An intelligent inspection method and system based on digital twinning and graph database

Abstract

The application relates to the technical field of intelligent inspection, and discloses an intelligent inspection method and system based on digital twinning and a graph database, which comprises the following steps: performing semantic analysis on an inspection task sheet to obtain a structured task table, inputting the device name in the structured task table into a graph database to obtain a device information table; constructing an initial digital twinning map of an inspection environment, and calibrating the initial digital twinning map according to the device theoretical coordinates in the device information table to obtain a digital twinning map; adopting a path planning algorithm based on density clustering to generate an accessible path, and performing inspection point sampling to obtain a safe inspection point set; and adopting a multi-target path planning algorithm to generate an optimal inspection path according to the safe inspection point set and an obstacle coordinate set. Through multi-modal data fusion and coordinate calibration, the application improves the accuracy of digital twinning mapping; and through a path planning algorithm based on dynamic environment perception, the application improves the path planning efficiency and the safety of the inspection path.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology, and in particular to an intelligent inspection method and system based on digital twins and graph databases. Background Technology

[0002] Currently, equipment inspection methods in power systems are mainly divided into two types: manual inspection and automated inspection. Manual inspection involves inspectors checking power equipment one by one according to a pre-defined inspection task list and a fixed route. For example, in a power substation inspection scenario, inspectors might divide the area into high-voltage equipment zone, low-voltage equipment zone, transformer room, etc., and inspect each type of equipment sequentially. The limitation of this method is that different inspectors may have differing interpretations of the inspection task list, leading to inconsistent inspection content and resulting in some equipment being missed or over-inspected. Furthermore, manual inspection routes are usually performed in a fixed sequence, lacking consideration for real-time environmental changes. When encountering unexpected situations such as equipment failure or on-site construction, inspectors find it difficult to adjust the route in time, resulting in low inspection efficiency and potentially increased safety risks. Automated inspection involves inspection robots performing intelligent inspections of power equipment according to pre-defined inspection paths. While this method can overcome the influence of human factors in manual inspections, its inspection paths are still based on static path planning in a fixed inspection environment. It cannot respond in real time to moving obstacles and equipment position shifts, which can easily lead to frequent path failures. It is difficult to balance path safety and operation and maintenance efficiency. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an intelligent inspection method and system based on digital twins and graph databases. This method solves the problem of low inspection efficiency caused by the disconnect between inspection path planning and the dynamic environment, which prevents dynamic avoidance of sudden obstacles. The goal is to improve the inspection efficiency of power equipment and ensure the safety of inspection paths.

[0004] In a first aspect, the present invention provides an intelligent inspection method based on digital twins and graph databases, the method comprising:

[0005] The inspection task book of power equipment is semantically parsed to obtain a structured task table. The equipment names in the structured task table are input into a preset graph database to obtain an equipment information table.

[0006] An initial digital twin map of the inspection environment is constructed, and the device point cloud coordinates in the initial digital twin map are calibrated according to the theoretical coordinates of the devices in the device information table to obtain the digital twin map.

[0007] Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, a path planning algorithm based on density clustering is used to generate reachable paths, and inspection points are sampled on the reachable paths to obtain a set of safety inspection points.

[0008] Based on the set of safety inspection points and the set of obstacle coordinates, a multi-objective path planning algorithm is used to generate the optimal inspection path.

[0009] Based on the optimal inspection path and the equipment information table, the inspection robot is controlled to perform equipment inspection, obtain equipment status data, and determine whether there is any abnormality in the power equipment based on the equipment status data.

[0010] Furthermore, the steps of constructing an initial digital twin map of the inspection environment and calibrating the device point cloud coordinates in the initial digital twin map according to the theoretical device coordinates in the device information table to obtain the digital twin map include:

[0011] Acquire sensor data of the inspection environment, and perform data fusion and filtering noise reduction on the sensor data to generate an initial digital twin map. The sensor data includes lidar point cloud data, infrared thermal imaging temperature data, and camera visual data.

[0012] The least squares method is used to align the device point cloud data in the initial digital twin map with the theoretical device coordinates in the device information table and remove noise to obtain calibrated device point cloud data.

[0013] The initial digital twin map is updated based on the calibrated device point cloud data to obtain a digital twin map.

[0014] Further, the step of generating reachable paths using a density-based clustering path planning algorithm based on the device coordinate set and obstacle coordinate set in the digital twin map, and sampling inspection points on the reachable paths to obtain a set of safety inspection points includes:

[0015] Density clustering algorithm is used to perform cluster analysis on the set of device coordinates in the digital twin map to obtain several densely populated areas of devices;

[0016] Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, a path planning algorithm is used to generate reachable paths in each densely populated area of ​​devices;

[0017] The accessible path is sampled at fixed intervals to obtain a set of safety inspection points.

