Dynamic planning method and system for intelligent patrol point location of power transformation equipment

Abstract

The invention discloses a dynamic planning method and system for an intelligent patrol point location of power transformation equipment, and belongs to the technical field of intelligent patrol of power systems, the dynamic planning method for the intelligent patrol point location of the power transformation equipment comprises the following steps: mapping dynamic features and static attributes to the same feature space and carrying out cross-modal association; constructing a defect severity model, and introducing defect severity in risk quantification calculation to generate a point location priority list; a transformer substation three-dimensional point cloud model is built, path nodes are initialized according to a point location priority list, and a greedy algorithm is utilized. A closed-loop decision-making system of multi-modal data fusion is constructed; according to the method, automatic planning driven by a risk quantification model based on defect history, generation of a three-dimensional space non-blind area coverage path, dynamic adjustment triggered by two factors of equipment change and inspection effect, deep collaborative analysis of machine account-defect-real-time data and continuous inspection of a complex scene are guaranteed by a semantic compensation mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent inspection of power systems, and particularly relates to a dynamic planning method and system for intelligent inspection points of power transformation equipment. BACKGROUND

[0002] The intelligent inspection system is deployed at the station end of a substation, perfects comprehensive sensing means in the station, and mainly comprises real-time sensors, image acquisition devices and the like. The intelligent inspection system performs real-time collection on device states and environmental information in the field, performs real-time diagnosis on the device states, effectively replaces manual inspection, and reduces the management drawbacks of excessive maintenance of main devices, thereby greatly improving the operation life of the devices. However, the current intelligent inspection system of power transformation equipment generally has the following deficiencies:

[0003] (1) Low intelligent degree: the inspection points depend on manual experience configuration, and need to set position and angle parameters one by one, and the deployment of a single substation takes more than 40 hours;

[0004] (2) Incomplete coverage of key components: the traditional path planning only considers the static importance weight of the equipment, and does not combine historical defect data, resulting in a missing detection rate of more than 30% for high-risk components such as arc extinguishing chambers of circuit breakers;

[0005] (3) Poor dynamic adaptability: the existing technology cannot automatically adjust the points according to equipment modification or inspection effect feedback;

[0006] (4) Lack of data collaboration: the device account, defect library and real-time sensor data are processed in isolation, and it is difficult to build a complete device health profile;

[0007] (5) Weak environmental robustness: when the preset path fails due to temporary obstructions, manual re-mapping is required.

[0008] It should be noted that the information disclosed in the background section of the present application is only intended to deepen the understanding of the general background of the present application, and should not be regarded as acknowledging or implying in any form that the information constitutes prior art known to those skilled in the art. SUMMARY

[0009] The present application aims to provide a dynamic planning method and system for intelligent inspection points of power transformation equipment, so as to solve the problems of low intelligent degree, incomplete coverage of key components, poor dynamic adaptability, lack of data collaboration and weak environmental robustness of the current intelligent inspection system of power transformation equipment.

[0010] To solve the above technical problems, the present application provides a dynamic planning method for intelligent inspection points of power transformation equipment, comprising the following steps:

[0011] The dynamic features and static attributes are extracted, mapped to the same feature space, and cross-modality correlated, a defect severity model is constructed based on the cross-modality correlated data, and a point priority list is generated by introducing the defect severity in risk quantification calculation.

[0012] A substation three-dimensional point cloud model is built, a path node is initialized according to the point priority list, an initial path is generated by using a greedy algorithm, and the covering angle is optimized by using an A* algorithm, and an optimal inspection path is output after collision detection.

[0013] The equipment account is monitored in real time, and the inspection effect is evaluated, and the point is dynamically adjusted and the inspection path is re-planned.

[0014] The data after cross-modality correlation of part of the dynamic features is obtained, the equipment health index matrix is calculated, and the real-time correction mechanism containing dynamic response and semantic compensation is used to correct the inspection path in real time under the triggering of the equipment health index matrix.

[0015] Preferably, the dynamic features are extracted from the historical defect library, and the dynamic features include defect occurrence frequency, defect severity, fault type, fault repair, maintenance record and real-time sensor data; the static attributes are extracted from the equipment account, and the static attributes include equipment type, voltage level, operation life and key component coordinates.

[0016] Preferably, mapping the dynamic features and static attributes to the same feature space and cross-modality correlation includes:

[0017] The dynamic features and static attributes are normalized;

[0018] The mapping relationship between the static attributes and the dynamic features is established by a feature correlation matrix.

[0019] Preferably, constructing a defect severity model based on cross-modality correlated data includes:

[0020] According to the influence degree of defects on equipment operation and power system safety, the defects are divided into different grades and given corresponding numerical values.

