Thermal fault diagnosis method and device for power equipment, storage medium and computer equipment

Abstract

The invention relates to the technical field of power equipment safety detection, and discloses a power equipment thermal fault diagnosis method and device, a storage medium and computer equipment, and the method comprises the steps: collecting on-site infrared image data of a transformer substation, constructing an infrared image data set, and carrying out the image enhancement processing; performing target identification and segmentation by using an instance segmentation model combining a cross-stage local network and an attention mechanism, and determining a power equipment region; inputting the gray value of the power equipment area into the deep neural network model to obtain a temperature value of the target power equipment, drawing a probability density curve of the temperature value by using a kernel density estimation method, obtaining a reference temperature of the target power equipment, and calculating a hot spot temperature; and calculating the relative temperature difference and determining a fault state based on a preset fault judgment standard. According to the method, the original data quality is improved, the power equipment is accurately identified and segmented, high-precision temperature prediction is carried out, and finally, full-automatic and reliable fault judgment of the power equipment is completed.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment safety detection, and in particular to a method, device, storage medium and computer equipment for diagnosing thermal faults of power equipment. Background Art

[0002] Traditional thermal fault detection of power equipment mainly relies on manual inspections and regular infrared inspections. Among them, manual inspections are too dependent on the experience and judgment of inspectors and are easily affected by subjective factors. At the same time, when large-scale power equipment needs to be inspected, manual inspections are less efficient. Although traditional infrared inspections can monitor equipment temperature through thermal imaging, manual analysis of thermal images is still required, resulting in problems such as detection lag and high misjudgment rate. With the development of intelligent inspection technology, infrared inspections based on equipment such as drones and intelligent robots have been gradually applied. However, existing infrared detection methods still lack efficient automatic analysis technology and cannot accurately distinguish abnormal temperature rises from normal temperature fluctuations, affecting the accuracy of fault identification.

[0003] In the existing technology, the thermal fault diagnosis method of power equipment is mainly based on target detection method, semantic segmentation method or instance segmentation method. Among them, target detection methods, such as Faster R-CNN and YOLO, can quickly identify the location information of the equipment, but cannot accurately extract the temperature information of the equipment; semantic segmentation methods can distinguish different equipment areas, but cannot distinguish multiple devices of the same category, and their applicability in complex scenarios is limited; instance segmentation methods, such as Mask R-CNN and SOLOv2, can achieve fine-grained equipment segmentation, but in environments such as substations, they are affected by factors such as equipment occlusion and lighting changes, and the recognition accuracy still needs to be improved; in addition, traditional temperature calculation methods, such as grayscale value fitting, find it difficult to accurately model the temperature distribution of power equipment, resulting in low reliability of fault identification.

[0004] Therefore, a method for thermal fault diagnosis of power equipment that can accurately evaluate the thermal status of power equipment and automatically diagnose faults is urgently needed. Summary of the Invention

[0005] In view of this, the present application provides a method, device, storage medium and computer equipment for diagnosing thermal faults of power equipment, the main purpose of which is to solve the technical problem of low reliability of fault identification in the existing method for diagnosing thermal faults of power equipment.

[0006] According to a first aspect of the present invention, a method for diagnosing thermal faults of electric power equipment is provided, the method comprising:

[0007] Collecting infrared image data of the substation site, constructing an infrared image dataset based on the infrared image data, and performing image enhancement processing on the infrared image dataset;

[0008] Using a preset instance segmentation model that combines a cross-stage local network and an attention mechanism to perform target recognition and segmentation on the infrared image dataset after image enhancement to determine the power equipment area;

[0009] Obtaining a grayscale value of the power equipment area, inputting the grayscale value into a deep neural network model to obtain a temperature value of the target power equipment, plotting a probability density curve of the temperature value using a kernel density estimation method, obtaining a reference temperature of the target power equipment based on the probability density curve, and calculating a hotspot temperature of the target power equipment;

[0010] A relative temperature difference is calculated according to the reference temperature and the hot spot temperature, and a fault state of the target power equipment is determined based on a preset fault judgment standard.

[0011] Optionally, performing image enhancement processing on the infrared image data set includes performing at least one operation of flipping, rotating, cropping and scaling on the infrared image data in the infrared image data set.

[0012] Optionally, the instance segmentation model includes a backbone network, a fast spatial pyramid pooling module, a neck structure and a head; the backbone network includes a convolutional layer, a convolutional block and a SA-AT module, and the backbone network is used to perform feature extraction, wherein the SA-AT module is arranged at the tail end of the backbone network, and the SA-AT module combines a cross-stage local network and an attention mechanism to enhance the feature extraction capability and output a feature map; the fast spatial pyramid pooling module performs multi-scale feature fusion based on maximum pooling and convolutional layers to enhance feature representation capabilities; the neck structure is used to optimize feature fusion; and the head is used to output target recognition and segmentation results based on the fused features.

[0013] Optionally, the process of the SA-AT module outputting a feature map includes: obtaining an original feature map; performing a global average pooling operation on the original feature map on a first path to generate a global feature vector, performing a linear transformation on the global feature vector, and generating a first channel attention weight through an activation function on the linear transformation result, performing a point-by-point multiplication operation on the first channel attention weight and the original feature map to obtain a first path result; normalizing the original feature map on a second path, performing a linear transformation on the normalized original feature map, and generating a second channel attention weight through an activation function on the linear transformation result, performing a point-by-point multiplication operation on the second channel attention weight and the original feature map to obtain a second path result; splicing and merging the first path result and the second path result to output a feature map.

[0014] Optionally, the grayscale value of the power equipment area is obtained, and the grayscale value is input into a deep neural network model to obtain the temperature value of the target power equipment, including: constructing an original data set and normalizing the data points in the original data set, wherein the data points include grayscale values and temperature values; selecting some data points in the original data set, and fitting some data points using a trained deep neural network to obtain a correspondence curve between temperature values and grayscale values, wherein the deep neural network model includes multiple hidden layers, and the hidden layer includes multiple neuron nodes, and the output of the neuron node is nonlinearly transformed through an activation function; obtaining the grayscale value of the power equipment area, and determining the temperature value of the target power equipment corresponding to the grayscale value in the correspondence curve.

