Temperature detection and identification method and system for power equipment

Abstract

The embodiment of the invention relates to the technical field of electric power inspection, and discloses a temperature detection and identification method for electric power equipment, which comprises the following steps: acquiring temperature image information detected by a corresponding temperature measurement camera according to information of the temperature measurement camera configured on a server; acquiring equipment monitoring conditions of each monitoring area in the power equipment; temperature parameters of all the temperature monitoring areas in the temperature image information are determined, the determined temperature parameters of all the temperature monitoring areas are compared with equipment monitoring conditions of all the corresponding monitoring areas, and if the equipment monitoring conditions are not met, early warning operation is carried out. According to the method in the embodiment of the invention, automatic monitoring and analysis of the temperature of the power equipment are realized, manual intervention is reduced, and the working intensity of operation and maintenance personnel is reduced. Meanwhile, through timely and accurate early warning, operation and maintenance personnel can be helped to carry out equipment maintenance and overhaul in a targeted manner, unnecessary inspection and maintenance work is avoided, the operation and maintenance efficiency is improved, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of power detection technology, and specifically to a method and system for temperature detection and identification of power equipment. Background Technology

[0002] With the continuous rise in global energy demand and the accelerated development of smart grids, power systems are evolving towards higher voltage, larger capacity, and higher density. Consequently, the level of automation and equipment operation and maintenance requirements in the power industry are constantly increasing. Against this backdrop, ensuring the safe and stable operation of power equipment has become a core element in maintaining grid reliability, improving labor productivity, and enhancing economic benefits. High-temperature detection and abnormal alarm systems for power equipment operation are key technological means to achieve this goal.

[0003] During long-term operation, electrical connection points of power equipment (such as cable joints, busbar connections, and circuit breaker contacts) are prone to overheating due to factors such as material aging, poor contact, and load fluctuations. In special environments such as substations and ring main units, if such temperature anomalies are not monitored and addressed in a timely manner, they can not only lead to a decline in equipment insulation performance and damage to mechanical parts, but may also cause serious safety accidents such as equipment combustion and explosion, resulting in large-scale power outages and significant economic losses. Therefore, designing a solution capable of rapid early warning has become a pressing technical problem for those skilled in the art. Summary of the Invention

[0004] To address the aforementioned shortcomings, this invention discloses a temperature detection and identification method for power equipment, which enables more comprehensive monitoring of the equipment status of the power system.

[0005] The first aspect of this invention discloses a temperature detection and identification method for power equipment, comprising: The temperature image information detected by the corresponding temperature camera is obtained based on the information of the temperature camera configured on the server; wherein, the temperature image information includes multiple temperature monitoring areas; Obtain the equipment monitoring conditions for each monitoring area in the power equipment; The temperature parameters of each temperature monitoring area in the temperature image information are determined, and the determined temperature parameters of each temperature monitoring area are compared with the equipment monitoring conditions of the corresponding monitoring areas. If the equipment monitoring conditions are not met, an early warning operation is performed.

[0006] As an optional implementation, in the first aspect of the present invention, determining the temperature parameters of each temperature monitoring area in the temperature image information includes: A mask is generated to display the current temperature image information, and temperature measurement data of each pixel in the corresponding mask area is extracted using the mask. The maximum, minimum, and average temperature values ​​of the corresponding mask area are determined based on the temperature measurement data of each pixel in the mask area. The temperature parameters of each determined temperature monitoring area are compared with the equipment monitoring conditions of the corresponding monitoring areas. If the equipment monitoring conditions are not met, an early warning operation is performed, including: The maximum temperature value of the corresponding mask area is compared with the primary monitoring conditions of each monitoring area. If it exceeds the primary monitoring conditions of the corresponding monitoring area, proceed to the next step. The minimum and average temperature values ​​of the masked area are compared with the secondary monitoring conditions of each monitoring area. If they exceed the secondary monitoring conditions of the corresponding monitoring area, it is determined that the equipment monitoring conditions are not met, and an early warning operation is performed.

[0007] As an optional implementation, in the first aspect of the present invention, the detection and identification method further includes: Temperature data of corresponding time series are obtained from multiple associated monitoring areas; Calculate the first-order difference of the temperature data at consecutive time points to obtain the temperature change rate vector; The temperature change rate vector is calculated to generate a filtered change rate vector; A time window of a set length is selected to calculate the filtered rate of change vectors of multiple historical data within the window to obtain the corresponding mean vector and covariance matrix. Based on the mean vector and the covariance matrix, calculate the Mahalanobis distance of the filtered rate of change vector at the current time. If the Mahalanobis distance meets the set judgment conditions, an early warning signal will be generated.

[0008] As an optional implementation, in a first aspect of the present invention, calculating the temperature change rate vector to generate a filtered change rate vector includes: The temperature change rate vector is calculated using the exponentially weighted average filtering formula to generate a filtered rate of change vector; the exponentially weighted average filtering formula is as follows: ,in, This is the filtered rate of change vector. For filtering weights, ; Let the vector be the rate of change at the current moment. This is the filtered temperature change rate vector from the previous moment; The calculation of the filtered rate-of-change vectors from multiple historical data within the window to obtain the corresponding mean vector and covariance matrix includes: The mean vector and covariance matrix are calculated based on the mean formula and covariance formula for multiple historical filtered rate-of-change vectors within the window. The formula for the mean is: ; The covariance formula is: ,in, Let covariance matrix be the variance matrix. The length of the time window, It is the mean vector. This is the filtered temperature change rate vector from the previous moment; The step of calculating the Mahalanobis distance of the filtered rate of change vector at the current time based on the mean vector and the covariance matrix includes: Based on the mean vector and the covariance matrix, the Mahalanobis distance of the filtered rate of change vector at the current time is calculated according to the Mahalanobis distance formula, which is: in, It is the inverse of the covariance matrix. The Mahalanobis distance, This is the filtered rate of change vector.

[0009] As an optional implementation, in the first aspect of the present invention, the setting judgment condition is: ,in, This is an empirical threshold for abnormal temperature behavior.

[0010] As an optional implementation, in the first aspect of the present invention, the setting judgment condition is: ,in, Follows chi-square distribution According to the chi-square distribution corresponding to the degrees of freedom n Table lookup settings value.

[0011] As an optional implementation, in the first aspect of the present invention, the detection and identification method further includes: In response to the generation of early warning signals, the group of power equipment corresponding to multiple associated monitoring areas is highlighted on the graphical user interface; The detection and identification method further includes: Abstract each monitoring area in the power equipment topology as a node in a graph structure; Historical normal temperature data are used as features of nodes; a graph neural network model is used to train the graph structure to learn complex, nonlinear temperature correlation patterns between nodes.

[0012] A second aspect of this invention discloses a temperature detection and identification system for power equipment, comprising: The first acquisition module is used to acquire temperature image information detected by the corresponding temperature measuring camera based on the information of the temperature measuring camera configured on the server; wherein, the temperature image information includes multiple temperature monitoring areas; The second acquisition module is used to acquire the equipment monitoring conditions of each monitoring area in the power equipment. Comparison module: Used to determine the temperature parameters of each temperature monitoring area in the temperature image information, and compare the determined temperature parameters of each temperature monitoring area with the corresponding equipment monitoring conditions of each monitoring area. If the equipment monitoring conditions are not met, an early warning operation is performed.

[0013] A third aspect of the present invention discloses an electronic device, comprising: a memory storing executable program code; a processor coupled to the memory; the processor calling the executable program code stored in the memory to execute the temperature detection and identification method for power equipment disclosed in the first aspect of the present invention.