[0021] Preferably, the point priority is calculated by using the following risk quantification formula, and the point priority list is output by sorting the priority:

[0022] P=α×F+β×S+γ×C

[0023] Wherein, P represents the component inspection priority, the value range is 0-1; F is the defect occurrence rate, the unit is times / year; S is the defect severity, the value range is 0-1; C is the fault repair cost, the unit is ten thousand yuan; α, β, γ are weight coefficients, α+β+γ=1.

[0024] Preferably, the optimization of the coverage angle by the A* algorithm comprises:

[0025] The coverage angle is incorporated in the heuristic function of the A* algorithm, which is as follows:

[0026] f(n)=g(n)+h(n)+w×c(n)

[0027] wherein g(n) is the actual distance from the starting node to the current node n, h(n) is the estimated distance from the current node n to the target node, c(n) is the coverage angle of the current node n to the key component, and w is the weight coefficient of the coverage angle.

[0028] Preferably, the real-time monitoring of the equipment account and the evaluation of the inspection effect, and the dynamic adjustment of the point and the re-planning of the inspection path comprise:

[0029] The change of the equipment account is monitored in real time, and after the change content is identified, the added or deleted inspection point is determined according to the changed equipment layout;

[0030] An evaluation model is constructed based on the inspection effect data, and when the inspection point does not meet the standard, the inspection point optimization process is triggered;

[0031] After the addition, deletion or optimization of the inspection point is completed, the adjusted point information is transmitted to the three-dimensional path planning module, and the inspection path is re-planned.

[0032] Preferably, the data obtained after the cross-modal association of part of the dynamic features is used to calculate the equipment health index matrix, which comprises:

[0033] The data obtained after the cross-modal association of part of the dynamic features and the account parameters are synchronized, wherein the part of the dynamic features includes the infrared thermal image, visible light image and millimeter wave radar data collected from the inspection scene;

[0034] The spatio-temporal dimensions are aligned by the cross-modal attention mechanism, and the infrared thermal image feature parameters, visible light image feature parameters, millimeter wave radar feature parameters and account parameters are fused by the time series attention mechanism to generate the equipment health index matrix.

[0035] Preferably, the equipment health index matrix triggers the use of a real-time correction mechanism containing dynamic response and semantic compensation to correct the inspection path in real time, which comprises:

[0036] When the equipment health index of the current equipment exceeds the preset threshold, the equipment health index of the current equipment is used to evaluate the abnormal level, and the corresponding response is performed according to the abnormal level;

[0037] When the device health index of the current device does not exceed the preset threshold, the data after the cross-modal association of the part of dynamic features is analyzed, the occlusion is detected and its position is located, the alternative path is found, when the environment mutation causes the three-dimensional point cloud model to be invalid, the semantic navigation module is started, and temporary path instructions are generated.

[0038] The application provides a transformer equipment intelligent patrol point dynamic planning system.

[0039] In the transformer equipment intelligent patrol point dynamic planning method provided by the application, through the construction of a multi-modal data fusion closed-loop decision system, the risk quantification model based on defect history is realized, automatic planning is driven, three-dimensional space blind area coverage path generation is realized, dynamic adjustment is realized based on equipment change and patrol effect double factors, deep collaborative analysis of account-defect-real-time data is realized, and a semantic compensation mechanism is used to guarantee continuous inspection in complex scenes, so that the intelligent patrol efficiency of the transformer equipment is significantly improved, the planning efficiency is greatly improved, the coverage capacity of high-risk components is enhanced, the missed inspection is effectively avoided, and the adaptability to complex scenes is enhanced.

[0040] The transformer equipment intelligent patrol point dynamic planning system provided by the application belongs to the same inventive concept as the transformer equipment intelligent patrol point dynamic planning method provided by the application, so the transformer equipment intelligent patrol point dynamic planning system provided by the application at least has all the advantages of the transformer equipment intelligent patrol point dynamic planning method provided by the application, and details are not repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0041] Those skilled in the art will understand that the provided drawings are used to better understand the application, and do not constitute any limitation on the scope of the application. Among them:

[0042] Figure 1 is the overall flowchart of an embodiment of the application;

[0043] Figure 2 is the account-defect data fusion flowchart of an embodiment of the application;

[0044] Figure 3 is the path planning schematic diagram of an embodiment of the application;

[0045] Figure 4 is the point dynamic adjustment flowchart of an embodiment of the application;

[0046] Figure 5 is the real-time path correction flowchart of an embodiment of the application;

[0047] Figure 6 is the dynamic path correction strategy schematic diagram of an embodiment of the application. DETAILED DESCRIPTION

[0048] In order to make the objects, advantages and features of the present application more clearly, the following further describes the present application in conjunction with the drawings and specific embodiments. It should be noted that the drawings are very simplified and not drawn to scale, and are only used to facilitate and clearly assist the purpose of describing the embodiments of the present application. In addition, the structures shown in the drawings are often part of the actual structures. In particular, the emphasis shown in each drawing is different, and sometimes different scales are used.