[0015] Optionally, the kernel density estimation method is used to draw a probability density curve of the temperature value, and the reference temperature of the target power equipment is obtained based on the probability density curve and the hotspot temperature of the target power equipment is calculated, including: drawing a probability density curve corresponding to the temperature value of the target power equipment using the kernel density estimation method; locating the maximum temperature point in the probability density curve, and using the temperature value corresponding to the maximum temperature point as the reference temperature; constructing a target area within a preset distance range of the maximum temperature point, and calculating the average value of the temperature values corresponding to all pixel points in the target area to obtain the hotspot temperature.

[0016] Optionally, the relative temperature difference is calculated based on the reference temperature and the hot spot temperature, and the fault state of the target power equipment is determined based on a preset fault judgment standard, including: calculating the difference between the hot spot temperature and the reference temperature to obtain a first temperature difference; obtaining the ambient temperature of the environment in which the target power equipment is located, and calculating the difference between the hot spot temperature and the ambient temperature to obtain a second temperature difference; calculating the ratio of the first temperature difference to the second temperature difference to obtain a relative temperature difference; obtaining a preset fault judgment standard, comparing the hot spot temperature and the relative temperature difference with the fault judgment standard, and determining the fault state of the target power equipment based on the comparison result.

[0017] According to a second aspect of the present invention, a device for diagnosing thermal faults of electric power equipment is provided, the device comprising:

[0018] A data acquisition module is used to collect infrared image data at the substation site, construct an infrared image dataset based on the infrared image data, and perform image enhancement processing on the infrared image dataset;

[0019] A target recognition module is used to identify and segment targets in the infrared image dataset after image enhancement using a preset instance segmentation model that combines a cross-stage local network and an attention mechanism to determine the area of power equipment.

[0020] a temperature acquisition module, configured to acquire a grayscale value of the power equipment area, input the grayscale value into a deep neural network model to obtain a temperature value of a target power equipment, plot a probability density curve of the temperature value using a kernel density estimation method, acquire a reference temperature of the target power equipment based on the probability density curve, and calculate a hotspot temperature of the target power equipment;

[0021] The result output module is used to calculate the relative temperature difference according to the reference temperature and the hot spot temperature, and determine the fault state of the target power equipment based on a preset fault judgment standard.

[0022] According to a third aspect of the present invention, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned power equipment thermal fault diagnosis method is implemented.

[0023] According to a fourth aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for diagnosing thermal faults of electric power equipment when executing the program.

[0024] The present invention provides a method, device, storage medium and computer equipment for diagnosing thermal faults of power equipment. First, infrared image data of the substation site is collected to construct an infrared image dataset, and image enhancement processing is performed on the infrared image dataset to improve data quality. Secondly, an instance segmentation model improved based on the attention mechanism is used to achieve accurate segmentation of the power equipment area, effectively improving segmentation accuracy. Then, the grayscale value of the power equipment area is input into a deep neural network model to obtain the temperature value of the target power equipment. The deep neural network model is used to fit the equipment temperature value, with better fitting accuracy and stronger generalization, and the error of fitting temperature is small. Then, the kernel density estimation method is used to draw the probability density curve of the temperature value, determine the reference temperature of the target power equipment and calculate the hot spot temperature, effectively improving the acquisition accuracy of temperature parameters. Finally, the relative temperature difference is calculated, and the fault state of the target power equipment is determined based on the preset fault judgment standard, realizing fully automatic judgment of the fault state of the power equipment, and improving judgment efficiency and judgment accuracy. The above method improves the quality of raw data, accurately identifies and segments power equipment, and performs high-precision temperature prediction, ultimately completing fully automatic and reliable fault judgment for power equipment.

[0025] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0027] Figure 1 A schematic diagram showing a flow chart of a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0028] Figure 2 A schematic diagram showing a flow chart of another method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention;

[0029] Figure 3 The following is a diagram showing a structure of an example segmentation model in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention;

[0030] Figure 4 A flowchart showing a characteristic diagram output by a SA-AT module in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0031] Figure 5 A schematic diagram of training parameters of an instance segmentation model in a method for diagnosing thermal faults of power equipment provided by an embodiment of the present invention is shown;

[0032] Figure 6 A schematic diagram showing the training results of an instance segmentation model in a method for diagnosing thermal faults of power equipment provided by an embodiment of the present invention is shown;

[0033] Figure 7 A schematic diagram of quantized data of an instance segmentation model in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0034] Figure 8 A diagram comparing the training results of an instance segmentation model and other models in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0035] Figure 9 A temperature fitting result diagram of a deep neural network model in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0036] Figure 10 A comparison chart of temperature fitting results between a deep neural network model and other models in a method for diagnosing thermal faults of power equipment provided by an embodiment of the present invention is shown;

[0037] Figure 11 A diagram showing the temperature fitting effect of a deep neural network model in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0038] Figure 12A fault diagnosis criterion diagram in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0039] Figure 13 A faulty device diagram in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0040] Figure 14 A diagram showing a fault diagnosis result in a method for diagnosing thermal faults of electric power equipment provided by an embodiment of the present invention is shown;

[0041] Figure 15 A schematic structural diagram of a thermal fault diagnosis device for electric power equipment provided by an embodiment of the present invention is shown;

[0042] Figure 16 A schematic diagram of the device structure of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0043] The following describes exemplary embodiments of the present application in more detail with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0044] The present application provides a method for diagnosing thermal faults in power equipment. Figure 1 As shown, the method includes the following steps:

[0045] 101. Collect infrared image data at the substation site, construct an infrared image dataset based on the infrared image data, and perform image enhancement processing on the infrared image dataset.

