A power distribution substation intelligent inspection method and system for urban power grids

Abstract

The present application relates to the technical field of intelligent detection of transformer substations, and more particularly to a power distribution substation intelligent inspection method and system for urban power grids, comprising: collecting multi-source inspection feature data and performing space-time dimension alignment, obtaining a target area corresponding to the to-be-tested equipment in infrared thermal imaging data, and associating and identifying a three-phase equipment group to which the target area belongs; constructing a thermal load evolution offset index of each sampling time, which is used to quantify the matching degree between the cumulative Joule heat effect generated by the current in a specific time period and the actual temperature rise at the sampling time. The present application constructs a thermal balance window in the time dimension to eliminate thermal inertia errors, and combines three-phase balance degree and image texture features in the space dimension to filter optical noise, thereby accurately identifying real high-temperature thermal defects, greatly improving the inspection accuracy of intelligent robots, and releasing grassroots operation and maintenance human resources.

Description

Technical Field

[0001] This invention relates to the field of intelligent substation detection technology. More specifically, this invention relates to an intelligent inspection method and system for distribution substations in urban power grids. Background Technology

[0002] With the acceleration of urbanization, the reliability of urban power grids is crucial to the operation of society. As the core hub of the power grid, the safe and stable operation of internal electrical equipment (such as transformers and circuit breakers) in distribution substations is of paramount importance. In recent years, to improve operation and maintenance efficiency, the use of intelligent inspection robots equipped with visible light cameras and infrared thermal imagers to automate substation inspections has become an important technological means in this field. This intelligent inspection system can collect real-time images of the equipment's appearance and surface temperature distribution, thereby determining whether the equipment has thermal defects or electrical faults.

[0003] In real-world urban power grid substation inspection scenarios, urban electricity demand often fluctuates dramatically throughout the day (such as during morning and evening peak hours), causing the current load on electrical equipment within the substation to change rapidly and frequently. Simultaneously, the physical environment of substations is relatively complex, containing numerous electrical connectors with metallic surfaces, making them highly susceptible to changes in external ambient light. The inspection robot operates under these real-world conditions of drastically fluctuating loads and significant ambient light interference, moving between various observation points to acquire cross-modal images and perform status diagnosis of target equipment using both infrared and visible light.

[0004] However, when using existing intelligent diagnostic technologies to address the issue of equipment thermal defect detection in this scenario, a significant practical pain point exists: a large number of false high-temperature alarms are easily generated, leading to poor system availability and increased maintenance burden. Specifically, due to the thermal inertia of electrical equipment, its temperature rise exhibits a significant lag effect relative to the instantaneous fluctuations in current load. In scenarios where urban electricity consumption peaks cause drastic load fluctuations, existing technologies that directly compare instantaneous temperature with instantaneous current thresholds are prone to generating false alarms before the thermal equilibrium process is complete. Furthermore, isolated reflective bright spots caused by complex lighting on the metal surface of equipment are easily mistaken by existing algorithms for real abnormal heat sources. These judgments, detached from the actual physical properties of the equipment (such as thermal evolution matching degree and spatial three-phase balance), cause the system to frequently output false alarms. This not only forces frontline maintenance personnel to spend countless hours performing ineffective manual verifications on-site, severely wasting human resources, but also causes genuine serious hidden dangers such as poor contact to be submerged in massive amounts of false alarm data, making it difficult for intelligent inspection systems to truly realize their practical value in ensuring power grid safety. Summary of the Invention

[0005] This invention provides an intelligent inspection method and system for distribution substations in urban power grids, aiming to solve the problem of frequent false alarms in related technologies. This not only forces grassroots maintenance personnel to spend countless hours manually verifying data on-site, resulting in a serious waste of human resources, but also causes genuine hidden dangers such as poor contact to be buried in massive amounts of false alarm data, making it difficult for intelligent inspection systems to truly be implemented and realize their practical value in ensuring power grid safety.