[0018] Furthermore, the step of using a density clustering algorithm to perform cluster analysis on the set of device coordinates in the digital twin map to obtain several densely populated areas of devices includes:

[0019] Based on the set of device coordinates, the Euclidean distance between adjacent devices is calculated using a sliding window mechanism to obtain a local spacing sequence corresponding to each sliding window;

[0020] Based on the local spacing sequence of each sliding window, calculate the local average spacing and local spacing standard deviation of each sliding window, and calculate the global average spacing.

[0021] Determine whether the ratio between the local spacing standard deviation and the global average spacing of each sliding window is less than a distance threshold. If it is, then the sliding window is designated as the first sliding window; otherwise, the sliding window is designated as the second sliding window.

[0022] Based on the global average spacing, the local average spacing, and the preset scaling factor, the first neighborhood radius of each power device in the first sliding window and the second neighborhood radius of each power device in the second sliding window are obtained.

[0023] Based on the first neighborhood radius and the second neighborhood radius, a density-based noise-based spatial clustering algorithm is used to perform cluster analysis on the power equipment, resulting in several densely populated equipment areas.

[0024] Further, the step of obtaining the first neighborhood radius of each power device within the first sliding window and the second neighborhood radius of each power device within the second sliding window based on the global average spacing, the local average spacing, and a preset scaling factor includes:

[0025] The product of the global average spacing and the preset scaling factor is used as the first neighborhood radius of each power device within the first sliding window.

[0026] The product of the local average spacing and the scaling factor is used as the second neighborhood radius of each power device within the second sliding window.

[0027] Further, the step of generating reachable paths in each densely populated area of ​​devices using a path planning algorithm based on the device coordinate set and obstacle coordinate set in the digital twin map includes:

[0028] Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, an improved fast search random tree algorithm is used to search for nodes and obtain the reachable path for each densely populated area of ​​devices.

[0029] The path cost function of the improved fast search random tree algorithm consists of the device path length and the obstacle risk value.

[0030] Furthermore, the improved fast search random tree algorithm employs a density-guided sampling strategy for random point sampling.

[0031] Furthermore, the step of generating the optimal inspection path using a multi-objective path planning algorithm based on the set of safety inspection points and the set of obstacle coordinates includes:

[0032] Based on the set of safety inspection points and the set of obstacle coordinates, calculate the obstacle risk value for each safety inspection point;

[0033] The optimal inspection path is obtained by using the objective functions of minimizing the path length and minimizing the sum of obstacle risk values ​​at each safety inspection point, and solving the problem with a multi-objective path planning algorithm.

[0034] Furthermore, the multi-objective path planning algorithm is any one of genetic algorithm, particle swarm optimization algorithm, and ant colony algorithm.

[0035] Secondly, the present invention provides an intelligent inspection system based on digital twins and graph databases, the system comprising:

[0036] The information query module is used to perform semantic parsing on the inspection task book of power equipment to obtain a structured task table. The equipment names in the structured task table are input into a preset graph database to obtain an equipment information table.

[0037] The digital twin module is used to construct an initial digital twin map of the inspection environment, and to calibrate the device point cloud coordinates in the initial digital twin map according to the theoretical coordinates of the devices in the device information table, so as to obtain the digital twin map.

[0038] The inspection point generation module is used to generate reachable paths based on the set of equipment coordinates and the set of obstacle coordinates in the digital twin map, using a density clustering-based path planning algorithm, and to sample inspection points on the reachable paths to obtain a set of safety inspection points.

[0039] The path planning module is used to generate the optimal inspection path based on the set of safety inspection points and the set of obstacle coordinates using a multi-objective path planning algorithm.

[0040] The intelligent inspection module is used to control the inspection robot to perform equipment inspection based on the optimal inspection path and the equipment information table, obtain equipment status data, and determine whether there is any abnormality in the power equipment based on the equipment status data.

[0041] This invention provides an intelligent inspection method and system based on digital twins and graph databases. Through multimodal data fusion and coordinate calibration, this invention improves the accuracy of digital twin mapping; through a path planning algorithm based on dynamic environment perception, it enhances path planning efficiency and the safety of inspection paths; and through intelligent data analysis, it improves the accuracy of fault analysis, thereby improving the operation and maintenance efficiency of power equipment and ensuring the safety and stability of power equipment operation. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the intelligent inspection method based on digital twins and graph databases in an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of the structure of an intelligent inspection system based on digital twins and graph databases in an embodiment of the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Please see Figure 1 The first embodiment of the present invention proposes an intelligent inspection method based on digital twins and graph databases, including steps S10-S50:

[0046] Step S10: Semantic parsing is performed on the inspection task book of the power equipment to obtain a structured task table. The equipment names in the structured task table are input into a preset graph database to obtain an equipment information table.

