Substation three-dimensional space reconstruction and path optimization method and system

Abstract

The invention provides a three-dimensional space reconstruction and path optimization method and system for a transformer substation, and the method comprises the steps: obtaining image data and point cloud data, and constructing a scene map of the transformer substation; performing semantic annotation on a substation scene map, performing scene conversion, analyzing regional semantic nodes by using a manual feature-based graph convolutional network to obtain semantic node tags, and performing back projection on the semantic node tags to realize three-dimensional space reconstruction; and based on the reconstructed three-dimensional space, defining a state space and a transition probability, using an LPA * algorithm to initialize a path, and combining with a state transition probability of a Markov chain to assist in heuristic function calculation to determine an optimal path. According to the invention, the navigation and path planning capability of the robot in the complex environment of the transformer substation can be improved under the limitation of the sparse scene visible light condition.

Description

Technical Field

[0001] The present invention belongs to the field of path planning, and in particular relates to a substation three-dimensional space reconstruction and path optimization method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Under the background of the current new power system construction and the rapid development of artificial intelligence, various technologies of robots based on artificial intelligence have brought new development opportunities for the transformation and upgrading of traditional power grid business. Through years of technical research and development and practical application, power equipment robot inspection technology has become one of the important ways to improve power inspection efficiency, coverage, detection accuracy, and reduce inspection costs.

[0004] At present, some areas have initially realized the application of power equipment inspection robots, and realized video image collection and analysis of equipment in different areas through inspection robot technology. However, there are also certain problems in the existing research and application of inspection robots.

[0005] Existing navigation and path planning technologies rely on pre-built map resources, which require a lot of time and human resources to pre-configure. Under the background of changes and uncertainties in the power environment, inspection robots cannot operate accurately. Specifically, they have the following shortcomings:

[0006] (1) The traditional substation inspection robot pre-positioning operation mode requires a huge amount of manual configuration work, and after long-term operation, the pre-positioning will deviate, resulting in a decrease in the quality of the collected images and an inability to ensure the efficiency of the inspection;

[0007] (2) In sparse scenarios, due to environmental changes and uncertainties, the pre-built map may not accurately reflect the state of the current environment, and the robot's perception ability is not strong;

[0008] (3) According to the environmental conditions of different substations, it is necessary to manually plan preset routes, which has poor adaptability to the environment and insufficient generalization ability. Summary of the invention

[0009] In order to solve the above problems, the present invention proposes a substation three-dimensional space reconstruction and path optimization method and system, which can improve the robot's navigation and path planning capabilities in the complex environment of the substation under the limitations of sparse scene visible light conditions.

[0010] According to some embodiments, the present invention adopts the following technical solutions:

[0011] A substation three-dimensional space reconstruction and path optimization method comprises the following steps:

[0012] Obtain image data and point cloud data to build a substation scene map;

[0013] Semantically annotate the substation scene map, perform scene conversion, use the graph convolution network based on manual features to parse the regional semantic nodes, obtain the semantic node labels, and back-project them to achieve three-dimensional space reconstruction;

[0014] Based on the reconstructed three-dimensional space, the state space and transition probability are defined, the LPA* algorithm is used to initialize the path, and the state transition probability of the Markov chain is combined to assist the heuristic function calculation to determine the optimal path.

[0015] As an optional implementation, the process of scene conversion includes: obtaining a plane projection map of the 3D scene using a projection method, setting the boundaries and corner points of each semantic node in the plane projection map as map nodes, and constructing a plane projection map structure based on the map nodes.

[0016] As a further feature, the semantic nodes include various power equipment, buildings and street lamps.

[0017] As an optional implementation, the process of parsing regional semantic nodes using a graph convolutional network based on manual features includes extracting consistency features based on a plane projection graph structure using a dynamic graph convolutional network model DGCN, extracting corner features of semantic nodes using SIFT which is good at extracting corner features, and associating the obtained corner features to the plane projection graph structure, and combining the SIFT with the dynamic graph convolutional network model DGCN into a SIFT-DGCN feature extraction layer.

[0018] As an optional implementation, the process of back-projecting the semantic node labels includes inputting the extracted features into the fully connected FC to obtain 2D semantic node labels, and projecting the 2D labels into the 3D scene to obtain 3D semantic node labels.