[0014] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the temperature detection and identification method for power equipment disclosed in the first aspect of the present invention.

[0015] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: The method described in this invention enables automated monitoring and analysis of power equipment temperature, reducing manual intervention and lowering the workload of maintenance personnel. Simultaneously, timely and accurate early warnings help maintenance personnel perform targeted equipment maintenance and repairs, avoiding unnecessary inspections and maintenance work, improving maintenance efficiency, and reducing maintenance costs. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of a temperature detection and identification method for power equipment disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of the process for performing correlation trend analysis disclosed in an embodiment of the present invention; Figure 3This is a system framework structure diagram disclosed in an embodiment of the present invention; Figure 4 This is a schematic diagram of frame-by-frame threshold analysis disclosed in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure for performing correlation trend analysis disclosed in an embodiment of the present invention; Figure 6 This is a schematic flowchart of the arc flashover identification method based on image recognition disclosed in an embodiment of the present invention; Figure 7 This is a schematic diagram of the AI ​​recognition process disclosed in an embodiment of the present invention; Figure 8 This is a schematic diagram of the specific process for flashover identification disclosed in an embodiment of the present invention; Figure 9 This is a diagram illustrating the image synthesis process disclosed in an embodiment of the present invention; Figure 10 This is a schematic diagram of AI model inference disclosed in an embodiment of the present invention; Figure 11 This is a schematic diagram of the model training process disclosed in an embodiment of the present invention; Figure 12 This is a schematic diagram of the structure of a temperature detection and identification system for power equipment provided in an embodiment of the present invention; Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that the terms first, second, third, fourth, etc., in the specification and claims of this invention are used to distinguish different objects, not to describe a specific order. The terms used in the embodiments of this invention include and have, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0020] Example 1 Please see Figure 1 , Figure 1This is a flowchart illustrating a temperature detection and identification method for power equipment disclosed in an embodiment of the present invention. The execution entity of the method described in this embodiment is an execution entity composed of software and / or hardware. This execution entity can receive relevant information via wired or / or wireless means and can send certain instructions. It may also have certain processing and storage functions. This execution entity can control multiple devices, such as remote physical servers or cloud servers and related software, or local hosts or servers and related software that perform related operations on equipment located in a certain location. In some scenarios, it can also control multiple storage devices, which may be placed in the same location as the equipment or in different locations. Figures 1 to 5 As shown, the temperature detection and identification method for power equipment includes the following steps: S101: Obtain temperature image information detected by the corresponding temperature measuring camera based on the information of the temperature measuring camera configured on the server; wherein, the temperature image information includes multiple temperature monitoring areas; S102: Obtain the equipment monitoring conditions for each monitoring area in the power equipment; S103: Determine the temperature parameters of each temperature monitoring area in the temperature image information, and compare the determined temperature parameters of each temperature monitoring area with the corresponding equipment monitoring conditions of each monitoring area. If the equipment monitoring conditions are not met, an early warning operation is performed.

[0021] This invention utilizes a temperature-measuring camera to acquire temperature image information, enabling real-time, non-contact monitoring of the temperature in various monitoring areas of electrical equipment. This method allows for timely acquisition of the equipment's temperature status without affecting its normal operation. Compared to traditional manual inspection methods, it offers higher efficiency and accuracy, precisely locating specific temperature monitoring areas and providing a reliable data foundation for subsequent analysis and processing.

[0022] By comparing the temperature parameters of each temperature monitoring area with the corresponding equipment monitoring conditions, accurate judgments can be made based on the specific requirements of different equipment and areas. Once it is found that the temperature parameters do not meet the equipment monitoring conditions, an immediate warning operation is initiated. This targeted warning mechanism effectively improves the accuracy and effectiveness of warnings, avoiding false alarms or missed alarms caused by uniform standards, and promptly alerting maintenance personnel to pay attention to equipment anomalies in specific areas. In particular, in this embodiment of the invention, images monitored by the same camera can be divided into monitoring areas, enabling precise zoning settings. For example, three areas can be fixed as corresponding monitoring areas within an image, which greatly improves detection efficiency and recognition accuracy.

[0023] The solution implemented in this invention allows for the timely detection and early warning of abnormal temperatures in power equipment, enabling maintenance personnel to take appropriate measures, such as adjusting equipment operating status and conducting equipment maintenance. This helps prevent equipment failures or damage caused by overheating, ensuring the safe and stable operation of power equipment. This is of great significance for ensuring the reliability and stability of the power system, reducing power outages caused by equipment failures, and improving power supply quality.

[0024] System framework such as Figure 3 As shown, the operation process is mainly divided into three stages: equipment deployment, configuration information (control), and real-time analysis.

[0025] First, temperature-measuring cameras are installed in the monitored equipment (such as switchgear) of the power system and their angles are adjusted. Unlike traditional temperature sensors, a single thermal infrared temperature-measuring camera can monitor a surface, covering multiple areas within its effective monitoring range. Then, the information for each temperature-measuring camera is configured on the application server. The monitoring area is set as a polygon in the real-time image of each camera, and an independent threshold is set for each area. Multiple detection areas can be drawn for the same camera. This configuration information is sent to the algorithm server via a pre-defined interface. Finally, upon receiving the configuration information for each camera, the algorithm server starts the corresponding thread to execute the necessary analysis algorithm, analyzing the real-time temperature data sent by the camera and sending alarm information to the application server when an anomaly is detected. Temperature anomaly monitoring employs two methods: threshold judgment based on independent frame-by-frame analysis, and correlation analysis of temperature change trends between multiple cameras, thus enabling more comprehensive monitoring of the power system's equipment status.

[0026] More preferably, determining the temperature parameters of each temperature monitoring area in the temperature image information includes: A mask is generated to display the current temperature image information, and temperature measurement data of each pixel in the corresponding mask area is extracted using the mask. The maximum, minimum, and average temperature values ​​of the corresponding mask area are determined based on the temperature measurement data of each pixel in the mask area. The temperature parameters of each determined temperature monitoring area are compared with the equipment monitoring conditions of the corresponding monitoring areas. If the equipment monitoring conditions are not met, an early warning operation is performed, including: The maximum temperature value of the corresponding mask area is compared with the primary monitoring conditions of each monitoring area. If it exceeds the primary monitoring conditions of the corresponding monitoring area, proceed to the next step. The minimum and average temperature values ​​of the masked area are compared with the secondary monitoring conditions of each monitoring area. If they exceed the secondary monitoring conditions of the corresponding monitoring area, it is determined that the equipment monitoring conditions are not met, and an early warning operation is performed.

[0027] By generating a mask and extracting the temperature data of the corresponding pixels, precise focusing on a specific monitoring area can be achieved. This eliminates interference from irrelevant data areas, extracting temperature information only for the core areas of the power equipment that need monitoring (such as joints and windings), reducing the impact of background or non-critical part temperatures on the judgment.

[0028] When implementing the technology, the maximum, minimum, and average temperature values ​​within the region are obtained simultaneously. Compared to a single temperature value, this provides a more comprehensive reflection of the overall temperature distribution and extreme temperature conditions in the region, offering richer data support for subsequent assessments.