[0049] As used in the present application, the singular forms "a", "an" and "the" include plural referents, the term "or" is generally used in the sense of "and / or", the term "several" is generally used in the sense of "at least one", the term "at least two" is generally used in the sense of "two or more", and in addition, the terms "first", "second", "third" are only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second", "third" can explicitly or implicitly include one or at least two of the features, the term "proximal" generally refers to the end closer to the operator, the term "distal" generally refers to the end closer to the patient, "one" and "the other" and "proximal" and "distal" generally refer to two parts corresponding to each other, which not only includes the end point, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or it can be integrated; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the internal communication or interaction relationship between two elements. In addition, as used in the present application, a component is provided in another component, which generally only indicates that there is a connection, coupling, cooperation or transmission relationship between the two components, and the two components can be directly or indirectly connected, coupled, cooperated or transmitted through an intermediate component, and cannot be understood as indicating or implying the spatial positional relationship between the two components, i.e. one component can be in any orientation inside, outside, above, below or one side of another component, unless the content is otherwise explicitly indicated. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0050] Research shows that the existing intelligent patrol system often relies on manual experience for patrol point arrangement, and has a series of problems such as long time consumption, low accuracy, and incomplete coverage of key components.

[0051] Based on this, the present application aims to solve the five core problems of low artificial planning efficiency, high-risk component monitoring blind area, dynamic response lag, data island effect and poor environmental adaptability in the prior art, by constructing a closed-loop decision system of multi-modal data fusion, realizing risk quantization model driven automatic planning based on defect history, three-dimensional space blind area free path generation, dynamic adjustment triggered by equipment change and patrol effect double factors, deep collaborative analysis of account-defect-real-time data and semantic compensation mechanism guaranteeing continuous inspection in complex scenes, and finally achieving the goal of 10 times patrol efficiency improvement and >99% high-risk component coverage rate.

[0052] Specifically, please refer to Figures 1-6 , which is a schematic diagram of an embodiment of the present application. As Figure 1 shows, a power transformation equipment intelligent patrol point dynamic planning method comprises the following steps:

[0053] Step one, extract dynamic features and static attributes, map to the same feature space and perform cross-modal correlation, build a defect severity model based on cross-modal correlated data, and introduce defect severity in risk quantization calculation to generate a point priority list.

[0054] Step two, build a three-dimensional point cloud model of the substation, initialize the path node according to the point priority list, use the greedy algorithm to generate the initial path, and then optimize the coverage angle by A* algorithm to output the optimal patrol path after collision detection.

[0055] Step three, real-time monitoring of equipment account and evaluation of patrol effect, dynamic adjustment of point and re-planning of patrol path.

[0056] Step four, obtain the data after cross-modal correlation of part of the dynamic features, calculate the equipment health index matrix, and under the trigger of the equipment health index matrix, use the real-time correction mechanism containing dynamic response and semantic compensation to correct the patrol path in real time.

[0057] As Figure 1 shows, the overall architecture flowchart, the data fusion engine first cleans, standardizes and spatiotemporal correlation processes the multi-source heterogeneous data, maps the static attributes such as equipment model, installation position, key component coordinates in the equipment account, dynamic features such as defect occurrence frequency, fault type, maintenance record in the historical defect library, and data such as real-time temperature, vibration amplitude, image clarity of real-time sensors to the unified data space; introduce defect severity in risk quantization calculation, calculate the patrol priority of each component, generate a point priority list; use three-dimensional space modeling technology to generate the optimal patrol path that meets the full-view coverage, complete the automatic division and preset of the point; realize the dynamic adjustment of the point through the double-factor trigger mechanism of equipment change monitoring and patrol effect evaluation, and combine the real-time correction algorithm of multi-modal fusion to deal with complex scenes such as temporary occlusion.

[0058] In step one, the account-defect data fusion and priority modeling are performed. Static attributes and dynamic features are extracted from the equipment account and the historical defect library respectively, converted to the same feature space by a feature mapping algorithm, and cross-modal correlation is performed. The attention mechanism is used to give higher weight to high-risk features. Finally, the inspection priority of each component is calculated based on the risk quantification formula, and the point priority list is output. The account-defect data fusion flowchart is shown in Figure 2 .

[0059] The dynamic features reflecting the equipment failure risk are extracted from the historical defect library, including defect occurrence frequency, defect severity, failure type, failure repair, maintenance record and real-time sensor data. It can be understood that the dynamic features reflecting the equipment failure risk are mined from the historical defect library, at least covering the defect occurrence rate, defect severity, failure repair cost. These features are dynamically updated with the change of equipment running time and working conditions, which can reflect the actual failure risk status of the equipment, making the priority evaluation more in line with the actual operation of the equipment.