[0019] As an optional implementation, based on the reconstructed three-dimensional space, the process of defining the state space and the transition probability includes: discretizing the substation environment and dividing it into a network, where each grid is a state;

[0020] Assign a unique identifier to each state, using two-dimensional coordinates (x, y);

[0021] Determine the attributes of each state, distinguishing between devices and paths;

[0022] Based on the actual dynamic characteristics of the environment and expert experience, the transition probabilities between states are determined.

[0023] The actual dynamic characteristics of the environment are the dynamic positional relationship between the substation inspection equipment and the surrounding environment (including trees, roads, debris, etc.).

[0024] As a further implementation, a state transition probability matrix P is created, whose size is the state space size multiplied by itself, and each element P(i, j) in the matrix represents the probability of transitioning from state i to state j;

[0025] For complex environments, the state space is hierarchical and the environment is divided into multiple regions. The states in each region have similar transition characteristics. The size of the state space is reduced by merging states with similar transition probabilities.

[0026] As an optional implementation, the LPA* algorithm is used to initialize the path, and the state transition probability of the Markov chain is combined to assist the heuristic function calculation. The process of determining the optimal path includes: setting the starting node and the target node, initializing the g value and rhs value of all nodes;

[0027] Combined with the state transition probability of the Markov chain to assist the heuristic function calculation in the LPA* algorithm, all paths from the current node to the target node are considered, and the expected cost is calculated according to the transition probability matrix P of the Markov chain. For each possible state s, the transition probability from the current node n to the state s is calculated, and all possible states that can be transferred to are accumulated and the calculated expected cost is used as part of the heuristic function.

[0028] Use a priority queue to store the nodes to be processed and sort them according to the k value of the nodes, where the k value is a combination of the g value and the heuristic function;

[0029] When the priority queue is not empty, pop the node u with the smallest k value from the priority queue. If u is the target node and rhs(u) is equal to g(u), the algorithm terminates and the shortest path is found. If g(u) is greater than rhs(u), update g(u) to rhs(u), remove u from the priority queue, and perform the following operations on each successor node s of u:

[0030] If rhs(s) is greater than g(u)+c(u, s), update the parent node of s to u, update rhs(s), and update the position of s in the priority queue;

[0031] Otherwise, if s is not the starting node and its parent is u, recalculate the rhs value of s and update its position in the priority queue.

[0032] As an optional implementation, the following steps are also included:

[0033] The optimal path is smoothed using the stationary distribution characteristics.

[0034] Specifically, after finding the shortest path from the starting node to the target node, the path is reconstructed, the path is smoothed using the stationary distribution characteristics of the Markov chain model, and the transition probability between adjacent nodes on the path is evaluated. If the probability of a certain transition is lower than the set value, an intermediate node is added or the path direction is changed until a smooth path is achieved.

[0035] A substation three-dimensional space reconstruction and path optimization system, comprising:

[0036] A map construction module is configured to acquire image data and point cloud data and construct a substation scene map;

[0037] The 3D space reconstruction module is configured to semantically annotate the substation scene map, perform scene conversion, parse regional semantic nodes using a graph convolutional network based on manual features, obtain semantic node labels, and back-project them to achieve 3D space reconstruction;

[0038] The path optimization module is configured to define the state space and transition probability based on the reconstructed three-dimensional space, use the LPA* algorithm to initialize the path, and combine the state transition probability of the Markov chain to assist the heuristic function calculation to determine the optimal path.

[0039] An electronic device comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above method are completed.

[0040] A robot comprises the system or electronic device described above, or comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein the computer instructions complete the steps in the method described above when executed by the processor.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention innovatively provides a three-dimensional space reconstruction and path optimization method for a substation, which integrates visual laser technology to reconstruct a three-dimensional space map. While ensuring the quality of image acquisition, the robot can autonomously configure environmental information. In sparse scenarios, visible light information is used to capture environmental conditions, and fine segmentation and semantic labeling are performed at the regional and target levels to improve the robot's environmental perception ability. The decision path is optimized by an algorithm, thereby improving the active adaptability and generalization ability of path planning.