[0046] Among them, infrared image data refers to the image of the surface temperature distribution of an object captured by an infrared imaging device. Unlike visible light images, infrared images show the intensity of infrared radiation emitted by the object, usually presented in grayscale or pseudo-color. Specifically in the substation environment, infrared image data can be used to detect hot spots or fault points of power equipment, that is, abnormal heating areas caused by electrical faults, wear or other problems; image enhancement processing aims to expand the number of images or improve the quality of images to enhance the generalization ability of the model.

[0047] In an embodiment of the present application, infrared imaging equipment is first used to collect infrared image data of power equipment at the substation site to reflect the temperature distribution on the surface of the equipment. Then, all the collected infrared images are organized into a data set to provide basic information for subsequent analysis and model training. Finally, a series of image enhancement operations are performed on the constructed data set to effectively expand the data set and enhance the generalization ability of the model.

[0048] 102. The preset instance segmentation model is used to perform target recognition and segmentation on the infrared image dataset after image enhancement processing to determine the power equipment area. The instance segmentation model combines the cross-stage local network and attention mechanism.

[0049] Among them, the instance segmentation model is a computer vision technology that can not only identify and classify different objects in the image, but also accurately mark the boundaries of each object to achieve segmentation. In an infrared image containing multiple power equipment, the instance segmentation model can not only identify the existence of all devices, but also create an accurate outline for each device separately; Cross Stage Partial Network (CSPNet) is an architectural design designed to improve network efficiency and accuracy while reducing the amount of computation. By directly passing part of the feature map of the base layer to the deep layer, the propagation of repeated gradient information is reduced, thereby reducing computational redundancy and enhancing the model's ability to learn features at different levels. It is particularly suitable for tasks that require efficient processing of high-resolution images; the attention mechanism allows the model to focus on the most important part when processing data. For example, in the diagnosis of thermal faults in power equipment, the attention mechanism can help the model focus more on hot spots where problems may exist, rather than evenly distributing resources across the entire image. This not only improves detection accuracy, but also enhances the model's resistance to complex background interference.

[0050] In an embodiment of the present application, a pre-trained instance segmentation model is used to analyze the enhanced infrared image dataset, which can identify and accurately divide the various power equipment areas in the image, and realize the precise positioning and contour extraction of the power equipment; the instance segmentation model adopts the CSPNet architecture to improve computing efficiency and performance, and incorporates the attention mechanism to optimize the model's attention to key features, so that the model can better cope with the complex substation environment and ensure that it can provide reliable diagnostic results.

[0051] 103. Obtain the grayscale value of the power equipment area, input the grayscale value into the deep neural network model, obtain the temperature value of the target power equipment, use the kernel density estimation method to draw the probability density curve of the temperature value, obtain the reference temperature of the target power equipment based on the probability density curve, and calculate the hot spot temperature of the target power equipment.

[0052] Among them, Kernel Density Estimation (KDE) is a non-parametric method used to estimate the probability density function of a random variable. Unlike parametric methods, KDE does not need to assume that the data obeys a specific distribution, such as a normal distribution. It estimates the overall probability density by placing a kernel function, usually a Gaussian kernel, around each data point and then superimposing all these kernel functions.

[0053] Specifically, a deep neural network model (DNN) combined with the kernel density estimation method is used to analyze the temperature values of power equipment. First, the grayscale value of the target power equipment area is extracted from the infrared image, and the grayscale value is sent as input to the trained deep neural network model, which can output the corresponding temperature value prediction. Then, based on the series of temperature values predicted by the DNN model, the kernel density estimation method is used to calculate and draw the probability density curve of these temperature values. Then, the reference temperature is determined according to the probability density curve, and the hotspot temperature is calculated using the reference temperature.

[0054] In an embodiment of the present application, based on the DNN model, the temperature value is predicted by obtaining the grayscale value, and the prediction result is processed by KDE, which not only improves the accuracy of temperature estimation but also enhances the robustness to noise. By drawing the probability density curve, it can help identify the reference temperature and calculate the hot spot temperature. Finally, fault judgment is made based on the relative temperature difference, providing an objective and quantitative standard, which helps to make more scientific and reasonable maintenance decisions. The method provided in this application is particularly suitable for complex environments such as substations. KDE does not require any specific probability distribution form to be preset and can flexibly respond to temperature distribution conditions under various actual conditions.

[0055] 104. Calculate the relative temperature difference based on the reference temperature and the hot spot temperature, and determine the fault state of the target power equipment based on a preset fault judgment standard.

[0056] Specifically, the preset fault judgment standard is a set of predefined rules or thresholds used to judge whether a device fails based on the relative temperature difference.

[0057] In this embodiment, by accurately calculating the relative temperature difference and applying preset standards, effective detection of early faults in power equipment can be achieved, and fault judgments are made based on quantitative indicators, providing a more scientific and objective basis and reducing errors caused by human judgment; after accurately identifying the problem equipment, a corresponding maintenance plan can be formulated according to the actual situation, thereby optimizing resource allocation, reducing unnecessary downtime, and reducing operating costs. Ultimately, timely discovery and handling of potential thermal faults can effectively prevent the occurrence of serious safety accidents such as fires, and ensure the safe and stable operation of substations and power grids. The method provided in this application is not only suitable for standardized ideal environments, but can also cope with complex situations in actual operations. By setting reasonable thresholds, it can flexibly adapt to fault diagnosis needs under various working conditions.

[0058] The present invention provides a method, device, storage medium and computer equipment for diagnosing thermal faults of power equipment. First, infrared image data of the substation site is collected to construct an infrared image dataset, and image enhancement processing is performed on the infrared image dataset to improve data quality. Secondly, an instance segmentation model improved based on the attention mechanism is used to achieve accurate segmentation of the power equipment area, effectively improving segmentation accuracy. Then, the grayscale value of the power equipment area is input into a deep neural network model to obtain the temperature value of the target power equipment. The deep neural network model is used to fit the equipment temperature value, with better fitting accuracy and stronger generalization, and the error of fitting temperature is small. Then, the kernel density estimation method is used to draw the probability density curve of the temperature value, determine the reference temperature of the target power equipment and calculate the hot spot temperature, effectively improving the acquisition accuracy of temperature parameters. Finally, the relative temperature difference is calculated, and the fault state of the target power equipment is determined based on the preset fault judgment standard, realizing fully automatic judgment of the fault state of the power equipment, and improving judgment efficiency and judgment accuracy. The above method improves the quality of raw data, accurately identifies and segments power equipment, and performs high-precision temperature prediction, ultimately completing fully automatic and reliable fault judgment for power equipment.

