Electric power monitoring system based on AI visual identification technology

Abstract

The invention relates to the technical field of target detection, in particular to an AI visual identification technology-based power monitoring system, which comprises a multispectral monitoring module, a thermal anomaly identification module, a dynamic region correction module, a structure trend prediction module and an abnormal mode judgment module. According to the method, the visible light image and the infrared thermal imaging data are subjected to joint analysis, the appearance contour, the component arrangement and the thermal feature information of the power equipment are extracted from multiple angles, equipment identification and temperature anomaly detection are realized, a sliding time window is constructed to extract the morphological change gradient, and the structural degradation trend of the equipment is accurately predicted. Through calculating the joint posterior probability distribution of the abnormal parameters such as the cable current-carrying capacity, the switch contact resistance value and the line power factor, and comparing the probability gradient value of the abnormal mode, the abnormal mode matching is realized, the equipment fault early warning capability is improved, and the accuracy of target detection, the precision of abnormal identification and the reliability of equipment state evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the field of target detection technology, and in particular to an electric power monitoring system based on AI visual recognition technology. Background Art

[0002] The field of target detection technology includes computer vision methods and model training schemes for automatically locating and identifying specific objects from image or video data. The core content of this technical field involves feature extraction, classification, positioning and segmentation of targets. Common technologies include target recognition methods based on convolutional neural networks, region proposal networks for target candidate region generation, and non-maximum suppression algorithms for screening the best target boxes. Target detection technology is widely used in intelligent monitoring, autonomous driving, medical image analysis and industrial detection. In the field of power monitoring, this technology is used to identify transmission line equipment, detect power grid faults, analyze thermal imaging data and other scenarios.

[0003] Among them, the power monitoring system refers to a system that analyzes the operating status of the power grid based on target detection technology, mainly covering technical matters such as power grid equipment target detection, power line fault identification, and thermal imaging anomaly analysis. The system usually uses a neural network model to detect images of power equipment, uses the attention mechanism to optimize feature extraction, improves the recognition ability of key components, and combines motion analysis methods based on optical flow to track dynamic targets. In addition, the system combines image registration technology to compare multiple time series images, and uses histogram equalization and super-resolution reconstruction to enhance image quality, thereby ensuring the accuracy of target detection. In terms of power line detection, the system uses a transformation-invariant feature detection algorithm for stable target identification in different environments, and combines a topological analysis method based on a graph convolutional neural network to optimize the transmission line defect detection capability.

[0004] Traditional monitoring systems mainly rely on single spectral data for target detection and thermal anomaly analysis. In complex environments, they are easily affected by lighting conditions, equipment materials, and background interference, resulting in reduced target recognition accuracy and increased high-temperature anomaly misjudgment rate. Traditional systems use a fixed threshold method to judge temperature anomalies, without considering the thermal conductivity characteristics of equipment materials and the dynamic changes in operating conditions, resulting in insufficient accuracy in determining high-temperature anomaly areas, which is prone to false alarms or missed alarms. In terms of monitoring equipment morphological changes, it mainly relies on static image comparison, and does not fully utilize time series data to extract morphological change trends, making it difficult to accurately predict the risk of structural degradation of equipment. In addition, in terms of abnormal pattern recognition, traditional systems mainly use rule matching to judge fault categories, lacking joint analysis of multi-dimensional parameters, resulting in limited recognition capabilities for complex fault patterns and affecting the accuracy of equipment abnormality warnings. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose an electric power monitoring system based on AI visual recognition technology.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: a power monitoring system based on AI visual recognition technology, the system comprising: The multi-spectral monitoring module uses the AI ​​visual recognition model to identify the type of power equipment based on the visible light image and infrared thermal imaging image data of the power monitoring area, extract the temperature distribution characteristics of each power equipment component, and generate equipment thermal characteristic data; The thermal anomaly identification module screens local temperature mutation points based on the thermal characteristic data of the equipment, calculates the thermal gradient change rate, sets the monitoring priority of each area according to the temperature mutation, and obtains high temperature abnormal area data; The dynamic area correction module calls the high temperature abnormal area data, calculates the ratio of the operating time to the temperature change, compares the thermal distribution curves of similar equipment, corrects the temperature abnormality judgment deviation, and generates the corrected abnormal area data; The structural trend prediction module calls the corrected abnormal area data, extracts the morphological parameters of the power equipment, calculates the morphological change gradient within a continuous time period, evaluates the morphological change trend, and generates equipment morphological change trend data; The abnormal mode identification module calls the device morphology change trend data, calculates the joint posterior probability distribution of each abnormal parameter, screens the matching abnormal mode category, and obtains abnormal mode monitoring identification information.