[0043] The present invention innovatively provides a SIFT and dynamic graph convolution DGC feature extraction algorithm, studies the environmental understanding of the inspection robot in a large-scale substation scenario, and recognizes and marks the substation environment and equipment semantic information in real time based on multi-source sensor information. It has the characteristics of strong environmental perception ability and high reliability.

[0044] The present invention innovatively provides an improved path optimization method, defines the environment state space, uses the LPA* algorithm to initialize the path, and combines the state transition probability of the Markov chain to assist the heuristic function calculation, and uses the stationary distribution characteristics to smooth the path. It has the characteristics of strong dynamic real-time performance and high inspection efficiency, and can realize reliable inspection of robots in field environments.

[0045] The technical solution of the present invention can be applied to electric robot products, improve the multi-scenario adaptability of artificial intelligence technology, identify environmental equipment information in different electric operation scenarios, etc., improve the engineering practical ability of artificial intelligence, and has a relatively broad application prospect.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0048] Figure 1 is a schematic diagram of a method flow of an embodiment;

[0049] Figure 2 is a schematic diagram of a 3D semantic annotation process of an embodiment;

[0050] Figure 3 is a schematic diagram of a path planning process of an embodiment;

[0051] Figure 4 The electronic device structure diagram of an embodiment is shown. DETAILED DESCRIPTION

[0052] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0053] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0054] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0055] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0056] Embodiment 1

[0057] A substation three-dimensional space reconstruction and path optimization method comprises the following steps:

[0058] Obtain image data and point cloud data to build a substation scene map;

[0059] Semantically annotate the substation scene map, perform scene conversion, use the graph convolution network based on manual features to parse the regional semantic nodes, obtain the semantic node labels, and back-project them to achieve three-dimensional space reconstruction;

[0060] Based on the reconstructed three-dimensional space, the state space and transition probability are defined, the LPA* algorithm is used to initialize the path, and the state transition probability of the Markov chain is combined to assist the heuristic function calculation to determine the optimal path.

[0061] In this embodiment, the construction of the substation scene map is carried out by integrating environmental parameters collected by multiple sensors such as three-dimensional lasers and cameras.

[0062] The process of preliminary substation scene mapping can use existing technology.

[0063] This embodiment mainly focuses on Figure 1 The semantic annotation and path planning parts in the paper have been improved.

[0064] First, the semantic annotation process is explained in detail.

[0065] like Figure 2 As shown, this embodiment performs semantic annotation on a complex power scene map constructed based on three-dimensional lasers and cameras, projects the 3D scene into a 2D scene, uses a graph convolutional network based on manual features to parse regional semantic nodes, uses visual images and three-dimensional point cloud clustering to jointly train semantic labels, and back-projects the obtained semantic node labels to the 3D scene.

[0066] Among them, the projection strategy can effectively avoid blind spots and occlusion problems in lens shooting; combining manual features and graph convolution, it can focus on extracting corner features within regional semantic nodes and consistency features between semantic nodes to improve parsing accuracy.

[0067] The specific process is as follows:

[0068] 1) Construct the plane projection graph structure. The scene plane projection graph can reflect the three-dimensional area information of each node, and the plane projection graph of the 3D scene is obtained by using the projection method; the boundaries and corner points of each semantic node (power equipment, buildings, street lights, etc.) in the plane projection graph are set as graph nodes, and the plane projection graph structure is constructed based on the graph nodes.

[0069] Among them, corner point is a term in image processing and computer vision, which refers to the point where two edges meet in the image to form an angle.

[0070] 2) Extraction of semantic node consistency features and corner features. The semantic nodes in the scene are seamlessly connected, and the node consistency features reflect the relative position relationship and seamless connection direction of the semantic nodes.

[0071] Based on the planar projection graph structure in step 1), the dynamic graph convolutional network DGCN is used to extract consistent features. However, consistent features help to seamlessly segment the semantic nodes, but there is still under-segmentation or over-segmentation of the boundaries. Therefore, in this embodiment, SIFT, which is good at extracting corner features, is used to extract the corner features of the semantic nodes, and the obtained corner features are associated with the planar projection graph structure in step 1).

[0072] SIFT is combined with the dynamic graph convolutional network model DGCN into the SIFT-DGCN feature extraction layer.

