points within the cluster can be inspected serially within a specified time window.

[0126] S1.4: Calculate the average risk index and geographical compactness index of each initial cluster to be inspected, so as to update the initial clusters to be inspected and obtain multiple clusters to be inspected;

[0127] Specifically, after obtaining the initial clusters, two types of evaluation metrics are calculated for each cluster:

[0128] A comprehensive risk indicator that reflects the importance of a task is aggregated based on the risk weight, alarm intensity and expiration urgency of devices within the cluster, and takes into account the shared impact of shared nodes in multiple clusters.

[0129] Another type of geographical compactness index that reflects the quality of spatial organization takes into account factors such as the average mutual reachability time within a cluster, the shape of the boundary envelope, and the distance to take-off / landing points, to assess whether a cluster is conducive to forming short and efficient inspection routes.

[0130] If the combination of risk level and compactness of a cluster does not meet the preset executable range, then perform a structural update:

[0131] For high-risk but dispersed clusters, priority is given to splitting them around candidate connectors to form multiple geographically more compact small clusters; for low-risk but spatially over-concentrated clusters that are close to neighboring clusters, they are merged according to accessibility thresholds and time window merging rules.

[0132] refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of the cluster to be inspected provided in an embodiment of this application.

[0133] Figure 4 Three clusters to be inspected are shown, namely Cluster 1, Cluster 2, and Cluster 3, each containing several power grid devices.

[0134] Figure 4 Multiple shared devices are further shown, including shared device 1, shared device 2 and shared device 3, where shared device 1 belongs to three clusters to be inspected, shared device 2 belongs to clusters to be inspected 1 and clusters to be inspected 2, and shared device 3 belongs to clusters to be inspected 2 and clusters to be inspected 3.

[0135] It is understandable that each cluster to be inspected contains at least one shared device, mainly due to the objective characteristics of the distribution of tasks on the power grid site. In actual power grid operation and maintenance scenarios, devices are often distributed in clusters along power supply circuits, feeder branches, or communication optical cable paths. Different inspection areas are not completely independent, but rather have overlapping relationships such as electrical connections, communication links, or geographical proximity. For example, the interconnecting switch, the external lead-out tower, or the ring main unit joint point at the outgoing node of a substation usually belong to different operation and maintenance units or regional responsibility boundaries. Therefore, when the task is generated, it will be jointly identified as the device node to be covered by two or more inspection areas, thus naturally forming shared devices.

[0136] It should be noted that, Figure 4 This is merely an illustrative example to demonstrate the logical relationship between the clusters to be inspected and the shared devices, and does not limit the actual number, type, or distribution of power grid equipment. In practical applications, the number of clusters to be inspected and the structural relationship of the shared devices can dynamically change according to the specific power grid scale, regional division, and task generation strategy, and this application does not impose any limitations on this.

[0137] Next, we will further elaborate on the technical content of the method of this application regarding overlapping cluster hypergraphs.

[0138] by Figure 4 Taking the cluster to be inspected as an example, refer to Figure 5 , Figure 5 This is a schematic diagram of the structure of the overlapping cluster hypergraph provided in the embodiments of this application.

[0139] like Figure 5 As shown, node 1, node 2 and node 3 correspond to cluster 1, cluster 2 and cluster 3 to be inspected, respectively. Similarly, superedge 1, superedge 2 and superedge 3 correspond to shared device 1, shared device 2 and shared device 3, respectively.

[0140] Understandable, Figure 5The basic mapping relationship of the overlapping cluster hypergraph is illustrated using only a three-node, three-hyperedge configuration, without limiting the actual number of nodes and hyperedges or their connection methods. Each node represents a cluster to be inspected, and each hyperedge represents a higher-order association between clusters caused by at least one shared device. When the same shared device belongs to two or more clusters to be inspected, a hyperedge is established between the corresponding nodes. When there is a set of shared devices, they can be represented by the same hyperedge or by multiple hyperedges, depending on whether the shared set needs to be distinguished in terms of business. Hyperedges can carry weights, directions, or timeliness attributes to reflect the risk weights, reachability windows, and resource occupancy status of shared devices. The connectivity between nodes and hyperedges is dynamically adjusted as the inspection task flow changes. In some optional implementations, the number of nodes can be extended to any positive integer, the degree of any node can be zero or greater than or equal to one, and hyperedges can connect two or more nodes and allow multiple overlaps, all of which are equivalent forms of the overlapping cluster hypergraph of this application.