[0047] Step S20: Construct an initial digital twin map of the inspection environment, and calibrate the device point cloud coordinates in the initial digital twin map according to the theoretical coordinates of the devices in the device information table to obtain the digital twin map;

[0048] Step S30: Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, a path planning algorithm based on density clustering is used to generate a reachable path, and inspection points are sampled on the reachable path to obtain a set of safety inspection points.

[0049] Step S40: Based on the set of safety inspection points and the set of obstacle coordinates, a multi-objective path planning algorithm is used to generate the optimal inspection path;

[0050] Step S50: Based on the optimal inspection path and the equipment information table, control the inspection robot to perform equipment inspection, obtain equipment status data, and determine whether there is any abnormality in the power equipment based on the equipment status data.

[0051] In this invention, the inspection task book for power equipment described in natural language is first subjected to semantic parsing and structuring. Specifically, a pre-trained language model (BERT model) is used to perform semantic analysis on the inspection task book, extracting the name of the power equipment to be inspected (e.g., transformer), inspection items (e.g., oil level detection), and inspection cycle (e.g., daily). For missing cycles (e.g., a certain device in the inspection task book is to be inspected periodically), regular expressions and context inference are used to supplement them with specific cycles (e.g., set to monthly). Then, the extracted equipment names, inspection items, and inspection cycles are structuring to obtain a structured task table.

[0052] Because power equipment typically involves multi-dimensional relationships such as spatial location and functional dependencies, traditional databases require multiple table joins for queries. This leads to a sharp performance drop when dealing with large datasets. Furthermore, when the relationships between devices change dynamically (e.g., adding new equipment or relocating devices), relational databases require frequent table structure modifications, resulting in high maintenance costs. Therefore, this embodiment employs a graph database for equipment management. Graph databases can handle complex relationships. In a graph database, entities such as equipment, location, and maintenance records are represented by nodes, and relationships are represented by edges. This allows for direct traversal without multiple table joins. When equipment locations are moved or new dependencies are added, only edge attributes need to be modified, without reconstructing the table structure.

[0053] For the device name in the structured task table, it is input into a pre-established graph database for association retrieval to obtain the device type, spatial coordinates, and functional description. Then, the functional description is segmented and vectorized to verify its consistency with the device type. After the consistency verification is passed, the device information table is output, which contains the device name, device type, theoretical coordinates of the device, and functional description of the device.

[0054] In power line inspection environments, dynamic obstacles such as material handling exist. Inspection paths generated solely based on the static coordinates of the equipment can lead to frequent collisions or stalls for the inspection robot. Therefore, in this embodiment, a digital twin map of the inspection environment is created through real-time data fusion to ensure that the inspection path always conforms to the actual scenario. Specific generation steps include:

[0055] Acquire sensor data of the inspection environment, and perform data fusion and filtering noise reduction on the sensor data to generate an initial digital twin map. The sensor data includes lidar point cloud data, infrared thermal imaging temperature data, and camera visual data.

[0056] The least squares method is used to align the device point cloud data in the initial digital twin map with the theoretical device coordinates in the device information table and remove noise to obtain calibrated device point cloud data.

[0057] The initial digital twin map is updated based on the calibrated device point cloud data to obtain a digital twin map.

[0058] In this embodiment, various types of sensor data about the inspection environment are first acquired, including lidar point cloud data, infrared thermal imaging temperature data, and camera visual data. Then, multi-sensor data fusion is performed. Specifically, the data acquisition timestamps of lidar, camera, and temperature sensor are aligned using a hardware trigger signal to ensure spatiotemporal consistency. The RGB-D image and lidar point cloud are aligned using the ICP (Iterative Closest Point) algorithm to generate a color point cloud. Then, the infrared thermal imaging data is associated with the point cloud coordinates to generate a three-dimensional point cloud with temperature attributes. Kalman filtering is then applied to the three-dimensional point cloud to reduce noise, resulting in a denoised three-dimensional point cloud, thus forming an initial digital twin map. The initial digital twin map contains at least point cloud data of the power equipment and obstacles around the equipment.

[0059] Then, the least squares method is used to align the device point cloud data in the initial digital twin map with the theoretical device coordinates in the device information table, resulting in aligned device point cloud data. This aligned point cloud data is then input into a trained SVM classification model to classify device points and background noise, removing points classified as background noise to obtain calibrated device point cloud data. Finally, the calibrated device point cloud data and real-time sensor data are used to reconstruct the digital twin map, resulting in the final digital twin map.