[0029] By comparing temperature parameters with primary and secondary monitoring conditions step by step, a dual-verification judgment logic is formed, significantly improving the accuracy of early warnings. First, the maximum temperature value is compared with the primary condition to quickly screen out areas with extreme high temperature risks, serving as the initial threshold for early warning and preventing missed reports due to localized high temperatures being overlooked. Then, the minimum and average temperature values ​​are compared with the secondary condition to further verify whether the overall temperature of the area is abnormal, preventing false alarms caused by errors in individual pixels (such as occasional interference from temperature-measuring cameras), and ensuring that early warnings only target genuine equipment temperature anomalies.

[0030] The tiered comparison design can adapt to the differentiated safety requirements of different power equipment and monitoring areas. For some areas sensitive to extreme high temperatures (such as equipment insulation layers), the maximum temperature can be strictly controlled through primary monitoring conditions, enabling rapid response to catastrophic risks. For areas where overall temperature trends need to be monitored (such as the inside of equipment cabinets), the average temperature value of secondary monitoring conditions can be used to determine whether overall heat dissipation is normal, and the minimum temperature value can be used to assist in verifying the validity of the data, making the early warning logic more aligned with the actual safety logic of equipment operation.

[0031] More preferably, the detection and identification method further includes: S104: Obtain temperature data of corresponding time series from multiple associated monitoring areas; S105: Calculate the first-order difference of the temperature data at consecutive time points to obtain the temperature change rate vector; S106: Calculate the temperature change rate vector to generate a filtered change rate vector; S107: Select a time window of a set length to calculate the filtered rate of change vectors of multiple historical data within the window to obtain the corresponding mean vector and covariance matrix; S108: Based on the mean vector and the covariance matrix, calculate the Mahalanobis distance of the filtered rate of change vector at the current time. S109: If the Mahalanobis distance meets the set judgment conditions, an early warning signal is generated.

[0032] The solution in this invention upgrades temperature monitoring from static threshold judgment to dynamic trend early warning by introducing time series analysis and Mahalanobis distance. The technical effect is mainly reflected in three key dimensions.

[0033] First, instead of relying solely on whether the current temperature exceeds the limit, analyzing temperature change trends through time-series data allows for earlier detection of abnormal temperature rises in equipment. Second, calculating the temperature change rate vector directly quantifies the magnitude of temperature rise and fall per unit time, avoiding the overlooking of potential rapid temperature increases due to a single instance where the temperature does not reach the threshold (e.g., poor internal contact may indicate a normal current temperature that is rapidly increasing). Third, data analysis based on continuous time points covers the transition phase from normal operation to anomaly, providing earlier warnings compared to static threshold judgments, giving maintenance personnel more time to respond. Through filtering and statistical calculations, the impact of noisy data on temperature trend analysis is effectively reduced, making anomaly detection more reliable.

[0034] In this embodiment of the invention, the temperature change rate vector is filtered to remove abnormal data points caused by environmental fluctuations (such as wind and changes in light) or occasional equipment interference during temperature measurement, thus preserving the true temperature change trend. The mean vector and covariance matrix are calculated using a time window of a set length, and a statistical model of normal change trends is established using historical data. This provides a more scientific benchmark for judging whether the current change is abnormal, rather than relying on fixed empirical values.

[0035] Leveraging the statistical properties of Mahalanobis distance, true anomalies can be identified more accurately, avoiding misjudging temperature fluctuations during normal equipment operation as malfunctions. Mahalanobis distance can be calculated by combining the correlation of temperature changes in each monitoring area (represented by the covariance matrix). For example, normal load fluctuations in the equipment may cause slight synchronous temperature changes in multiple related areas, which Mahalanobis distance will determine as normal; while when the temperature of a single area changes independently and rapidly, it will be identified as an anomaly, adapting to the actual scenario of multi-area collaborative operation of equipment.

[0036] Instead of setting judgment conditions based on statistical models, this approach can automatically adapt to the normal fluctuation range of different equipment and operating conditions, significantly reducing false alarms caused by changes in operating conditions (such as the natural rise in equipment temperature during peak electricity consumption periods) and reducing ineffective operation and maintenance costs.

[0037] More preferably, the calculation of the temperature change rate vector to generate a filtered change rate vector includes: The temperature change rate vector is calculated using the exponentially weighted average filtering formula to generate a filtered rate of change vector; the exponentially weighted average filtering formula is as follows: ,in, This is the filtered rate of change vector. For filtering weights, ; Let the vector be the rate of change at the current moment. This is the filtered temperature change rate vector from the previous moment; The calculation of the filtered rate-of-change vectors from multiple historical data within the window to obtain the corresponding mean vector and covariance matrix includes: The mean vector and covariance matrix are calculated based on the mean formula and covariance formula for multiple historical filtered rate-of-change vectors within the window. The formula for the mean is: ; The covariance formula is: ,in, Let covariance matrix be the variance matrix. The length of the time window, It is the mean vector. This is the filtered temperature change rate vector from the previous moment; The step of calculating the Mahalanobis distance of the filtered rate of change vector at the current time based on the mean vector and the covariance matrix includes: Based on the mean vector and the covariance matrix, the Mahalanobis distance of the filtered rate of change vector at the current time is calculated according to the Mahalanobis distance formula, which is: in, It is the inverse of the covariance matrix. The Mahalanobis distance, This is the filtered rate of change vector.

[0038] Specifically, by assigning weights, the current raw data and historical filtered data are weighted and fused to achieve smooth processing of the temperature change rate vector.

[0039] Dynamically balances real-time performance and stability: α can be flexibly adjusted (e.g., when α is 0.3, it focuses more on historical trends), which can not only quickly respond to real temperature changes, but also filter out occasional interference such as camera misreading and environmental fluctuations (e.g., false temperature rise data caused by instantaneous changes in light). Relying solely on the filtering results from the previous moment, it eliminates the need to store large amounts of historical raw data, reducing server computational and storage pressure and adapting to the low-latency requirements of real-time power equipment monitoring. It addresses the pain points of traditional static solutions, which lack data preprocessing and are susceptible to interference, providing a clean data source for subsequent calculations of the mean vector and covariance matrix. This avoids statistical model bias caused by interfering data, improving early warning accuracy from the source.

[0040] It solves the pain points of traditional static solutions, which lack data preprocessing and are susceptible to interference. It provides a clean data source for subsequent calculations of mean vector and covariance matrix, avoids statistical model bias caused by interference data, and improves the accuracy of early warning from the source.

[0041] In this embodiment of the invention, based on historical data within a time window, the correlation (covariance matrix) between the average level of temperature change under normal operating conditions and the temperature change in multiple related regions is quantified, and a statistical model of normal trends is established.

[0042] It can be adjusted according to the type of equipment (e.g., ω=24 hours for transformers to cover daily load fluctuations). The model updates in real time as the window slides, adapting to differences in operating conditions such as peak / off-peak electricity consumption and seasonal changes, avoiding the problem that fixed thresholds in traditional solutions cannot adapt to operating conditions. The solution of this invention can capture regional correlation: the covariance matrix Σ, which can reflect the synergy of temperature changes in multiple related regions (such as transformer windings and cores) (such as when the normal load increases, the temperature of multiple regions changes slightly in sync, and the covariance is positive), providing a regional correlation benchmark for subsequent Mahalanobis distance calculation and avoiding false alarms caused by isolated judgment of a single region.

[0043] More preferably, the set judgment condition is: ,in, This is an empirical threshold for abnormal temperature behavior.

[0044] More preferably, the set judgment condition is: ,in, Follows chi-square distribution According to the chi-square distribution corresponding to the degrees of freedom n Table lookup settings value.