[0141] In one example, generating the overlapping cluster hypergraph includes:

[0142] S2.1: Take multiple clusters to be inspected obtained by clustering algorithm as a vertex set, and the elements in each vertex correspond to the power grid equipment to be inspected in the cluster.

[0143] S2.2: Traverse the vertex elements and extract the set of shared devices in each vertex. If at least two vertices have the same shared device, establish a hyperedge between the corresponding vertices. The hyperedge is used to represent the shared inspection association between the clusters to be inspected.

[0144] S2.3: Form the overlapping cluster hypergraph by combining all hyperedges and their corresponding vertex sets, wherein each hyperedge connects at least two vertices, the hyperedge corresponds to at least one shared device, and the hyperedge is weighted according to the demand information of the shared device.

[0145] In yet another example, the process of obtaining a candidate set by calculating the initial upper bound of the value of the hyperedge includes:

[0146] S3.1: For each hyperedge in the overlapping cluster hypergraph, extract the set of shared devices corresponding to the hyperedge and the cluster center coordinates of the multiple clusters to be inspected to which each shared device in the set of shared devices belongs;

[0147] S3.2: Calculate the overall risk-return value of the shared device set based on the weights corresponding to the hyperedges;

[0148] S3.3: Based on the geographical distance between the coordinates of the cluster centers, the average energy consumption within each cluster, and the preset resource parameters, calculate the driving cost between the vertices connected by the hyperedge, wherein the resource parameters include the range, energy capacity, and operating time window of the inspection equipment.

[0149] S3.4: The difference between the comprehensive risk return value and the driving cost value is used as the initial value upper bound of the hyperedge, wherein the initial value upper bound is used to characterize the maximum net return obtained by forming a joint inspection path by multiple vertices connected by the hyperedge under the constraint conditions.

[0150] S3.5: Sort the values ​​of the initial value upper bound from high to low and generate a candidate set.

[0151] In this embodiment, to rationally select hyperedges that can preferentially participate in path game within the structure of the overlapping cluster hypergraph, it is first necessary to quantitatively model the set of shared devices corresponding to each hyperedge. Each hyperedge represents a shared association between multiple clusters to be inspected, and the strength of this association is determined by the number of shared devices, risk level, and business coupling degree. For ease of measurement, the set of shared devices corresponding to the hyperedge is extracted during calculation, and the cluster center coordinates of the cluster to which each shared device belongs are determined. The cluster center coordinates are obtained by weighted averaging of the coordinates of devices within each cluster to be inspected. The weighting factor can be the risk weight or task priority of each device, ensuring that the cluster center reflects not only geographical centrality but also business centrality. This allows the geographical distance between cluster centers to approximate the average spatial cost of cross-cluster movement in the subsequent calculation of travel costs, thereby reducing computational complexity.

[0152] Furthermore, for each hyperedge, its risk-reward characteristics are analyzed, and its comprehensive risk-reward value is calculated by combining the risk parameters, historical failure rate, and equipment importance of the shared equipment set. The comprehensive risk-reward value can be considered as the maximum business gain obtainable if joint inspections are performed among the clusters connected by the hyperedge. During the calculation, the risk weights of the shared equipment are first non-linearly compressed to avoid excessive local dominance by high-risk nodes. Then, based on the repeated affiliation relationships of the shared equipment in different clusters, a sharing coefficient is set to proportionally distribute the risk-reward of the shared nodes among different hyperedges. In this way, when a shared equipment appears on multiple hyperedges simultaneously, its revenue contribution is not counted repeatedly, maintaining global consistency in risk-reward evaluation.

[0153] In some optional implementations, the overall risk-reward value can also take into account the time decay factor. The more urgent the response time of the inspection task, the greater the corresponding risk-reward will be according to the preset curve, thereby guiding the agent to prioritize joint inspection of time-sensitive areas.

