Substation equipment defect early warning method and system based on deep learning algorithm

Abstract

The invention discloses a substation equipment defect early warning method and system based on a deep learning algorithm, and relates to the technical field of equipment early warning, and the method comprises the steps: collecting temperature image data, and calculating a temperature difference value; binarizing the temperature difference value, marking a mask pixel, segmenting an area where the mask pixel is located into super-pixel blocks, calculating a comprehensive abnormal index, and marking a temperature abnormal area according to the comprehensive abnormal index; the temperature gradient of the area marked as the temperature anomaly is calculated, high-gradient area merging is carried out according to the safety standard and the Euclidean distance, and report output is carried out; constructing a deep learning model, and performing anomaly prediction on the comprehensive anomaly region to obtain prediction information; and transmitting the prediction information to an early warning terminal so as to carry out equipment early warning. The method can effectively identify and position the temperature abnormal area, guarantees the reliability of the detection result, fully combines the feature information of temperature distribution, can flexibly adapt to the local sudden change of the temperature, is not liable to be interfered by noise, and improves the early warning efficiency.

Description

Technical Field

[0001] The present invention relates to the field of equipment early warning technology, and in particular to a substation equipment defect early warning method and system based on a deep learning algorithm. Background Art

[0002] With the continuous development of power systems and the increasing complexity of substation equipment, ensuring the stable operation of equipment and fault warning has become an important part of power safety management. Infrared thermal imaging technology has gradually been applied in substation equipment monitoring as a non-destructive testing method. By collecting the temperature image of the equipment, it can effectively identify equipment abnormalities caused by heating, thereby reducing the risk of power outages caused by equipment failures.

[0003] At present, the application of existing deep learning-based detection systems in substation equipment defect detection still has certain shortcomings. It is often difficult to effectively identify and segment multi-scale temperature anomaly areas, lacks accurate positioning of the shape and boundaries of the abnormal areas, and cannot guarantee the reliability of the detection results. Secondly, some algorithms do not fully combine the characteristic information of temperature distribution and cannot flexibly adapt to local temperature mutations, making the screening of high temperature anomalies susceptible to noise interference, reducing the early warning efficiency. Summary of the invention

[0004] In view of the problems existing in the above-mentioned existing substation equipment defect early warning method and system based on deep learning algorithm, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that the application of the existing deep learning-based detection system in substation equipment defect detection still has certain shortcomings. It is often difficult to effectively identify and segment multi-scale temperature anomaly areas, lacks accurate positioning of the shape and boundaries of the abnormal areas, and cannot guarantee the reliability of the detection results. Secondly, some algorithms do not fully combine the characteristic information of temperature distribution and cannot flexibly adapt to local temperature mutations, which makes the screening of high temperature anomalies susceptible to noise interference, reducing the early warning efficiency.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a substation equipment defect early warning method based on deep learning algorithm, comprising:

[0007] Collect temperature image data of substation equipment and calculate the temperature difference between each pixel and its neighboring pixels;

[0008] Binarize the temperature difference value, mark the mask pixel according to the comparison result between the processed temperature difference value and the monitoring threshold, segment the area where the mask pixel is located into super-pixel blocks through the SLIC segmentation algorithm, calculate the comprehensive anomaly index of all super-pixel blocks, and mark the temperature anomaly area according to the comprehensive anomaly index;

[0009] Calculate the temperature gradient of the superpixel blocks corresponding to the temperature anomaly area, merge the high-gradient areas according to the safety standard and Euclidean distance, and output the merged comprehensive anomaly areas in a report;

[0010] Construct an FCOS deep learning model, and predict the anomaly classification probability, the coordinates of the bounding box of the anomaly region, and the coordinates of the center of the anomaly region based on the FCOS deep learning model to obtain prediction information;

[0011] The prediction information is encrypted and transmitted to the maintenance personnel's early warning terminal for equipment prediction.

[0012] As a preferred solution of the substation equipment defect early warning method based on deep learning algorithm described in the present invention, wherein: the temperature image data of the substation equipment is collected, and the temperature difference value between each pixel and the neighboring pixels is calculated, including:

[0013] Deploy infrared thermal imaging cameras based on the substation equipment environment to collect temperature image data of substation equipment in real time;

[0014] Perform data denoising and normalization processing on temperature image data;

[0015] Based on the processed temperature image data, the temperature difference between each pixel and its neighboring pixels is calculated, which is expressed as:

[0016]

[0017] Where ΔTi,j represents the temperature difference value of pixel (i,j), Tsm,i,j and Tsm,m,n represent the temperature values ​​of pixels (i,j) and (m,n) respectively, and N represents the number of neighboring pixels of pixel (i,j).