[0060] The static attributes related to equipment inspection are extracted from the equipment account, including equipment type, voltage level, running time and key component coordinates. The fixed feature information related to equipment inspection is selected and extracted from the equipment account, including equipment type, voltage level, running time, key component coordinates, etc. These information are relatively stable during the operation of the equipment, which is the key basis for evaluating the basic importance of the equipment and provides basic data support for subsequent priority calculation.

[0061] Specifically, the dynamic features and static attributes are mapped to the same feature space and cross-modal correlation includes:

[0062] The dynamic features and static attributes are normalized;

[0063] The mapping relationship between the static attributes and the dynamic features is established by a feature correlation matrix.

[0064] Among them, the static attributes and dynamic features are normalized respectively to eliminate the dimensional difference of different data types, and the normalization formula is as follows:

[0065]

[0066] x is the original data, and x' is the normalized data. Then the mapping relationship between the static attributes and the dynamic features is established by a feature correlation matrix, and the feature correlation matrix is constructed by calculating the Pearson correlation coefficient of the static attributes and the dynamic features, and the formula is:

[0067]

[0068] wherein x i is static attribute data, y i is dynamic feature data, and are their mean values respectively, and r is a correlation coefficient, thereby forming a mapping matrix to enable information originally belonging to different data domains to be compared and fused in the same dimension.

[0069] Specifically, the construction of the defect severity model based on cross-modal correlation includes:

[0070] According to the influence of the defect on the operation of the equipment and the safety of the power system, the defect is divided into different levels and is given corresponding numerical values.

[0071] For example, according to the influence of the defect on the operation of the equipment and the safety of the power system, the defect is divided into different levels and is given corresponding numerical values, so as to accurately reflect the difference in the severity of the defect in the risk quantification calculation and provide a clear basis for priority assessment. The characteristics and influence range of defects of different severity are defined by the defect severity division rule, so that the quantification of defect severity is more standardized and accurate. The defect severity division rule table is shown in Table 1 as follows:

[0072] Table 1: Defect severity division rule

[0073]

[0074] Based on the fusion of the data, the severity of the defect is quantified, and a point priority assessment model is constructed. The point priority is calculated using the following risk quantification formula, and a point priority list is output for priority sorting:

[0075] P = α × F + β × S + γ × C

[0076] Wherein, P represents the component inspection priority, the value range is 0-1, the larger the value, the higher the priority; F is the defect occurrence rate, unit is times / year; S is the defect severity, the value range is 0-1, 1 represents a fatal defect; C is the fault repair cost, unit is ten thousand yuan; α, β, γ are weight coefficients, α+β+γ=1, which are obtained by training historical data.

[0077] Through the cross-modal data correlation algorithm, the deep fusion of the equipment account and the historical defect data is realized, and a point priority assessment model based on risk quantification is constructed, thereby providing accurate priority basis for the planning of intelligent inspection points of the power transformation equipment, ensuring that high-risk and important equipment components can be monitored first, and improving the pertinence and effectiveness of the inspection.

[0078] In step two, a three-dimensional space of the substation is built, and a path with full-view coverage is planned. Based on three-dimensional point cloud modeling and path optimization algorithms, an optimal inspection path is generated that can achieve blind area coverage of key components, ensuring that the inspection equipment can monitor key components from multiple angles, improving the accuracy of equipment operation state evaluation, while reducing the redundancy of the inspection path and improving the efficiency of the inspection.

[0079] As shown in Figure 3 , a three-dimensional point cloud model of the substation is constructed using a laser radar, path nodes are initialized according to a point priority list, a greedy algorithm is used to generate an initial path, and an improved A* algorithm is used to optimize the path to achieve full-view coverage. After collision detection to ensure no obstruction, the optimal inspection path is output.

[0080] Building a three-dimensional point cloud model of the substation includes: using a laser radar to perform a full-range scan of the substation scene to obtain three-dimensional coordinate information of all objects in the scene, and performing processing steps such as point cloud filtering, registration, and segmentation to construct a three-dimensional point cloud model containing device positions, sizes, and obstruction distributions. This model can accurately reflect the actual spatial layout of the substation, providing real environmental data support for subsequent path planning.

[0081] First, the path nodes are initialized. According to the point priority list, the center points of components with higher priority are selected as initial path nodes. These nodes are key stopping points for the inspection path, ensuring that high-priority components are monitored first.

[0082] Second, a greedy algorithm is used to generate an initial path: starting from the starting node, the distances from the current node to all unvisited nodes are calculated, and the node with the shortest distance is selected as the next visited node. It is added to the path and marked as visited, and the above steps are repeated until all nodes are visited, forming a preliminary inspection path. This algorithm can quickly generate a feasible path, but may have problems such as long path length or incomplete coverage.