[0007] As a further solution of the present invention, the thermal characteristic data of the equipment include the surface temperature distribution of the equipment, the thermal gradient change rate and the coordinates of the thermal anomaly area, the high-temperature abnormal area data include the local temperature mutation point, the thermal gradient change rate and the abnormal area weight, the corrected abnormal area data are specifically thermal distribution time series data, temperature change ratio and detection weight adjustment value, the equipment morphology change trend data include edge profile parameter change, crack length growth rate and contact flatness offset, and the abnormal pattern classification data specifically refers to the joint posterior probability distribution, abnormal pattern probability gradient value and abnormal category matching result.

[0008] As a further solution of the present invention, the multi-spectral monitoring module includes: The equipment contour extraction submodule uses the AI ​​visual recognition model to pre-process the image data based on the visible light image and infrared thermal imaging image data of the power monitoring area, obtain edge features, calculate the curvature change rate based on the edge gradient, and select edge pixels that meet the boundary continuity conditions to generate equipment appearance contour data; The device type identification submodule calls the AI ​​visual recognition model based on the device appearance profile data, compares the power equipment classification database, and performs matching calculations on the profile features using the formula: ; Calculate the device feature matching value, select the device type with the highest matching degree, and obtain the device classification result; in, Represents the device feature matching degree, The first character vector representing the observed profile of the device Quantity, The first character representing the contour feature vector of the corresponding type in the equipment classification database Quantity, represents the feature vector dimension, Representative The weight of the component, Represents the preset maximum feature matching degree; The device thermal signature analysis submodule extracts the temperature distribution information in the infrared thermal imaging image data based on the device classification results, calculates the average temperature value of each device component area, analyzes the temperature gradient change, screens the area that exceeds the device standard temperature threshold, and obtains the device thermal signature data.

[0009] As a further solution of the present invention, the thermal anomaly identification module includes: The temperature mutation screening submodule extracts the local temperature data of the transformer winding, the circuit breaker contact and the busbar connection point based on the thermal characteristic data of the equipment, calculates the temperature difference between adjacent pixels in each area, screens the pixels exceeding the temperature change threshold, marks the temperature mutation points, and obtains the temperature mutation point data; The thermal gradient calculation submodule uses the formula based on the local temperature mutation point data: ; Calculate the thermal gradient change rate of each area to obtain thermal gradient change data; in, represents the rate of change of thermal gradient, Representing the region The temperature value of each measuring point, Represents the temperature value of adjacent measuring points, Represents the total number of measurement points; The monitoring priority setting submodule sets the monitoring priority of each area based on the thermal gradient change data and the set temperature mutation priority standard, sorts them according to the monitoring priority, screens high-priority areas, and obtains high-temperature abnormal area data.

[0010] As a further solution of the present invention, the dynamic area correction module includes: The temperature change calculation submodule calls the thermal response time parameters of the transformer oil tank, switch cabinet copper busbar and cable connector based on the abnormal high temperature area data, calculates the temperature change rate during the operating time, obtains the temperature change ratio of each area, and obtains the temperature change ratio data; The heat distribution comparison submodule calls the heat distribution curve of similar equipment based on the temperature change ratio data, extracts the temperature change trend of the corresponding time series, compares the temperature change ratio of each area, and uses the formula: ; Calculate the corrected temperature value and generate temperature deviation data; in, Represents the corrected temperature value, Represents the original measured temperature value, Representative The measured temperature value at each time series point, Represents the reference temperature value of similar equipment. Represents the number of time series points, Represents the thermal conductivity correction coefficient of the material; The abnormal region correction submodule corrects the temperature abnormality determination deviation caused by the difference in thermal conductivity characteristics of the materials based on the temperature deviation data, adjusts the identification boundary of the high temperature abnormal region, and obtains the corrected abnormal region data.

[0011] As a further solution of the present invention, the structural trend prediction module includes: The morphological parameter extraction submodule extracts edge profile parameters, crack length parameters, and contact flatness parameters of the transformer core, support insulator, and line joint based on the corrected abnormal area data to generate morphological parameter data; The morphological change calculation submodule sets a sliding time window based on the morphological parameter data and calculates the morphological change gradient within a continuous time period using the formula: ; Calculate the trend value of morphological changes; in, Represents the trend value of morphological change, and Respectively represent and Morphological parameters within a time window, is the width of the sliding window, is the attenuation factor, is the base of natural logarithms; The trend assessment submodule assesses the device morphology change trend based on the morphology change trend value, determines the device morphology stability, and obtains the device morphology change trend data.