[0073] 3) Obtain 3D node labels. Input the scene features (i.e., corner features and consistency features) obtained in step 2) into the fully connected FC to obtain 2D semantic node labels. Project the 2D labels into the 3D scene to obtain 3D semantic node labels. This completes the 3D space reconstruction.

[0074] The path planning part is introduced below, which includes the following steps:

[0075] (1) Define state space and transition probability.

[0076] 1) First, analyze the substation environment where the robot operates (here is a map with labeled information) and identify all possible locations. Since the environment is continuous, it needs to be discretized to fit the Markov chain model. Therefore, according to the grid world division, each cell can be a state;

[0077] Markov chain: Markov chain refers to a discrete event random process with Markov properties in mathematics. In each step, the system can change from one state to another according to the probability distribution, or it can maintain the current state. The change of state is called transition.

[0078] 2) Assign a unique identifier to each state, using two-dimensional coordinates (x, y);

[0079] 3) Determine the attributes of each state and distinguish which are devices and which are paths;

[0080] In some embodiments, this step can be performed in combination with semantic nodes and human analysis.

[0081] 4) Determine the transition probability between states based on the actual dynamic characteristics of the environment and expert experience;

[0082] Among them, the actual dynamic characteristics of the environment are the dynamic position relationship between the substation inspection equipment and the surrounding environment (including trees, roads, debris, etc.), which generally changes little.

[0083] Specifically, the transition probability of each state can be determined by combining the working experience of the front-line operation site environment and the execution mode of the inspection robot.

[0084] If there is equipment next to the road, then the probability of defining the state from road to equipment is relatively low, that is, the inspection robot is unlikely to move from the road to the equipment.

[0085] For example, at a road intersection, the state transition between the two roads is relatively high, which means that the inspection robot can pass through.

[0086] (2) Construct state transfer matrix and optimize it.

[0087] 1) Create a state transition probability matrix P, whose size is the state space size multiplied by itself. Each element P(i, j) in the matrix represents the probability of transitioning from state i to state j.

[0088] 2) In complex environments, the state space may be relatively large. We can consider hierarchical state space and divide the environment into multiple regions. The states in each region have similar transition characteristics. The size of the state space can be reduced by merging states with similar transition probabilities.

[0089] (3) LPA* algorithm

[0090] 1) Set the starting node s start and the target node s goal , initialize the g values ​​(the actual cost from the starting node to the current node) and rhs values ​​(the one-step lookahead value based on the g value) of all nodes to infinity, except that the rhs value of the starting node is set to 0.

[0091] Combine the state transition probability of the Markov chain to assist the calculation of the heuristic function h(n) in the LPA* algorithm, considering the transition from the current node n to the target node s goalFor all paths, the expected cost is calculated based on the transition probability matrix P of the Markov chain. For each possible transition state s, the transition probability from the current node n to state s is calculated, and the expected cost calculated is used as part of h(n) by accumulating all possible transition states. c(s′, s) refers to the cost function between two nodes. pred(s) refers to the previous generation node of node s (or the parent node, that is, the node that has been passed from the starting node to the current node).

[0092]

[0093] 2) Build a priority queue: Use a priority queue (minimum heap) to store the nodes to be processed and sort them according to the k value of the nodes (combined with the g value and the value of the heuristic function h(n)).

[0094] 3) Search loop: When the priority queue is not empty, perform the following operations: Pop the node u with the smallest k value from the priority queue. If u is the target node and rhs(u) is equal to g(u), the algorithm terminates and the shortest path is found.

[0095] 4) Update the node: If g(u) is greater than rhs(u), update g(u) to rhs(u) and remove u from the priority queue. Then perform the following operations for each successor node s of u:

[0096] If rhs(s) is greater than g(u)+c(u,s), update the parent node of s to u, update rhs(s), and update the position of s in the priority queue.

[0097] Otherwise, if s is not the starting node and its parent is u, recalculate the rhs value of s and update its position in the priority queue.