[0059] The present application embodiment provides another method for diagnosing thermal faults of power equipment, such as Figure 2 As shown, the method includes the following steps:

[0060] 201. Collect infrared image data of the substation site and construct an infrared image dataset, and perform image enhancement processing on the infrared image data in the infrared image dataset.

[0061] Specifically, at least one operation of flipping, rotating, cropping and scaling is performed on the infrared image data in the infrared image data set.

[0062] In the embodiment of the present application, an infrared image dataset is constructed using infrared images of circuit breakers taken on-site at a substation. This is primarily due to the fact that the thermal imaging characteristic of a faulty circuit breaker is that the temperature difference between the flange and the top cap is significant, allowing for clear distinction. A total of 201 infrared images were obtained from the on-site photography. Image enhancement processing methods such as flipping and rotation can effectively expand the infrared image dataset, resulting in a dataset containing 622 infrared images.

[0063] 202. The YOLOv8 instance segmentation model is used to perform target recognition and segmentation of power equipment, and the SA-AT module is introduced based on YOLOv8.

[0064] Among them, the instance segmentation model includes a backbone network, a fast spatial pyramid pooling module, a neck structure and a head; the backbone network includes convolutional layers, convolutional blocks and SA-AT modules. The backbone network is used for feature extraction. Among them, the SA-AT module is set at the tail end of the backbone network. The SA-AT module combines cross-stage local networks and attention mechanisms to enhance feature extraction capabilities and output feature maps; the fast spatial pyramid pooling module performs multi-scale feature fusion based on maximum pooling and convolutional layers to enhance feature representation capabilities; the neck structure is used to optimize feature fusion; and the head is used to output target recognition and segmentation results based on the fused features.

[0065] Specifically, the YOLOv8 instance segmentation model structure that introduces the SA-AT module is as follows: Figure 3As shown in the figure, the main part includes Backbone (backbone network), Neck (neck structure) and YOLO Head (head structure), in addition to SPPF (Spatial Pyramid Pooling with Fusion, fast spatial pyramid pooling module), where the backbone network includes multiple convolutional layers (Conv) for preliminary feature extraction and C2 blocks for multi-level feature extraction. The C2 block specifically includes multiple convolutional layers and residual connections. Each C2 block may contain multiple convolutional layers and BN (BatchNormalization, normalization module), SiLU activation function, and a SA-AT module is embedded at the tail end of the backbone network. The SA-AT module combines channel attention (Channel Attention) and spatial attention mechanism (SpatialAttention) to enhance the ability to extract key features, and the channel shuffling operation (Spatial Attention) promotes information interaction between different feature channels to avoid information loss. At the same time, the SA-AT module does not introduce additional convolution calculations, keeping the computational complexity within a reasonable range; the fast spatial pyramid pooling module uses maximum pooling (Maxpool) and convolution layers for multi-scale feature fusion, further enhancing feature representation capabilities; the PAN (Path Aggregation Network) in the neck structure further optimizes feature fusion to ensure that features at each level are fully utilized; the head structure is responsible for the final prediction output, including target classification and bounding box regression, and usually contains multiple convolutional layers and a specific output layer to generate the final detection results.

[0066] In an embodiment of the present application, the overall process of using the YOLOv8 instance segmentation model to perform target recognition and segmentation of power equipment is as follows: the input image is first subjected to feature extraction through the backbone network, and the features are multi-scale fused through the path aggregation network in the neck structure to generate a richer feature representation, and the head structure performs final target recognition and segmentation based on the fused features.

[0067] Furthermore, the process of the SA-AT module outputting a feature map includes: obtaining the original feature map; performing a global average pooling operation on the original feature map on the first path to generate a global feature vector, performing a linear transformation on the global feature vector, and generating a first channel attention weight by passing the linear transformation result through an activation function, performing a point-by-point multiplication operation on the first channel attention weight and the original feature map to obtain a first path result; normalizing the original feature map on the second path, performing a linear transformation on the normalized original feature map, and generating a second channel attention weight by passing the linear transformation result through an activation function, performing a point-by-point multiplication operation on the second channel attention weight and the original feature map to obtain a second path result; splicing and merging the first path result and the second path result to output the feature map.

[0068] In the embodiment of the present application, the specific process of the SA-AT module outputting the feature map is as follows: Figure 4 As shown in the figure, the feature map is first input and then divided into two paths. One path directly enters the normalization layer, then undergoes linear transformation, and then passes through the sigmoid activation function to obtain the channel attention weight. The other path generates a global feature vector through global average pooling, and then passes through a linear transformation and then passes through the sigmoid activation function to obtain another channel attention weight. The channel attention weights generated by the two paths are respectively multiplied point by point with the original feature map to obtain two weighted feature maps. The two weighted feature maps are merged through a splicing operation to form the final output feature map. The final output is a feature map processed by the channel attention mechanism. This feature map is selectively enhanced or suppressed on each channel to better capture key features.