[0045] The judgment condition of the embodiment of the present invention utilizes the statistical characteristic that the squared Mahalanobis distance follows a chi-square distribution (χ²(k), where k is the number of monitored areas, i.e., degrees of freedom n), and determines the value of γ by consulting the chi-square distribution table. For example, when monitoring 4 related areas (n=4) and the confidence level is set to 99%, the table shows χ²(4)=13.28, that is, γ=13.28 is set. The probability of a value greater than 13.28 is only 1%, which is statistically significant enough to classify it as an anomaly. Based on practical experience, temperature anomalies caused by equipment failures in power systems are usually quite pronounced; therefore, a confidence level of 99.9% can be set accordingly. It has a high tolerance for small perturbations.

[0046] Compared to the subjectivity of experience-based thresholds relying on human judgment, the chi-square distribution lookup method determines the γ value through statistical probability, making the early warning standard more objective. The confidence level of the chi-square distribution directly corresponds to the probability that normal data is misjudged as abnormal, and the confidence level can be flexibly selected according to the reliability requirements of the power system. Simultaneously, by adapting the degrees of freedom n to the number of monitoring areas (e.g., when monitoring 5 areas, the n=5 column of the χ² distribution table is consulted), the γ value is ensured to match the monitoring dimensions, avoiding early warning bias caused by changes in the number of areas.

[0047] The setting of γ values ​​based on statistical distribution can be automatically achieved through algorithms (the system automatically queries the chi-square distribution table to generate γ values ​​based on the number of monitoring areas n and the preset confidence level), without manual intervention, thus realizing automated and standardized management of early warning standards. For example, when the monitoring system adds one temperature monitoring area (n changes from 3 to 4), the system can automatically update the degrees of freedom and look up the new γ value (such as χ²0 at a 95% confidence level). 05 (4) = 9.49), adapting to the dynamic adjustment of the monitoring range, reflecting the intelligent advantages of the system.

[0048] More preferably, the detection and identification method further includes: In response to the generation of early warning signals, the group of power equipment corresponding to multiple associated monitoring areas is highlighted on the graphical user interface; Once the system generates an early warning signal, it will highlight the power equipment groups corresponding to multiple associated monitoring areas on the graphical user interface (GUI). Essentially, through visual interaction optimization, it transforms abstract early warning information into intuitive equipment location guidance, helping maintenance personnel to quickly locate abnormal associated equipment clusters.

[0049] Traditional early warning systems often only provide textual alerts about abnormal temperatures in a specific area. Maintenance personnel must manually refer to the equipment topology diagram to locate the corresponding devices, which can lead to location delays, especially in densely populated environments like substations. This new approach, however, uses a user interface highlighting (e.g., using red highlights or flashing borders to mark abnormal device groups) to allow maintenance personnel to visually see the location and extent of the associated abnormal devices the moment the alert is triggered. This reduces location time from minutes to seconds, saving valuable time for fault handling and minimizing the risk of further deterioration of equipment malfunctions.

[0050] Multiple interconnected monitoring areas often correspond to equipment groups that operate collaboratively (such as transformers, circuit breakers, and disconnectors on a power supply line), and anomalies can propagate between devices within the group. The interface highlighting feature not only locates individual devices but also presents clusters of interconnected devices, preventing maintenance personnel from neglecting to check related equipment and resulting in incomplete localized troubleshooting. For example, when a transformer temperature monitoring area triggers an alarm, the interface simultaneously highlights the associated cooling system equipment group, helping maintenance personnel comprehensively investigate the root cause of the fault and reducing secondary failures caused by incomplete investigations.

[0051] In collaborative scenarios involving multiple maintenance personnel, a clear and intuitive interface can serve as a unified benchmark for anomaly localization, avoiding misunderstandings caused by inaccurate textual descriptions. For example, dispatch center personnel and on-site inspection personnel can communicate based on highlighted equipment group information within the same GUI interface, quickly reaching a consensus on handling issues, reducing information transmission costs, and improving collaborative efficiency.

[0052] The detection and identification method further includes: Abstract each monitoring area in the power equipment topology as a node in a graph structure; Historical normal temperature data are used as features of nodes; a graph neural network model is used to train the graph structure to learn complex, nonlinear temperature correlation patterns between nodes.

[0053] The solution of this invention abstracts each monitoring area in the power equipment topology as a graph structure node, uses historical normal temperature data as node features, and trains a graph neural network (GNN) to learn complex nonlinear temperature correlation patterns between nodes. The core is to use the modeling advantages of GNN for structured data to mine the implicit temperature correlation patterns between devices that are difficult to capture by traditional methods, and provide deeper pattern support for anomaly identification.

[0054] Traditional temperature correlation analysis often relies on manually preset linear correlation rules (such as a positive correlation between transformer temperature and load current), which cannot handle complex nonlinear correlations between devices (such as the dynamic changes in the temperature correlation coefficient between transformers and cooling fans under different seasons and loads). In contrast, GNNs can be trained through multi-layer neural networks to automatically learn hidden nonlinear correlation patterns in historical normal data (such as in summer under high load, for every 1°C increase in transformer temperature, the cooling fan temperature should rise by 0.3°C to be considered normal; deviations indicate anomalies), covering correlation dimensions missed by traditional methods and improving the comprehensiveness of anomaly identification.

[0055] Power equipment topology naturally possesses graph structure characteristics (nodes represent monitoring areas, and edges represent physical / functional connections between devices). GNNs can fully utilize this structural information to learn the coordinated temperature change patterns of a cluster of devices under normal operating conditions. When a device exhibits a minor anomaly but does not reach the warning threshold, if the temperature changes of its associated devices deviate from the normal correlation pattern learned by the GNN, the system can identify the precursor to cluster anomalies in advance, preventing a single device anomaly from spreading into a cascading failure. For example, if the temperature of a circuit breaker on a certain line rises slightly, even if it does not exceed the threshold, the GNN can detect that the temperature changes of its associated bus deviate from the normal coordination pattern, providing an early warning and preventing subsequent circuit breaker failures from causing bus outages.

[0056] As power equipment ages and its operating conditions change (such as adding loads or replacing components), the temperature correlation patterns between devices dynamically evolve. By continuously inputting new historical normal temperature data to iteratively train the GNN model, the model can continuously update the learned correlation patterns and automatically adapt to changes in equipment status. Compared to the traditional method of fixing correlation rules and requiring periodic manual adjustments, this method significantly reduces manual maintenance costs while ensuring that the correlation pattern judgment always matches the actual operating status of the equipment, avoiding false alarms and missed alarms due to outdated correlation rules.

[0057] The temperature correlation patterns learned by GNN can serve as a feature library for normal equipment operation. Subsequent systems can further realize intelligent fault diagnosis (such as comparing the correlation patterns during abnormal events with those during normal events to locate the fault type) and predictive maintenance (predicting potential faults based on the trend of correlation pattern changes) based on this feature library, promoting the upgrade of power equipment operation and maintenance from passive response to proactive prevention, and laying the foundation for building an intelligent power operation and maintenance system.

[0058] In this invention, real-time data is input into a trained graph neural network model. By calculating the reconstruction error or Mahalanobis distance of the hidden layer features output by the model, deviations from the learned normal association patterns are detected. When an anomaly is detected, the interpretability of the graph neural network is used to locate which edge (relationship between devices) or which node (device) in the graph contributes the most to the anomaly, thereby assisting in the root cause analysis of the fault.