[0154] Furthermore, in the cost-of-travel calculation stage, to accurately reflect the resource consumption of multi-cluster collaborative inspection, this embodiment introduces energy consumption constraints and operational resource parameters. The average energy consumption within each cluster to be inspected is estimated from the typical travel cost of its internal paths; the cross-cluster travel cost is based on the shortest feasible path length between cluster centers and corrected according to environmental parameters such as terrain slope, wind speed, and obstacle density. Resource parameters include the inspection equipment's range, energy capacity, and operational time window, used to introduce constraint boundaries when calculating costs. For example, if any resource parameter is insufficient to support cross-cluster inspection, the cost-of-travel value will be set to infeasible and removed from the candidate set. Finally, the difference between the comprehensive risk-reward value and the cost-of-travel value is used as the initial upper bound of the hyperedge value, representing the maximum net benefit achievable by joint inspection under given resource conditions. The candidate set is obtained by sorting the values ​​of this indicator from high to low. In this way, the generation of the candidate set not only takes into account the balance between economic efficiency and risk-benefit, but also takes into account the feasibility of equipment and time constraints, so that subsequent game optimization can be carried out on a smaller, higher value hyperedge set, thereby improving the overall solution efficiency and path rationality.

[0155] Next, we will further elaborate on the technical content of iterative calculation in the method of this application.

[0156] In one example, the order of selecting the current shared device from the candidate set is top-down, that is, the shared device with the largest upper bound of the initial value is selected as the current shared device.

[0157] In one example, the determination of the joint inspection path that maximizes revenue through cost game theory includes:

[0158] S4.1: Using the current shared device as the anchor point, determine multiple clusters to be inspected, including the anchor point, as participants. Each participant generates multiple candidate sub-path packages based on the demand information of the power grid equipment within the cluster. Each candidate sub-path in the candidate sub-path package satisfies the following: the candidate sub-path includes the current shared device, and the covered power grid equipment comes from the power grid equipment within the corresponding participant's cluster; the candidate sub-path satisfies preset constraints and includes corresponding travel time, energy consumption, and waiting time estimates.

[0159] Specifically, candidate sub-paths are generated starting from or through anchor points. Feasible corridors are constructed on the traversability map based on time windows and safe distances. Bounded enumeration and pruning are performed within these corridors, prioritizing the expansion of high-risk and time-sensitive equipment nodes. When encountering restricted areas, height-restricted areas, or energized safety boundaries, alternative transfer points are used for detours. Each expansion simultaneously maintains three types of cumulative quantities: arrival time, remaining range and energy capacity, and remaining time window for the pending operation. If any cumulative quantity becomes unfeasible, the process immediately reverts to ensure that the generated sequence naturally satisfies the operational constraints.

[0160] Those skilled in the art will understand that the specific methods for selecting candidate sub-paths can be implemented using different strategies based on the spatial density of the inspection objects and the urgency of the task. For example, they can be achieved through constraint-based heuristic search or rolling window search methods based on local optimization. This application will not elaborate on these methods here.

[0161] It should be noted that in this application, the candidate sub-paths are only the paths generated within the cluster to be inspected corresponding to the anchor point. Figure 4 For example, shared device 2 belongs to cluster 1 and cluster 2 to be inspected. Candidate sub-paths are generated only within cluster 1 and cluster 2 to be inspected. In addition, the candidate sub-paths generated within each cluster to be inspected are not unique. Multiple candidate sub-paths are aggregated to generate the candidate sub-path package corresponding to the cluster to be inspected.

[0162] S4.2: Price the candidate sub-paths in each candidate sub-path package through an auction mechanism, and submit the priced candidate sub-paths as auction items to the autonomous planning agent. The price of each candidate sub-path is determined by the risk-reward value and execution cost of the candidate sub-path. The risk-reward value is determined based on the sum of the fault prevention benefits of the power grid equipment in the candidate sub-path, and the execution cost includes travel time, energy consumption and waiting time.

[0163] Specifically, the pricing process first calculates two components for each candidate sub-path: coverage revenue and execution cost. Coverage revenue is derived from a weighted combination of equipment risk weight, expected health improvement, and task timeliness. If the same equipment appears repeatedly in multiple candidate sub-paths, only its revenue on the highest-revenue path is retained, and it is proportionally reduced on other paths to avoid double-counting. Execution cost is represented by a standardized resource equivalent value, with travel time, energy consumption, and waiting time mapped to three resource categories: time occupation, energy consumption, and operational congestion, respectively. These three categories have different conversion coefficients under different carrier types. To curb aggressive bidding, pricing considers both uncertainty buffers and penalties: when a path contains many boundary devices or traverses high-risk corridors, an uncertainty buffer is automatically added; when the path's occupancy of the interlocking sections on both sides of the anchor point is nearly saturated, a congestion penalty is added to prevent subsequent splicing from failing to meet the time window.