[0069] Specifically, the SA-AT (Shuffle Attention-Attention Transformer) module is located at the tail end of the backbone network. By introducing the attention mechanism to weight the feature maps, it enhances the expression of key features and improves the model's ability to capture global information. It should be further explained that the SA-AT module combines the spatial attention mechanism, the channel attention mechanism, and the channel shuffling operation. The attention module can learn the importance weights of the feature maps and weight the input feature maps accordingly. A good attention module can improve the model's ability to extract key feature information. Spatial attention extracts the target position contained in each feature map, and channel attention extracts the position of the channel with richer feature information. The channel attention branch uses global pooling to reduce the dimension of the feature map and uses a linear function to implement channel attention weighting. The spatial attention branch compresses the channels of the feature map through group normalization and uses a linear function to implement spatial attention weighting. Finally, channel shuffling promotes the flow of feature information between channels. Based on the above operations, focusing on spatial and channel information is achieved. In addition, the SA-AT module does not use convolution operations, and its size is lighter than other attention mechanism modules.

[0070] Based on this, the SA-AT module generates channel attention weights through global average pooling, normalization, linear transformation and sigmoid activation function, and applies these weights to the original feature map through point-by-point product and splicing operations, thereby achieving selective enhancement of important information in the feature map, helping the model to more effectively capture and utilize key features in the image.

[0071] Further illustrate the superiority of the YOLOv8 instance segmentation model introduced by the SA-AT module provided in this application. The training parameters of the model are as follows: Figure 5 As shown in Figure 2, the stochastic gradient descent (SGD) optimization algorithm is used for training. The training and testing of the experimental data are both performed on the same deep learning server with the Ubuntu 18.04 operating system. In order to accurately evaluate the training results of the model, the precision (P), recall (R), and mean average precision (mAP) are used to measure the segmentation effect of the model. Among them, mAP uses mAP@0.5 and mAP@0.5-0.95 to reflect the performance of the model at different accuracy thresholds. The training results of the model are shown in Figure 2. Figure 6 As shown, Box and Mask represent the target box and mask output by the model respectively. Specifically, the quantized data of the model is Figure 7As shown in the figure, the model has only 3.39M parameters, 12.1GFLOPs of floating-point operations, and a detection speed of 11ms, which has the advantages of lightweight and low latency.

[0072] Furthermore, in order to prove the superiority of the improved instance segmentation algorithm, the segmentation results of the model on the power equipment dataset are compared with the original YOLOv8 model and its variants. The results are as follows: Figure 8 As shown in the table, YOLO-NAM, YOLO-TR, and YOLO-C represent the models with the addition of the normalization-based attention module (NAM), Transformer module, and convolutional block attention module (CBAM), respectively. YOLOv8seg is the original YOLOv8 model, and according to Figure 8 It can be seen from the results that the performance of the original YOLOv8 model still has room for improvement. The accuracy of the model with the NAM module added is slightly improved, and the accuracy of the model with the Transformer module added is slightly decreased, which may be related to the small number of parameters of the Transformer module itself. The model with the CBAM attention module added has poor accuracy in all aspects, which may be related to the overfitting of the model. The model of this application has the highest recognition accuracy at different accuracy thresholds, which is significantly better than other models. Compared with the original model, the improved model significantly improves the segmentation accuracy by introducing the attention mechanism module. The average accuracy of the mask and detection box is increased by 3.6% and 2.5% respectively, which proves the effectiveness of the improvement. The comparison results with other attention mechanisms show that the model improvement results provided by this application are better.

[0073] 203. Use a deep neural network model to fit the temperature value, input the grayscale value of the power equipment area, and output the temperature value of the target power equipment.

[0074] Specifically, an original data set is constructed, and the data points in the original data set are normalized, wherein the data points include grayscale values and temperature values; some data points are selected from the original data set, and the trained deep neural network is used to fit some of the data points to obtain a correspondence curve between the temperature value and the grayscale value, wherein the deep neural network model includes multiple hidden layers, the hidden layer includes multiple neuron nodes, and the output of the neuron node is nonlinearly transformed through an activation function; the grayscale value of the power equipment area is obtained, and the temperature value of the target power equipment corresponding to the grayscale value is determined in the correspondence curve.

[0075] In an embodiment of the present application, a deep neural network model (DNN) is used to fit the temperature value, an 11-layer DNN network is built, the grayscale value of the power equipment area is input, and the temperature value of the target power equipment is output. Specifically, in order to solve the problems that the traditional function fitting method cannot handle noisy data, has poor generalization ability and overfitting, a device temperature fitting method based on a deep neural network is proposed. Through large-scale data training, the deep neural network can resist overfitting to a certain extent and show good generalization ability for new samples, thereby improving the robustness of the model. At the same time, the deep neural network can more accurately represent nonlinear mapping through multi-layer abstract feature representation, improve the model's anti-interference ability to noise, and the parallel processing capability of the neural network also improves the efficiency of temperature fitting.

[0076] First, a temperature-grayscale dataset was created for fitting the deep neural network. The dataset contains 95,628 data points. In order to obtain a better fitting effect, all grayscale values and temperature values were first normalized. The temperature value normalization formula is:

[0077] T n =(TT min ) / (T max -T min )

[0078] Where: Tn is the normalized temperature value, T is the original temperature value, Tmax and Tmin are the upper and lower limits of the temperature value respectively.

[0079] The grayscale normalization formula is:

[0080] G n =G / 255

[0081] Where: Gn is the normalized grayscale value; G is the pixel grayscale value.

[0082] It should be noted that the constructed deep neural network model contains 11 hidden layers, each of which has 10 neuron nodes. The output of each layer of neuron nodes is fed into the Leaky ReLU activation function, specifically using the mean square error (MSE) loss function and the Adam optimizer.

[0083] The function curve of the device grayscale value and temperature value is called the TG curve. 256 data points were extracted from the temperature-grayscale dataset, and the TG curve was fitted using the trained DNN network. The results are as follows: Figure 9 As shown in Figure 2, after fitting the TG curve using the DNN model, the grayscale value obtained is mapped to the curve to obtain the fitting temperature of the power equipment. Figure 10The temperature fitting results of various functions on the data set are given in . Taking the average value of the fitting temperature error as the evaluation index, it can be seen that the DNN algorithm has the smallest error.

[0084] 204. Use the kernel density estimation method to analyze the probability density distribution of the target power equipment temperature, extract the reference temperature of the target power equipment based on the maximum temperature point of the probability density curve, and calculate the hot spot temperature of the target power equipment.