[0012] As a further solution of the present invention, the abnormal mode identification module includes: The abnormal parameter acquisition submodule obtains the cable current carrying capacity, switch contact resistance, line power factor and thermal abnormality area data based on the equipment morphology change trend data, extracts the time series distribution of each abnormal parameter, and generates abnormal parameter data; The joint posterior calculation submodule adopts the formula based on the abnormal parameter data: ; Calculate the joint posterior probability distribution of each abnormal parameter, obtain the probability gradient value of the abnormal pattern, and generate abnormal pattern probability data; in, Represents a given anomaly parameter matrix At that time, The posterior probability of an abnormal pattern, For the The likelihood probability of the parameter matrix under the abnormal mode, For the The prior probability of abnormal patterns, is the total number of abnormal patterns, For the parameter values, For the The mean of the parameters, For the The standard deviation of the parameters, is the number of abnormal parameters; The abnormal pattern matching submodule screens the abnormal pattern category with the highest matching probability based on the abnormal pattern probability data, performs pattern classification and labeling, and obtains abnormal pattern monitoring identification information.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by jointly analyzing visible light images and infrared thermal imaging data, the appearance contour, component arrangement and thermal characteristic information of power equipment are extracted from multiple angles to realize equipment identification and temperature anomaly detection. By calculating the thermal gradient change rate, the local temperature mutation point is accurately located, and the monitoring priority is set according to the temperature change trend to improve the screening accuracy of high-temperature abnormal areas. By correcting the temperature anomaly judgment deviation, the misjudgment caused by differences in material properties is reduced. The edge contour, crack length and contact flatness of the equipment are analyzed, and a sliding time window is constructed to extract the morphological change gradient to accurately predict the structural degradation trend of the equipment. By calculating the joint a posteriori probability distribution of abnormal parameters such as cable current carrying capacity, switch contact resistance and line power factor, and comparing the probability gradient value of the abnormal mode, abnormal mode matching is realized, the equipment fault warning capability is improved, and the accuracy of target detection, the accuracy of abnormal identification and the reliability of equipment status assessment are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the multi-spectral monitoring module of the present invention; Figure 3 This is a flow chart of the thermal anomaly identification module of the present invention; Figure 4 is a flow chart of a dynamic area correction module of the present invention; Figure 5 It is a flow chart of the structural trend prediction module of the present invention; Figure 6 It is a flow chart of the abnormal mode identification module of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0016] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0017] Embodiment 1 See also Figure 1 , an electric power monitoring system based on AI visual recognition technology, the system includes: The multi-spectral monitoring module uses the AI ​​visual recognition model to extract the appearance, component arrangement and identification information of the equipment based on the visible light image and infrared thermal imaging image data of the power monitoring area, compares it with the power equipment classification database, identifies the type of power equipment, including cables, transformers, switch cabinets and circuit breakers, extracts the temperature distribution characteristics of each power equipment component, and generates equipment thermal feature data; The thermal anomaly identification module selects local temperature mutation points of transformer windings, circuit breaker contacts and busbar connection points based on the thermal characteristic data of the equipment, calculates the thermal gradient change rate using the thermal imaging gradient, sets the monitoring priority of each area according to the temperature mutation, and obtains the high temperature abnormal area data; The dynamic area correction module calls the high temperature abnormal area data, calculates the operating time and temperature change ratio based on the thermal response time parameters of the transformer oil tank, switch cabinet copper busbar and cable connector, compares the thermal distribution curves of similar equipment, calls the thermal distribution time series data, corrects the temperature abnormality judgment deviation caused by the difference in material thermal conductivity characteristics, and generates the corrected abnormal area data; The structural trend prediction module calls the corrected abnormal area data, extracts the edge profile parameters, crack length parameters, and contact flatness parameters of the transformer core, supporting insulator, and line joint, sets the sliding time window, calculates the morphological change gradient within a continuous time period, evaluates the morphological change trend, and generates equipment morphological change trend data; The abnormal mode identification module calls the equipment morphology change trend data to obtain the cable current carrying capacity, switch contact resistance, line power factor and thermal abnormality area data, calculates the joint posterior probability distribution of each abnormal parameter, compares the probability gradient value of the abnormal mode, screens the matching abnormal mode category, and obtains the abnormal mode monitoring identification information.

[0018] The thermal characteristic data of the equipment include the surface temperature distribution of the equipment, the thermal gradient change rate and the coordinates of the thermal anomaly area. The high-temperature abnormal area data include the local temperature mutation point, the thermal gradient change rate and the abnormal area weight. The corrected abnormal area data specifically include the thermal distribution time series data, the temperature change ratio and the detection weight adjustment value. The equipment morphology change trend data include the edge profile parameter change, the crack length growth rate and the contact flatness offset. The abnormal pattern classification data specifically refers to the joint posterior probability distribution, the abnormal pattern probability gradient value and the abnormal category matching result.