[0018] As a preferred solution of the substation equipment defect early warning method based on deep learning algorithm described in the present invention, wherein: the temperature difference value is binarized, mask pixels are marked according to the comparison result between the processed temperature difference value and the monitoring threshold, the area where the mask pixels are located is divided into super pixel blocks by SLIC segmentation algorithm, the comprehensive abnormal index of all super pixel blocks is calculated, and the temperature abnormal area is marked according to the comprehensive abnormal index, including:

[0019] The temperature difference value is binarized, and the sum of the temperature difference mean value calculated based on historical data and the table standard deviation is used as the monitoring threshold. If the processed temperature difference value is greater than the monitoring threshold, the corresponding pixel is marked as the mask pixel Mp(i,j); if the processed temperature difference value is less than or equal to the monitoring threshold, the corresponding pixel is not marked;

[0020] Count the mask pixels and generate high temperature areas on the temperature image data;

[0021] The SLIC segmentation algorithm is used to segment the high temperature area into multiple adjacent areas, and each area is defined as a superpixel block. The superpixel block is marked with a serial number according to its position. x represents the row of the superpixel block, and y represents the column of the superpixel block. The average temperature feature of each superpixel block is extracted, which is expressed as;

[0022]

[0023] Where Sx,y represents the pixel temperature variance of the (x,y)th superpixel block, M represents the number of pixels in the current superpixel block, Tsh,q represents the temperature value of the qth pixel in the superpixel block, and Tavg represents the average pixel temperature of the superpixel block;

[0024] Based on the ratio of the pixel with the highest pixel temperature, the pixel with the lowest pixel temperature, and the pixel average temperature of the superpixel block, the temperature concentration of the superpixel block is calculated, which is expressed as:

[0025]

[0026] Where Gx,y represents the temperature concentration of the (x,y)th superpixel block, Tmax,x,y and Tmin,x,y represent the maximum pixel temperature and the minimum pixel temperature of the (x,y)th superpixel block, respectively, and Tavg,x,y represents the average pixel temperature of the (x,y)th superpixel block;

[0027] The temperature concentration and temperature variance are combined to calculate the comprehensive anomaly index, which is expressed as:

[0028] Ax,y=Sx,y×(1-Cx,y);

[0029] Where Ax,y represents the comprehensive abnormality index of the (x,y)th superpixel block;

[0030] The mean and double standard deviation of the comprehensive anomaly index of all superpixel blocks are calculated as the distribution threshold. If the comprehensive anomaly index of the superpixel block is greater than or equal to the distribution threshold, it is judged as a temperature anomaly area.

[0031] As a preferred solution of the substation equipment defect early warning method based on deep learning algorithm described in the present invention, wherein: the temperature gradient of the superpixel block marked as the temperature abnormality area is calculated, and the high gradient area is merged according to the safety standard and Euclidean distance, including:

[0032] Based on the average diameter, maximum pixel temperature and minimum pixel temperature of the superpixel block in the marked temperature anomaly area, the temperature gradient of the superpixel block is calculated, which is expressed as:

[0033]

[0034] Among them, Fx,y represents the temperature gradient of the superpixel block in the marked temperature anomaly area, and dx,y represents the average diameter of the superpixel block;

[0035] The sum of the temperature gradient average and standard deviation calculated based on historical data is used as the gradient threshold. If the calculated temperature gradient is greater than or equal to the gradient threshold, the superpixel block marked as the temperature anomaly area is judged as a high-gradient area, and the temperature difference between the pixel average temperature of the high-gradient area and the adjacent superpixel block is calculated;

[0036] Based on the temperature safety standard, the adjacent superpixel blocks with a temperature difference less than the temperature safety standard are marked as adjacent areas that meet the temperature consistency check;

[0037] The adjacent areas that meet the temperature consistency check are merged with the current high gradient area, and the Euclidean distance of the adjacent areas is calculated. The maximum historical distance of the pixel space connection is used as the distance threshold. If the Euclidean distance of the adjacent areas is greater than the distance threshold, the regions are connected to form a comprehensive abnormal area.

[0038] As a preferred solution of the substation equipment defect early warning method based on deep learning algorithm of the present invention, wherein: the report output of the combined comprehensive abnormal area includes:

[0039] Assign a unique number to the merged comprehensive anomaly area, and calculate the overall characteristics of the merged comprehensive anomaly area, the overall characteristics including temperature mean, temperature variance, and temperature gradient;

[0040] The timestamp of the temperature image data of the merged comprehensive abnormal area is recorded, the position is marked based on the pixel coordinates of the center point of the comprehensive abnormal area, and a temperature anomaly report of the comprehensive abnormal area is generated using a chart tool.

[0041] As a preferred solution of the substation equipment defect early warning method based on deep learning algorithm described in the present invention, wherein: the FCOS deep learning model is constructed, and the abnormal classification probability, the coordinates of the abnormal area boundary box and the center coordinates of the abnormal area are predicted for the comprehensive abnormal area based on the FCOS deep learning model to obtain prediction information, including:

[0042] Build the FCOS deep learning model and use the pixel data of the merged comprehensive abnormal area as the input image;

[0043] Use the feature pyramid network FPN to extract multi-scale features from the input image and obtain feature maps of different scales to help locate comprehensive abnormal areas of different sizes;

[0044] At each pixel point in the feature map, the classification branch is used to classify the pixel point into two categories: abnormal or not, and the classification loss is obtained. At the same time, the centrality loss of the centrality branch is calculated, which is expressed as:

[0045]

[0046] Among them, Lcls represents classification loss, uk represents the actual category label, pk represents the probability of category k, K represents the total number of categories, Lcen represents the center loss, g, r, t, b represent the distance from the center pixel to the left, right, top and bottom boundaries respectively;

[0047] Use the regression branch to perform bounding box regression on each synthetic abnormal region and obtain the regression loss, which is expressed as:

[0048]

[0049] Among them, Lreg represents regression loss, Smo represents SmoothL1, that is, smooth L1 loss, ηi and Represents the coordinate values ​​of the upper, lower, left, and right sides of the predicted bounding box and the true bounding box respectively;

[0050] The total loss function of FCOS is calculated by integrating classification loss, center loss and regression loss, which is expressed as:

[0051] Lto=Lcls+λ1Lreg+λ2Lcen;

[0052] Among them, Lto represents the total loss of LCOS, Lcls represents the classification loss, Lreg represents the regression loss, and Lcen represents the centrality loss;

[0053] The model is trained based on the training set. The temperature image data is input into the FCOS deep learning model to generate classification scores, bounding box positions, and center predictions. The total loss Lto is calculated based on the prediction results and actual labels. The Adam optimizer is used for gradient descent optimization to update the parameters of the FCOS deep learning model until the model loss no longer decreases significantly during continuous iterations.