[0085] Among them, the kernel density estimation method is used to draw the probability density curve corresponding to the temperature value of the target power equipment; the maximum temperature point is located in the probability density curve, and the temperature value corresponding to the maximum temperature point is used as the reference temperature; the target area is constructed within the preset distance range of the maximum temperature point, and the average temperature value corresponding to all pixel points in the target area is calculated to obtain the hotspot temperature.

[0086] In this embodiment, kernel density estimation is a parameter-free probability density function estimation method. Specifically, a kernel function is placed at each sample point. When calculating the probability density of any point in the interval, the average value of all kernel functions at that point is taken as the probability density of that point. The parameter-free nature of kernel density estimation enables it to be used to estimate data of various distribution types and give a smooth curve of the probability density.

[0087] Specifically, the formula for kernel density estimation is:

[0088]

[0089] Where x is any point in the interval, n is the total number of sample points, h is the interval width, xj is the jth sample point among n, and K is the kernel function.

[0090] When the Gaussian function is selected as the kernel function, the formula is:

[0091]

[0092] The value of h when the Gaussian function is selected as the kernel function is:

[0093]

[0094] Where σ is the standard deviation of the data.

[0095] Specifically, the device temperature probability density function curve can be obtained through kernel density estimation. Figure 11This is the kernel density estimation result of the circuit breaker. By observing the shape of the probability density curve, it can be seen that the curve has two maximum points A and B, which represent that the equipment temperature is concentrated near these two temperature values. The probability density value of point A is larger, and the temperature value of point B is higher. From a physical point of view, the temperature corresponding to point A is the most likely value of the equipment temperature. Therefore, it is considered as the representative temperature of the equipment's normal temperature zone, that is, the reference temperature. In addition, the hot spot temperature of the power equipment needs to be extracted. The hot spot temperature can be simply considered as the maximum value of the equipment temperature value. When shooting, the infrared camera will generate noise in the image due to the influence of the environment, causing the grayscale value of the pixel point to offset, thereby affecting the fitting temperature value. In order to reduce the impact of single-point fitting error on the results, the calculation of the equipment hot spot temperature should consider nearby pixels, and the average temperature of a small area should be used as the hot spot temperature.

[0096] Furthermore, a temperature extraction method is proposed. Kernel density estimation is used to obtain a smooth curve of the device temperature probability density function. The temperature corresponding to the maximum value of the curve is taken as the reference temperature. The device hotspot temperature is obtained by calculating the average temperature of the area within a rectangular box with a side length of 3 pixels near the maximum temperature point.

[0097] 205. Calculate the relative temperature difference and match it with the preset fault judgment standard to determine the fault status of the target power equipment.

[0098] Among them, the fault judgment standard adopted in this application specifically uses the national power industry standard "DL / T 664-2016 Infrared Diagnosis Application Specification for Live Equipment" to set fault diagnosis rules, calculate the relative temperature difference δ, and match the industry standard to automatically determine whether the equipment has a fault and the fault level.

[0099] Specifically, for current-induced heating devices, the diagnostic criteria mainly rely on the relative temperature difference and the hotspot temperature. The calculation formula for the relative temperature difference is:

[0100]

[0101] Where: δ is the relative temperature difference, T1 is the temperature of the hot point, T2 is the temperature of the reference point, and T0 is the ambient temperature.

[0102] Further, Figure 12 A diagnostic criterion for poor contact finger crimping of circuit breaker contacts is presented. Hotspot temperature T1 and relative temperature difference δ are used as the diagnostic criteria for defects in current-induced heating equipment. These two parameters are used in defect diagnosis for different equipment, so the algorithm can be extended to other equipment such as disconnectors and current transformers.

[0103] Furthermore, we show examples of fault diagnosis. Figure 13 is the fault circuit breaker in the dataset, Figure 14The diagnosis results of the circuit breakers are given. By observing the infrared thermal image, it can be intuitively found that the tops of the circuit breakers on both sides are obviously hot. Specifically, the leftmost device is named circuit breaker 1 and the rightmost device is circuit breaker 2. FLIR Tools is used to manually select representative temperature points, and the temperature information is marked in the figure. The hotspot temperature of circuit breaker 1 is 30.7°C and the reference temperature is 7.7°C; the hotspot temperature of circuit breaker 2 is 32.2°C and the reference temperature is 7.8°C. The diagnosis results in Table 8 show that the hotspot temperature of circuit breaker 1 is 32.1°C and the reference temperature is 7.71°C; the hotspot temperature of circuit breaker 2 is 33.91°C and the reference temperature is 7.89°C. Compared with the manual temperature extraction results, it can be seen that the temperature extraction algorithm in this paper can more accurately identify the device temperature. The fault diagnosis results show that the relative temperature difference between the two circuit breakers is large, and the device status is seriously defective. Based on this, the method provided in this application can more accurately extract device temperature parameters and provide equipment fault level classification, providing a reference for maintenance work.

[0104] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides a thermal fault diagnosis device for power equipment, such as Figure 15 As shown, the device includes: a data acquisition module 301, a target recognition module 302, a temperature acquisition module 303 and a result output module 304.

[0105] The data acquisition module 301 is used to collect infrared image data of the substation site, construct an infrared image dataset based on the infrared image data, and perform image enhancement processing on the infrared image dataset;

[0106] The target recognition module 302 is used to perform target recognition and segmentation on the infrared image dataset after image enhancement using a preset instance segmentation model, wherein the instance segmentation model combines a cross-stage local network and an attention mechanism to determine the power equipment area;

[0107] The temperature acquisition module 303 is used to obtain the grayscale value of the power equipment area, input the grayscale value into the deep neural network model, obtain the temperature value of the target power equipment, use the kernel density estimation method to draw the probability density curve of the temperature value, obtain the reference temperature of the target power equipment based on the probability density curve, and calculate the hot spot temperature of the target power equipment;

[0108] The result output module 304 is used to calculate the relative temperature difference according to the reference temperature and the hot spot temperature, and determine the fault state of the target power equipment based on a preset fault judgment standard.