[0059] Specifically, select a GNN model suitable for structured data and association learning (such as GAT (Graph Attention Network) or GCN (Graph Convolutional Network)) and train it with the goal of reconstructing normal temperature features. The specific steps are as follows: The preferred GAT model automatically learns edge weights through an attention mechanism (without requiring manual preset of fixed weights), which can more accurately capture dynamic relationships between devices (such as the relationship weight between coolers and transformers being higher in summer than in winter), adapting to changes in the operating conditions of power equipment. Dataset partitioning: The processed graph structure data (including nodes, edges, and features) is divided into a training set (70% historical normal data) and a validation set (30% historical normal data) in a 7:3 ratio. Label definition: Since the training objective is to learn the normal pattern, all training samples are labeled as normal (label=0). Training objective: To encode and decode the input node features X using a GNN model, minimize the error between the reconstructed features output by the model and the original normal features (using the MSE loss function), and enable the model to learn the association rules of node features under normal conditions. Input layer: Receives node feature matrix X (dimension N×F, where F is the number of features) and graph adjacency matrix A (dimension N×N, indicating the relationship between edges between nodes); Graph convolutional layer: Through the attention heads of GAT (e.g., 8 attention heads), calculate the attention weight of each node to its neighboring nodes, aggregate neighbor features, and generate the association feature representation of the nodes (capturing the temperature association between devices). Decoding layer: Decodes the aggregated features into reconstructed features X′ with the same dimensions as the original features; Iterative optimization: Using the Adam optimizer, iteratively train until the reconstruction error of the validation set converges (if the MSE loss on the validation set does not decrease after 10 consecutive rounds, stop training) to obtain the normal association pattern model M.

[0060] The method described in this invention enables automated monitoring and analysis of power equipment temperature, reducing manual intervention and lowering the workload of maintenance personnel. Simultaneously, timely and accurate early warnings help maintenance personnel perform targeted equipment maintenance and repairs, avoiding unnecessary inspections and maintenance work, improving maintenance efficiency, and reducing maintenance costs.

[0061] Example 2 like Figures 6 to 11 As shown, the image recognition-based arc flashover identification method includes the following steps: S101: The video data acquisition module obtains the corresponding real-time video stream from the camera component through a real-time streaming protocol; S102: The real-time video stream is decoded using the hardware encoding and decoding module of the edge computing box, and the decoded video frames are output in NV12 format, and the output data is placed in the buffer queue; the NV12 format includes luminance channel data, which is used to characterize the luminance information of all pixels in the image; S103: Extract the luminance channel data from the NV12 format data of the cache array, and calculate the signed difference image of adjacent frames based on the extracted luminance channel data. The signed difference image is a luminance difference image with a luminance change direction. S104: Analyze the parameters of each difference pixel in the signed difference image to determine the total area of ​​changed pixels, the area of ​​positive change, and the ratio between the two in the signed difference image; S105: Compare the total area of ​​changed pixels, the area of ​​positive change, and the ratio between the two in the signed difference image with the set conditions. If the set conditions are met, it is determined that an arc flashover is suspected.

[0062] The solution of this invention acquires real-time video streams from a camera component via a real-time streaming protocol and performs fast decoding using an edge computing box. This enables real-time monitoring of arc flashovers, timely detection of potential arc flashover phenomena, and provides timely assurance for subsequent processing and early warning. It helps to take appropriate measures in the early stages of arc flashovers, reducing potential damage.

[0063] Specifically, the video stream is decoded using the hardware encoding / decoding module of the edge computing box, and the decoded video frames are output in NV12 format and placed into a buffer queue. This hardware encoding / decoding method can fully leverage the performance advantages of edge computing devices, quickly process video data, improve the efficiency of the entire recognition process, and meet the needs of scenarios with high real-time requirements. Compared with pure software decoding, it significantly shortens the decoding time, allowing more time for subsequent image analysis. The luminance channel data of the NV12 format data in the buffer array is extracted, and the signed difference image of adjacent frames is calculated. The presence of an electric arc flashover is determined by analyzing the differential pixel parameters. This method based on luminance channel data change analysis can effectively capture the image brightness change characteristics caused by electric arc flashover. By quantifying parameters such as the total area of ​​changed pixels, the area of ​​positive change, and the ratio between the two, and comparing them with set conditions, the accuracy and reliability of electric arc flashover identification are improved, reducing the possibility of false alarms and missed alarms.

[0064] Specifically, the video data acquisition module obtains real-time video streams from dual-light cameras via the RTSP protocol and efficiently decodes the video stream using the hardware encoding / decoding module of the edge computing box. The decoded video frames are output in NV12 format and placed in a buffer queue. Unlike common three-channel RGB / BGR format images, NV12 is a planar storage format in the YUV color space, with the following characteristics: Y channel: stores luminance information, directly reflecting the brightness and darkness details of the image, and serves as input for subsequent inter-frame difference analysis and detection models. UV channels: store chromaticity information, which requires no additional processing in this application, thus reducing computational overhead.

[0065] In practical implementation, the module maintains a fixed-length circular queue as a frame buffer, the functions of which include: Temporal feature extraction: providing three consecutive frames for the YOLOv8 flashover detection model Synthetic input data is used to capture the dynamic characteristics of arc flashover.

[0066] Alarm Short Video Generation: Upon detecting an anomaly, the module automatically extracts relevant frames from the queue and generates a short video clip as an alarm attachment for subsequent review. By directly reusing the NV12's Y-channel data, the module avoids unnecessary format conversions, significantly improving real-time performance and meeting the low-latency requirements of edge devices.

[0067] More preferably, the calculation of the signed difference image of adjacent frames includes: The signed difference image between adjacent frames is calculated according to the luminance difference calculation formula, which is: ; in, This refers to the brightness parameter of the corresponding pixel in the current frame. This refers to the brightness parameter of the corresponding pixel in the previous frame. This represents the brightness difference between corresponding pixels in a signed difference image; The step of analyzing the parameters of each difference pixel in the signed difference image to determine the total area of ​​changed pixels, the area of ​​positive change, and the ratio between the two in the signed difference image includes: The parameters of each difference pixel in the signed difference image are analyzed according to the threshold analysis formula to determine the total area of ​​changed pixels and the area of ​​positive change in the signed difference image. The threshold analysis formula is as follows: , ;in, The total area of ​​the changed pixels. For positively changing area, To set the comparison threshold; The ratio between the two is calculated using the ratio calculation formula; the ratio calculation formula is: ,in, This represents the ratio between the two.

[0068] The solution of this invention directly quantifies the brightness changes of corresponding pixels in adjacent frames through a clear brightness difference calculation formula, while retaining the direction of change (positive / negative). This allows the sudden strong light increment features (usually manifested as positive brightness changes) during arc flashover to be accurately captured, avoiding feature loss caused by the mutual cancellation of positive and negative changes in unsigned difference, and providing more reliable raw data for subsequent analysis.

[0069] The threshold analysis formula divides the difference pixels into total changed pixels and positively changed pixels by setting a comparison threshold (and then uses the following formulas to define the threshold values ​​respectively). and Quantifying its area. This quantification method transforms the abstract brightness changes of an image into calculable numerical parameters, reducing the interference of subjective human judgment and making the judgment criteria for arc flashover more objective and consistent.

[0070] The proportion calculation formula in this invention focuses on the percentage of positively changing pixels in the total changing pixels. Since the core characteristic of an electric arc flashover is a sudden burst of intense light (a surge in positive brightness), its positive change percentage is typically significantly higher than other interferences (such as shadow movement, object occlusion, etc., which may be accompanied by negative changes). This proportion parameter effectively distinguishes between the brightness change patterns of electric arc flashovers and non-flashovers, further reducing the false alarm rate and improving the specificity of the identification.