[0054] The pixel data of the merged comprehensive abnormal area is input into the FCOS deep learning model to detect the abnormal area, and the abnormal classification probability of each pixel point, the bounding box coordinates of each comprehensive abnormal area and the center coordinates of each comprehensive abnormal area are obtained respectively.

[0055] As a preferred solution of the substation equipment defect early warning method based on deep learning algorithm described in the present invention, wherein: the prediction information is encrypted and transmitted to the early warning terminal of the maintenance personnel to perform equipment early warning, including:

[0056] Generate a device temperature change trend graph through a chart tool based on the anomaly classification probability, the coordinates of the integrated anomaly region bounding box, and the coordinates of the integrated anomaly region center;

[0057] The equipment temperature change trend chart is encrypted using the TLS protocol using the AES-256 symmetric encryption algorithm for network transmission and sent to the maintenance personnel's early warning terminal, so that the maintenance personnel can quickly check the defect warning information caused by abnormal temperature of substation equipment through the equipment temperature change trend chart and perform early warning operations.

[0058] Another object of the present invention is to provide a system for early warning of substation equipment defects based on a deep learning algorithm, which is applied to the above-mentioned method for early warning of substation equipment defects based on a deep learning algorithm, and comprises:

[0059] The data acquisition module is used to collect temperature image data of substation equipment and calculate the temperature difference value between each pixel and its neighboring pixels;

[0060] The temperature anomaly recognition module is used to binarize the temperature difference value, mark the mask pixel according to the comparison result between the processed temperature difference value and the monitoring threshold, divide the area where the mask pixel is located into super-pixel blocks through the SLIC segmentation algorithm, calculate the comprehensive anomaly index of all super-pixel blocks, and mark the temperature anomaly area according to the comprehensive anomaly index;

[0061] The high gradient region merging module is used to calculate the temperature gradient of superpixel blocks marked as temperature anomaly areas and merge high gradient regions according to safety standards and Euclidean distance;

[0062] A report output module is used to output the combined comprehensive abnormal area into a report;

[0063] FCOS deep learning detection module, used to build FCOS deep learning model, and predict the abnormal classification probability, abnormal area bounding box coordinates and abnormal area center coordinates of the comprehensive abnormal area based on the FCOS deep learning model to obtain prediction information;

[0064] The secure transmission module is used to encrypt and transmit the prediction information to the maintenance personnel's early warning terminal for equipment early warning.

[0065] A computer device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of the above-mentioned substation equipment defect early warning method based on a deep learning algorithm when executing the computer program.

[0066] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned substation equipment defect early warning method based on a deep learning algorithm.

[0067] The beneficial effects of the present invention are as follows: through binarization processing, mask pixels with abnormal temperature can be accurately screened out, and mask pixels that meet the abnormal conditions are clustered into high-temperature areas. The SLIC superpixel segmentation algorithm divides the high-temperature area in the temperature image into multiple adjacent and continuous areas, thereby improving the resolution of anomaly detection. The comprehensive anomaly index combines the temperature concentration and temperature variance of the superpixel block, thereby effectively identifying areas with drastic temperature changes and eliminating evenly distributed high-temperature areas. By merging high-gradient areas, local or large-scale abnormal areas can be simultaneously identified to ensure the reliability of the detection results. In addition, by fully combining the characteristic information of temperature distribution, the local temperature mutation can be flexibly adapted, so that the screening of high-temperature anomalies is not easily affected by noise, reducing the early warning efficiency, and thus facilitating a clearer identification of faults. Through the FCOS model, the detection of these areas is further optimized in the feature space of deep learning, providing reliable temperature distribution information with physical significance, so that the system can more accurately locate the fault hotspots and boundaries. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0069] Figure 1 A schematic diagram of a process flow of a substation equipment defect early warning method based on a deep learning algorithm provided by the present invention;

[0070] Figure 2 A structural schematic diagram of a substation equipment defect early warning system based on a deep learning algorithm provided by the present invention. DETAILED DESCRIPTION

[0071] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0072] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0073] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0074] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and provides a substation equipment defect early warning method based on a deep learning algorithm, comprising:

[0075] S1, collect temperature image data of substation equipment and calculate the temperature difference value between each pixel and its neighboring pixels;

[0076] Preferably, collecting temperature image data of substation equipment and calculating the temperature difference between each pixel and neighboring pixels include:

[0077] Deploy infrared thermal imaging cameras based on the substation equipment environment to collect temperature image data of substation equipment in real time;

[0078] Perform data denoising and normalization processing on temperature image data;

[0079] Based on the processed temperature image data, the temperature difference between each pixel and its neighboring pixels is calculated, which is expressed as:

[0080]

[0081] Where ΔTi,j represents the temperature difference of pixel (i,j), Tsm,i,j and Tsm,m,n represent the temperature values ​​of pixels (i,j) and (m,n) respectively, and N represents the number of neighboring pixels of pixel (i,j). Represents the sum of the temperature values ​​of all pixels.