[0109] In a specific application scenario, the data acquisition module 301 may be specifically configured to perform at least one of flipping, rotating, cropping, and scaling on the infrared image data in the infrared image data set.

[0110] In a specific application scenario, the instance segmentation model in the target recognition module 302 includes a backbone network, a fast spatial pyramid pooling module, a neck structure and a head; the backbone network includes a convolutional layer, a convolutional block and a SA-AT module, and the backbone network is used for feature extraction, wherein the SA-AT module is set at the tail end of the backbone network, and the SA-AT module combines the cross-stage local network and the attention mechanism to enhance the feature extraction capability and output the feature map; the fast spatial pyramid pooling module performs multi-scale feature fusion based on maximum pooling and convolutional layers to enhance the feature representation capability; the neck structure is used to optimize feature fusion; and the head is used to output target recognition and segmentation results based on the fused features.

[0111] In a specific application scenario, the process of the SA-AT module in the target recognition module 302 outputting a feature map includes: obtaining the original feature map; performing a global average pooling operation on the original feature map on the first path to generate a global feature vector, performing a linear transformation on the global feature vector, and generating a first channel attention weight by an activation function on the linear transformation result, performing a point-by-point multiplication operation on the first channel attention weight and the original feature map to obtain a first path result; normalizing the original feature map on the second path, performing a linear transformation on the normalized original feature map, and generating a second channel attention weight by an activation function on the linear transformation result, performing a point-by-point multiplication operation on the second channel attention weight and the original feature map to obtain a second path result; splicing and merging the first path result and the second path result to output a feature map.

[0112] In a specific application scenario, the temperature acquisition module 303 can be used to construct an original data set and normalize the data points in the original data set, wherein the data points include grayscale values and temperature values; select some data points in the original data set, and use the trained deep neural network to fit some data points to obtain a correspondence curve between temperature values and grayscale values, wherein the deep neural network model includes multiple hidden layers, the hidden layer includes multiple neuron nodes, and the output of the neuron node is nonlinearly transformed through an activation function; obtain the grayscale value of the power equipment area, and determine the temperature value of the target power equipment corresponding to the grayscale value in the correspondence curve.

[0113] In a specific application scenario, the temperature acquisition module 303 can also be used to draw a probability density curve of the temperature value using the kernel density estimation method, obtain the reference temperature of the target power equipment based on the probability density curve and calculate the hotspot temperature of the target power equipment, including: drawing the probability density curve corresponding to the temperature value of the target power equipment using the kernel density estimation method; locating the maximum temperature point in the probability density curve, and using the temperature value corresponding to the maximum temperature point as the reference temperature; constructing a target area within a preset distance range of the maximum temperature point, and calculating the average temperature values corresponding to all pixel points in the target area to obtain the hotspot temperature.

[0114] In a specific application scenario, the result output module 304 can be used to calculate the difference between the hot spot temperature and the reference temperature to obtain a first temperature difference; obtain the ambient temperature of the environment in which the target power equipment is located, and calculate the difference between the hot spot temperature and the ambient temperature to obtain a second temperature difference; calculate the ratio of the first temperature difference to the second temperature difference to obtain a relative temperature difference; obtain a preset fault judgment standard, compare the hot spot temperature and the relative temperature difference with the fault judgment standard, and determine the fault state of the target power equipment based on the comparison result.

[0115] It should be noted that for other corresponding descriptions of the functional units involved in the power equipment thermal fault diagnosis device provided in this embodiment, please refer to Figure 1 and Figure 2 The corresponding description in will not be repeated here.

[0116] Based on the above Figure 1 The method shown, accordingly, this embodiment also provides a storage medium, which stores a computer program, and when the program is executed by a processor, it implements the above-mentioned power equipment thermal fault diagnosis method.

[0117] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product. The software product to be identified can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the thermal fault diagnosis method of power equipment in various implementation scenarios of the present application.

[0118] Based on the above Figure 1 and Figure 2 The method shown, and Figure 15 In order to achieve the above-mentioned purpose, the embodiment of the power equipment thermal fault diagnosis device shown in FIG. Figure 16As shown, this embodiment also provides a physical device for diagnosing thermal faults in power equipment. The device includes a communication bus, a processor, a memory, and a communication interface. It may also include input and output interfaces and a display device. The various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor is configured to execute the program stored in the memory and perform the method for diagnosing thermal faults in power equipment described in the above embodiment.

[0119] Optionally, the physical device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Wi-Fi interface), etc.

[0120] Those skilled in the art will understand that the structure of a power equipment thermal fault diagnosis entity device provided in this embodiment does not constitute a limitation on the entity device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0121] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the physical device hardware and the software resources to be identified, supporting the execution of the information processing program and other software and / or programs to be identified. The network communication module is used to enable communication between components within the storage medium and with other hardware and software in the physical information processing device.

[0122] Through the description of the above embodiments, it can be clearly understood by those skilled in the art that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. By applying the technical solution of the present application, infrared image data of the substation site is first collected to construct an infrared image dataset, and the infrared image dataset is subjected to image enhancement processing to improve data quality. Secondly, an instance segmentation model based on an improved attention mechanism is used to accurately segment the power equipment area, effectively improving segmentation accuracy. Then, the grayscale value of the power equipment area is input into a deep neural network model to obtain the temperature value of the target power equipment. The deep neural network model is used to fit the equipment temperature value, with better fitting accuracy and stronger generalization, and the error of fitting temperature is small. Then, the kernel density estimation method is used to draw the probability density curve of the temperature value, determine the reference temperature of the target power equipment and calculate the hot spot temperature, effectively improving the acquisition accuracy of the temperature parameter. Finally, the relative temperature difference is calculated, and the fault state of the target power equipment is determined based on the preset fault judgment standard, realizing fully automatic judgment of the fault state of the power equipment, and improving judgment efficiency and judgment accuracy. The above method improves the quality of the original data, accurately identifies and segments the power equipment, and performs high-precision temperature prediction, and finally completes fully automatic and reliable fault judgment for the power equipment.

