A method for abnormal monitoring of electrical equipment in a substation

Abstract

The present invention discloses a method and device for abnormal monitoring of substation electrical equipment. The method includes the following steps: Step 100, collect infrared thermal imaging images and visible light images of substation electrical equipment; Step 200, use the trained improved SSD model to perform thermal fault detection on the above thermal infrared images. If the detection result shows a thermal fault, send the fault information to the client. If the detection result is normal, execute Step 300; Step 300, input the above visible light images into the trained improved Yolov5s model for smoke detection and recognition. If the detection result shows the generation of smoke, send the fault information to the client. The method and device for abnormal monitoring of substation electrical equipment of the present invention achieve the technical effects of high efficiency, high safety, more accurate and more timely monitoring of abnormal faults of substation electrical equipment through the above step-by-step fault monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of substation monitoring, and particularly to a method and device for abnormal monitoring of substations. Background Art

[0002] Substations are an important part of the power grid system. The safe and stable operation of electrical equipment such as transformers, instrument transformers, and switchgear in substations is an important factor to ensure the safety and reliability of power supply. In substations, fire accidents caused by abnormal high temperatures occur frequently. Therefore, timely detection of abnormal situations in substations is of great significance for taking early measures to avoid serious consequences and ensuring the safe and reliable operation of the power grid.

[0003] Currently, abnormal situations in substations are mainly detected in a timely manner by inspection personnel or inspection robots. However, this method has problems such as poor safety, low efficiency, and insufficient timeliness and accuracy.

[0004] In view of this, it is necessary to provide a method and device for abnormal monitoring of substation electrical equipment to solve the above defects. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the defects of poor safety, low efficiency, and insufficient timeliness and accuracy in the abnormal monitoring of substation electrical equipment in the prior art, so as to provide a method and device for abnormal monitoring of substation electrical equipment.

[0006] To achieve the above object, the first aspect embodiment of the present application proposes a method for abnormal monitoring of substation electrical equipment, including the following steps:

[0007] Step 100, collecting infrared thermal imaging images and visible light images of substation electrical equipment;

[0008] Step 200, using a trained improved SSD model to perform thermal fault detection on the above thermal infrared images. If the detection result shows a thermal fault, the fault information is sent to the client. If the detection result is normal, step 300 is executed;

[0009] Step 300, inputting the above visible light images into a trained improved Yolov5s model for smoke detection and recognition. If the detection result shows the generation of smoke, the fault information is sent to the client;

[0010] According to an embodiment of the present application, the use of a trained improved SSD model to perform thermal fault detection on the above thermal infrared images further includes the following steps:

[0011] Step 210, improve the SSD model: The input image will generate 7 convolutional feature layers. An attention mechanism is introduced between the second convolutional feature layer and the third convolutional feature layer, and the new feature map obtained is used as the output of the second convolutional feature layer and is simultaneously input into the third convolutional feature layer; at the same time, an SPP network is introduced between the third convolutional feature layer and the fourth convolutional feature layer, and the output of the SPP network is used as the input of the fourth convolutional feature layer; the attention mechanism includes channel attention and spatial attention. Regarding channel attention, first, average pooling operation and max pooling operation are performed on the input feature map to generate channel weight matrices of size 1*1*C respectively, then they are learned through the same MLP to obtain the channel attention weights, and then they are normalized through the Sigmoid function; the weights are then added to the original feature map through multiplication to obtain the feature result with the channel attention mechanism added. In spatial attention, first, average pooling operation and max pooling operation are performed on the input feature, then 7*7 convolution and Relu activation function are used to reduce the dimension of the feature map, and then it is restored to the original dimension after another convolution, and finally, it is standardized through the Sigmoid activation function to obtain the spatial attention weights, and the spatial attention weights are added to the feature result of the channel attention mechanism through multiplication to obtain the feature result with the spatial attention mechanism added; improve the localization loss in the loss function of the SSD model, and use the following localization loss:

[0012]

[0013] In the formula, π 2 (b, b gt ) represents the square of the distance between the center points of the predicted box and the ground truth box, D 2 represents the square of the length of the diagonal of the minimum bounding rectangle of the predicted box and the ground truth box, π 2 (w, w gt ) and π 2 (h, h gt ) represent the square of the width difference and the square of the height difference between the predicted box and the ground truth box respectively; D w 2 and represent the squares of the width and height of the bounding rectangle respectively, and IOU is an index commonly used in object detection tasks to characterize the localization accuracy;

[0014] Step 220, use the training sample set to train the improved SSD model to obtain the trained improved SSD model;