[0164] In this embodiment, the auction mechanism adopts a sealed submission process, where participants only submit the final pricing and structural summary of the sub-path, without submitting internal estimation details, thus avoiding strategic learning.

[0165] S4.3: Select multiple candidate sub-paths based on the pricing of each candidate sub-path to generate a candidate joint inspection path;

[0166] Specifically, the selection problem is modeled as a compatible set selection with resource constraints:

[0167] The compatibility requirements are that there is no overlap between the coverage sets, the plug-in segments can be aligned, the total resource consumption does not exceed the upper limit of the carrier, and the target is selected to maximize the total benefit after combination.

[0168] In this embodiment, the solution process adopts a two-stage strategy of main path-replacement enhancement:

[0169] First, the sub-path with the highest price and compatibility with existing options is selected from the priority queue as the main path. Then, starting from the anchor point, the main path is extended outwards, and other sub-paths are used to replace or splice non-critical sections of the main path according to the marginal benefit from high to low. The marginal benefit calculation takes into account the new coverage and reduced waiting time and energy consumption after replacement, and deducts the splicing disturbance. When multiple sub-paths compete for the same splicing section, the decision is made based on the unit resource benefit and the impact on anchor point congestion.

[0170] refer to Figure 6 , Figure 6 This is a schematic diagram illustrating the principle of main path replacement provided in the embodiments of this application.

[0171] like Figure 6 As shown, with the anchor point as the reference point, there exists a main path and several candidate sub-paths (including candidate sub-path one and candidate sub-path two). Each path contains several power grid equipment nodes and a shared anchor point. The power grid equipment on different paths is distinguished by different grayscale fill methods in the diagram to facilitate identification of their respective paths. The main path represents the path structure with the highest price in the current auction round that has been selected. Candidate sub-paths are the set of paths that can potentially be used to replace or connect sections of the main path. The dashed lines represent the logical process of judging marginal benefits.

[0172] It is understood that the path replacement process in this application is not a simple one-to-one correspondence as shown in the diagram, but rather a dynamic game-theoretic matching process centered on anchor points and based on marginal benefits. During actual replacement, the system will partially replace, extend, or concurrently splice non-critical sections of the main path based on the spatial location of nodes in each path, time window alignment, and resource consumption. This improves overall benefit efficiency and coverage completeness while ensuring temporal continuity and resource constraints are met. Figure 6 The illustrations are for illustrative purposes only and do not limit the number of actual replacement paths, node correspondences, or spatial distribution.

[0173] by Figure 6 For example, refer to Figure 7 , Figure 7 This is a schematic diagram of the main path replacement result provided in an embodiment of this application.

[0174] like Figure 7 As shown, some power grid devices in the main path are replaced with power grid devices in the candidate sub-paths.

[0175] In one example, the step of selecting multiple candidate sub-paths based on the pricing of each candidate sub-path to generate a candidate joint inspection path includes:

[0176] Prioritize each candidate sub-path based on its pricing, and select the candidate sub-path with the highest pricing as the main path;

[0177] Starting from the anchor point in the main path, each candidate power grid device in the main path is traversed sequentially, and a replacement device that can replace the candidate power grid device is selected from the remaining candidate sub-paths. The selection of the replacement device is based on the marginal benefit of the replacement device, which represents the difference in pricing on the main path after the replacement device replaces the candidate power grid device.

[0178] Candidate joint inspection paths are obtained by replacing the corresponding candidate power grid equipment with all alternative equipment.

[0179] Specifically, the generation process of candidate joint inspection paths uses the main path as the core structure and optimizes and integrates the paths by introducing a dynamic replacement mechanism driven by marginal revenue. In the priority ranking stage, the unit resource yield of each candidate sub-path is first calculated based on its pricing result, and a priority queue is generated under global resource constraints. The path with the highest price is determined as the main path, and the power grid equipment nodes it contains constitute the initial inspection framework. At this point, the anchor point of the main path serves as a fixed starting point and convergence point to maintain the overall topological continuity and executability of the path. Subsequently, at each power grid equipment node on the main path, the system calculates its contribution to the overall path revenue and compares it with the replaceable nodes in the corresponding areas of the remaining candidate sub-paths. If a replaceable node can complete the same inspection task with higher revenue density or lower execution cost, its marginal revenue is calculated as the basis for replacement priority.