[0006] In a first aspect, the present invention provides an intelligent inspection method and system for distribution substations in urban power grids, comprising: collecting multi-source inspection feature data and aligning it in a spatiotemporal dimension; obtaining the target area corresponding to the device under test in infrared thermal imaging data and identifying the three-phase equipment group to which the target area belongs; constructing a thermal load evolution offset index at each sampling time to quantify the degree of matching between the cumulative Joule heating effect generated by the current in a specific time period and the actual temperature rise at that sampling time; constructing a spatial topology compensation factor to quantify the temperature imbalance between the three-phase equipment groups and determining the compensation factor in conjunction with the analysis of the image texture mutation features inside the target area; calculating the final anomaly risk score at each sampling time based on the isolated forest algorithm, and performing anomaly warning based on the magnitude of the final anomaly risk score, wherein the final anomaly risk score is negatively correlated with the product of the average path length of the corresponding sampling point in the isolated forest and the spatial topology compensation factor, and positively correlated with the thermal load evolution offset index. Compared to existing technologies that rely solely on comparing instantaneous temperature and instantaneous current thresholds, which easily leads to numerous false high-temperature alarms, this invention, under real-world inspection conditions with drastic load fluctuations in urban power grids and significant ambient light interference, integrates thermal load evolution (to address thermal hysteresis), spatial topology compensation (to eliminate bright spots caused by reflected light), and the isolated forest algorithm. This enables it to accurately identify genuine equipment thermal defects and potential contact problems, significantly reducing the false alarm rate and preventing frontline maintenance personnel from exhausting themselves with ineffective manual on-site verification.

[0007] Furthermore, a thermal load evolution offset index is constructed for each sampling time. This index is positively correlated with the difference between the highest temperature of the target area and the ambient temperature at the sampling time, and negatively correlated with the mean square current, humidity correction factor, and standard thermal resistance of the equipment within the corresponding thermal equilibrium time window. By introducing the thermal equilibrium time window and the cumulative Joule heating effect, in scenarios with frequent load fluctuations such as morning and evening peak electricity consumption, the diagnostic interference caused by the lag time difference between load spikes and actual temperature rises is effectively eliminated, ensuring the accuracy of the thermodynamic assessment.

[0008] Furthermore, the spatial topology compensation factor is calculated, including: the spatial topology compensation factor is negatively correlated with the sum of the average temperature differences of the three-phase equipment group and positively correlated with the texture mutation index of the target region. Combining the temperature imbalance of the three-phase equipment with the texture mutation characteristics inside the image, it can reliably filter out optical noise and lock the real heat source with outward spread characteristics when facing complex sunlight environment in outdoor or semi-outdoor substations.

[0009] Furthermore, the method for obtaining the texture abrupt change index of the target region includes: calculating the second derivative of the pixel grayscale matrix within the target region, i.e., the Laplacian operator, as the texture abrupt change index of the target region, which reflects the texture abrupt change characteristics of the image. By introducing the Laplacian operator to quickly extract the second derivative of the image texture, a computationally lightweight and efficient feature extraction method is provided for the system, enabling the system to distinguish between real gradient hotspots and false sharp reflective brightspots at the edge in real time.

[0010] Furthermore, the formula for calculating the texture mutation index is as follows: In the formula, Indicates the spatial topology compensation factor; Indicates the target region and its group. The average temperature of the three corresponding regions; This indicates the standard for the maximum permissible interphase temperature difference during normal operation of this type of equipment; Indicates the texture abruptness index of the target region; This indicates a sensitivity adjustment bias. This method enables the system to maintain numerical stability even when encountering extreme pixel abrupt changes caused by strong metallic reflections, further improving the robustness and detection sensitivity of anomaly truncation determination in harsh and complex substation environments.

[0011] Furthermore, anomaly warnings are generated based on the magnitude of the final anomaly risk score, including: triggering an alarm and highlighting the anomaly type on the human-machine interface when the final anomaly risk score exceeds a set threshold for multiple consecutive moments. This method of issuing warnings for multiple consecutive moments filters out interference caused by transient data jitter, and the intuitive highlighting of anomalies on the human-machine interface allows on-site repair personnel to understand the problem immediately, significantly shortening the time for fault location and troubleshooting.

[0012] Furthermore, alarm triggering also includes: retrieving historical load data of the power equipment within the corresponding thermal balance time window; if there is an extreme load exceeding the rated range, the alarm is suppressed; otherwise, the alarm is confirmed. Combining historical load curves eliminates reasonable temperature rises caused by short-term rated overloads, giving the system the intelligent ability to identify compliant extreme operating conditions, preventing the misjudgment of normal physical phenomena as equipment defects, and further purifying the alarm data source.