[0083] Third, an improved A* algorithm is used to optimize the path. In the A* algorithm, the coverage angle is introduced, and the A* algorithm is used to optimize the coverage angle, including:

[0084] The coverage angle is included in the heuristic function of the A* algorithm, which is as follows:

[0085] f(n) = g(n) + h(n) + w x c(n)

[0086] where g(n) is the actual distance from the start node to the current node n, h(n) is the estimated distance from the current node n to the target node, c(n) is the coverage angle of the current node n to the key component, and w is the weight coefficient of the coverage angle.

[0087] The heuristic function of the traditional A* algorithm only considers the distance factor. The new algorithm proposed in the application incorporates the coverage angle factor into the heuristic function, taking the shortest path length and the most comprehensive coverage angle as the dual optimization goals, guiding the algorithm to preferentially select a better path direction in the search process, and considering the monitoring needs of different angles of the key component to ensure that each node on the path can achieve multi-dimensional coverage of the component.

[0088] Further, the path segments that may collide with the device or the obstruction are identified, and these path segments are locally adjusted according to the spatial information in the three-dimensional point cloud model to re-plan the bypass route, thereby finally ensuring the full-angle coverage of the key component. The core code is as follows:

[0089] Input: three-dimensional point cloud model M, priority list P

[0090] Output: optimal inspection path R

[0091] 1. Initialize the path node set N = {p | p is the component center point with the top k priority in P}

[0092] 2. Calculate the Euclidean distance matrix D between nodes

[0093] 3. Generate the initial path R0 using the greedy algorithm:

[0094] a. Select the start node s and mark it as visited

[0095] b. Current node current = s

[0096] c. while there is an unvisited node:

[0097] i. Find the unvisited node next closest to current

[0098] ii. Add next to R0 and mark it as visited

[0099] iii. current = next

[0100] 4. Iteratively optimize R0:

[0101] for i = 1 to n:

[0102] Randomly select path segment R0[j:j+2]

[0103] Calculate the locally optimal sub-path R' using the improved A* algorithm

[0104] If R' length < R0[j:j+2] length and no occlusion:

[0105] R0 = R0[0:j] + R' + R0[j+2:]

[0106] 5. Output optimized path R = R0

[0107] It can be understood that in the optimization process of the path, evaluation indexes are set from three key dimensions of path efficiency, coverage effect and safety, to provide clear goals and measurement standards for path optimization. The path length reflects the patrol efficiency, the coverage angle reflects the monitoring effect on the key components, and the non-collision safety ensures the safety of the patrol process. Through these indexes, the advantages and disadvantages of the path can be effectively evaluated to guide the path optimization process. The path optimization evaluation index table is shown in Table 2.

[0108] Table 2: Path optimization evaluation index

[0109] Indicator Name Indicator Definition Optimization Goal Path Length Total Distance of Inspection Path Minimize Coverage Angle Angle Range in Which Critical Components Are Monitored Maximize Collision-Free Whether the Safety Distance of the Path to Equipment, Obstructions Is Up to Standard Meet Safety Threshold

[0110] In step three, device change monitoring and patrol effect evaluation are introduced as two influencing factors for adjusting the patrol point, real-time monitoring of device account change and evaluation of patrol effect, when detecting device modification or continuous substandard patrol point, adding, deleting or optimizing the point, and then re-planning the path and updating the point deployment. The point dynamic adjustment process is shown in Figure 4 .

[0111] Specifically, real-time monitoring of device account and evaluation of patrol effect, dynamic adjustment of point and re-planning of patrol path includes:

[0112] Real-time monitoring of device account change, after identifying the change content, determine the added or deleted patrol point according to the changed device layout;

[0113] Based on the patrol effect data, build an evaluation model, when the patrol point is substandard, trigger the patrol point optimization process;

[0114] After completing the addition, deletion or optimization of the patrol point, the adjusted point information is transmitted to the three-dimensional path planning module to re-plan the patrol path.

[0115] In one embodiment, real-time monitoring of device account change, when adding interval, device retirement and other device modification events occur, the system automatically identifies the change content, and determines the added or deleted patrol point according to the changed device layout. The added point needs to meet the requirement of covering the key components of the new device, and the deleted point is for the retired device or the area that no longer needs to be monitored.

[0116] The patrol effect evaluation is used for optimizing the point, and a more optimal evaluation model is constructed based on the patrol effect to periodically evaluate the defect recognition accuracy and missed detection rate of each patrol point. When a point fails to meet the standard for several times, it indicates that the setting of the point may have problems, such as poor angle or too far distance, and the point optimization process is triggered at this time.

[0117] In the read patrol data, the key indicators that can directly reflect the effectiveness of the patrol point are selected, and the threshold is set to determine whether the patrol point can meet the monitoring requirements. When the indicators do not meet the standard, the point optimization process is triggered to ensure that the patrol point can always play an effective role and improve the patrol effect. The patrol effect evaluation indicators and thresholds are shown in Table 3.