[0180] In this embodiment, the calculation of marginal revenue considers not only the direct revenue difference but also the indirect impact of the replacement node on time window coordination, energy consumption balance, and path connectivity. To ensure the rationality of the replacement, each replacement must be verified by the feasibility verification module to determine whether it meets three types of constraints:

[0181] First, there is a temporal continuity constraint, meaning that after the replacement node is inserted, it must complete the task and return to the main path within a specified time.

[0182] Second, there is an energy redundancy constraint, meaning that the total energy consumption of the alternative path does not exceed the equipment's endurance limit.

[0183] Third, there is the path connectivity constraint, which states that the spatial distance between the replacement node and its adjacent nodes does not exceed the maximum operation span. If any constraint is not met, the replacement operation is canceled and recorded as an infeasible replacement to avoid subsequent repeated calculations.

[0184] Understandably, the process of replacing the main path can strike a balance between benefit orientation and feasibility and safety, thus forming a dynamic optimization process of multi-path collaboration.

[0185] In some optional implementations, to prevent path structure imbalance caused by replacement, the system first generates a set of candidate replacement pairs within a local area. Then, it identifies node groups that can be replaced simultaneously through local topology analysis to reduce interference between replacements. The replacement order is determined by the marginal benefit gradient of the replacement nodes. When the benefits of multiple candidate nodes are similar, the system prioritizes node combinations with higher connectivity and better spatial distribution. After replacement, the benefits, costs, and feasibility indicators of the main path are updated in real time, forming a new joint inspection path. Through this marginal benefit-driven dynamic replacement mechanism, optimal allocation and collaborative scheduling of tasks among multiple clusters can be achieved while ensuring path continuity, thereby effectively improving the overall benefits and resource utilization of the entire inspection plan.

[0186] S4.4: Settle and allocate the costs of the candidate joint inspection paths through the VCG payment mechanism to obtain the joint inspection paths;

[0187] Specifically, during the settlement phase, the externalities of each participant are calculated based on the combined revenue of candidate joint inspection paths: for each participant, its selected sub-path is temporarily removed, and a compatible set selection is re-executed under the same resources and constraints. The difference between the total revenue after re-selection and the original total revenue is taken as the impact of that participant on others. The impact is compared with its own execution cost to obtain the amount to be paid or compensated, which is used to ensure individual rationality and true pricing. To reduce the cost of repeated solutions, the recalculation process reuses the previous priority queue and compatibility cache, and only performs local re-selection around the affected insertion segment.

[0188] In one example, the settlement and allocation of the candidate joint inspection path through the VCG payment mechanism to obtain the joint inspection path includes:

[0189] For each participant, calculate the marginal contribution to the candidate joint inspection path, where the marginal contribution is the proportion of the risk-reward value and execution cost of the power grid equipment included in the participant in the candidate joint inspection path;

[0190] The total technical benefits of the candidate joint inspection path are calculated based on the marginal contributions of all participants.

[0191] Based on the total technical benefits, a payment is allocated to each participant through the VCG payment mechanism to adjust the payment cost for each participant. The final path is determined based on the payment allocation results, and a joint inspection path is output.

[0192] Specifically, the VCG (Vickrey Clarke Groves) payment mechanism is used to achieve fair sharing and incentive compatibility of path benefits and costs among multiple participants. The mechanism uses marginal contribution as its core calculation basis, aiming to ensure that each participant maximizes its own interests without disrupting the globally optimal outcome by exaggerating costs or concealing benefits. Specifically, after the candidate joint inspection path is generated, for each participant's corresponding cluster to be inspected, the set of selected power grid equipment in that path is extracted, and the benefit contribution of this set in the entire joint path is calculated, including indicators such as equipment failure prevention benefits, inspection risk reduction ratio, and path connectivity improvement coefficient. Simultaneously, combined with the energy consumption, travel time, and scheduling waiting resources required for each participant's execution, a corresponding execution cost curve is formed. The difference between the two is the participant's net marginal contribution to the joint path.

[0193] In this embodiment, the VCG payment mechanism follows the principle of externality compensation. The system calculates the difference between the optimal total revenue achievable by the remaining participants when a participant is temporarily removed and the optimal total revenue when that participant is included. This difference reflects the actual impact of the participant on improving overall revenue and is used to determine the compensation or payment amount it should receive. If the participant's presence significantly improves path revenue, its payment amount will be appropriately reduced to reflect the incentive effect; conversely, if it increases the overall execution cost, it will bear the corresponding compensation cost.