[0082] By using infrared thermal imaging cameras to collect temperature image data of substation equipment in real time, the temperature changes of the equipment can be continuously monitored. The dimension of the temperature data is unified through normalization processing to ensure the comparability of the temperature image data under different ambient temperatures or lighting conditions. By calculating the temperature difference between each pixel and its neighboring pixels, local temperature mutation areas can be effectively identified, that is, small-scale hot spots or uneven temperature distribution on the surface of the equipment can be detected.

[0083] S2, binarize the temperature difference value, mark the mask pixel according to the comparison result between the processed temperature difference value and the monitoring threshold, segment the superpixel block in the area where the mask pixel is located by SLIC segmentation algorithm, calculate the comprehensive anomaly index of all superpixel blocks, and mark the temperature anomaly area according to the comprehensive anomaly index;

[0084] Preferably, the temperature difference value is binarized, mask pixels are marked according to the comparison result between the processed temperature difference value and the monitoring threshold, super pixel blocks are segmented in the area where the mask pixels are located by the SLIC segmentation algorithm, the comprehensive abnormality index of all super pixel blocks is calculated, and the temperature abnormality area is marked according to the comprehensive abnormality index, including:

[0085] The temperature difference value is binarized, and the sum of the temperature difference mean value calculated based on historical data and the table standard deviation is used as the monitoring threshold. If the processed temperature difference value is greater than the monitoring threshold, the corresponding pixel is marked as the mask pixel Mp(i,j); if the processed temperature difference value is less than or equal to the monitoring threshold, the corresponding pixel is not marked;

[0086] Count the mask pixels and generate high temperature areas on the temperature image data;

[0087] The SLIC segmentation algorithm is used to segment the high temperature area in the temperature image into multiple adjacent areas, and each area is defined as a superpixel block. The superpixel block is marked with a serial number according to its position in the temperature image. x represents the row of the temperature image where the superpixel block is located, and y represents the column of the temperature image where the superpixel block is located. The average temperature feature of each superpixel block is extracted, which is expressed as:

[0088]

[0089] Where Sx,y represents the pixel temperature variance of the (x,y)th superpixel block, M represents the number of pixels in the current superpixel block, Tsh,q represents the temperature value of the qth pixel in the superpixel block, and Tavg represents the average pixel temperature of the superpixel block;

[0090] Based on the pixel temperature of the superpixel block, the ratio of the pixel with the lowest pixel temperature and the pixel average temperature, the temperature concentration of the superpixel block is calculated, which is expressed as:

[0091]

[0092] Where Gx,y represents the temperature concentration of the (x,y)th superpixel block, Tmax,x,y and Tmin,x,y represent the maximum pixel temperature and the minimum pixel temperature of the (x,y)th superpixel block, respectively, and Tavg,x,y represents the average pixel temperature of the (x,y)th superpixel block;

[0093] The temperature concentration and temperature variance are combined to calculate the comprehensive anomaly index, which is expressed as:

[0094] Ax,y=Sx,y×(1-Cx,y);

[0095] Where Ax,y represents the comprehensive abnormality index of the (x,y)th superpixel block;

[0096] The mean and the sum of two times the standard deviation of the comprehensive anomaly index of all superpixel blocks are calculated as the distribution threshold. If the comprehensive anomaly index of a superpixel block is greater than or equal to the distribution threshold, the superpixel block is judged as a temperature anomaly area.

[0097] By dynamically setting the monitoring threshold based on the mean and standard deviation of temperature differences, the binarization process can accurately screen out the mask pixels with abnormal temperatures, avoiding false alarms caused by small-scale temperature fluctuations. The mask pixels that meet the abnormal conditions are aggregated into high-temperature areas, which can identify the actual hotspot locations on the surface of the equipment, making the monitoring results more consistent with the physical characteristics of the equipment and laying the foundation for temperature anomaly positioning. The SLIC superpixel segmentation algorithm divides the high-temperature area in the temperature image into multiple adjacent and continuous areas, so that the temperature image can be refined into multiple small areas (superpixel blocks) with similar temperature characteristics, greatly improving the resolution of anomaly detection. The average temperature, temperature variance and temperature concentration of superpixel blocks: The average temperature, variance and concentration characteristics are extracted for each superpixel block, which can finely characterize the temperature state of each area. The comprehensive anomaly index combines the temperature concentration and temperature variance of the superpixel block, which can effectively identify areas with drastic temperature changes and eliminate evenly distributed high-temperature areas. By marking abnormal areas, real-time monitoring of the equipment can be achieved, so that equipment managers can promptly know and locate fault hotspots, so as to take corresponding maintenance measures to avoid greater damage.