[0060] After the digital twin map is constructed, based on the set of device coordinates and obstacle coordinates in the map, a density-based clustering-based path planning algorithm is used to generate reachable paths. Inspection points are then sampled along these reachable paths to obtain a set of safety inspection points. Specific steps include:

[0061] Density clustering algorithm is used to perform cluster analysis on the set of device coordinates in the digital twin map to obtain several densely populated areas of devices;

[0062] Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, a path planning algorithm is used to generate reachable paths in each densely populated area of ​​devices;

[0063] The accessible path is sampled at fixed intervals to obtain a set of safety inspection points.

[0064] In this embodiment, density clustering algorithms are first used to identify densely distributed areas of devices, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise) or OPTICS (Ordering Points To Identify the Clustering Structure). Then, in each densely distributed area, path planning algorithms are used to generate reachable paths within the area, such as RRT (Rapidly-exploring RandomTree) or PRM (Probabilistic Roadmap). Finally, uniform sampling is performed along the reachable paths to obtain security inspection points. The security inspection points in each area are then combined into a set of securely reachable inspection points, thus obtaining the security inspection point set.

[0065] This embodiment identifies densely populated equipment areas and generates inspection points in each area, ensuring that inspection points cover all critical equipment. It also filters out points blocked by obstacles through reachable paths, improving the safety of inspection points. Furthermore, it transforms the full-space search problem into a local path optimization problem, reducing the computational complexity of inspection point generation.

[0066] In a preferred embodiment, the present invention employs an improved DBSCAN algorithm for cluster analysis, the specific steps of which include:

[0067] Based on the set of device coordinates, the Euclidean distance between adjacent devices is calculated using a sliding window mechanism to obtain a local spacing sequence corresponding to each sliding window;

[0068] Based on the local spacing sequence of each sliding window, calculate the local average spacing and local spacing standard deviation of each sliding window, and calculate the global average spacing.

[0069] Determine whether the ratio between the local spacing standard deviation and the global average spacing of each sliding window is less than a distance threshold. If it is, then the sliding window is designated as the first sliding window; otherwise, the sliding window is designated as the second sliding window.

[0070] Based on the global average spacing, the local average spacing, and the preset scaling factor, the first neighborhood radius of each power device in the first sliding window and the second neighborhood radius of each power device in the second sliding window are obtained.

[0071] Based on the first neighborhood radius and the second neighborhood radius, a density-based noise-based spatial clustering algorithm is used to perform cluster analysis on the power equipment, resulting in several densely populated equipment areas.

[0072] In this embodiment, the DBSCAN algorithm (also known as the density-based noise spatial clustering algorithm) is a density-based clustering algorithm that can discover clusters of arbitrary shapes and effectively process noisy data. Its core idea is to define the density by the neighborhood radius and the minimum number of points (min_samples) to divide the data points into core points, boundary points and noise points.

[0073] Here, a core point is a sample whose neighborhood contains at least `min_samples` points; a boundary point is a point located in the neighborhood of a core point but not satisfying the core point condition itself; a noise point is a sample that is neither a core point nor a boundary point; density reachability and density connectivity refer to forming clusters through recursive expansion of the core point's neighborhood. The algorithm flow includes: traversing all unvisited points and determining if they are core points; if they are core points, recursively expanding their neighborhoods and grouping density reachable points into the same cluster; marking non-core points that cannot be expanded as noise.

[0074] The difference between the improved DBSCAN algorithm in this embodiment and the conventional DBSCAN algorithm lies in the definition of the neighborhood radius. In the conventional DBSCAN algorithm, the neighborhood radius is a fixed value preset by the user, which is also a global fixed value. In the actual scenario of power equipment inspection, the equipment distribution is uneven, and the global fixed value cannot adapt to the changes in data density, resulting in over-segmentation or under-segmentation. Over-segmentation refers to the dense area being split into multiple small clusters, while under-segmentation refers to the equipment in the sparse area not being effectively clustered.

[0075] To address the issue of uneven segmentation, this embodiment employs a dynamic neighborhood radius calculation method. By judging the differences in device spacing in local areas, the neighborhood radius is dynamically adjusted. Specifically, a sliding window is first defined, assuming the window size is k and the step size is l. That is, each time the spacing of k consecutive devices is analyzed, the window slides sequentially in the device sequence, moving the position of l devices each time.

[0076] Based on the sliding window mechanism, the Euclidean distance between adjacent devices is calculated to obtain the local spacing sequence within each sliding window. Based on the local spacing sequences of all sliding windows, the global average spacing between all devices, as well as the local average spacing and local spacing standard deviation for each sliding window, are calculated. The ratio of the local spacing standard deviation to the global average spacing is used to determine whether the spacing difference within a local region is too large. If the ratio is greater than a spacing threshold, the spacing difference is considered significant, and the neighborhood radius of all device points within that window needs dynamic adjustment. Otherwise, no dynamic adjustment of the neighborhood radius is required. The sliding window that does not need adjustment is defined as the first sliding window, and the sliding window that needs adjustment is defined as the second sliding window. Different methods are used to determine the neighborhood radius of device points within different sliding windows. The specific steps include:

[0077] The product of the global average spacing and the preset scaling factor is used as the first neighborhood radius of each power device within the first sliding window.