[0019] See also Figure 2 , the multi-spectral monitoring module includes: The equipment contour extraction submodule uses the AI ​​visual recognition model to pre-process the image data based on the visible light image and infrared thermal imaging image data of the power monitoring area, obtain edge features, calculate the curvature change rate based on the edge gradient, and select edge pixels that meet the boundary continuity conditions to generate equipment appearance contour data; The equipment contour extraction submodule is based on the visible light image and infrared thermal imaging image data of the power monitoring area. The process of obtaining edge features involves using the AI ​​visual recognition model to normalize the image data first to standardize the image data format and reduce the impact of lighting. The image edge is extracted by the Sobel operator, and then the Canny algorithm is used to enhance edge detection to obtain a clear edge feature map. For each point in the image, its gradient amplitude and direction on the edge are calculated. Based on these gradient information, the non-maximum suppression technology is used to eliminate the stray responses caused by edge detection. Finally, the double threshold method is used to separate the real edge and the false edge, determine the final edge continuity, and determine the edge pixel points that match the equipment shape to form a device appearance contour dataset. The device appearance contour dataset is used for subsequent equipment type identification and analysis. The process effectively improves the extraction accuracy and reliability of the equipment contour.

[0020] The equipment type identification submodule uses the AI ​​visual recognition model based on the equipment appearance profile data, compares it with the power equipment classification database, and performs matching calculations on the profile features using the formula: ; Calculate the device feature matching value, select the device type with the highest matching degree, and obtain the device classification result; in, Represents the device feature matching degree, The first vector representing the observed profile feature of the device Quantity, The first character representing the contour feature vector of the corresponding type in the equipment classification database Quantity, represents the feature vector dimension, Representative The weight of the components, Represents the preset maximum feature matching degree; The device type identification submodule performs matching calculations on the profile features based on the device appearance profile data, such as setting the maximum feature matching degree. 100 points is used as the theoretical benchmark value for perfect matching. The device type is set to have 5 characteristic dimensions, namely , extract the standard profile feature vector of each type of equipment from the power equipment classification database For example, a typical transformer has eigenvectors of 20, 30, 15, 25, 10; for the observed device profile, its eigenvector 18, 32, 12, 22, 15; each feature dimension The weight given It reflects the importance of the feature in the device, such as the weight vector 0.4, 0.3, 0.1, 0.1, 0.1; By calculation formula:

[0021] Calculated device feature matching value The value is 99.5, and then all the calculation results are compared to select the device type with the highest matching degree as the recognition result. This process ensures high-precision recognition of device types through precise weight allocation and feature differentiation calculation.

[0022] The equipment thermal signature analysis submodule extracts the temperature distribution information from the infrared thermal imaging image data based on the equipment classification results, calculates the average temperature value of each equipment component area, analyzes the temperature gradient change, and screens the area that exceeds the equipment standard temperature threshold to obtain the equipment thermal signature data; The equipment thermal signature analysis submodule extracts temperature distribution information from infrared thermal imaging image data based on equipment classification results. The process starts with obtaining the temperature value of each pixel from the infrared image and calculating the average temperature value of each equipment component area. The calculation involves accumulating the temperature values ​​by component partition and dividing it by the number of pixels in the area. Then, the temperature gradient changes of each component are analyzed and compared with the temperature records during equipment operation to determine whether there are abnormal temperature fluctuations. The standard deviation of the temperature values ​​is calculated to evaluate the uniformity of the temperature distribution. Areas that exceed the standard temperature threshold will be marked as potential failure or overheating areas. This threshold is the upper limit of the safe temperature set according to the equipment manufacturing and operation standards. It can detect possible failure points of the equipment and prevent safety accidents caused by equipment overheating.

[0023] See also Figure 3 ,The thermal anomaly recognition module includes: The temperature mutation screening submodule extracts the local temperature data of transformer windings, circuit breaker contacts and busbar connection points based on the equipment thermal feature data, calculates the temperature difference between adjacent pixels in each area, screens the pixels that exceed the temperature change threshold, marks the temperature mutation points, and obtains the temperature mutation point data; The function of the temperature mutation screening submodule is to extract the local temperature data of the specified equipment area and identify abnormal temperature changes. By monitoring the temperature data of the transformer windings, circuit breaker contacts and busbar connection points, the temperature values ​​of the pixels in each area are extracted. For example, the temperature data set of the transformer winding in normal operation may be , calculate the temperature difference between adjacent pixels, such as , set the temperature change threshold to 4°C. This threshold is set based on the temperature fluctuation range during normal operation of the equipment and historical fault data. Any temperature difference exceeding this threshold, such as , will be identified and marked as temperature mutation points. The temperature mutation point data will be used for further analysis and fault warning. This process provides key information for equipment operation and maintenance through detailed monitoring and data processing, and effectively identifies possible faults or abnormalities.