[0098] S3, calculate the temperature gradient of the superpixel block marked as the temperature anomaly area, merge the high gradient areas according to the safety standard and Euclidean distance, and output the merged comprehensive anomaly area in a report;

[0099] Preferably, the temperature gradient of the superpixel block marked as the temperature abnormality area is calculated, and the high gradient area is merged according to the safety standard and the Euclidean distance, including:

[0100] Based on the average diameter of the superpixel block in the marked temperature anomaly area and the maximum pixel temperature and the minimum pixel temperature of the superpixel block, the temperature gradient of the superpixel block is calculated, which is expressed as:

[0101]

[0102] Among them, Fx,y represents the temperature gradient of the superpixel block in the marked temperature anomaly area, and dx,y represents the average diameter of the superpixel block;

[0103] The sum of the temperature gradient average and standard deviation calculated based on historical data is used as the gradient threshold. If the calculated temperature gradient is greater than or equal to the gradient threshold, the superpixel block marked as the temperature anomaly area is judged as a high-gradient area, and the temperature difference between the pixel average temperature of the high-gradient area and the adjacent superpixel block is calculated;

[0104] Based on the temperature safety standard, the adjacent superpixel blocks with a temperature difference less than the temperature safety standard are marked as adjacent areas that meet the temperature consistency check;

[0105] The adjacent areas that meet the temperature consistency check are merged with the current high gradient area, and the Euclidean distance of the adjacent areas is calculated. The maximum historical distance of the pixel space connection is used as the distance threshold. If the Euclidean distance of the adjacent areas is greater than the distance threshold, the regions are connected to form a comprehensive abnormal area.

[0106] By calculating the temperature gradient of the superpixel block, it is possible to effectively distinguish between areas with large temperature changes (high gradient areas) and areas with small temperature changes. By checking the temperature difference threshold of the temperature safety standard, it is ensured that only adjacent superpixel blocks that meet the safety conditions will be merged with the current high gradient area, avoiding misjudging areas with relatively low temperatures as abnormal areas, thereby improving the rationality of the merging process. In the process of superpixel block merging, by setting the maximum distance threshold based on historical data, it is ensured that the merged superpixel blocks are spatially connected without over-expansion, ensuring that each comprehensive abnormal area has spatial integrity. Through the dual check of temperature difference and Euclidean distance, it is possible to effectively reduce small fragment areas that are misjudged as abnormal, and concentrate the detection results in high-temperature areas with obvious fault characteristics, thereby improving the efficiency of fault location.

[0107] In the process of merging high-gradient areas, local or large-scale abnormal areas can be identified at the same time, and comprehensive high-temperature area annotations can be formed for them, making fault identification more complete, especially when large-scale faults occur, which helps to identify the fault more clearly.

[0108] Furthermore, the combined abnormal area is output as a report, including:

[0109] Assign a unique number to the merged comprehensive anomaly area, and calculate the overall characteristics of the merged comprehensive anomaly area, the overall characteristics including temperature mean, temperature variance, and temperature gradient;

[0110] The timestamp of the temperature image data of the merged comprehensive abnormal area is recorded, the position is marked based on the pixel coordinates of the center point of the comprehensive abnormal area, and a temperature anomaly report of the comprehensive abnormal area is generated using a chart tool.

[0111] By assigning a unique number to each merged comprehensive abnormal area, it is possible to ensure that the specific fault area is accurately located in the subsequent data recording, analysis and maintenance work to avoid data confusion. The recording of temperature mean, temperature variance and temperature gradient can help identify the severity of the fault area and provide a reference for equipment health assessment. The temperature report of the abnormal area is generated through chart tools, and data such as temperature mean, variance, gradient and timestamp are visualized, which is convenient for managers to intuitively analyze abnormal trends and evaluate the health status of the equipment.

[0112] S4, building a FCOS deep learning model, and predicting the abnormal classification probability, the coordinates of the abnormal region bounding box, and the coordinates of the center of the abnormal region for the comprehensive abnormal region based on the FCOS deep learning model to obtain prediction information;

[0113] Preferably, an FCOS deep learning model is constructed, and based on the FCOS deep learning model, the abnormal classification probability, the coordinates of the abnormal region bounding box, and the coordinates of the center of the abnormal region are predicted for the comprehensive abnormal region to obtain prediction information, including:

[0114] Build the FCOS deep learning model and use the pixel data of the merged comprehensive abnormal area as the input image;

[0115] Use the feature pyramid network FPN to extract multi-scale features from the input image and convert the input image into feature maps of different scales to help locate comprehensive abnormal areas of different sizes;

[0116] At each pixel point in the feature map, the classification branch is used to classify the pixel point into two categories: abnormal or not, and the classification loss is obtained. At the same time, the centrality loss of the centrality branch is calculated, which is expressed as:

[0117]

[0118] Among them, Lcls represents classification loss, uk represents the actual category label, pk represents the probability of category k, K represents the total number of categories, which is 2 here, Lcen represents the center loss, g, r, t, and b represent the distances from the center pixel to the left, right, upper and lower boundaries respectively;

[0119] Use the regression branch to perform bounding box regression on each synthetic abnormal region and obtain the regression loss, which is expressed as:

[0120]

[0121] Among them, Lreg represents regression loss, Smo represents SmoothL1, that is, smooth L1 loss (used to balance the robustness and accuracy of regression), ηi and Represents the coordinate values ​​of the upper, lower, left, and right sides of the predicted bounding box and the true bounding box respectively;

[0122] The total loss function of FCOS is calculated by integrating classification loss, center loss and regression loss, which is expressed as:

[0123] Lto=Lcls+λ1Lreg+λ2Lcen;

[0124] Among them, Lto represents the total loss of LCOS, Lcls represents the classification loss, Lreg represents the regression loss, and Lcen represents the centrality loss;

[0125] Model training is performed based on the training set. The temperature image data is input into the FCOS deep learning model to generate classification scores, bounding box positions, and center predictions. The total loss Lto is calculated based on the prediction results and actual labels. The Adam optimizer is used for gradient descent optimization to update the parameters of the FCOS deep learning model until the loss of the FCOS deep learning model no longer decreases significantly during continuous iterations. That is, if the loss of the model no longer decreases significantly during continuous iterations, the iteration output model parameters are stopped to update the model.