[0013] Furthermore, obtaining the target area corresponding to the device under test in the infrared thermal imaging data includes: using a pre-trained lightweight target detection model to perform real-time analysis on the high-definition images acquired by the visible light camera, identifying the device under test and outputting its bounding box pixel coordinate range in the visible light image; obtaining the coordinate system transformation matrix between the visible light camera and the infrared thermal imager based on pre-completed binocular vision calibration, and using this matrix to map the pixel coordinate range of the bounding box onto the infrared thermal imaging matrix, thereby locking the target area that completely corresponds to the device under test in the thermal imaging data. By utilizing the lightweight target detection and binocular vision calibration matrix, it is ensured that the inspection robot can accurately map the equipment in the high-definition images to the infrared temperature measurement matrix at each docking observation point, guaranteeing the absolute accuracy of the underlying temperature acquisition objects (such as circuit breaker connectors).

[0014] Furthermore, identifying the three-phase equipment group to which the target area belongs includes: automatically identifying the electrical phase of the current target area based on the equipment ledger information and topology database in the background; and automatically matching the target areas corresponding to the other two phases belonging to the same equipment group as the current target area in the same frame image or adjacent observation points. These three areas together constitute a three-phase equipment group. In urban substations with limited space and severe obstruction, even if the robot's single observation angle is limited and cannot capture all three-phase equipment, the system can automatically piece together a complete three-phase physical association group based on the background ledger and images of adjacent points, ensuring the smooth operation of the spatial temperature comparison logic in complex terrain.

[0015] In a second aspect, an intelligent inspection system for distribution substations in urban power grids is also provided, comprising a processor and a memory, the memory storing a computer program, the processor executing the computer program to implement the intelligent inspection method for distribution substations in urban power grids as described in any of the above embodiments.

[0016] Beneficial effects: To address the pain points of massive false alarms caused by temperature rise lag due to drastic fluctuations in current load and false heating reflections on metal surfaces caused by complex lighting during the inspection of urban power distribution stations, an improved isolated forest algorithm based on physical weighting is proposed. This method constructs a thermal balance window in the time dimension to eliminate thermal inertia error, and combines three-phase balance and image texture features in the spatial dimension to filter optical noise, thereby accurately identifying real high-temperature thermal defects, significantly improving the inspection accuracy of intelligent robots, and freeing up grassroots operation and maintenance human resources. Attached Figure Description

[0017] Figure 1 This is a schematic illustration of an embodiment according to the present invention; Figure 2This is a schematic diagram illustrating the change of the final abnormal risk score with the sampling time under conditions of drastic load fluctuations. Detailed Implementation

[0018] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] S101: Acquisition of multi-source inspection feature data.

[0020] In this embodiment, the inspection robot first moves to the optimal observation point of the device under test (such as a transformer or circuit breaker) according to a preset path. This point ensures that the robot's visible light camera and infrared thermal imager can clearly capture the key connection points of the target device.

[0021] Specifically, this step is implemented as follows: Target region acquisition method: A pre-trained lightweight target detection model is used to perform real-time analysis of high-resolution images acquired by a visible light camera. As a preferred solution, this model can... This series of models was trained using supervised learning on a dataset containing a large number of labeled images of power equipment such as "circuit breaker joints," "cable terminals," and "three-phase bushings." During training, the following methods were employed: The optimizer has the following initial learning rate: and conduct indivual The model undergoes iterative training. When used online, it can quickly identify the device under test in a visible light image and output its pixel coordinate range, i.e., the bounding box.

[0022] Infrared mapping: Based on pre-completed binocular vision calibration, a coordinate transformation matrix is ​​obtained between the visible light camera and the infrared thermal imager. Using this matrix, the coordinates of the device bounding box identified in the visible light image are accurately mapped onto the infrared thermal imaging matrix, thereby locating the target area in the thermal imaging data that perfectly corresponds to the device under test. This accurate cross-modal mapping is the foundation for subsequent accurate temperature rise analysis.

[0023] Three-phase topology association: The system automatically identifies the electrical phase of the current target area based on the equipment ledger information and topology database in the background. Mutually, phase or (Phase 1). Simultaneously, within the same image frame or adjacent observation points, the system automatically matches the target regions corresponding to the other two phases belonging to the same equipment group as the current target region. These three physically and electrically related regions together constitute a three-phase equipment group, providing a benchmark for spatial coherence verification in subsequent steps.

[0024] S102: Construct the thermal load evolution offset index at each sampling time.