[0071] All formulas are based on simple pixel-level arithmetic operations (difference, absolute value comparison, summation, and division), resulting in low computational complexity and efficient execution on resource-constrained hardware such as edge computing boxes. Combined with the real-time decoding and caching mechanisms of edge computing, millisecond-level processing of video streams can be achieved, meeting the high real-time requirements of arc flashover monitoring (avoiding missed detections due to computational delays).

[0072] The comparison threshold can be flexibly adjusted according to the actual scene (such as light intensity and camera parameters). For example, it can be increased in strong light environments to filter background noise and decreased in low light environments to avoid missed detections. This adjustability allows the method to adapt to the lighting conditions of different industrial scenarios (such as substations and power distribution rooms), enhancing the flexibility of actual deployment.

[0073] More preferably, the setting conditions include: in, For the variable area threshold, The threshold value for the direction of change.

[0074] The solution in this embodiment of the invention introduces a variable area threshold (T) simultaneously. a ) and the threshold for the proportion of change direction (T) This forms a dual judgment criterion: only when the area of ​​the positively changing pixel is large enough (exceeding T) a And the proportion of positive changes is high enough (exceeding T). Only when the condition is met is it considered a suspected arc flashover. This dual constraint can effectively filter out two types of interference: one is a sudden change in brightness in a local small area (such as electric sparks, flashing lights, etc., which are not arc flashovers but small-scale strong light), and the other is a scene with a large area but a low proportion of positive changes (such as light and shadow mixing changes caused by object movement), which significantly improves the accuracy of recognition.

[0075] The conditions are defined using mathematical expressions to clarify the judgment logic, avoiding vague qualitative descriptions. In actual deployment, T can be calibrated experimentally based on specific scenarios (such as device type, monitoring distance, and ambient lighting). a and T Specific values ​​(e.g., for arc flashover in high-voltage equipment, a larger T value can be set) a And T close to 1 This makes the judgment criteria operable and consistent, facilitating engineering implementation.

[0076] The typical characteristics of electric arc flashover are large-area, strong positive brightness abrupt changes, and the set conditions specifically amplify these two characteristics: T a Ensure the range of variation reaches the scale of an electric arc, T The design ensures that the direction of change is primarily positive (a surge in brightness). This design closely matches the physical characteristics of electric arc flashover, further enhancing the method's specificity in identifying the target event and reducing confusion with other brightness change events.

[0077] The judgment of the set conditions depends only on the calculated values. The comparison between ρ and the threshold is a simple logical operation that adds almost no additional computation. Combined with the previously efficient differential calculation and parameter extraction steps, the entire recognition process can be completed quickly on an edge computing device, ensuring millisecond-level response to real-time video streams and meeting the timeliness requirements of arc flashover monitoring (avoiding the escalation of accidents due to judgment delays).

[0078] T a and T It can be dynamically adjusted according to the actual environment: for example, in a workshop scene with complex lighting, the T value can be appropriately increased. a To filter stray light interference; in low-light environments at night, it can reduce T a Simultaneously improve T This ensures sensitive detection of even weak electric arcs. This flexibility allows the method to adapt to the monitoring needs of different industrial scenarios, expanding its application scope.

[0079] More preferably, after determining that an arc flashover is suspected, the method further includes: The system automatically extracts related frames from the cache queue, generates short video clips, and sends them as alarm attachments to the corresponding terminal devices.

[0080] Specifically, once a suspected arc flashover is detected, the system automatically extracts related frames before and after the arc flashover to generate a short video clip as an alarm attachment. This provides managers or monitoring personnel with intuitive on-site video evidence. Compared to a simple alarm signal, the short video clearly shows the changes in the scene before and after the arc flashover, helping to quickly confirm whether it is a real arc flashover, reducing the difficulty and time cost of manual verification, and improving decision-making efficiency.

[0081] Short video clips are simultaneously sent to the terminal device as alarm attachments, ensuring that the alarm information not only includes a notification of a suspected incident but also key video footage of the incident. This is of great value for tracing the cause of the incident and analyzing the occurrence patterns of arc flashovers (such as frequency, location, and environmental factors), providing data support for equipment maintenance and safety improvements.

[0082] The process of automatically extracting relevant frames from the cache queue and generating short videos requires no manual intervention. It can be completed quickly by relying on the local processing capabilities of the edge computing box, avoiding delays caused by manual operation. Combined with a real-time alarm mechanism, key images can be pushed to the terminal immediately after a suspected incident, ensuring that relevant personnel are aware of the situation on site as soon as possible, buying time for emergency response.

[0083] Instead of storing the entire video stream, relevant frames are extracted and short videos are generated only after a suspected event is identified, significantly reducing unnecessary storage usage. The edge computing box's cache queue only temporarily stores recent frames, extracting them as needed to generate concise alarm attachments. This ensures the retention of key information while reducing the consumption of local or cloud storage resources.

[0084] More preferably, after determining that an arc flashover is suspected, the method further includes: S106: Extract three consecutive frames of image data from the buffer queue, and perform image synthesis on the three consecutive frames of image data in the order of blue channel, green channel and red channel respectively to obtain the corresponding synthesized image information; S107: The synthesized image information is input into a pre-constructed flashover recognition model for recognition to determine and output a rectangular target box, target category and confidence level, wherein the flashover recognition model is a YOLOv8 detection model.

[0085] The solution in this invention extracts three consecutive frames of image data and synthesizes image information in the order of blue, green, and red channels, transforming the dynamic changes in the time dimension into color feature fusion in the spatial dimension. This method can preserve the dynamic brightness change trajectory during an electric arc flashover (such as the flashing and diffusion process of the arc), and more richly depicts the dynamic characteristics of the arc compared to a single-frame image, providing more effective input information for subsequent model recognition and reducing the influence of static interference (such as a fixed light source).

[0086] The YOLOv8 detection model is used to identify the synthetic image, directly outputting a rectangular target box (locating the arc position), target category (confirming whether it is an arc flashover), and confidence score (quantifying the reliability of the identification). Compared with the preliminary judgment based on pixel statistics, this step achieves an upgrade from regional change detection to precise target-level identification, which can more accurately distinguish between arc flashovers and similar interference, and significantly reduces the false alarm rate.

[0087] Building upon the initial step of identifying suspected electric arcs through brightness difference analysis, a secondary verification based on a deep learning model is added, forming a dual mechanism of preliminary screening and precise identification. This progressive judgment logic utilizes the efficiency of traditional algorithms for initial filtering while leveraging the strong feature learning capabilities of YOLOv8 to achieve fine classification, significantly improving the reliability of the overall identification system. It is particularly suitable for scenarios with extremely high safety requirements, such as those involving power.

[0088] The YOLOv8 model itself has the characteristic of fast inference. Combined with the hardware acceleration capabilities of edge computing boxes (such as GPU or NPU support), it can efficiently complete the recognition and processing of synthetic images locally, avoiding the latency caused by relying on cloud computing. At the same time, model inference is triggered only after a suspected arc is detected, rather than processing all frames, which further optimizes the consumption of computing resources and ensures that the system as a whole meets the real-time monitoring requirements.

[0089] The rectangular target bounding box, category, and confidence level output by the scheme model in this embodiment of the invention are structured information that can be directly used for subsequent linkage operations (such as locating faulty equipment, triggering partial power outages, etc.). Compared with simple alarm signals, structured results can provide maintenance personnel with more specific references on the location and severity of faults, improve the pertinence and efficiency of emergency response, and enhance the engineering practicality of the system.