[0024] The thermal gradient calculation submodule is based on the local temperature mutation point data and uses the formula: ; Calculate the thermal gradient change rate of each area to obtain thermal gradient change data; in, represents the rate of change of thermal gradient, Representing the region The temperature value of each measuring point, Represents the temperature value of adjacent measuring points, Represents the total number of measurement points; The core task of the thermal gradient calculation submodule is to calculate the thermal gradient change rate of each area in order to more accurately monitor and analyze the thermal state of the equipment. The calculation formula used is: , suppose there are 4 measuring points in a transformer winding area, and its temperature value is The temperature differences of adjacent measuring points are , the total temperature difference is 20°C, the total number of measuring points is 4, so the thermal gradient change rate in this area Calculated as , this value represents the average temperature change rate of each measuring point. This data helps to determine whether there is an overheating risk or potential failure in the area. The thermal gradient data calculated by this formula provides decision support for the maintenance team and helps determine the areas that need to be monitored or repaired.

[0025] The monitoring priority setting submodule sets the monitoring priority of each area based on the thermal gradient change data and the set temperature mutation priority standard, sorts them by monitoring priority, screens high-priority areas, and obtains high-temperature abnormal area data; The responsibility of the monitoring priority setting submodule is to set the monitoring priority of each area according to the thermal gradient change data, and to evaluate the temperature mutation priority of each area according to the preset standards. The standards are defined based on the thermal gradient change rate and the importance of the area. For example, if the thermal gradient change rate of an area exceeds 5°C, it is regarded as a high-priority area and the area will be monitored first. By screening the data of high-temperature abnormal areas, the operation and maintenance team can focus resources and attention on where they are most needed, thereby more effectively preventing equipment failures and extending equipment service life, ensuring the stable operation and safety of the power system.

[0026] See also Figure 4 , the dynamic area correction module includes: The temperature change calculation submodule uses the thermal response time parameters of the transformer oil tank, switch cabinet copper busbar and cable connector based on the data of the abnormal high temperature area, calculates the temperature change rate during the operating time, obtains the temperature change ratio of each area, and obtains the temperature change ratio data; The task of the temperature change calculation submodule is to analyze the temperature changes of specific power equipment components such as transformer tanks, switch cabinet copper bars and cable connectors during operation. This module uses the thermal response time parameters of the equipment to calculate the temperature change rate of each area. For example, the thermal response time of the transformer tank is 2 hours, the starting temperature during operation is 30°C, and the ending temperature is 60°C. The calculated temperature change rate is This rate shows the heating rate of the oil tank. Then, based on this rate and similar calculations for other components, the temperature change ratio of each component is obtained. For example, the ratio of the transformer oil tank to the switchgear copper busbar may be 1.5, indicating that the temperature rise rate of the oil tank is 1.5 times that of the copper busbar. The data helps technicians evaluate the thermal stability of the equipment and the potential overheating risks, and effectively guide maintenance and fault diagnosis.

[0027] The heat distribution comparison submodule calls the heat distribution curve of similar equipment based on the temperature change ratio data, extracts the temperature change trend of the corresponding time series, and compares the temperature change ratio of each area using the formula: ; Calculate the corrected temperature value and generate temperature deviation data; in, Represents the corrected temperature value, Represents the original measured temperature value, Representative The measured temperature value at each time series point, Represents the reference temperature value of similar equipment. Represents the number of time series points, Represents the thermal conductivity correction coefficient of the material; The heat distribution comparison submodule analyzes and diagnoses the heat distribution of the equipment based on the temperature change ratio data, using the formula Perform temperature correction, use the thermal distribution curve of similar equipment as a reference, and set the number of time series points is 5, the original measured temperature value of the device The reference temperature is 50°C. They are , the measured temperature value for , calculate the absolute value of the temperature difference and sum it to get 8°C, the material thermal conductivity correction coefficient Set it to 0.1, and calculate the corrected temperature value according to the formula This value provides the temperature deviation compared with the standard of similar equipment, helping technicians evaluate whether the thermal distribution of the equipment is normal. Through this method, it can more accurately determine whether the equipment has problems such as local overheating or insufficient cooling, providing a scientific basis for equipment maintenance and fault prevention.

[0028] The abnormal area correction submodule corrects the temperature anomaly judgment deviation caused by the difference in thermal conductivity characteristics of the material based on the temperature deviation data, adjusts the identification boundary of the high temperature abnormal area, and obtains the corrected abnormal area data.