[0025] In electrical engineering, the temperature rise of equipment is not an instantaneous response to changes in current load, but rather follows the laws of thermodynamics, exhibiting a significant physical hysteresis, or thermal inertia. Directly comparing instantaneous temperature with instantaneous current can easily lead to false high-temperature alarms under conditions of drastic load fluctuations (such as peak urban electricity consumption) because the thermal equilibrium process has not yet been completed. Therefore, this embodiment constructs a thermal load evolution offset index to eliminate the thermal hysteresis effect and achieve precise alignment of load and temperature rise on the physical time axis.

[0026] To construct this metric, this step introduces the concept of a thermal equilibrium time window. This window is based on the current sampling time. As the endpoint, backtrack one device characteristic thermal time constant. The cycle, that is By integrating the current load within this time window, the instantaneously changing load is converted into effective heat energy that truly reflects the cumulative heat generation effect of the equipment.

[0027] The thermal load evolution offset index is constructed based on the following principle: quantifying the cumulative Joule heating effect generated by the current within a specific time period using an integral sliding window, and then normalizing and comparing it with the actual temperature rise at that moment to assess the degree of matching between the temperature rise and the load. Its calculation formula is as follows: In the formula, for The thermal load evolution offset index at any given time. for The temperature of the highest pixel in the target area is extracted at any time through the infrared thermal imaging matrix. for The ambient reference temperature at any given time can be obtained from a temperature sensor deployed near the device. For The thermal equilibrium time window, with the endpoint at a certain time, is subject to a delay in temperature response relative to current changes due to the device's heat capacity. The value of this thermal equilibrium time window is determined based on the specific heat capacity parameters of the device. Minutes, when the window is smaller At a time window of 1 minute, the cumulative thermal effect of the current may not be fully apparent; when the window is larger than 1 minute... At the minute level, critical transient anomalies may be smoothed out. For time window Real-time current value within. for Real-time ambient humidity. This is a humidity correction factor used to compensate for the impact of ambient humidity on the heat dissipation efficiency of the equipment. In this embodiment, its value is an empirical constant. . This is the standard thermal resistance constant under rated healthy conditions. This parameter reflects the inherent heat dissipation performance of the equipment and is usually calibrated by the equipment's factory specifications or historical healthy operating data. The integral term in the formula... The mean square of the current within the thermal equilibrium time window was calculated to characterize the equivalent average load driving the temperature rise.

[0028] As shown in the formula above, the denominator calculates the average equivalent heat generation power within the thermal equilibrium time window through integration, taking into account the influence of ambient humidity. When parameters such as equipment contact resistance are normal, its actual temperature rise... It should be proportional to the equivalent heat generation power, so that The value is close to zero. If the resistance increases due to poor contact or other reasons, the actual temperature rise will exceed the expectation under the normal physical model, leading to... The value increases significantly, thus indicating a potential physical anomaly.

[0029] S103: Construct spatial topology compensation factors.

[0030] During normal operation, the temperature distribution among the three phases of the three-phase electrical equipment in a substation should exhibit extremely high symmetry and coherence, i.e., three-phase balance. Furthermore, the actual heat generation of the equipment should, in an image, present a textured feature with a heat source center gradually spreading outwards, rather than isolated, high-gradient noise pixels caused by factors such as metal surface reflection. To incorporate these physical priors into the isolated forest algorithm, this embodiment constructs a spatial topological isolated path compensation factor.

[0031] The construction of this indicator is based on: quantifying the temperature imbalance between three-phase devices and analyzing the image texture within the target area to assign a physical weight to statistical outliers, thereby correcting their path length in the isolated forest. The calculation formula is as follows: In the formula, Indicates the spatial topology compensation factor; Indicates the target region and its group. The average temperature of the three corresponding regions; This represents the maximum permissible interphase temperature difference standard for this type of equipment during normal operation, serving as an engineering empirical value; The texture abrupt change index represents the second derivative of the pixel grayscale matrix within the target region, also known as the Laplacian operator. This operator is used to detect subtle textures and noise in an image. A true gradient hotspot has a small second derivative value, while a sharp brightspot formed by reflection has a very large second derivative value. This indicates the sensitivity adjustment bias, and its value is... This is used to prevent the denominator from being too small, which could lead to drastic changes in the factors.