[0015] Step 230, use the trained improved SSD model to perform thermal fault detection on the above thermal infrared image;

[0016] According to an embodiment of the present application, inputting the above visible light image into the trained improved Yolov5s model for smoke detection and recognition includes the following steps:

[0017] Step 310, improve the Yolov5s model: In the Yolov5s model, retain the ordinary convolutions in the backbone feature extraction network, and replace the ordinary feature layers in the neck network with MSConv convolutions. The MSConv convolution combines ordinary convolution, depthwise separable convolution, and shuffle operation. The specific implementation method is as follows: First, generate high-dimensional features through ordinary convolution, then use depthwise separable convolution to transform the high-dimensional features, then splice the two features obtained from the above two steps, and through the shuffle operation, shuffle the feature maps of different groups together to enhance the information interaction between different groups; Add a large-scale detection layer on the basis of multi-scale detection of Yolov5s, so as to realize the recognition of large, medium, small, and smaller targets in the picture using 4 different sizes of feature maps.

[0018] Step 320, use the training sample set to train the improved Yolov5s model to obtain the trained improved Yolov5s model;

[0019] Step 330, input the above visible light image into the trained improved Yolov5s model for smoke detection and recognition.

[0020] According to a substation electrical equipment abnormal monitoring method of the present application, it comprehensively analyzes the fault problems of electrical equipment by using the trained improved SSD model to detect thermal faults in thermal infrared images and by using the trained improved Yolov5s model to detect and recognize smoke, thereby overcoming the defects of poor safety, low efficiency, and insufficient timeliness and accuracy in the abnormal monitoring of substation electrical equipment in the prior art.

[0021] According to an embodiment of the present application, after the detection result is a thermal fault or the detection result is the generation of smoke, it further includes: controlling the automatic circuit breaker to cut off the power supply electrically connected to the electrical equipment.

[0022] According to an embodiment of the present application, after the detection result is the generation of smoke, it further includes: controlling the fire extinguisher nozzle to spray fire extinguishing agent towards the electrical equipment, and sending an alarm signal or an alarm bell.

[0023] According to an embodiment of the present application, the fire extinguisher is a dry powder fire extinguisher. Among them, each electrical equipment is correspondingly provided with a fire extinguisher.

[0024] According to an embodiment of the present application, sending the fault information to the client includes: the client receiving the fault information.

[0025] To achieve the above object, a power grid fault detection device proposed in the second aspect embodiment of the present application includes: a memory, and

[0026] an acquisition module for collecting infrared thermal imaging images and visible light images of substation electrical equipment;

[0027] a first processing module that uses a trained improved SSD model to perform thermal fault detection on the above thermal infrared image. If the detection result shows a thermal fault, the fault information is sent to the client. If the detection result is normal, step 300 is executed;

[0028] a second processing module that inputs the above visible light image into a trained improved Yolov5s model for smoke detection and recognition. If the detection result shows the generation of smoke, the fault information is sent to the client.

[0029] The above abnormal monitoring method and device for substation electrical equipment can perform thermal fault detection on the above thermal infrared image by using a trained improved SSD model, determine whether to input the visible light image into a trained improved Yolov5s model for smoke detection and recognition according to the monitoring result, and send the fault information to the client according to the monitoring result. Specifically, when improving the SSD model, first, an attention model is introduced to obtain the attention of the feature map in the channel and space through channel attention and spatial attention, thereby enhancing the spatial and semantic features, obtaining higher attention to the temperature features in the infrared image, and reducing the influence of the image background on the image recognition process; second, the loss function is improved. By introducing factors such as the diagonal length, width, and height of the circumscribed rectangle, the difference between the predicted value and the true value during the training process can be better measured, which can improve the convergence speed and positioning accuracy of the model; third, the ordinary convolution in the backbone feature extraction network is retained, and the ordinary feature layers in the neck network are replaced with MSConv convolution, where the MSConv convolution combines ordinary convolution, depthwise separable convolution, and shuffle operations to improve the accuracy. Finally, by gradually monitoring the abnormal conditions of substation electrical equipment, the technical effects of high safety, high efficiency, timeliness, and accuracy in abnormal monitoring of substation electrical equipment are achieved. Description of the Drawings

[0030] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0031] Figure 1Flowchart of the abnormal monitoring method for substation electrical equipment in the first embodiment of the present invention. Detailed implementation manners

[0032] In order to more clearly understand the above objects, features and advantages of the present application, the present application will be further described in detail below with reference to the drawings and specific implementation manners. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other.