[0090] Conventional video analytics uses models such as LSTM and 3D-CNN to process consecutive frames. These models are computationally intensive and structurally complex, making them unsuitable for edge devices. The standard YOLOv8 is a single-frame image object detection model, which cannot detect motion or change.

[0091] The solution in this invention cleverly folds information from the temporal dimension (three consecutive frames) into the spatial dimension, transforming a complex temporal analysis problem into a standard single-frame image object detection problem. Therefore, the lightweight and efficient YOLOv8 model can be used directly without modifying its structure.

[0092] Suppose we have three consecutive Y-channel images, all of which are grayscale images, with each pixel having a value ranging from 0 to 255. For example, I... t-1 (Previous frame): Recorded the state before the flashover occurred. t (Current frame): Records the brightest peak state of the flashover. It+1 (Next frame): Records the state after flashover decay.

[0093] The synthesis process is as follows: Assign color channels: Blue channel = I t-1 Green channel = I t And the red channel = I t+1 Merge into a new image: Create an empty three-channel BGR image I BGR .

[0094] Will I t-1 Copy the data to image I BGR The B (blue) channel will... t Data copied to I BGR The G (green) channel will... t+1 Data copied to I BGR The R (red) channel.

[0095] An explanation based on a typical lifecycle of an electric arc flashover: At time t-1: the electric arc has not yet occurred or is very weak, and the image is dark. In the composite image: the information for this frame is in the blue channel. If a region is dark at this time, then the blue component of that region in the composite image will be very weak.

[0096] At time t: The electric arc erupts violently, producing an extremely bright light. In the composite image: this frame's information is in the green channel. The core region of the electric arc is extremely bright in this frame, meaning that the green component of this region is extremely strong in the composite image.

[0097] At time t+1: The arc rapidly decays, and the brightness decreases. In the composite image: the information for this frame is in the red channel. The decrease in brightness in the arc region means that the red component in this region is moderate in the composite image.

[0098] In the background area: the brightness remained unchanged at all three time points. Therefore, the values ​​of the B, G, and R channels were essentially equal.

[0099] In the core region of the electric arc: the previous frame is very weak, the current frame is extremely strong, and the next frame is moderate. When green is extremely strong and blue and red are relatively weak, the resulting mixture is a bright green, which may even appear yellowish / white due to excessive green intensity. This is because the above method optimizes a complex temporal analysis problem that requires recursion or 3D model processing into a simple and mature single-frame image classification / detection problem.

[0100] For the model: Traditional time series models: require understanding motion and change, the model is complex and computationally intensive.

[0101] The YOLOv8 model in this embodiment of the invention does not require any changes. Its task becomes finding a specific color in the image, that is, finding the bright green / yellowish-white patch (weak-strong-weak temporal pattern) within a gray (unchanging background). This task is much more intuitive and simpler.

[0102] For the system: Significantly reduced computational requirements: The YOLOv8 model, optimized for single-frame images, can be used directly, eliminating the need for 3D-CNN or RNN / LSTM models with several times higher computational demands. It can leverage the pre-trained models, optimization techniques, and deployment tools accumulated across the industry in 2D object detection.

[0103] The core of this invention lies in proposing a static image synthesis method based on temporal feature color coding to solve the problem of insufficient computing power for temporal event detection on edge devices.

[0104] Specifically, this method encodes the continuous temporal dynamic changes in brightness (weak-strong-weak) into the color distribution of a static composite image through a specific color channel mapping (BGR). This transforms the dynamic features that originally existed in the time dimension into a stable and dominant color feature in the spatial dimension.

[0105] In this way, the present invention successfully transforms the detection of a time-series event, electric arc flashover, into the detection of a static target with specific green characteristics. This not only avoids dependence on complex time-series models and greatly reduces the computational load, but also enables the present solution to directly utilize mature and efficient single-frame target detection networks (such as YOLOv8), thereby achieving high-precision real-time analysis with limited edge computing power.

[0106] In this embodiment of the invention, the YOLOv8 model uses different sample data during training than conventional single-frame samples. Based on the principle that training and inference data are distributed similarly, training samples also need to be synthesized from three adjacent frames. The process is as follows: Figure 6 As shown.

[0107] More preferably, after determining the target category as arc flashover, the method further includes: The target bounding box output by the flashover recognition model is drawn onto the current image frame and used as an alarm indication image. The frames cached in the circular buffer queue are used to construct an alarm short video; Determine the current camera ID and timestamp information, and send the alarm indication image, alarm short video, current camera ID and timestamp information to the corresponding smart terminal.

[0108] The solution of this invention draws the target box output by the flashover identification model onto the current image frame to form an alarm indication image, which can intuitively mark the specific location of the arc flashover. This allows managers to quickly locate the fault area through terminal devices, avoiding manual searching in complex images, significantly shortening the time from alarm to fault location confirmation, and saving valuable time for emergency response.

[0109] By constructing alarm short videos using a circular buffer queue's frame queue, the dynamic process before and after an arc flashover can be fully recorded. This includes not only the critical moments of the flashover but also the trajectory of the event (such as the process from occurrence to extinction). This provides complete visual evidence for subsequent analysis of flashover causes (such as equipment aging, operational errors, etc.) and assessment of the fault's impact range, facilitating the optimization of equipment maintenance strategies and safety control measures.

[0110] By integrating alarm indication images, short alarm videos, camera IDs, and timestamp information, a complete alarm data packet is formed, including location markers, dynamic processes, occurrence location, and occurrence time. This multi-dimensional information combination can comprehensively reconstruct the event scene, avoiding misjudgments caused by incomplete information (such as confusion of coverage areas from different cameras, unclear timelines, etc.), and improving the reliability of alarm information and its decision-making reference value.

[0111] Camera ID information can be directly linked to a specific monitoring location (such as a switch cabinet in a substation or a circuit in a power distribution room). Combined with timestamps, it enables event tracing and collaborative processing in multi-camera monitoring scenarios. For example, when multiple cameras alarm simultaneously, the ID and timestamp can be used to quickly determine whether it is a spread of the same event or an independent event in different areas, improving the scheduling efficiency of large-scale monitoring systems.

[0112] The structured alarm data (images, videos, IDs, timestamps) in this embodiment of the invention can be directly connected to the backend intelligent management system, supporting automatic triggering of linkage operations (such as remotely disconnecting faulty circuits, dispatching nearby maintenance personnel, etc.). This standardized output method enables the arc flashover identification system to be seamlessly integrated into the industrial safety monitoring system, realizing a closed loop from alarm identification to automatic handling, and improving the overall level of intelligent safety control.

[0113] More preferably, the camera assembly is a dual-light camera, which includes infrared temperature measurement data, and the infrared temperature measurement data is used to characterize the temperature parameters at the corresponding location.

[0114] The infrared temperature measurement data provided by the dual-light camera can directly obtain the temperature parameters of the monitored area. The core characteristics of electric arc flashover are not only strong light, but also instantaneous high temperature. This design can verify the visual recognition results through temperature data.

[0115] If visual recognition identifies a suspected electric arc, but infrared thermography does not detect abnormally high temperatures, the probability of false alarms can be reduced. If visual recognition and high-temperature data are triggered simultaneously, the reliability of electric arc flashover detection can be significantly improved, avoiding missed or false alarms caused by a single visual feature.