[0032] From this formula, it can be seen that when the three phases are in temperature equilibrium (the numerator term...) When the exponent approaches zero, the exponent term approaches 0. The value will remain high. If the three-phase temperature difference increases significantly, and the texture conforms to the characteristics of heat spread ( (If the value is small), the exponent term will increase sharply due to the large numerator and small denominator. The value will decrease rapidly. This decrease shortens the equivalent path of the sample in the isolated forest, thereby accelerating the algorithm's truncation of the physically real anomaly and improving detection sensitivity. Conversely, if isolated bright pixels caused by metallic reflection appear locally in the image, even if the bright spot causes an increase in the fluctuation of the extracted three-phase temperature difference, the drastic texture change ( (the value is extremely large), the denominator is extremely large This will forcibly lower the value of the entire exponential term, making The value of , on the contrary, continues to remain high, thus effectively suppressing false alarms of such high-gradient noise outliers.

[0033] From this formula, it can be seen that when the three phases are in temperature equilibrium (the numerator term...) Approaching zero) and local texture smoothing ( When the value is small, The value will remain high, close to If the three-phase temperature difference increases significantly, and the texture conforms to the characteristics of heat spread ( If the value increases, then The value will decrease. This decrease shortens the equivalent path of the sample in the isolated forest, thereby accelerating the algorithm's truncation of the physical anomaly and improving detection sensitivity. Conversely, if the three-phase temperature is balanced, but only a single pixel experiences a sudden change due to reflection ( (the value is extremely large) The value will remain high, effectively suppressing false alarms of such noise outliers.

[0034] S104: Construct an improved isolated forest scoring model based on physical weighting.

[0035] Traditional isolation forest algorithms only evaluate the isolation degree of samples from a statistical perspective. The core improvement of this embodiment lies in incorporating the aforementioned constructed temporal dimension features, which contain physical meaning. and spatial dimensional features By deeply integrating the statistical path length of the isolated forest, a multi-dimensional final anomaly risk score model is constructed.

[0036] The model is built upon the premise that a real equipment failure should exhibit coordinated anomalies simultaneously across three dimensions: time (heat load mismatch), space (three-phase imbalance), and statistics (sparse data distribution). The final score is calculated using the following formula: In the formula, express The final abnormal risk score at any given moment; express The average path length of all trees in the isolated forest at a given time. This indicates that for a given number of samples The average path length constant of a binary search tree is used to... Perform normalization; for The thermal load evolution offset index at any given time; : Represents the hyperbolic tangent function, used to shift the heat load evolution exponent. The values ​​are smoothly mapped to The interval serves as a normalization mechanism.

[0037] As can be seen from the above formula, when the statistical path length of a sampling point... Very short, poor spatial coherence, that is The value is small, and the heat load shift is significant. When the value is large, the value inside the parentheses in the first term of the formula will approach the value of the first term. Meanwhile, the second item The value also approaches The combined effect of these three factors will lead to the final anomaly risk score. close to This allows for a high degree of confidence in indicating the presence of substantial thermal defects in the equipment.

[0038] S105: Intelligent inspection results realization and equipment early warning control.

[0039] The system will associate the final anomaly risk score calculated in real time with the equipment ledger and execute the following intelligent diagnosis and feedback logic.

[0040] Specifically, when the final abnormal risk score exceeds a preset warning threshold for multiple consecutive time points, the system automatically triggers a trend warning. These multiple time points must be at least three, and the warning threshold is 0.7. Simultaneously, the system will retrieve and analyze the load curve of the equipment within the corresponding thermal equilibrium time window. If extreme loads exceeding the rated range are present, the alarm will be suppressed; otherwise, the alarm will be confirmed. This eliminates instantaneous score fluctuations caused by extreme loads (such as short-term overloads) and ensures the accuracy of the warning. Once an anomaly is confirmed, the system will proceed based on the following steps... The three-phase topology coordinates identified in the system are highlighted on the human-machine interface to display the specific faulty phase, enabling accurate fault location and guiding maintenance personnel to respond quickly.

[0041] like Figure 2 As shown in the figure, this diagram illustrates the comparison between the final anomaly risk scores calculated by the two algorithms and the sampling time under conditions of drastic load fluctuations. The horizontal axis represents the sampling time, and the vertical axis represents the final anomaly risk score. The score curve of the traditional isolated forest algorithm exhibits significant oscillations during load surges and easily exceeds the warning threshold, generating false high-temperature alarms. In contrast, the improved isolated forest score model proposed in this invention exhibits smoother fluctuations, with the highest score significantly lower than that of the traditional method and consistently remaining within the safe range of the threshold. This indicates that the method of this invention effectively eliminates false alarms caused by thermal hysteresis.

[0042] The present invention also provides an intelligent inspection system for distribution substations in urban power grids. The system includes a processor and a memory, the memory storing computer program instructions. When the processor executes the computer program instructions, it implements the intelligent inspection method for distribution substations in urban power grids according to the first aspect of the present invention.