[0029] The function of the abnormal area correction submodule is to adjust the identification boundary of the high-temperature abnormal area according to the temperature deviation data to eliminate the judgment error caused by the difference in the thermal conductivity characteristics of the material. By analyzing the material properties and thermal response data of each component of the equipment, for example, if the temperature deviation of a certain material area frequently exceeds the standard threshold, the module will investigate whether it is caused by the difference in thermal conductivity coefficient, set the thermal conductivity correction coefficient of the material, and then recalculate the temperature deviation of the area, and adjust the abnormal judgment boundary according to the corrected data. For example, if the original high-temperature abnormal threshold is 55°C, and the actual temperature after correction is 53°C, the area will be removed from the high-temperature abnormal list. This precise correction helps avoid false alarms and missed alarms, ensuring the safe operation of the equipment and the accuracy of maintenance work.

[0030] See also Figure 5 , the structural trend prediction module includes: The morphological parameter extraction submodule extracts the edge profile parameters, crack length parameters, and contact flatness parameters of the transformer core, support insulator, and line joint based on the corrected abnormal area data, and generates morphological parameter data; The operation of the morphological parameter extraction submodule involves extracting edge contour parameters, crack length parameters and contact flatness parameters from key components such as transformer core, supporting insulator, line joints, etc. based on the corrected abnormal area data. The data is crucial for judging the health of the equipment. For example, when inspecting the transformer core, edge recognition technology can be used to determine whether its contour is deformed or displaced. The measurement of crack length uses digital image processing technology to locate and quantify the specific size of the crack, while the contact flatness parameter is evaluated through 3D scanning of the contact surface. All morphological parameters are stored uniformly to form morphological parameter data, which provides a quantitative basis for subsequent equipment maintenance and fault prediction.

[0031] The morphological change calculation submodule sets a sliding time window based on the morphological parameter data and calculates the morphological change gradient within a continuous time period using the formula: ; Calculate the trend value of morphological changes; in, Represents the trend value of morphological change, and Respectively represent and Morphological parameters within a time window, is the width of the sliding window, is the attenuation factor, is the base of natural logarithms; The function of the morphology change calculation submodule is to calculate the change gradient of the device morphology parameters within the set sliding time window. The formula is Used to quantitatively analyze the trend of morphological changes, for example, setting a sliding window Set to 3, the attenuation factor is 5. Morphological parameters The three consecutive values ​​of , the values ​​represent the morphological parameters of three consecutive time points.

[0032] Substitute the change into the formula for calculation: ; Calculated trend value of the pattern change The value is 0.1216, which indicates the average change trend of the morphological parameters within the given three time windows, providing quantitative information about the rate of change of the equipment morphology. This information is crucial for monitoring the health status of the equipment and predicting potential problems, helping the maintenance team make timely maintenance or adjustment decisions.

[0033] The trend assessment submodule assesses the device morphology change trend based on the morphology change trend value, determines the device morphology stability, and obtains the device morphology change trend data; The responsibility of the trend assessment submodule is to evaluate the stability of the equipment morphology based on the morphology change trend value. This module compares the morphology change trend value obtained by continuous calculation with the historical data and the set stability threshold. If the morphology change trend value is stable in a small range, such as between 0.5 and 0.7, it means that the equipment morphology is relatively stable. If the morphology change trend value fluctuates greatly, such as above 1.0, it indicates that some parts of the equipment have abnormal deformation or damage. This dynamic trend analysis helps technicians to discover potential problems in time and take corresponding maintenance measures to ensure the stable operation of the equipment and extend its service life.

[0034] See also Figure 6 , the abnormal pattern identification module includes: The abnormal parameter acquisition submodule obtains the cable current carrying capacity, switch contact resistance, line power factor and thermal abnormality area data based on the equipment morphology change trend data, extracts the time series distribution of each abnormal parameter, and generates abnormal parameter data; The abnormal parameter acquisition submodule operation includes extracting key performance indicators from the equipment morphology change trend data, such as cable current carrying capacity, switch contact resistance, line power factor and thermal abnormality area data. Parameters are crucial for diagnosing the health status of equipment. For example, by analyzing the current carrying capacity of the cable, the cable usage status and possible overload conditions can be monitored. The increase in switch contact resistance may indicate contact wear or corrosion, and the abnormal change in power factor may reflect the decline in power system efficiency or the increase in nonlinear loads. By monitoring these parameters in real time and analyzing their time series data, anomalies and trends in operation can be identified, thereby generating detailed abnormal parameter data. The data helps the operation and maintenance team to diagnose faults and provide a basis for formulating maintenance plans.