[0078] The product of the local average spacing and the scaling factor is used as the second neighborhood radius of each power device within the second sliding window.

[0079] In this embodiment, for the first sliding window, the first neighborhood radius of the device points within the window is calculated based on the global average spacing:

[0080]

[0081] In the formula, Represents the radius of the first neighborhood. The value represents the global average spacing, and α represents the scaling factor; preferably, α = 0.5.

[0082] For the second sliding window, the second neighborhood radius of the device points within the window is calculated based on the local average spacing:

[0083]

[0084] In the formula, Indicates the radius of the second neighborhood. This indicates the local average spacing.

[0085] Finally, based on the neighborhood radii of different sliding windows—that is, based on the first neighborhood radius of device points within the first sliding window and the second neighborhood radius of device points within the second sliding window—the DBSCAN algorithm is used for cluster analysis to obtain multiple densely populated device regions. The specific clustering steps are the same as the standard steps of the DBSCAN algorithm and will not be elaborated upon here.

[0086] In this embodiment, by calculating local statistics through a sliding window, the neighborhood radius is adaptively adjusted, which can reduce the radius of dense areas and expand the radius of sparse areas, thereby avoiding over-segmentation and under-segmentation and improving the rationality of region division.

[0087] In another preferred embodiment, the present invention employs an improved RRT* algorithm for reachability path planning. The RRT* (Rapidly-exploring Random Tree Star) algorithm is an improved version of the RRT (Rapidly-exploring Random Tree Star) algorithm. By introducing reconnection and optimization steps, the generated path is gradually optimized, resulting in a higher quality final path. The reachable path generation steps of the conventional RRT* algorithm include: constructing a structured tree, starting from the root node, and defining a search space; generating random points by randomly sampling in the search space; finding the node closest to the random point in the tree, expanding from the nearest node towards the random point to generate candidate nodes, and checking whether the straight path from the nearest node to the candidate node intersects with an obstacle. If so, random sampling is performed again; otherwise, a dynamic neighborhood radius is defined based on the tree nodes, and all neighboring nodes of the candidate node are searched within the neighborhood radius. The node that minimizes the path cost is selected as the parent node of the candidate node. For each neighboring node, it is checked whether the path to the neighboring node through the candidate node is better. If so, the parent node of the neighboring node is changed to the candidate node, and its path cost is updated. If the candidate node enters the target area, the complete path is returned.

[0088] The difference between the improved RRT* algorithm and the conventional RRT* algorithm in this embodiment is that the path cost function in the conventional algorithm is represented by the total path length, consisting of the cumulative path length from the starting point to the parent node and the Euclidean distance from the parent node to the candidate node. The optimization objective of the conventional path cost function is to minimize the total path length, without considering the importance of different devices in the actual inspection environment and the risk impact of dynamic obstacles around the devices. Therefore, in this embodiment, based on the conventional path cost function, an obstacle risk value based on device weight is added for each candidate node, where the obstacle risk value can be expressed as:

[0089]

[0090] In the formula, Risk(p) represents the obstacle risk value of node p. This represents the obstacle density around node p, obtained by dividing the number of obstacles within a preset radius around node p by the area of ​​that radius. This represents the weight of the device closest to node p. This weight is a preset value and is positively correlated with the historical failure rate of the device. a represents the weight coefficient and b represents the density coefficient.

[0091] By adjusting the path cost function based on the obstacle risk value, we can obtain:

[0092]

[0093] In the formula, Cost(p) represents the path cost of node p. f ) represents the parent node p f The cumulative path cost is λ, where λ represents the risk weight coefficient and Risk(p) represents the obstacle risk value of node p.

[0094] In the path cost function described above, Cost(p) f Cost(p) is the historical cumulative cost from the starting point to the parent node. Using the parent node's cost as a basis ensures that the path cost of the new node, i.e., the candidate node p, inherits global optimality, avoids isolated optimization of path segments, and ensures that the overall path cost from the starting point to the end point is minimized. Since the parent node's Cost(p) is... f The algorithm integrates all target parameters from the starting point to the parent node. The cost of a new node only needs to be accumulated with an incremental term. While ensuring global optimality, it reduces redundant calculations and improves algorithm efficiency. In the parent node cost, the parent node path has already prioritized covering high-weight devices. In the incremental term, the physical distance is accumulated through Euclidean distance, and the obstacle density and device status of the new node are quantified in real time through risk values ​​to avoid the accumulation of risks on fixed paths. The path cost function provided in this embodiment can ensure that the expansion of new nodes takes into account both global optimality and dynamic environment adaptability. It should be noted that the expressions for the risk values ​​and the path cost function mentioned above will be normalized to convert the parameters into dimensionless parameters during actual calculation, and then each dimensionless parameter will be calculated. Of course, conventional path planning algorithms can also be used for reachable path planning, such as the fast random tree algorithm, A* algorithm, Dijkstra's algorithm, etc. The improved fast random tree algorithm mentioned above is only a preferred method and not a specific limitation.