[0116] Infrared thermography data can independently monitor equipment temperature changes. Before an arc flashover occurs, some equipment may exhibit abnormal temperature increases (such as poor line contact or insulation aging). The system can provide early warnings of temperature anomalies through infrared data, extending the warning process from post-event alarms to pre-event alerts. This gives maintenance personnel more time to troubleshoot and reduces the likelihood of arc flashovers at their source. Combining infrared thermography data with visual images, timestamps, camera IDs, and other information can form a more complete event data chain.

[0117] In subsequent analysis, not only can the arc generation process be reviewed through video, but the temperature change curve of the flashover point can also be viewed to help determine the cause of the arc flashover (such as whether it is caused by local overheating), providing more comprehensive data support for equipment fault tracing and maintenance strategy optimization.

[0118] In the solution of this invention embodiment, since the NPU computing power of the AI ​​computing chip is limited, it usually cannot meet the high-concurrency analysis requirements under multiple high-resolution inputs. Therefore, a framework of real-time computing and event-triggered AI analysis is adopted. The real-time analysis part uses OpenCV to perform change analysis on the {xt-1,xt} data of the current frame and the previous frame. Only when there is a significant difference between the previous and subsequent frames is it passed to the AI ​​model for secondary judgment. This can not only meet the requirements of high concurrency and low latency, but also improve the overall anti-interference capability.

[0119] The image recognition-based arc flashover identification method in this invention does not rely on complex deep learning models, has relatively low hardware requirements, and exhibits strong adaptability and deployability. It can be applied in resource-constrained environments, such as remote monitoring points in power systems, while also reducing system construction and operating costs.

[0120] Example 3 Please see Figure 12 , Figure 12 This is a schematic diagram of the structure of a temperature detection and identification system for power equipment disclosed in an embodiment of the present invention. Figure 12 As shown, the temperature detection and identification system for power equipment may include: First acquisition module 21: used to acquire temperature image information detected by the corresponding temperature measuring camera based on the information of the temperature measuring camera configured on the server; wherein, the temperature image information includes multiple temperature monitoring areas; Second acquisition module 22: used to acquire the equipment monitoring conditions of each monitoring area in the power equipment; Comparison module 23: Used to determine the temperature parameters of each temperature monitoring area in the temperature image information, and compare the determined temperature parameters of each temperature monitoring area with the corresponding equipment monitoring conditions of each monitoring area. If the equipment monitoring conditions are not met, an early warning operation is performed.

[0121] The method described in this invention enables automated monitoring and analysis of power equipment temperature, reducing manual intervention and lowering the workload of maintenance personnel. Simultaneously, timely and accurate early warnings help maintenance personnel perform targeted equipment maintenance and repairs, avoiding unnecessary inspections and maintenance work, improving maintenance efficiency, and reducing maintenance costs.

[0122] The above provides a detailed description of the temperature detection and identification method, system, electronic device, and storage medium for power equipment disclosed in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A temperature detection identification method for power equipment, characterized in that, The method comprises the following steps: obtaining temperature image information detected by a temperature measurement camera according to information of the temperature measurement camera configured on a server, wherein the temperature image information comprises a plurality of temperature monitoring areas; obtaining device monitoring conditions of each monitoring area in a power device; determining temperature parameters of each temperature monitoring area in the temperature image information, and comparing the determined temperature parameters of each temperature monitoring area with device monitoring conditions of each corresponding monitoring area, and if the device monitoring conditions are not met, a warning operation is performed.

2. The temperature detection identification method for the power device according to claim 1, wherein, The method comprises the following steps: generating a mask for the current temperature image information, and extracting temperature measurement data of each pixel point in a corresponding mask area through the mask; determining a maximum temperature value, a minimum temperature value and an average temperature value of the corresponding mask area according to the temperature measurement data of each pixel point in the mask area; The method comprises the following steps: comparing the maximum temperature value of the corresponding mask area with first-level monitoring conditions of each monitoring area, and if the first-level monitoring conditions are exceeded, the next step is performed; comparing the minimum temperature value and the average temperature value of the mask area with second-level monitoring conditions of each monitoring area, and if the second-level monitoring conditions are exceeded, it is determined that the device monitoring conditions are not met, and a warning operation is performed.

3. The temperature detection identification method for the power device according to claim 1, wherein, The method further comprises the following steps: obtaining temperature data of a corresponding time sequence from a plurality of associated monitoring areas; calculating a first-order difference of the temperature data at consecutive time points to obtain a temperature change rate vector; calculating the temperature change rate vector to generate a filtered change rate vector; selecting a time window of a set length to calculate a plurality of historical filtered change rate vectors within the window to obtain a corresponding mean vector and a covariance matrix; based on the mean vector and the covariance matrix, calculating the Mahalanobis distance of the filtered change rate vector at the current time; if the Mahalanobis distance meets a set judgment condition, a warning signal is generated.

4. The temperature detection identification method for the power device according to claim 3, wherein, The method comprises the following steps: The temperature rate of change vector is calculated according to an exponential weighted average filtering formula to generate a filtered rate of change vector; the exponential weighted average filtering formula is: wherein, is the filtered rate of change vector, is a filtering weight, ; is the rate of change vector at a current time, is a filtered temperature rate of change vector at a previous time. The method comprises the following steps: based on the mean vector and the covariance matrix, calculating the Mahalanobis distance of the filtered change rate vector at the current time according to the Mahalanobis distance formula, wherein the Mahalanobis distance formula is: The mean formula is: ; The covariance formula is: wherein, is a covariance matrix, is a length of a time window, is a mean vector, is a filtered temperature rate of change vector of a previous time. The method further comprises the following steps: in response to generating a warning signal, highlighting a power device group corresponding to the plurality of associated monitoring areas on a graphical user interface; wherein is the inverse of the covariance matrix, is the Mahalanobis distance, is the filtered rate of change vector.

5. The temperature detection identification method for the power device according to claim 4, wherein, The set judgment condition is: wherein, is an empirical threshold value for temperature behavior anomalies.

6. The temperature detection identification method for the power device according to claim 4, wherein, The setting judgment condition is: wherein, subject to chi-square distribution , according to the chi-square distribution corresponding to the degree of freedom n table setting value.

7. The temperature detection and identification method for power equipment according to claim 4, wherein, The method further comprises the following steps: ​ ​ Each monitoring area in the power equipment topology is abstracted as a node in a graph structure; The historical normal temperature data is taken as the feature of the node; the graph structure is trained using a graph neural network model to learn the complex and nonlinear temperature correlation mode between the nodes.

8. A temperature detection identification system for electrical equipment, characterized by, The method comprises the steps of: The first acquisition module is configured to acquire temperature image information detected by the temperature measurement camera according to the information of the temperature measurement camera configured on the server; wherein the temperature image information comprises a plurality of temperature monitoring areas; The second acquisition module is configured to acquire equipment monitoring conditions of each monitoring area in the power equipment; The comparison module is configured to determine temperature parameters of each temperature monitoring area in the temperature image information, and compare the determined temperature parameters of each temperature monitoring area with the equipment monitoring conditions of the corresponding monitoring area; if the equipment monitoring conditions are not met, a warning operation is performed.

9. An electronic device, comprising: The method comprises the steps of: The memory stores executable program codes; The processor is coupled to the memory; The processor calls the executable program codes stored in the memory to execute the temperature detection and identification method for the power equipment according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, wherein the computer program enables the computer to execute the temperature detection and identification method for the power equipment according to any one of claims 1 to 7.

Images