[0043] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.

[0044] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0045] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for intelligent inspection of distribution substations in urban power grids, characterized in that, include: Collect multi-source inspection feature data and perform spatiotemporal alignment to obtain the target area of ​​the device under test in the infrared thermal imaging data, and identify the three-phase equipment group to which the target area belongs; A thermal load evolution offset index is constructed for each sampling time to quantify the degree of matching between the cumulative Joule heating effect generated by the current within the time period and the actual temperature rise at that sampling time. A spatial topology compensation factor is constructed to quantify the temperature imbalance between the three-phase equipment groups, and the compensation factor is determined in combination with the analysis of the image texture abrupt change features within the target area. The final anomaly risk score at each sampling time is calculated based on the isolated forest algorithm. Anomaly warning is given based on the magnitude of the final anomaly risk score. The final anomaly risk score is negatively correlated with the product of the average path length and spatial topology compensation factor of the corresponding sampling point in the isolated forest, and positively correlated with the thermal load evolution offset index.

2. The intelligent inspection method for distribution substations in urban power grids according to claim 1, characterized in that, Construct the heat load evolution offset index at each sampling time, including: The heat load evolution offset index is positively correlated with the difference between the highest temperature of the target area and the ambient temperature at the sampling time, and negatively correlated with the square mean of the current, the humidity correction factor, and the standard thermal resistance of the equipment within the thermal equilibrium time window corresponding to the sampling time.

3. The intelligent inspection method for distribution substations in urban power grids according to claim 1, characterized in that, Calculate the spatial topology compensation factor, including: The spatial topology compensation factor is negatively correlated with the sum of the average temperature differences of the three-phase equipment group and positively correlated with the texture mutation index of the target region.

4. The intelligent inspection method for distribution substations in urban power grids according to claim 3, characterized in that, Methods for obtaining the texture mutation index of the target region include: The second derivative of the pixel grayscale matrix within the target region is calculated, i.e., the Laplacian operator, and is used as the texture mutation index of the target region, which reflects the texture mutation characteristics of the image.

5. The intelligent inspection method for distribution substations in urban power grids according to claim 1 or 3, characterized in that, The formula for calculating the spatial topology compensation factor is: ； In the formula, Indicates the spatial topology compensation factor; Indicates the target region and its group. The average temperature of the three corresponding regions; This indicates the standard for the maximum permissible interphase temperature difference during normal operation of this type of equipment; Indicates the texture abruptness index of the target region; This indicates the sensitivity adjustment bias.

6. The intelligent inspection method for distribution substations in urban power grids according to claim 1, characterized in that, Anomaly warnings are issued based on the final anomaly risk score, including: If the final abnormal risk score exceeds a set threshold for multiple consecutive moments, an alarm is triggered and the abnormal phase is highlighted on the human-computer interaction interface.

7. The intelligent inspection method for distribution substations in urban power grids according to claim 6, characterized in that, Triggering an alarm also includes: Retrieve historical load data of the power equipment within the corresponding thermal balance time window; if there is an extreme load exceeding the rated range, suppress the alarm; otherwise, confirm the alarm.

8. The intelligent inspection method for distribution substations in urban power grids according to claim 1, characterized in that, Obtain the target area of ​​the device under test in the infrared thermal imaging data, including: Using a pre-trained lightweight target detection model, high-definition images captured by a visible light camera are analyzed in real time to identify the device under test and output the range of its bounding box pixel coordinates in the visible light image. Based on the pre-completed binocular vision calibration, the coordinate system transformation matrix between the visible light camera and the infrared thermal imager is obtained. This matrix is ​​then used to map the pixel coordinate range of the bounding box onto the infrared thermal imaging matrix, thereby locking the target area that completely corresponds to the device under test in the thermal imaging data.

9. The intelligent inspection method for distribution substations in urban power grids according to claim 8, characterized in that, Identifying the three-phase equipment group to which the target area belongs includes: Based on the equipment ledger information and topology database in the background, the electrical phase of the current target area is automatically identified. In the same frame of image or adjacent observation points, the target areas corresponding to the other two phases belonging to the same equipment group as the current target area are automatically matched. These three areas together constitute a three-phase equipment group.

10. An intelligent inspection system for distribution substations in urban power grids, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the intelligent inspection method for distribution substations in urban power grids as described in any one of claims 1-9.

Images