[0095] Furthermore, another difference between the improved RRT* algorithm in this embodiment and the conventional RRT* algorithm is that the conventional RRT* algorithm uses random sampling. However, due to the uneven distribution of equipment in actual inspection environments, completely random sampling may lead to circuitous paths or the omission of key areas. Therefore, in this embodiment, the improved RRT* algorithm adopts density-guided sampling, combining the clustering results of the DBSCAN algorithm to select different sampling areas with a certain probability, thereby prioritizing path generation in densely populated equipment areas. Specifically, a uniformly distributed random number c∈[0,1] is generated for each sampling. If c≤c1, assuming c1=0.8, a cluster is randomly selected from the clustering results, i.e., a densely populated equipment area is randomly selected, and random points are generated within its bounding box. If c>c1, random points are randomly generated in the free space of the entire map. The value of c1 can be adjusted according to actual needs to adapt to different scenarios. For example, in high-security scenarios, increasing c1 can increase the sampling probability within the area and reduce the risk of the robot entering unknown areas. In high-dynamic scenarios, decreasing c1 can reduce the sampling probability within the area and enhance global exploration capabilities. The density-guided sampling strategy in this embodiment can ensure priority coverage of densely populated areas of equipment, increase the detection frequency of key equipment, and retain the ability to discover new paths, thus coping with dynamic obstacles and complex environments.

[0096] For each reachable path within a device density area, inspection points are sampled along the reachable path at fixed intervals to obtain a set of safety inspection points. Then, based on the set of safety inspection points and the set of obstacle coordinates, a dynamic multi-objective path planning algorithm is used to generate the optimal inspection path. The specific steps include:

[0097] Based on the set of safety inspection points and the set of obstacle coordinates, calculate the obstacle risk value for each safety inspection point;

[0098] The optimal inspection path is obtained by using the objective functions of minimizing the path length and minimizing the sum of obstacle risk values ​​at each safety inspection point, and solving the problem with a multi-objective path planning algorithm.

[0099] In this embodiment, the obstacle risk value Risk(s) at each safety inspection point is first considered. i The calculation formula is the same as the obstacle risk value calculation formula mentioned above. Then, a multi-objective optimization function is established by minimizing the path length and the sum of obstacle risk values:

[0100]

[0101] In the formula, m represents the total number of safety inspection points. Indicates safety inspection point s i+1 and security inspection points iEuclidean distance between them, Risk(s) i ) indicates a safety inspection point s i The obstacle risk value. It should be noted that if the path planning algorithm described above uses a conventional path planning algorithm and does not employ the improved fast search random tree algorithm, the multi-objective optimization function here can still be constructed based on the obstacle risk value calculation formula in the above embodiment.

[0102] For the above objective function, conventional multi-objective path planning algorithms can be used to solve it, such as genetic algorithms, particle swarm optimization algorithms, or ant colony optimization algorithms. Taking the NSGA-II algorithm (i.e., non-dominated sorting genetic algorithm) as an example, its algorithm steps include: initializing the population and randomly generating path sequences; non-dominated sorting and hierarchical classification according to the Pareto front; crossover and mutation, exchanging path segments and randomly perturbing them; iterative optimization until convergence, thereby obtaining the optimal inspection path. For specific solution steps, refer to the conventional solution steps of the NSGA-II algorithm, which will not be repeated here.

[0103] After generating the optimal inspection path, a control strategy for the inspection robot is generated based on the optimal inspection path and the equipment information table. This strategy enables the inspection robot to inspect the power equipment listed in the equipment information table according to the optimal inspection path, thereby obtaining equipment status data. The inspection robot then transmits this equipment status data to an external server, where it is analyzed to determine whether any abnormalities exist in the power equipment during the current inspection phase. When analyzing the equipment status data, a data analysis model based on a neural network model can be used, such as a Long Short-Term Memory Neural Network (LSTM) or a Deep Convolutional Neural Network (DCN). The specific model chosen depends on the type of equipment status data and can be one or more appropriate neural network models; no further limitations are imposed here.

[0104] In this embodiment, the digital twin map is dynamically updated based on the collected real-time sensor data to ensure the timeliness of the digital twin map, so that the inspection path of each inspection can respond in a timely manner to the dynamic changes of the inspection environment, thereby improving the inspection efficiency of power equipment.