[0033] In the following description, many specific details are set forth in order to fully understand the present application. However, the present application may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0034] Refer to Figure 1 , this embodiment provides an abnormal monitoring method for substation electrical equipment, including the following steps:

[0035] Step 100, collect infrared thermal imaging images and visible light images of substation electrical equipment;

[0036] Step 200, use the trained improved SSD model to perform thermal fault detection on the above thermal infrared images. If the detection result shows a thermal fault, send the fault information to the client. If the detection result is normal, execute step 300;

[0037] Step 300, input the above visible light images into the trained improved Yolov5s model for smoke detection and recognition. If the detection result shows smoke generation, send the fault information to the client.

[0038] Specifically, the electrical equipment in the substation includes transformers, instrument transformers, switchgear, lightning arresters, etc. Each electrical equipment is provided with a corresponding infrared sensor capable of collecting its infrared thermal imaging image, and a camera capable of collecting visible light images inside the substation. The processor is connected to the infrared sensor and the camera, and it can obtain the infrared thermal imaging image and visible light image of the substation electrical equipment. Use the trained improved SSD model to perform thermal fault detection on the above thermal infrared images. If the detection result shows a thermal fault, give an alarm prompt. If the detection result is normal, input the above visible light images into the trained improved Yolov5s model for smoke detection and recognition. If the detection result shows smoke generation, give an alarm prompt. Otherwise, the electrical equipment is normal. Through the above technical solution, when a fault occurs in the substation electrical equipment, the user can timely and accurately discover the internal fault location of the electrical equipment when the electrical equipment is closed, so as to give an early warning and perform maintenance in time, improving the safety of inspecting high-voltage electrical equipment.

[0039] According to an embodiment of the present application, the use of the trained improved SSD model to perform thermal fault detection on the above thermal infrared image further includes the following steps:

[0040] Step 210, improve the SSD model: The input image will generate 7 convolutional feature layers. An attention mechanism is introduced between the second convolutional feature layer and the third convolutional feature layer, and the obtained new feature map is used as the output of the second convolutional feature layer and input into the third convolutional feature layer at the same time; at the same time, an SPP network is introduced between the third convolutional feature layer and the fourth convolutional feature layer, and the output of the SPP network is used as the input of the fourth convolutional feature layer; the attention mechanism includes channel attention and spatial attention. Regarding channel attention, first perform average pooling operation and max pooling operation on the input feature map, respectively generating channel weight matrices of size 1*1*C, then learn through the same MLP to obtain the channel attention weights, and then perform normalization operation through the Sigmoid function; the weights are then added to the original feature map through multiplication to obtain the feature result with the channel attention mechanism added. In spatial attention, first, perform average pooling operation and max pooling operation on the input features, then reduce the dimension of the feature map through a 7*7 convolution and the Relu activation function, restore to the original dimension after another convolution, and finally obtain the spatial attention weights after normalization through the Sigmoid activation function. The spatial attention weights are added to the feature result of the channel attention mechanism through multiplication to obtain the feature result with the spatial attention mechanism added; improve the localization loss in the loss function of the SSD model, and use the following localization loss:

[0041]

[0042] In the formula, π 2 (b, b gt ) represents the square of the distance between the center points of the predicted box and the ground truth box, D 2 represents the square of the diagonal length of the minimum bounding rectangle of the predicted box and the ground truth box, π 2 (w, w gt ) and π 2 (h, h gt ) respectively represent the square of the width difference and the square of the height difference between the predicted box and the ground truth box; D w 2 and respectively represent the squares of the width and height of the bounding rectangle, and IOU is a commonly used indicator to characterize the localization accuracy in object detection tasks;

[0043] Step 220, use the training sample set to train the improved SSD model to obtain the trained improved SSD model;

[0044] Step 230: Use the trained improved SSD model to perform thermal fault detection on the above thermal infrared image.

[0045] Specifically, when improving the SSD model, first introduce an attention model to obtain the attention of the feature map in the channel and space through channel attention and spatial attention, so as to enhance the spatial and semantic features, obtain higher attention to the temperature features in the infrared image, and reduce the influence of the image background on the image recognition process. Secondly, improve the loss function. By introducing factors such as the diagonal length, width, and height of the circumscribed rectangle, it is possible to better measure the difference between the predicted value and the true value during the training process, and improve the convergence speed and positioning accuracy of the model.