[0123] Those skilled in the art will understand that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required to implement the present application. Those skilled in the art will understand that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0124] The serial numbers of the above application are for descriptive purposes only and do not represent the advantages or disadvantages of the implementation scenarios. The above disclosure only discloses several specific implementation scenarios of the present application, but the present application is not limited thereto. Any changes that can be conceived by those skilled in the art should fall within the scope of protection of the present application.

Claims

1. A method for diagnosing thermal faults in power equipment, characterized in that: The method comprises: Collecting infrared image data of the substation site, constructing an infrared image dataset based on the infrared image data, and performing image enhancement processing on the infrared image dataset; Using a preset instance segmentation model that combines a cross-stage local network and an attention mechanism to perform target recognition and segmentation on the infrared image dataset after image enhancement to determine the power equipment area; Obtaining a grayscale value of the power equipment area, inputting the grayscale value into a deep neural network model to obtain a temperature value of the target power equipment, plotting a probability density curve of the temperature value using a kernel density estimation method, obtaining a reference temperature of the target power equipment based on the probability density curve, and calculating a hotspot temperature of the target power equipment; A relative temperature difference is calculated according to the reference temperature and the hot spot temperature, and a fault state of the target power equipment is determined based on a preset fault judgment standard.

2. The method according to claim 1, characterized in that The performing image enhancement processing on the infrared image data set includes: At least one operation of flipping, rotating, cropping, and scaling is performed on the infrared image data in the infrared image data set.

3. The method according to claim 1, characterized in that The instance segmentation model includes a backbone network, a fast spatial pyramid pooling module, a neck structure and a head; The backbone network includes a convolutional layer, a convolutional block, and a SA-AT module. The backbone network is used for feature extraction. The SA-AT module is arranged at the tail end of the backbone network. The SA-AT module combines a cross-stage local network and an attention mechanism to enhance feature extraction capabilities and output feature maps. The fast spatial pyramid pooling module performs multi-scale feature fusion based on maximum pooling and convolutional layers to enhance feature representation capabilities; The neck structure is used to optimize feature fusion; The head is used to output target recognition and segmentation results based on the fused features.

4. The method according to claim 3, characterized in that The process of the SA-AT module outputting the feature map includes: Get the original feature map; Performing a global average pooling operation on the original feature map on the first path to generate a global feature vector, performing a linear transformation on the global feature vector, and applying an activation function to the linear transformation result to generate a first channel attention weight, performing a point-by-point product operation on the first channel attention weight and the original feature map to obtain a first path result; Normalizing the original feature map on the second path, performing a linear transformation on the normalized original feature map, applying an activation function to the linear transformation result to generate a second channel attention weight, performing a point-by-point multiplication operation on the second channel attention weight and the original feature map to obtain a second path result; The first path result and the second path result are concatenated and merged to output a feature map.

5. The method according to claim 1, wherein The step of obtaining a grayscale value of the power equipment area and inputting the grayscale value into a deep neural network model to obtain a temperature value of a target power equipment includes: Constructing an original data set and performing normalization processing on data points in the original data set, wherein the data points include grayscale values and temperature values; Selecting some data points from the original data set, fitting the some data points using a trained deep neural network to obtain a corresponding relationship curve between temperature values and grayscale values, wherein the deep neural network model includes multiple hidden layers, each of the hidden layers includes multiple neuron nodes, and the outputs of the neuron nodes are nonlinearly transformed through an activation function; A grayscale value of the power equipment area is acquired, and a temperature value of a target power equipment corresponding to the grayscale value is determined in the corresponding relationship curve.

6. The method according to claim 1, characterized in that The method of drawing a probability density curve of the temperature value by using a kernel density estimation method, obtaining a reference temperature of the target power device based on the probability density curve, and calculating a hot spot temperature of the target power device includes: Use the kernel density estimation method to draw the probability density curve corresponding to the temperature value of the target power equipment; Locating a maximum temperature point in the probability density curve, and using a temperature value corresponding to the maximum temperature point as a reference temperature; A target area is constructed within a preset distance range of the maximum temperature point, and the average temperature value corresponding to all pixel points in the target area is calculated to obtain the hotspot temperature.

7. The method according to claim 1, characterized in that The calculating the relative temperature difference according to the reference temperature and the hot spot temperature, and determining the fault state of the target power equipment based on a preset fault judgment standard, includes: Calculating the difference between the hotspot temperature and the reference temperature to obtain a first temperature difference; Acquiring the ambient temperature of the environment in which the target power equipment is located, and calculating the difference between the hotspot temperature and the ambient temperature to obtain a second temperature difference; Calculating the ratio of the first temperature difference to the second temperature difference to obtain a relative temperature difference; A preset fault judgment standard is obtained, the hot spot temperature and the relative temperature difference are compared with the fault judgment standard, and the fault state of the target power equipment is determined according to the comparison result.

8. A thermal fault diagnosis device for electric power equipment, characterized in that: The device comprises: A data acquisition module is used to collect infrared image data at the substation site, construct an infrared image dataset based on the infrared image data, and perform image enhancement processing on the infrared image dataset; A target recognition module is used to identify and segment targets in the infrared image dataset after image enhancement using a preset instance segmentation model that combines a cross-stage local network and an attention mechanism to determine the area of power equipment. a temperature acquisition module, configured to acquire a grayscale value of the power equipment area, input the grayscale value into a deep neural network model to obtain a temperature value of a target power equipment, plot a probability density curve of the temperature value using a kernel density estimation method, acquire a reference temperature of the target power equipment based on the probability density curve, and calculate a hotspot temperature of the target power equipment; The result output module is used to calculate the relative temperature difference according to the reference temperature and the hot spot temperature, and determine the fault state of the target power equipment based on a preset fault judgment standard.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images