[0105] This embodiment provides an intelligent inspection method based on digital twins and graph databases. The invention improves the accuracy of digital twin mapping through multimodal data fusion and coordinate calibration, enhances path planning efficiency and inspection path safety through a path planning algorithm based on dynamic environment perception, and improves the accuracy of fault analysis through intelligent data analysis. This improves the operation and maintenance efficiency of power equipment and ensures the safety and stability of power equipment operation.

[0106] Please see Figure 2Based on the same inventive concept, the second embodiment of this invention proposes an intelligent inspection system based on digital twins and graph databases, comprising:

[0107] The information query module 10 is used to perform semantic parsing on the inspection task book of power equipment to obtain a structured task table, and input the equipment name in the structured task table into a preset graph database to obtain an equipment information table.

[0108] The digital twin module 20 is used to construct an initial digital twin map of the inspection environment, and to calibrate the device point cloud coordinates in the initial digital twin map according to the theoretical coordinates of the devices in the device information table, so as to obtain the digital twin map.

[0109] The inspection point generation module 30 is used to generate reachable paths based on the set of equipment coordinates and the set of obstacle coordinates in the digital twin map, using a path planning algorithm based on density clustering, and to sample inspection points on the reachable paths to obtain a set of safety inspection points.

[0110] The path planning module 40 is used to generate the optimal inspection path based on the set of safety inspection points and the set of obstacle coordinates using a multi-objective path planning algorithm.

[0111] The intelligent inspection module 50 is used to control the inspection robot to perform equipment inspection according to the optimal inspection path and the equipment information table, obtain equipment status data, and determine whether there is any abnormality in the power equipment based on the equipment status data.

[0112] The technical features and effects of the intelligent inspection system based on digital twins and graph databases proposed in this invention are the same as those of the method proposed in this invention, and will not be repeated here. Each module in the aforementioned intelligent inspection system based on digital twins and graph databases can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0113] In summary, the present invention proposes an intelligent inspection method and system based on digital twins and graph databases. The method involves semantically parsing the inspection task sheet of power equipment to obtain a structured task table; inputting the equipment names from the structured task table into a preset graph database to obtain an equipment information table; constructing an initial digital twin map of the inspection environment; calibrating the equipment point cloud coordinates in the initial digital twin map based on the theoretical coordinates of the equipment in the equipment information table to obtain a digital twin map; generating reachable paths using a density-based clustering path planning algorithm based on the set of equipment coordinates and the set of obstacle coordinates in the digital twin map; sampling inspection points on the reachable paths to obtain a set of safe inspection points; generating an optimal inspection path using a multi-objective path planning algorithm based on the set of safe inspection points and the set of obstacle coordinates; controlling an inspection robot to inspect the equipment based on the optimal inspection path and the equipment information table, obtaining equipment status data, and determining whether there are any abnormalities in the power equipment based on the equipment status data. This invention improves the accuracy of digital twin mapping through multimodal data fusion and coordinate calibration, enhances path planning efficiency and inspection path safety through a path planning algorithm based on dynamic environment perception, and improves the accuracy of fault analysis through intelligent data analysis. In this way, it improves the operation and maintenance efficiency of power equipment and ensures the safety and stability of power equipment operation.

[0114] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0115] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. An intelligent inspection method based on digital twins and graph databases, characterized in that, include: The inspection task book of power equipment is semantically parsed to obtain a structured task table. The equipment names in the structured task table are input into a preset graph database to obtain an equipment information table. An initial digital twin map of the inspection environment is constructed, and the device point cloud coordinates in the initial digital twin map are calibrated according to the theoretical coordinates of the devices in the device information table to obtain the digital twin map. Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, a path planning algorithm based on density clustering is used to generate reachable paths, and inspection points are sampled on the reachable paths to obtain a set of safety inspection points. Based on the set of safety inspection points and the set of obstacle coordinates, a multi-objective path planning algorithm is used to generate the optimal inspection path. Based on the optimal inspection path and the equipment information table, the inspection robot is controlled to perform equipment inspection, obtain equipment status data, and determine whether there is any abnormality in the power equipment based on the equipment status data. The step of generating reachable paths using a density-based clustering path planning algorithm based on the device coordinate set and obstacle coordinate set in the digital twin map, and sampling inspection points on the reachable paths to obtain a set of safety inspection points includes: Density clustering algorithm is used to perform cluster analysis on the set of device coordinates in the digital twin map to obtain several densely populated areas of devices; Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, a path planning algorithm is used to generate reachable paths in each densely populated area of ​​devices; The accessible path is sampled at fixed intervals to obtain a set of safety inspection points. The step of using a density clustering algorithm to perform cluster analysis on the set of device coordinates in the digital twin map to obtain several densely populated areas of devices includes: Based on the set of device coordinates, the Euclidean distance between adjacent devices is calculated using a sliding window mechanism to obtain a local spacing sequence corresponding to each sliding window; Based on the local spacing sequence of each sliding window, calculate the local average spacing and local spacing standard deviation of each sliding window, and calculate the global average spacing. Determine whether the ratio between the local spacing standard deviation and the global average spacing of each sliding window is less than a distance threshold. If it is, then the sliding window is designated as the first sliding window; otherwise, the sliding window is designated as the second sliding window. Based on the global average spacing, the local average spacing, and the preset scaling factor, the first neighborhood radius of each power device in the first sliding window and the second neighborhood radius of each power device in the second sliding window are obtained. Based on the first neighborhood radius and the second neighborhood radius, a density-based noise-based spatial clustering algorithm is used to perform cluster analysis on the power equipment, resulting in several densely populated equipment areas.