[0046] According to an embodiment of the present application, inputting the above visible light image into the trained improved Yolov5s model for smoke detection and recognition includes the following steps:

[0047] Step 310: Improve the Yolov5s model. In the Yolov5s model, retain the ordinary convolution in the backbone feature extraction network, and replace the ordinary feature layers in the neck network with MSConv convolution. The MSConv convolution combines ordinary convolution, depthwise separable convolution, and shuffle operation. The specific implementation method is as follows: First, generate high-dimensional features through ordinary convolution, then use depthwise separable convolution to transform the high-dimensional features, then splice the two features obtained in the above two steps, and through the shuffle operation, shuffle the feature maps of different groups together to enhance the information interaction between different groups. Add a large-scale detection layer on the basis of the multi-scale detection of Yolov5s to realize the recognition of large, medium, small, and smaller targets in the picture using 4 different sizes of feature maps.

[0048] Step 320: Use the training sample set to train the improved Yolov5s model to obtain the trained improved Yolov5s model.

[0049] Step 330: Input the above visible light image into the trained improved Yolov5s model for smoke detection and recognition.

[0050] According to an embodiment of the present application, after the detection result is a thermal fault or the detection result is the generation of smoke, it further includes: controlling the automatic circuit breaker to cut off the power supply electrically connected to the electrical equipment.

[0051] Specifically, the substation is provided with an automatic circuit breaker, and the automatic circuit breaker is connected to the processor. When it is judged that there is a thermal fault or the detection result is the generation of smoke, the automatic circuit breaker cuts off the power supply connected to the electrical equipment.

[0052] According to an embodiment of the present application, after the detection result is the generation of smoke, it further includes: controlling the fire extinguisher nozzle to spray the fire extinguishing agent towards the electrical equipment, and sending an alarm signal or an alarm bell.

[0053] Specifically, when the result is the generation of smoke, it indicates that a fire has occurred in the electrical equipment due to reasons such as short - circuit of the circuit, electrical aging, or high ambient temperature. If only the management personnel are notified, there is a risk that the management personnel may delay the best time for handling due to being too far away. At this time, by setting fire extinguishers in the substation, under the control of the processor, that is, after the detection result is the generation of smoke, the fire extinguishers are automatically controlled to open in a timely manner to avoid the deterioration of the fire situation of the electrical equipment.

[0054] According to an embodiment of the present application, the fire extinguisher is a dry - powder fire extinguisher. Among them, a fire extinguisher is correspondingly provided for each electrical equipment.

[0055] According to an embodiment of the present application, sending the fault information to the client includes: the client receiving the fault information.

[0056] Specifically, the processor is also connected to a remote client through a wireless transceiver unit, and can send the fault information to the management personnel or the control center at the remote end in a timely manner, so as to facilitate the management personnel or the control center to handle it in a timely manner, avoid the expansion of the dangerous situation, and further control and reduce the losses caused by the fault, and at the same time reduce the work risks of the staff.

[0057] Corresponding to the substation electrical equipment abnormal monitoring methods provided by the above - mentioned several embodiments, an embodiment of the present application also provides a substation electrical equipment abnormal monitoring device. Since the substation electrical equipment abnormal monitoring device provided by the embodiment of the present application corresponds to the substation electrical equipment abnormal monitoring methods provided by the above - mentioned several embodiments, the implementation manners of the foregoing substation electrical equipment abnormal monitoring methods are also applicable to the substation electrical equipment abnormal monitoring device provided by this embodiment, and will not be described in detail in this embodiment.

[0058] A power grid fault detection device provided by an embodiment of the present invention includes: a memory, and

[0059] An acquisition module, configured to collect infrared thermal imaging images and visible - light images of substation electrical equipment;

[0060] A first processing module, using a trained improved SSD model to perform thermal fault detection on the above - mentioned thermal infrared images. If the detection result is the occurrence of a thermal fault, the fault information is sent to the client. If the detection result is normal, step 300 is executed;

[0061] The second processing module inputs the above visible light image into the trained improved Yolov5s model for smoke detection and recognition. If the detection result is that smoke is generated, the fault information will be sent to the client.

[0062] Specifically, the acquisition module mainly includes an infrared sensor for collecting its infrared thermal imaging image, and a camera capable of collecting visible light images inside the substation. The processor includes a first processing module and a second processing module.

[0063] It should be understood that various forms of the processes shown above can be used, reordering, adding, or deleting steps. For example, the steps described in the present invention can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and no limitations are imposed herein.