[0035] The joint posterior calculation submodule is based on the abnormal parameter data and uses the formula: ; Calculate the joint posterior probability distribution of each abnormal parameter, obtain the probability gradient value of the abnormal pattern, and generate abnormal pattern probability data; in, Represents a given anomaly parameter matrix At that time, The posterior probability of an abnormal pattern, For the The likelihood probability of the parameter matrix under the abnormal mode, For the The prior probability of abnormal patterns, is the total number of abnormal patterns, For the parameter values, For the The mean of the parameters, For the The standard deviation of the parameters, is the number of abnormal parameters; The core task of the joint posterior calculation submodule is to determine the likelihood of various abnormal patterns through complex probability calculations. This process uses the formula For example, there are 4 abnormal modes, namely , abnormal number of parameters (such as cable current carrying capacity, contact resistance and power factor), the measured parameter value at a certain point in time for 75, the mean of each parameter for , standard deviation for , prior probability and the likelihood Set to example value: , ; , ; , .

[0036] Calculate the denominator: ; Calculate each part of the formula: ; Compute the sum of squares term: ; Combine to get the posterior probability: ; Therefore, the The posterior probability of the abnormal pattern is calculated to be 2.1865, which reflects the This method can be used to quantitatively analyze and compare different fault causes, thereby assisting the decision-making process and improving the accuracy of fault diagnosis and prevention. The abnormal pattern matching submodule selects the abnormal pattern category with the highest matching probability based on the abnormal pattern probability data, classifies and labels the pattern, and obtains abnormal pattern monitoring and identification information; The abnormal pattern matching submodule performs pattern classification and labeling based on the probability data of various abnormal patterns. The process involves analyzing and comparing the probability values ​​of different abnormal patterns and selecting the pattern with the highest matching probability as the most likely cause of the fault. For example, if the joint a posteriori calculation results show that the probability of the cable current carrying capacity abnormal pattern is the highest, this abnormal pattern is marked as the current main focus. This helps the operation and maintenance team to conduct targeted troubleshooting and maintenance to ensure the stable operation of the power system. Through such pattern recognition and probability calculation, the accuracy and efficiency of fault diagnosis are effectively improved.

[0037] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A power monitoring system based on AI visual recognition technology, characterized in that: The system comprises: The multi-spectral monitoring module uses the AI ​​visual recognition model to identify the type of power equipment based on the visible light image and infrared thermal imaging image data of the power monitoring area, extract the temperature distribution characteristics of each power equipment component, and generate equipment thermal characteristic data; The thermal anomaly identification module screens local temperature mutation points based on the thermal characteristic data of the equipment, calculates the thermal gradient change rate, sets the monitoring priority of each area according to the temperature mutation, and obtains high temperature abnormal area data; The dynamic area correction module calls the high temperature abnormal area data, calculates the ratio of the operating time to the temperature change, compares the thermal distribution curves of similar equipment, corrects the temperature abnormality judgment deviation, and generates the corrected abnormal area data; The structural trend prediction module calls the corrected abnormal area data, extracts the morphological parameters of the power equipment, calculates the morphological change gradient within a continuous time period, evaluates the morphological change trend, and generates equipment morphological change trend data; The abnormal mode identification module calls the device morphology change trend data, calculates the joint posterior probability distribution of each abnormal parameter, screens the matching abnormal mode category, and obtains abnormal mode monitoring identification information.

2. The power monitoring system based on AI visual recognition technology according to claim 1 is characterized in that: The equipment thermal characteristic data includes the equipment surface temperature distribution, thermal gradient change rate and thermal anomaly area coordinates; the high-temperature anomaly area data includes the local temperature mutation point, thermal gradient change rate and anomaly area weight; the corrected anomaly area data specifically includes thermal distribution time series data, temperature change ratio and detection weight adjustment value; the equipment morphology change trend data includes edge profile parameter change, crack length growth rate and contact flatness offset; the anomaly pattern classification data specifically refers to the joint posterior probability distribution, anomaly pattern probability gradient value and anomaly category matching result.