2. The intelligent inspection method based on digital twins and graph databases according to claim 1, characterized in that, The steps of constructing an initial digital twin map of the inspection environment and calibrating the device point cloud coordinates in the initial digital twin map according to the theoretical coordinates of the devices in the device information table to obtain the digital twin map include: Acquire sensor data of the inspection environment, and perform data fusion and filtering noise reduction on the sensor data to generate an initial digital twin map. The sensor data includes lidar point cloud data, infrared thermal imaging temperature data, and camera visual data. The least squares method is used to align the device point cloud data in the initial digital twin map with the theoretical device coordinates in the device information table and remove noise to obtain calibrated device point cloud data. The initial digital twin map is updated based on the calibrated device point cloud data to obtain a digital twin map.

3. The intelligent inspection method based on digital twins and graph databases according to claim 1, characterized in that, The step of obtaining the first neighborhood radius of each power device within the first sliding window and the second neighborhood radius of each power device within the second sliding window based on the global average spacing, the local average spacing, and a preset scaling factor includes: The product of the global average spacing and the preset scaling factor is used as the first neighborhood radius of each power device within the first sliding window. The product of the local average spacing and the scaling factor is used as the second neighborhood radius of each power device within the second sliding window.

4. The intelligent inspection method based on digital twins and graph databases according to claim 1, characterized in that, The step of generating reachable paths in each densely populated area of ​​devices using a path planning algorithm based on the device coordinate set and obstacle coordinate set in the digital twin map includes: Based on the set of device coordinates and the set of obstacle coordinates in the digital twin map, an improved fast search random tree algorithm is used to search for nodes and obtain the reachable path for each densely populated area of ​​devices. The path cost function of the improved fast search random tree algorithm consists of the device path length and the obstacle risk value.

5. The intelligent inspection method based on digital twins and graph databases according to claim 4, characterized in that, The improved fast search random tree algorithm employs a density-guided sampling strategy for random point sampling.

6. The intelligent inspection method based on digital twins and graph databases according to claim 5, characterized in that, The step of generating the optimal inspection path using a multi-objective path planning algorithm based on the set of safety inspection points and the set of obstacle coordinates includes: Based on the set of safety inspection points and the set of obstacle coordinates, calculate the obstacle risk value for each safety inspection point; The optimal inspection path is obtained by using the objective functions of minimizing the path length and minimizing the sum of obstacle risk values ​​at each safety inspection point, and solving the problem with a multi-objective path planning algorithm.

7. The intelligent inspection method based on digital twins and graph databases according to claim 6, characterized in that, The multi-objective path planning algorithm can be any one of genetic algorithm, particle swarm optimization algorithm, and ant colony algorithm.

8. An intelligent inspection system based on digital twins and graph databases, characterized in that, The system is applied to the method as described in any one of claims 1 to 7, comprising: The information query module is used to perform semantic parsing on the inspection task book of power equipment to obtain a structured task table. The equipment names in the structured task table are input into a preset graph database to obtain an equipment information table. The digital twin module is used to construct an initial digital twin map of the inspection environment, and to calibrate the device point cloud coordinates in the initial digital twin map according to the theoretical coordinates of the devices in the device information table, so as to obtain the digital twin map. The inspection point generation module is used to generate reachable paths based on the set of equipment coordinates and the set of obstacle coordinates in the digital twin map, using a density clustering-based path planning algorithm, and to sample inspection points on the reachable paths to obtain a set of safety inspection points. The path planning module is used to generate the optimal inspection path based on the set of safety inspection points and the set of obstacle coordinates using a multi-objective path planning algorithm. The intelligent inspection module is used to control the inspection robot to perform equipment inspection based on the optimal inspection path and the equipment information table, obtain equipment status data, and determine whether there is any abnormality in the power equipment based on the equipment status data.

Images