[0118] Table 3 Patrol effect evaluation indicators and thresholds

[0119]

[0120] After completing the addition, deletion or optimization of the point, the adjusted point information is transmitted to the three-dimensional path planning module to re-plan the patrol path, so that the new path can adapt to the adjusted point distribution, and ensure the continuity and effectiveness of the patrol work.

[0121] Through the dual factors of equipment change monitoring and patrol effect evaluation, the real-time dynamic adjustment of the patrol point is realized, so that the patrol point can adapt to changes such as equipment modification and patrol effect feedback, and ensure that the patrol work can always effectively cover the equipment components that need to be monitored, and continuously maintain the effectiveness of the patrol.

[0122] In step four, a real-time path correction mechanism is provided to ensure continuous inspection of complex scenes and timely response to equipment abnormalities. First, the image acquisition device of the inspection terminal and various sensors complete data acquisition, then the lightweight fusion inference engine processes the multi-source data in parallel, aligns the time and space dimensions through the cross-modal attention mechanism, and generates the equipment health index (EHI) matrix; if the EHI exceeds the threshold, the dynamic path correction strategy is started, and the corresponding response is executed according to the abnormal level; at the same time, the collected data is analyzed to detect and locate the position of the shielding object, and when the environment mutates and the three-dimensional model is invalid, the semantic map matching and alternative path generation process is started to ensure the completion of the monitoring of the target component. The real-time path correction process is shown in Figure 5 .

[0123] The real-time data of the image acquisition device of the above patrol terminal and various sensors is one of the dynamic features in step one, and this real-time sensor data is also standardized and normalized for processing, and after feature extraction and weight allocation, it is analyzed in a unified data space with the account and defect data, and then mapped and fused through a cross-modal attention mechanism. The specific process is: data cleaning → normalization → cross-modal attention mechanism for spatio-temporal alignment → generation of unified feature representation (EHI matrix).

[0124] Specifically, the data obtained after the cross-modal association of part of the dynamic features is used to calculate the equipment health index matrix, which includes:

[0125] The data obtained after the cross-modal association of part of the dynamic features and the account parameters are obtained synchronously, and the part of the dynamic features includes the infrared thermal image, visible light image and millimeter wave radar data collected in the patrol scene;

[0126] The spatio-temporal dimensions are aligned through the cross-modal attention mechanism, and the infrared thermal image feature parameters, visible light image feature parameters, millimeter wave radar feature parameters and account parameters are fused using the time series attention mechanism to generate the equipment health index matrix.

[0127] As can be understood, this part of the dynamic features is captured by the image acquisition device and the sensor deployed on the patrol equipment, which captures the infrared thermal image, visible light image and millimeter wave radar data in the patrol scene in real time, and synchronously obtains the equipment parameters in the account, thereby providing raw data for subsequent data processing and analysis.

[0128] The multi-source data parallel processing of this type of dynamic feature and the generation of the EHI matrix include: deploying a lightweight fusion inference engine on the inspection terminal, constructing a multi-source data parallel processing channel, and synchronously processing the collected infrared thermal image, visible light image, millimeter wave radar and account parameters. The spatio-temporal dimensions are aligned through the cross-modal attention mechanism, and the infrared thermal image feature parameters, visible light image feature parameters, millimeter wave radar feature parameters and account parameters are fused using the time series attention mechanism to generate the equipment health index (EHI) matrix. The infrared thermal image feature parameters are the temperature field T(x, y, t), the visible light image feature parameters are the appearance feature V, the millimeter wave radar feature parameters are the deformation AS, and the account parameters are the rated temperature rise Tmax. The equipment health index matrix formula is as follows:

[0129] EHI = σ(W T ·||T-T max ||+W V ·V damage +W S ·ΔS)

[0130] Where σ is an activation function, W T , W VW S These are the weighting coefficients for temperature, appearance, and deformation characteristics, respectively, V. damage These are the characteristic values ​​for appearance damage.

[0131] Specifically, triggered by the device health index matrix, a real-time correction mechanism including dynamic response and semantic compensation is used to correct the inspection path in real time, including:

[0132] When the current device's health index exceeds a preset threshold, the current device's health index is used to assess the anomaly level, and the corresponding response is executed according to the anomaly level.

[0133] In one implementation, the EHI (Equipment Health Index) value is used to determine the level of abnormality. When it is detected that the EHI of a device exceeds the threshold, the corresponding three-level response strategy is triggered.