[0064] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for monitoring abnormality of electrical equipment in a substation, comprising the following steps: Step 100, collecting infrared thermal imaging images and visible light images of electrical equipment in the substation; Step 200, using the trained improved SSD model to perform thermal fault detection on the infrared thermal imaging image. If the detection result is a thermal fault, the fault information is sent to the client. If the detection result is normal, step 300 is executed. Step 300, input the above visible light image into the trained improved Yolov5s model for smoke detection and recognition. If the detection result is that smoke is generated, the fault information is sent to the client; The method of using the trained improved SSD model to perform thermal fault detection on the infrared thermal imaging image also includes the following steps: Step 210, improving the SSD model: the input image will generate 7 convolutional feature layers, and an attention mechanism is introduced between the second convolutional feature layer and the third convolutional feature layer, and the obtained new feature map is used as the output of the second convolutional feature layer and input into the third convolutional feature layer at the same time; at the same time, an SPP network is introduced between the third convolutional feature layer and the fourth convolutional feature layer, and the output of the SPP network is used as the input of the fourth convolutional feature layer; the attention mechanism includes channel attention and spatial attention. Regarding channel attention, the input feature map is firstly subjected to average pooling and maximum pooling operations to generate channel weight matrices of size 1*1*C respectively, and then the same MLP is used for learning to obtain the attention weight of the channel, and then the channel is normalized by the Sigmoid function; The weights are then multiplied and added to the original feature map to obtain the feature result with the channel attention mechanism added. In spatial attention, first, the input features are average pooled and max pooled, then 7*7 convolution and Relu activation function are used to reduce the dimension of the feature map, and then convolution is performed again to restore the original dimension. Finally, the spatial attention weight is obtained after normalization by Sigmoid activation function. The spatial attention weight is added to the feature result of the channel attention mechanism by multiplication, thereby obtaining the feature result with the spatial attention mechanism added. The positioning loss in the loss function of the SSD model is improved, and the following positioning loss is used: In the formula, π 2 (b,b gt ) represents the square of the distance between the center point of the predicted box and the real box, D 2 Represents the square of the diagonal length of the minimum enclosing rectangle of the predicted box and the real box, π 2 (w,w gt ) and π 2 (h,h gt ) represent the square of the difference in width and height between the predicted box and the true box respectively; D w 2 and Respectively represent the square of the width and height of the circumscribed rectangular box. IOU is a commonly used indicator for representing positioning accuracy in target detection tasks. Step 220, training the improved SSD model using the training sample set to obtain a trained improved SSD model; Step 230: Use the trained improved SSD model to perform thermal fault detection on the infrared thermal imaging image.

2. The method for monitoring abnormality of electrical equipment in a substation according to claim 1, characterized in that: The step of inputting the above visible light image into the trained improved Yolov5s model for smoke detection and recognition includes the following steps: Step 310, improving the Yolov5s model: in the Yolov5s model, retain the ordinary convolution in the backbone feature extraction network, and replace the ordinary feature layer in the neck network with MSConv convolution, wherein the MSConv convolution combines ordinary convolution, depthwise separable convolution and shuffle operation, and is specifically implemented as follows: first, generate high-dimensional features through ordinary convolution, then use depthwise separable convolution to transform the high-dimensional features, then concatenate the two features obtained in the above two steps, and shuffle the feature maps of different groups together through shuffle operation to enhance information interaction between different groups; Add a large-scale detection layer based on Yolov5s multi-scale detection, so as to realize the recognition of large, medium, small and relatively small objects in the image using 4 different size feature maps; Step 320, training the improved Yolov5s model using the training sample set to obtain a trained improved Yolov5s model; Step 330: Input the visible light image into the trained improved Yolov5s model for smoke detection and recognition.

3. The method for monitoring abnormality of electrical equipment in a substation according to claim 1, characterized in that: After the detection result is a thermal fault or the detection result is smoke generation, the method further includes: controlling an automatic circuit breaker to cut off a power supply electrically connected to the electrical device.

4. The method for monitoring abnormality of electrical equipment in a substation according to claim 3, characterized in that: After the detection result is that smoke is generated, the method further includes: controlling the fire extinguisher nozzle to spray fire extinguishing agent toward the electrical equipment and sending out an alarm signal or an alarm ring.

5. The method for monitoring abnormality of electrical equipment in a substation according to claim 4, characterized in that: The fire extinguisher is a dry powder fire extinguisher, wherein each electrical equipment is provided with a corresponding fire extinguisher.

6. The method for monitoring abnormality of electrical equipment in a substation according to claim 5, characterized in that: The sending of the fault information to the client includes: the client receiving the fault information.

Images