[0126] The pixel data of the merged comprehensive abnormal area is input into the FCOS deep learning model to detect the abnormal area, and the abnormal classification probability of each pixel point, the bounding box coordinates of each comprehensive abnormal area and the center coordinates of each comprehensive abnormal area are obtained respectively.

[0127] The feature pyramid network (FPN) is used to extract multi-scale features from the input image, which can identify abnormal areas of different sizes at different scales, ensuring that the model can simultaneously detect large-scale equipment surface abnormalities and small-scale local hotspots. The classification branch is used to classify each pixel as abnormal or non-abnormal, and the abnormal classification probability is obtained. The use of classification loss enables the model to focus more on judging whether the pixel has abnormal temperature, further improving the accuracy of abnormal area identification. The regression branch is used for bounding box regression, which enables the model to accurately locate the boundaries of the abnormal area, obtain clear bounding box coordinates, and clarify the scope of the abnormal area. By calculating the center loss of the center branch, the model can better align the center of the detected abnormal area with the core area of ​​the actual abnormal area. The total loss function of the comprehensive classification loss, regression loss and center loss ensures the balanced performance of the model in classification, positioning and center prediction, making the detection results more stable.

[0128] Based on the temperature difference calculation, SLIC segmentation and region merging process, potential high temperature anomaly areas are preliminarily identified in the infrared temperature image. After the comprehensive abnormal areas are marked, these areas are used as the input of the FCOS model, which narrows the detection range of the model. This not only reduces the calculation amount of the model, but also concentrates the model resources on areas where anomalies are more likely to exist, improving the detection efficiency. The process of preliminarily identifying and merging abnormal areas can filter out noise and non-abnormal areas in the image, reducing the risk of possible false alarms of the model, and providing a guarantee for the quality of input data of the FCOS model, thereby improving the overall accuracy of anomaly detection;

[0129] The features such as temperature difference, temperature variance, temperature gradient and temperature concentration calculated in the previous article provide reliable temperature distribution information in physical sense for the abnormal areas. The FCOS model further optimizes the detection of these areas in the feature space of deep learning, thereby achieving the complementary effect of traditional features and deep features, improving the precision of detection, and enabling the system to locate fault hotspots and boundaries more accurately. Users can combine the early physical features with the later deep learning detection results to clearly understand the characteristics and fault types of each abnormal area.

[0130] S5, encrypt and transmit the prediction information to the early warning terminal of the maintenance personnel to provide equipment early warning;

[0131] Preferably, the prediction information is encrypted and transmitted to the early warning terminal of the maintenance personnel to perform equipment early warning, including:

[0132] Generate a device temperature change trend graph through a chart tool based on the anomaly classification probability, the coordinates of the integrated anomaly region bounding box, and the coordinates of the integrated anomaly region center;

[0133] The temperature change trend chart is encrypted using the TLS protocol using the AES-256 symmetric encryption algorithm for network transmission and sent to the maintenance personnel's early warning terminal, so that the maintenance personnel can quickly check the defect warning information caused by abnormal temperature of substation equipment through the temperature change trend chart and perform early warning operations.

[0134] The equipment temperature change trend chart can intuitively display the temperature change over time, helping maintenance personnel to quickly identify the temperature fluctuation trend in abnormal areas. Through the trend chart, maintenance personnel can clearly understand the time nodes and trends of fault development, so as to more reasonably arrange the maintenance frequency and maintenance priority, and improve the scientific nature of decision-making. The TLS protocol protects the integrity of data through identity authentication and encryption, further reducing the risk of malicious tampering of data during transmission, and providing security for the equipment early warning system. AES-256 encryption provides extremely high security, making it difficult for unauthorized personnel to decrypt early warning data, thereby effectively protecting the privacy of substation data. By sending abnormal area information and temperature change trend charts to the maintenance personnel's early warning terminal in real-time, the maintenance team can respond more quickly, reduce the response time of the fault, and avoid further expansion of the fault.

[0135] Example 2, reference Figure 2 , which is the second embodiment of the present invention, and which is different from the previous embodiment, provides a substation equipment defect early warning system based on a deep learning algorithm, including:

[0136] The data acquisition module is used to collect temperature image data of substation equipment and calculate the temperature difference value between each pixel and its neighboring pixels;

[0137] The temperature anomaly recognition module is used to binarize the temperature difference value, mark the mask pixel according to the comparison result between the processed temperature difference value and the monitoring threshold, divide the area where the mask pixel is located into super-pixel blocks through the SLIC segmentation algorithm, calculate the comprehensive anomaly index of all super-pixel blocks, and mark the temperature anomaly area according to the comprehensive anomaly index;

[0138] The high gradient region merging module is used to calculate the temperature gradient of superpixel blocks marked as temperature anomaly areas and merge high gradient regions according to safety standards and Euclidean distance;

[0139] A report output module is used to output the combined comprehensive abnormal area into a report;

[0140] FCOS deep learning detection module, used to build FCOS deep learning model, and predict the abnormal classification probability, abnormal area bounding box coordinates and abnormal area center coordinates of the comprehensive abnormal area based on the FCOS deep learning model to obtain prediction information;

[0141] The secure transmission module is used to encrypt and transmit the prediction information to the maintenance personnel's early warning terminal for equipment early warning.