[0134] like Figure 6 As shown, if the EHI value slightly exceeds the threshold, a primary response is activated, supplementing the observation points for the equipment, such as adding three preset observation points around the equipment (elevation angle 30°-45°), and initiating multi-angle high-definition imaging and infrared thermometry. If the EHI value is significantly abnormal, a medium response is activated, calling up similar equipment failure cases from the historical defect database, comparing discharge characteristic spectra, and pushing diagnostic suggestions to the backend. If the EHI value severely exceeds the threshold, it is determined to be an emergency defect, the path is adjusted in real time, low-priority tasks are suspended, and an optimal return path is generated. This tiered mechanism ensures that the response strategy matches the severity of the anomaly, guaranteeing both the continuity of inspections and improving the efficiency of emergency handling. Obviously, a more refined anomaly level assessment for the EHI value can also be provided, such as setting four or more levels, and refining the response strategy accordingly, which will not be elaborated here.

[0135] When the device health index of the current device does not exceed the preset threshold, the data after cross-modal correlation of the dynamic features is analyzed, occlusions are detected and their positions are located, and alternative paths are found. When environmental changes cause the 3D point cloud model to fail, the semantic navigation module is activated to generate temporary path instructions.

[0136] If the device health index does not exceed the preset threshold, then the obstruction is detected and located. The image recognition algorithm is used to analyze the collected image data to detect whether there is a temporary obstruction, and to determine the type and approximate range of the obstruction. Combined with GPS positioning and sensor data such as gyroscope angle, the specific position coordinates of the obstruction in three-dimensional space are accurately calculated.

[0137] According to the size and characteristics of the shield, different types of shield are classified, and corresponding alternative path generation strategies are developed for different types of shield, so as to ensure that suitable paths can be found to bypass the shield under various shielding conditions, and effective monitoring of key components is ensured, and the adaptability of the system in complex scenes is improved. For example, the shield type and coping strategy proposed in Table 4 below.

[0138] Table 4 Shield type and coping strategy

[0139]

[0140] A pre-constructed semantic map containing semantic information and spatial position relationship of equipment, roads, fixed facilities, etc. in the substation is called, the position coordinates of the shield are matched with the semantic map, and the alternative path that can bypass the shield and cover the target component is found. When the environment mutates and the three-dimensional model fails, the semantic navigation module is started, the equipment nameplate text is analyzed by a large language model, and temporary path instructions are generated by combining the topological relationship in the account, so that the inspection equipment can avoid the shield and complete the monitoring of the key components.

[0141] Fusion of multi-source sensor data and image recognition technology, construction of real-time correction mechanism containing dynamic response and semantic compensation, both can trigger hierarchical path adjustment through equipment health state evaluation, and can cope with complex scenes such as temporary shielding, to ensure the continuity of the inspection process, the timeliness of equipment anomaly monitoring and the efficiency of emergency handling.

[0142] Based on the same technical concept, the present application also provides a dynamic planning system for intelligent inspection points of power transformation equipment, which arranges the inspection points by using the dynamic planning method for intelligent inspection points of power transformation equipment as described above.

[0143] As shown in Figure 1 The data fusion engine in the dynamic planning system for intelligent inspection points of power transformation equipment is used to clean, standardize and space-time correlation process multi-source heterogeneous data, map static parameters such as equipment model, installation position, key component coordinates in the equipment account, dynamic characteristics such as defect frequency, fault type, maintenance record in the historical defect library, and real-time data such as real-time temperature, vibration amplitude, image clarity of real-time sensors to a unified data space; based on the fusion data, a point priority evaluation model is constructed, and the inspection priority of each component is calculated by a risk quantification algorithm; the three-dimensional space modeling technology is used to generate the optimal inspection path that meets the full-view coverage, and the automatic division and preset of the point are completed; the dynamic adjustment of the point is realized through a double-factor triggering mechanism, and the real-time correction algorithm of multi-modal fusion is used to cope with complex scenes such as temporary shielding.

[0144] The application provides a substation equipment patrol point adaptive planning and dynamic correction technology of fusion of equipment account, historical defect library and multi-source sensor data, adopts a four-layer progressive architecture of "data fusion engine-> priority modeling-> three-dimensional path planning-> dynamic closed-loop adjustment" as a whole, realizes space-time alignment of equipment account, historical defect library and multi-source sensor data through a cross-modal attention mechanism, continuously optimizes point deployment strategies in combination with a reinforcement learning algorithm, and forms a complete closed loop from data input to dynamic adjustment. Through multi-modal data fusion and closed-loop decision mechanism, the intelligent patrol efficiency of the substation equipment can be significantly improved, the deployment time of a single substation is reduced from 40 hours to less than 5 minutes, manual intervention operation is reduced by 90%, the planning efficiency is greatly improved, the coverage rate of key components such as circuit breaker arc chambers breaks through 95%, the defect recognition rate is stable at more than 95%, the coverage ability of high-risk components is enhanced, the point update delay after equipment modification is compressed from a week to less than 1 second, the real-time adjustment demand is met, and the dynamic response speed is improved, in the case of temporary shielding and other emergencies, the patrol continuity is kept above 98%, the missed detection is effectively avoided, and the adaptability to complex scenes is enhanced, and through actual substation application verification, the loss of equipment failure caused by missed detection of a single station per year is reduced by more than 2 million yuan, and the economic benefit is remarkable.