3. The power monitoring system based on AI visual recognition technology according to claim 1 is characterized in that: The multi-spectral monitoring module comprises: The equipment contour extraction submodule uses the AI ​​visual recognition model to pre-process the image data based on the visible light image and infrared thermal imaging image data of the power monitoring area, obtain edge features, calculate the curvature change rate based on the edge gradient, and select edge pixels that meet the boundary continuity conditions to generate equipment appearance contour data; The device type identification submodule calls the AI ​​visual recognition model based on the device appearance profile data, compares the power equipment classification database, and performs matching calculations on the profile features using the formula: ； Calculate the device feature matching value, select the device type with the highest matching degree, and obtain the device classification result; in, Represents the device feature matching degree, The first vector representing the observed profile feature of the device Quantity, The first character representing the contour feature vector of the corresponding type in the equipment classification database Quantity, represents the feature vector dimension, Representative The weight of the components, Represents the preset maximum feature matching degree; The device thermal signature analysis submodule extracts the temperature distribution information in the infrared thermal imaging image data based on the device classification results, calculates the average temperature value of each device component area, analyzes the temperature gradient change, screens the area that exceeds the device standard temperature threshold, and obtains the device thermal signature data.

4. The power monitoring system based on AI visual recognition technology according to claim 1 is characterized in that: The thermal anomaly recognition module includes: The temperature mutation screening submodule extracts the local temperature data of the transformer winding, the circuit breaker contact and the busbar connection point based on the thermal characteristic data of the equipment, calculates the temperature difference between adjacent pixels in each area, screens the pixels exceeding the temperature change threshold, marks the temperature mutation points, and obtains the temperature mutation point data; The thermal gradient calculation submodule uses the formula based on the local temperature mutation point data: ； Calculate the thermal gradient change rate of each area to obtain thermal gradient change data; in, represents the rate of change of thermal gradient, Representing the region The temperature value of each measuring point, Represents the temperature value of adjacent measuring points, Represents the total number of measurement points; The monitoring priority setting submodule sets the monitoring priority of each area based on the thermal gradient change data and the set temperature mutation priority standard, sorts them according to the monitoring priority, screens high priority areas, and obtains high temperature abnormal area data.

5. The power monitoring system based on AI visual recognition technology according to claim 1 is characterized in that: The dynamic area correction module comprises: The temperature change calculation submodule calls the thermal response time parameters of the transformer oil tank, switch cabinet copper busbar and cable connector based on the abnormal high temperature area data, calculates the temperature change rate during the operating time, obtains the temperature change ratio of each area, and obtains the temperature change ratio data; The heat distribution comparison submodule calls the heat distribution curve of similar equipment based on the temperature change ratio data, extracts the temperature change trend of the corresponding time series, compares the temperature change ratio of each area, and uses the formula: ； Calculate the corrected temperature value and generate temperature deviation data; in, Represents the corrected temperature value, Represents the original measured temperature value, Representative The measured temperature value at each time series point, Represents the reference temperature value of similar equipment. Represents the number of time series points, Represents the thermal conductivity correction coefficient of the material; The abnormal region correction submodule corrects the temperature abnormality determination deviation caused by the difference in thermal conductivity characteristics of the materials based on the temperature deviation data, adjusts the identification boundary of the high temperature abnormal region, and obtains the corrected abnormal region data.

6. The power monitoring system based on AI visual recognition technology according to claim 1 is characterized in that: The structural trend prediction module includes: The morphological parameter extraction submodule extracts edge profile parameters, crack length parameters, and contact flatness parameters of the transformer core, support insulator, and line joint based on the corrected abnormal area data to generate morphological parameter data; The morphological change calculation submodule sets a sliding time window based on the morphological parameter data and calculates the morphological change gradient within a continuous time period using the formula: ； Calculate the trend value of morphological changes; in, Represents the trend value of morphological change, and Respectively represent and Morphological parameters within a time window, is the width of the sliding window, is the attenuation factor, is the base of natural logarithms; The trend assessment submodule assesses the device morphology change trend based on the morphology change trend value, determines the device morphology stability, and obtains the device morphology change trend data.

7. The power monitoring system based on AI visual recognition technology according to claim 1 is characterized in that: The abnormal mode identification module comprises: The abnormal parameter acquisition submodule obtains the cable current carrying capacity, switch contact resistance, line power factor and thermal abnormality area data based on the equipment morphology change trend data, extracts the time series distribution of each abnormal parameter, and generates abnormal parameter data; The joint posterior calculation submodule adopts the formula based on the abnormal parameter data: ； Calculate the joint posterior probability distribution of each abnormal parameter, obtain the probability gradient value of the abnormal pattern, and generate abnormal pattern probability data; in, Represents a given anomaly parameter matrix At that time, The posterior probability of an abnormal pattern, For the The likelihood probability of the parameter matrix under the abnormal mode, For the The prior probability of abnormal patterns, is the total number of abnormal patterns, For the parameter values, For the The mean of the parameters, For the The standard deviation of the parameters, is the number of abnormal parameters; The abnormal pattern matching submodule screens the abnormal pattern category with the highest matching probability based on the abnormal pattern probability data, performs pattern classification and labeling, and obtains abnormal pattern monitoring identification information.

Images