[0194] In one example, updating the overlapping cluster hypergraph and the candidate set includes:

[0195] S4.5: For the deleted clusters to be inspected, recalculate the number of devices, average risk weight, geographic center coordinates and resource occupancy status of each cluster to update the feature information of the clusters to be inspected.

[0196] S4.6: Based on the updated cluster feature information to be inspected, perform local structural adjustments on the overlapping cluster hypergraph, wherein the local structural adjustments include: deleting the corresponding hyperedges; and recalculating the shared device set and the comprehensive risk-return value corresponding to the shared device set based on the remaining hyperedges.

[0197] S4.7: Update the initial value upper bound and reorder the hyperedges to generate an updated candidate set. If the updated candidate set is not empty, input the overlapping cluster hypergraph with local structure adjustment and the candidate set into the next iteration.

[0198] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent, applied to an inspection planning system, characterized in that: The inspection planning system includes an autonomous planning agent, multiple power grid devices, and inspection equipment. The autonomous planning agent is configured with a global map. The method includes: Based on the global map and the demand information of the power grid equipment to be inspected, a multi-objective clustering is performed using a clustering algorithm to obtain multiple clusters to be inspected, wherein each cluster to be inspected includes at least one shared device. Using the cluster to be inspected as the vertex, if there is a shared device between at least two clusters to be inspected, the shared device is used as a hyperedge to connect the corresponding clusters to be inspected, generating an overlapping cluster hypergraph, and the candidate set is obtained by calculating the upper bound of the initial value of the hyperedge. Perform iterative operations, selecting the current shared device from the candidate set. Based on the clusters to be inspected connected by the hyperedge corresponding to the current shared device, determine the joint inspection path that maximizes the benefit through cost game theory. Simultaneously delete the power grid devices in the joint inspection path from the clusters to be inspected, and update the overlapping cluster hypergraph and the candidate set until the candidate set is empty, wherein: The joint inspection path that maximizes benefits through cost game theory includes: Using the current shared device as the anchor point, multiple clusters to be inspected, including the anchor point, are identified as participants. Each participant generates multiple candidate sub-path packages based on the demand information of the power grid equipment within the cluster. Each candidate sub-path in the candidate sub-path package satisfies the following: the candidate sub-path includes the current shared device, and the covered power grid equipment comes from the power grid equipment within the corresponding participant's cluster; the candidate sub-path satisfies preset constraints and includes corresponding travel time, energy consumption, and waiting time estimates. The candidate sub-paths in each candidate sub-path package are priced through an auction mechanism, and the priced candidate sub-paths are submitted to the autonomous planning agent as auction items. The price of each candidate sub-path is determined by the risk-reward value and execution cost of the candidate sub-path. The risk-reward value is determined based on the sum of the fault prevention benefits of the power grid equipment in the candidate sub-path, and the execution cost includes travel time, energy consumption and waiting time. Based on the pricing of each candidate sub-path, multiple candidate sub-paths are selected to generate candidate joint inspection paths; The candidate joint inspection path is settled and allocated through the VCG payment mechanism to obtain the joint inspection path. The final inspection path is generated by summarizing the joint inspection paths from each iteration and the power grid equipment to be inspected that has not been deleted from the cluster to be inspected.

2. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 1, characterized in that, The required information for the power grid equipment to be inspected includes at least one of the following: Equipment operating status parameters, including voltage, current, temperature, partial discharge signal, and historical health index; Equipment risk weight parameters are used to characterize the potential failure probability, environmental risk level, and load importance of equipment; Inspection timeliness parameters are used to limit the latest response time for equipment inspection tasks; Geographic constraint parameters include the geographical coordinates of the equipment, the boundaries of its region, and safety distance limits.

3. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 2, characterized in that, The multi-objective clustering method using a clustering algorithm yields multiple clusters to be inspected, including: Calculate the geographical distance matrix between power grid equipment based on the geographical constraint parameters, and construct a multidimensional feature vector corresponding to the geographical distance matrix based on the equipment risk weight parameters; The different feature dimensions of the multidimensional feature vector are normalized and weighted using the equipment operating status parameters and the inspection timeliness parameters to obtain a weighted feature vector. Based on the weighted feature vector, the power grid equipment is clustered using a density clustering algorithm to obtain multiple initial clusters to be inspected. During the clustering process, a sharing threshold is preset. When the feature similarity of any power grid equipment relative to at least two initial clusters to be inspected is greater than or equal to the sharing threshold, the power grid equipment is simultaneously assigned to the corresponding initial cluster to be inspected and is used as a shared equipment. Calculate the average risk index and geographical compactness index of each initial cluster to be inspected to update the initial clusters to be inspected, resulting in multiple clusters to be inspected.

4. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 1, characterized in that, The generation of the overlapping cluster hypergraph includes: Multiple clusters to be inspected, obtained through clustering algorithms, are used as a set of vertices, and the elements in each vertex correspond to the power grid equipment to be inspected in the cluster. Traverse the vertex elements and extract the set of shared devices in each vertex. If at least two vertices have the same shared device, establish a hyperedge between the corresponding vertices. The hyperedge is used to represent the shared inspection association between the clusters to be inspected. The overlapping cluster hypergraph is formed by combining all hyperedges and their corresponding vertex sets, wherein each hyperedge connects at least two vertices, the hyperedge corresponds to at least one shared device, and the hyperedge is weighted according to the demand information of the shared device.

5. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 4, characterized in that, The process of obtaining the candidate set by calculating the initial upper bound of the superedge value includes: For each hyperedge in the overlapping cluster hypergraph, extract the set of shared devices corresponding to the hyperedge and the cluster center coordinates of multiple clusters to be inspected to which each shared device belongs in the set of shared devices; Calculate the overall risk-return value of the shared device set based on the weights corresponding to the hyperedges; Based on the geographical distance between the cluster center coordinates, the average energy consumption within each cluster, and the preset resource parameters, the driving cost between the vertices connected by the hyperedge is calculated, wherein the resource parameters include the inspection equipment's range, energy capacity, and operating time window. The difference between the comprehensive risk-return value and the driving cost value is used as the initial value upper bound of the hyperedge, wherein the initial value upper bound is used to characterize the maximum net return obtained by forming a joint inspection path by multiple vertices connected by the hyperedge under the constraint conditions. Sort the values ​​of the initial value upper bound from high to low and generate a candidate set.

6. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 1, characterized in that, The order in which the current shared device is selected from the candidate set is top-down.

7. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 1, characterized in that, The step of selecting multiple candidate sub-paths based on the pricing of each candidate sub-path to generate a candidate joint inspection path includes: Prioritize each candidate sub-path based on its pricing, and select the candidate sub-path with the highest pricing as the main path; Starting from the anchor point in the main path, each candidate power grid device in the main path is traversed sequentially, and a replacement device that can replace the candidate power grid device is selected from the remaining candidate sub-paths. The selection of the replacement device is based on the marginal benefit of the replacement device, which represents the difference in pricing on the main path after the replacement device replaces the candidate power grid device. Candidate joint inspection paths are obtained by replacing the corresponding candidate power grid equipment with all alternative equipment.

8. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 1, characterized in that, The process of settling and allocating costs for the candidate joint inspection paths through the VCG payment mechanism to obtain the joint inspection paths includes: For each participant, calculate the marginal contribution to the candidate joint inspection path, where the marginal contribution is the proportion of the risk-reward value and execution cost of the power grid equipment included in the participant in the candidate joint inspection path; The total technical benefits of the candidate joint inspection path are calculated based on the marginal contributions of all participants. Based on the total technical benefits, a payment is allocated to each participant through the VCG payment mechanism to adjust the payment cost for each participant. The final path is determined based on the payment allocation results, and a joint inspection path is output.

9. The method for optimizing power grid equipment inspection paths driven by an autonomous planning intelligent agent according to claim 1, characterized in that, Updating the overlapping cluster hypergraph and the candidate set includes: For the clusters to be inspected after deletion, the number of devices, average risk weight, geographic center coordinates and resource occupancy status of each cluster to be inspected are recalculated to update the characteristic information of the clusters to be inspected. Based on the updated cluster feature information to be inspected, the overlapping cluster hypergraph is locally structurally adjusted, wherein the local structural adjustment includes: deleting the corresponding hyperedges; and recalculating the shared device set and the comprehensive risk-reward value corresponding to the shared device set based on the remaining hyperedges. The initial value upper bound is updated, and the hyperedges are reordered to generate an updated candidate set. If the updated candidate set is not empty, the overlapping cluster hypergraph with local structure adjustment and the candidate set are input into the next iteration.

Images