[0142] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0143] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0144] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0145] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0146] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A substation equipment defect early warning method based on deep learning algorithm, characterized by: include, Collect temperature image data of substation equipment and calculate the temperature difference between each pixel and its neighboring pixels; Binarize the temperature difference value, mark the mask pixel according to the comparison result between the processed temperature difference value and the monitoring threshold, segment the area where the mask pixel is located into super-pixel blocks through the SLIC segmentation algorithm, calculate the comprehensive anomaly index of all super-pixel blocks, and mark the temperature anomaly area according to the comprehensive anomaly index; Calculate the temperature gradient of the superpixel blocks corresponding to the temperature anomaly area, merge the high-gradient areas according to the safety standard and Euclidean distance, and output the merged comprehensive anomaly areas in a report; Construct an FCOS deep learning model, and predict the anomaly classification probability, the coordinates of the bounding box of the anomaly region, and the coordinates of the center of the anomaly region based on the FCOS deep learning model to obtain prediction information; The prediction information is encrypted and transmitted to the maintenance personnel's early warning terminal for equipment early warning.

2. The substation equipment defect early warning method based on deep learning algorithm according to claim 1 is characterized in that: The temperature image data of the substation equipment is collected, and the temperature difference value between each pixel and the neighboring pixels is calculated. include, Deploy infrared thermal imaging cameras based on the substation equipment environment to collect temperature image data of substation equipment in real time; Perform data denoising and normalization processing on temperature image data; Based on the processed temperature image data, the temperature difference between each pixel and its neighboring pixels is calculated, which is expressed as: Where, ΔT i,j Represents the temperature difference value of pixel (i, j), T sm,i,j and T sm,m,n They represent the temperature values ​​of pixels (i, j) and (m, n) respectively, and N represents the number of neighboring pixels of pixel (i, j).

3. The substation equipment defect early warning method based on deep learning algorithm according to claim 2 is characterized in that: Binarize the temperature difference value, mark the mask pixel according to the comparison result between the processed temperature difference value and the monitoring threshold, segment the super pixel block in the area where the mask pixel is located by SLIC segmentation algorithm, calculate the comprehensive abnormal index of all super pixel blocks, and mark the temperature abnormal area according to the comprehensive abnormal index, including: The temperature difference value is binarized, and the sum of the temperature difference mean value calculated based on historical data and the table standard deviation is used as the monitoring threshold. If the processed temperature difference value is greater than the monitoring threshold, the corresponding pixel is marked as the mask pixel M. p (i, j), if the processed temperature difference value is less than or equal to the monitoring threshold, the corresponding pixel is not marked; Count the mask pixels and generate high temperature areas on the temperature image data; The SLIC segmentation algorithm is used to segment the high temperature area into multiple adjacent areas, and each area is defined as a superpixel block. The superpixel blocks are marked with serial numbers according to their positions. x represents the row of the superpixel block, and y represents the column of the superpixel block. The average temperature feature of each superpixel block is extracted, which is expressed as: Among them, S x,y represents the pixel temperature variance of the (x, y)th superpixel block, M represents the number of pixels in the current superpixel block, and T sh,q It is represented as the temperature value of the qth pixel in the superpixel block, T avg represents the average pixel temperature of the superpixel block; Based on the ratio of the maximum pixel temperature, the minimum pixel temperature and the average pixel temperature of the superpixel block, the temperature concentration of the superpixel block is calculated, which is expressed as: Among them, G x,y represents the temperature concentration of the (x,y)th superpixel block, T max,x,y and T min,x,y Respectively represent the maximum pixel temperature and the minimum pixel temperature of the (x, y)th superpixel block, T avg,x,y represents the average pixel temperature of the (x,y)th superpixel block; The temperature concentration and temperature variance are combined to calculate the comprehensive anomaly index, which is expressed as: A x,y =S x,y ×(1-C x,y ); Among them, A x,y represents the comprehensive abnormality index of the (x,y)th superpixel block; The mean and double standard deviation of the comprehensive anomaly index of all superpixel blocks are calculated as the distribution threshold. If the comprehensive anomaly index of the superpixel block is greater than or equal to the distribution threshold, it is judged as a temperature anomaly area.

4. The transformer substation equipment defect early warning method based on deep learning algorithm according to claim 3 is characterized in that: The temperature gradient of the superpixel block marked as the temperature anomaly area is calculated, and the high gradient area is merged according to the safety standard and the Euclidean distance, including: Based on the average diameter, maximum pixel temperature, and minimum pixel temperature of the superpixel block in the marked temperature anomaly area, the temperature gradient of the superpixel block is calculated, which is expressed as: Among them, F x,y The temperature gradient of the superpixel block representing the marked temperature anomaly area, d x,y represents the average diameter of superpixel blocks; The sum of the temperature gradient average and standard deviation calculated based on historical data is used as the gradient threshold. If the calculated temperature gradient is greater than or equal to the gradient threshold, the superpixel block marked as the temperature anomaly area is judged as a high-gradient area, and the temperature difference between the pixel average temperature of the high-gradient area and the adjacent superpixel block is calculated; Based on the temperature safety standard, the adjacent superpixel blocks with a temperature difference less than the temperature safety standard are marked as adjacent areas that meet the temperature consistency check; The adjacent areas that meet the temperature consistency check are merged with the current high gradient area, and the Euclidean distance of the adjacent areas is calculated. The maximum historical distance of the pixel space connection is used as the distance threshold. If the Euclidean distance of the adjacent areas is greater than the distance threshold, the regions are connected to form a comprehensive abnormal area.