[0098] 5) Path reconstruction and smoothing: After finding the shortest path from the starting node to the target node, the path is reconstructed. Then the path is smoothed using the stationary distribution characteristics of the Markov chain model. The stationary distribution of the Markov chain means that in the long run, the state transition probability will tend to a fixed probability distribution. It can be considered that the best choice of path should conform to this stationary distribution. The transition probability between adjacent nodes on the path is evaluated. If the probability of a certain transition is low, an intermediate node can be added or the path direction can be changed to avoid such a low-probability transition, and finally a smooth path direction can be achieved.

[0099] Embodiment 2

[0100] A substation three-dimensional space reconstruction and path optimization system, comprising:

[0101] A map construction module is configured to acquire image data and point cloud data and construct a substation scene map;

[0102] The 3D space reconstruction module is configured to semantically annotate the substation scene map, perform scene conversion, parse regional semantic nodes using a graph convolutional network based on manual features, obtain semantic node labels, and back-project them to achieve 3D space reconstruction;

[0103] The path optimization module is configured to define the state space and transition probability based on the reconstructed three-dimensional space, use the LPA* algorithm to initialize the path, and combine the state transition probability of the Markov chain to assist the heuristic function calculation to determine the optimal path.

[0104] It can be understood that the above-mentioned units / modules can be separately or completely combined into one or several other units / modules, or one (some) of the units can be further divided into multiple functionally smaller units, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application.

[0105] The above modules of the system are divided based on logical functions. In practical applications, the function of one module can also be implemented by multiple modules, or the functions of multiple modules can be implemented by one module.

[0106] Embodiment 3

[0107] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method provided in the first embodiment are completed.

[0108] Alternatively, the electronic device includes a processor 1001, a communication interface 1002, and a computer-readable storage medium 1003. The processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 may be connected via a bus or other means.

[0109] Among them, the communication interface 1002 is used to receive and send data, the computer-readable storage medium 1003 can be stored in the memory of the electronic device, the computer-readable storage medium 1003 is used to store a computer program, the computer program includes program instructions, and the processor 1001 is used to execute the program instructions stored in the computer-readable storage medium 1003.

[0110] Processor 1001 (or CPU (Central Processing Unit)) is the computing core and control core of the electronic device, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to implement corresponding method processes or corresponding functions.

[0111] The processor 1001 is configured to execute the following process:

[0112] Obtain image data and point cloud data to build a substation scene map;

[0113] Semantically annotate the substation scene map, perform scene conversion, use the graph convolution network based on manual features to parse the regional semantic nodes, obtain the semantic node labels, and back-project them to achieve three-dimensional space reconstruction;

[0114] Based on the reconstructed three-dimensional space, the state space and transition probability are defined, the LPA* algorithm is used to initialize the path, and the state transition probability of the Markov chain is combined to assist the heuristic function calculation to determine the optimal path.

[0115] The electronic device may be a navigator or a navigation system, etc.

[0116] The electronic device may be mounted on a robot.

[0117] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0121] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention without creative labor shall be included in the protection scope of the present invention.

Claims

1. A method for three-dimensional space reconstruction and path optimization of a substation, characterized in that: The following steps are involved: Obtain image data and point cloud data to build a substation scene map; Semantically annotate the substation scene map, perform scene conversion, use the graph convolution network based on manual features to parse the regional semantic nodes, obtain the semantic node labels, and back-project them to achieve three-dimensional space reconstruction; Based on the reconstructed three-dimensional space, the state space and transition probability are defined, the LPA* algorithm is used to initialize the path, and the state transition probability of the Markov chain is combined to assist the heuristic function calculation to determine the optimal path.

2. A substation three-dimensional space reconstruction and path optimization method as claimed in claim 1, characterized in that: The process of scene conversion includes: obtaining a plane projection map of the 3D scene using a projection method, setting the boundaries and corner points of each semantic node in the plane projection map as map nodes, and constructing a plane projection map structure based on the map nodes.

3. A substation three-dimensional space reconstruction and path optimization method as claimed in claim 2, characterized in that: The semantic nodes include various power equipment, buildings and street lamps.

4. A substation three-dimensional space reconstruction and path optimization method as claimed in claim 1, characterized in that: The process of parsing regional semantic nodes using a graph convolutional network based on manual features includes extracting consistency features based on a plane projection graph structure using a dynamic graph convolutional network model DGCN, extracting corner features of semantic nodes using SIFT, which is good at extracting corner features, and associating the obtained corner features to the plane projection graph structure. The SIFT and the dynamic graph convolutional network model DGCN are combined into a SIFT-DGCN feature extraction layer.