[0145] The above description is only a description of the preferred embodiments of the application, and does not limit the scope of the application in any way, and any modification or modification of the application by a person skilled in the art according to the above disclosure is within the protection scope of the technical scheme of the application.

Claims

1. A dynamic planning method for intelligent inspection points of substation equipment, characterized in that, Includes the following steps: Dynamic features and static attributes are extracted, mapped to the same feature space, and cross-modal association is performed. Based on the cross-modal association data, a defect severity model is constructed, and a point priority list is generated by introducing defect severity in risk quantification calculation. A 3D point cloud model of the substation is built, path nodes are initialized according to the point priority list, an initial path is generated using a greedy algorithm, the coverage angle is optimized using the A* algorithm, and the optimal inspection path is output after collision detection. Real-time monitoring of equipment records and evaluation of patrol effectiveness; dynamic adjustment of monitoring points and replanning of patrol routes. Data obtained after cross-modal correlation of some dynamic features is acquired, and the equipment health index matrix is ​​calculated. Under the trigger of the equipment health index matrix, a real-time correction mechanism including dynamic response and semantic compensation is used to correct the inspection path in real time.

2. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 1, characterized in that, Dynamic features are extracted from the historical defect database, including defect occurrence frequency, defect severity, fault type, fault repair, maintenance records, and real-time sensor data. Extract static attributes from the equipment ledger. These static attributes include equipment type, voltage level, years of operation, and coordinates of key components.

3. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 1, characterized in that, Mapping dynamic features and static attributes to the same feature space and performing cross-modal association includes: The dynamic features and static attributes are normalized. The mapping relationship between the static attributes and dynamic features is established by using a feature association matrix.

4. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 1, characterized in that, Defect severity models constructed based on cross-modal correlation data include: Based on the degree of impact of defects on equipment operation and power system safety, defects are classified into different levels and assigned corresponding numerical values.

5. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 4, characterized in that, The following risk quantification formula is used to calculate the site priority, and the site priority list is output after sorting the priorities: P = α × F + β × S + γ × C Where P represents the component inspection priority, with a value range of 0-1; F is the defect occurrence rate, in units of times / year; S is the defect severity, with a value range of 0-1; C is the fault repair cost, in units of ten thousand yuan; α, β, and γ are weighting coefficients, α+β+γ=1.

6. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 1, characterized in that, Optimizing the coverage angle using the A* algorithm includes: The coverage angle is incorporated into the heuristic function of the A* algorithm, and the heuristic function is as follows: f(n) = g(n) + h(n) + w × c(n) Where g(n) is the actual distance from the starting node to the current node n, h(n) is the estimated distance from the current node n to the target node, c(n) is the coverage angle of the current node n to the key component, and w is the weighting coefficient of the coverage angle.

7. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 1, characterized in that, Real-time monitoring of equipment records and evaluation of inspection effectiveness, dynamic adjustment of inspection points and replanning of inspection routes include: Real-time monitoring of changes in equipment records; upon identifying changes, determining new or deleted inspection points based on the revised equipment layout. An evaluation model is built based on the inspection results data, and an inspection point optimization process is triggered when the inspection points fail to meet the standards. After adding, deleting, or optimizing inspection points, the adjusted point information is transmitted to the 3D path planning module to replan the inspection path.

8. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 1, characterized in that, The data obtained after cross-modal correlation of some dynamic features are used to calculate the device health index matrix, including: The system synchronously acquires data and ledger parameters after cross-modal correlation of some dynamic features, including infrared thermal images, visible light images and millimeter-wave radar data of the patrol scene. By aligning the spatiotemporal dimensions through a cross-modal attention mechanism, and fusing infrared thermal imaging feature parameters, visible light image feature parameters, millimeter-wave radar feature parameters, and ledger parameters using a temporal attention mechanism, an equipment health index matrix is ​​generated.

9. The method for dynamic planning of intelligent inspection points for substation equipment according to claim 8, characterized in that, Triggered by the equipment health index matrix, a real-time correction mechanism including dynamic response and semantic compensation is used to correct the inspection path in real time, including: When the current device's health index exceeds the preset threshold, the current device's health index is used to assess the anomaly level, and the corresponding response is executed according to the anomaly level. When the device health index of the current device does not exceed the preset threshold, the data after cross-modal correlation of the dynamic features is analyzed, occlusions are detected and their positions are located, and alternative paths are found. When environmental changes cause the 3D point cloud model to fail, the semantic navigation module is activated to generate temporary path instructions.

10. A dynamic planning system for intelligent inspection points of substation equipment, characterized in that, The inspection points are arranged using the dynamic planning method for intelligent inspection points of substation equipment as described in any one of claims 1-9.

Images