5. The transformer substation equipment defect early warning method based on deep learning algorithm according to claim 4, characterized in that: The combined abnormal area is output as a report, including: Assign a unique number to the merged comprehensive abnormal area, calculate the overall characteristics of the merged comprehensive abnormal area, The overall characteristics include temperature mean, temperature variance, and temperature gradient; The timestamp of the temperature image data of the merged comprehensive abnormal area is recorded, the position is marked based on the pixel coordinates of the center point of the comprehensive abnormal area, and a temperature anomaly report of the comprehensive abnormal area is generated using a chart tool.

6. The method for early warning of substation equipment defects based on a deep learning algorithm according to claim 5, characterized in that: The FCOS deep learning model is constructed, and based on the FCOS deep learning model, the abnormal classification probability, the coordinates of the abnormal region boundary box, and the coordinates of the center of the abnormal region are predicted for the comprehensive abnormal region to obtain prediction information, including: Build the FCOS deep learning model and use the pixel data of the merged comprehensive abnormal area as the input image; Use the feature pyramid network FPN to extract multi-scale features from the input image and obtain feature maps of different scales to help locate comprehensive abnormal areas of different sizes; At each pixel point in the feature map, the classification branch is used to classify the pixel point into two categories: abnormal or not, and the classification loss is obtained. At the same time, the centrality loss of the centrality branch is calculated, which is expressed as: Among them, L cls represents the classification loss, u k represents the actual category label, p k represents the probability of category k, K represents the total number of categories, L cen represents the center loss, g, r, t, and b represent the distances from the center pixel to the left, right, top, and bottom boundaries respectively; Use the regression branch to regress the bounding box of the comprehensive abnormal area and obtain the regression loss, which is expressed as: Among them, L reg represents regression loss, Smo represents SmoothL1, i.e. smooth L1 loss, η i and Represents the coordinate values ​​of the upper, lower, left, and right sides of the predicted bounding box and the true bounding box respectively; The total loss function of FCOS is calculated by integrating classification loss, center loss and regression loss, which is expressed as: L to =L cls +λ1L reg +λ2L cen ; Among them, L to represents the total loss of LCOS, L cls represents the classification loss, L reg represents the regression loss, L cen represents the centrality loss; The model is trained based on the training set. The temperature image data is input into the FCOS deep learning model to generate classification scores, bounding box positions and center predictions. The total loss L is calculated based on the prediction results and actual labels. to , use the Adam optimizer to perform gradient descent optimization and update the parameters of the FCOS deep learning model until the loss of the FCOS deep learning model no longer decreases significantly during continuous iterations; The pixel data of the merged comprehensive abnormal area is input into the FCOS deep learning model to detect the abnormal area, and the abnormal classification probability of each pixel point, the bounding box coordinates of each comprehensive abnormal area and the center coordinates of each comprehensive abnormal area are obtained respectively.

7. The transformer substation equipment defect early warning method based on deep learning algorithm according to claim 6, characterized in that: The encrypted transmission of the prediction information to the early warning terminal of the maintenance personnel for equipment early warning includes: Generate a device temperature change trend graph through a chart tool based on the anomaly classification probability, the coordinates of the integrated anomaly region bounding box, and the coordinates of the integrated anomaly region center; The equipment temperature change trend chart is encrypted using the TLS protocol using the AES-256 symmetric encryption algorithm for network transmission and sent to the maintenance personnel's early warning terminal, so that the maintenance personnel can quickly check the defect warning information caused by abnormal temperature of substation equipment through the equipment temperature change trend chart and perform early warning operations.

8. A substation equipment defect early warning system based on a deep learning algorithm, applied to a substation equipment defect early warning method based on a deep learning algorithm as described in any one of claims 1 to 7, characterized in that: include, The data acquisition module is used to collect temperature image data of substation equipment and calculate the temperature difference value between each pixel and its neighboring pixels; The temperature anomaly recognition module is used to binarize the temperature difference value, mark the mask pixel according to the comparison result between the processed temperature difference value and the monitoring threshold, divide the area where the mask pixel is located into super-pixel blocks through the SLIC segmentation algorithm, calculate the comprehensive anomaly index of all super-pixel blocks, and mark the temperature anomaly area according to the comprehensive anomaly index; The high gradient region merging module is used to calculate the temperature gradient of superpixel blocks marked as temperature anomaly areas and merge high gradient regions according to safety standards and Euclidean distance; A report output module is used to output the combined comprehensive abnormal area into a report; FCOS deep learning detection module, used to build FCOS deep learning model, and predict the abnormal classification probability, abnormal area bounding box coordinates and abnormal area center coordinates of the comprehensive abnormal area based on the FCOS deep learning model to obtain prediction information; The secure transmission module is used to encrypt and transmit the prediction information to the maintenance personnel's early warning terminal for equipment early warning.

9. A computer device comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the substation equipment defect early warning method based on a deep learning algorithm as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the substation equipment defect early warning method based on a deep learning algorithm described in any one of claims 1 to 7 are implemented.

Images