5. A substation three-dimensional space reconstruction and path optimization method as claimed in claim 1, characterized in that: The process of back-projecting the semantic node labels includes inputting the extracted features into the fully connected FC to obtain 2D semantic node labels, and projecting the 2D labels into the 3D scene to obtain 3D semantic node labels.

6. A substation three-dimensional space reconstruction and path optimization method as claimed in claim 1, characterized in that: Based on the reconstructed three-dimensional space, the process of defining the state space and transition probability includes: discretizing the substation environment and dividing it into networks, with each grid being a state; Assign a unique identifier to each state, using two-dimensional coordinates (x, y); Determine the attributes of each state, distinguishing between devices and paths; Based on the actual dynamic characteristics of the environment and expert experience, the transition probabilities between states are determined. The actual dynamic characteristics of the environment are the dynamic positional relationship between the substation inspection equipment and the surrounding environment.

7. A method for three-dimensional space reconstruction and path optimization of a substation as claimed in claim 6, characterized in that: Create a state transition probability matrix P, whose size is the state space size multiplied by itself. Each element P(i, j) in the matrix represents the probability of transitioning from state i to state j. For complex environments, the state space is hierarchical and the environment is divided into multiple regions. The states in each region have similar transition characteristics. The size of the state space is reduced by merging states with similar transition probabilities.

8. A method for three-dimensional space reconstruction and path optimization of a substation as claimed in claim 6, characterized in that: Using the LPA* algorithm to initialize the path and combining the state transition probability of the Markov chain to assist the heuristic function calculation, the process of determining the optimal path includes: setting the starting node and the target node, initializing the g value and rhs value of all nodes; Combined with the state transition probability of the Markov chain to assist the heuristic function calculation in the LPA* algorithm, all paths from the current node to the target node are considered, and the expected cost is calculated according to the transition probability matrix P of the Markov chain. For each possible state s, the transition probability from the current node n to the state s is calculated, and all possible states that can be transferred to are accumulated and the calculated expected cost is used as part of the heuristic function. Use a priority queue to store the nodes to be processed and sort them according to the k value of the nodes, where the k value is a combination of the g value and the heuristic function; When the priority queue is not empty, pop the node u with the smallest k value from the priority queue. If u is the target node and rhs(u) is equal to g(u), the algorithm terminates and the shortest path is found. If g(u) is greater than rhs(u), update g(u) to rhs(u), remove u from the priority queue, and perform the following operations on each successor node s of u: If rhs(s) is greater than g(u)+c(u, s), update the parent node of s to u, update rhs(s), and update the position of s in the priority queue; Otherwise, if s is not the starting node and its parent is u, recalculate the rhs value of s and update its position in the priority queue.

9. A substation three-dimensional space reconstruction and path optimization method as claimed in claim 1, characterized in that: The following steps are also included: The optimal path is smoothed using the stationary distribution characteristics. Specifically, after finding the shortest path from the starting node to the target node, the path is reconstructed, the path is smoothed using the stationary distribution characteristics of the Markov chain model, and the transition probability between adjacent nodes on the path is evaluated. If the probability of a certain transition is lower than the set value, an intermediate node is added or the path direction is changed until a smooth path is achieved.

10. A substation three-dimensional space reconstruction and path optimization system, characterized in that: include: A map construction module is configured to acquire image data and point cloud data and construct a substation scene map; The 3D space reconstruction module is configured to semantically annotate the substation scene map, perform scene conversion, parse regional semantic nodes using a graph convolutional network based on manual features, obtain semantic node labels, and back-project them to achieve 3D space reconstruction; The path optimization module is configured to define the state space and transition probability based on the reconstructed three-dimensional space, use the LPA* algorithm to initialize the path, and combine the state transition probability of the Markov chain to assist the heuristic function calculation to determine the optimal path.

11. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps in the method according to any one of claims 1 to 9 are completed.

12. A robot, characterized in that: The invention comprises the system as claimed in claim 8 or the electronic device as claimed in claim 9, or comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps in the method as claimed in any one of claims 1 to 9 are completed.

Images