Automatic alarm method and system for oil leakage of transformer
Through deep learning network and computer vision technology, combined with robot process automation, automatic detection and early warning of transformer oil leakage is achieved, and the problems of false alarms and missed alarms in traditional technologies are solved, detection accuracy and response speed are improved, and new transformers are adapted to characteristics requirements.
Patent Information
- Application Number
- CN202411476809.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-01-10
AI Technical Summary
Traditional transformer oil leakage alarm technology relies on manual inspection, and there are problems of false alarms and omissions. The detection performance is not ideal in complex environments, lacks intelligence and automation, and cannot fully adapt to the characteristics and needs of new transformers, resulting in insufficient accuracy. and slow response time.
Deep learning networks are used to combine computer vision and robot process automation to collect and process videos from cameras around the transformer to detect oil leakage in real time, use data augmentation technology to train convolutional neural network models, identify oil leakage characteristics, and detect oil leakage. automatically alert is issued.
It realizes accurate identification and timely warning of transformer oil leakage, improves detection accuracy and response speed, reduces equipment damage and safety hazards, and enhances the intelligence and automation level of monitoring.
Smart Images
Figure CN119296005A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of transformer monitoring, and in particular to an automatic alarm method and system for transformer oil leakage. Background Art
[0002] Transformer oil leakage not only leads to deterioration of equipment performance, but may also cause fire or environmental pollution. Timely detection and warning of transformer oil leakage can prevent equipment damage. Through early warning, enterprises can take timely measures to reduce risks and potential losses. It can be seen that transformer oil leakage alarm is very important.
[0003] Traditional transformer oil leakage alarm technology relies on manual inspections, which is prone to false alarms and missed alarms. False alarms increase the workload of operation and maintenance personnel, while missed alarms may result in leakage not being handled in a timely manner, leading to greater failures. At the same time, the detection performance in complex environments is often not ideal, and the maintenance consumption is high. It lacks intelligence and automation, which limits its flexibility and adaptability. It cannot fully adapt to the characteristics and needs of new transformers, and has problems such as insufficient accuracy, slow response time, and poor adaptability.
[0004] Therefore, in order to solve the above problems, a transformer oil leakage automatic alarm method and system are needed, which can detect transformer oil leakage in real time, accurately identify the oil leakage situation, and issue an oil leakage warning in time. Summary of the invention
[0005] In view of this, the purpose of the present invention is to overcome the defects in the prior art and provide a transformer oil leakage automatic alarm method and system, which can detect transformer oil leakage in real time, accurately identify the oil leakage situation, and issue an oil leakage warning in time.
[0006] The transformer oil leakage automatic alarm method of the present invention comprises:
[0007] Collecting transformer oil leakage status images; the oil leakage status images include oil leakage images and non-oil leakage images;
[0008] Perform data enhancement processing on the oil spill state image to obtain a processed image;
[0009] Input the processed image into the deep learning network model to train the network model and obtain a trained model;
[0010] According to the set interval, a video stream of the transformer oil leakage status is collected in real time, and the video stream is preprocessed to obtain a processed frame image;
[0011] The processed frame image is input into the trained model and the oil leakage probability is output; if the oil leakage probability exceeds the preset threshold, oil leakage occurs and an alarm is issued.
[0012] Furthermore, the oil leak image includes an image of the oil leak dripping process.
[0013] Furthermore, the data enhancement processing includes rotation, flipping and scaling;
[0014] The rotation is used to rotate the image around the center point by a certain angle to obtain a new image;
[0015] The flipping is used to perform an axisymmetric transformation of the image along a certain axis, including horizontal flipping and vertical flipping;
[0016] The scaling is used to enlarge or reduce the image while maintaining the aspect ratio of the image.
[0017] Furthermore, the processed image is input into the deep learning network model for training the network model, which specifically includes:
[0018] Dividing the processed images into positive samples and negative samples; the positive samples are oil leakage images, and the negative samples are non-oil leakage images;
[0019] The divided images are used as training sets to train the convolutional neural network model, so that the network model can learn the oil leakage characteristics of the transformer; wherein the convolutional neural network model is set as a binary classifier; the oil leakage characteristics include the shape of the oil droplets, the dripping trajectory and the color change of the oil droplets; during the training process, the cross entropy loss function is used to optimize the model weights, and the optimization algorithm is used for back propagation.
[0020] Further, the video stream is preprocessed, specifically including:
[0021] The video stream is read frame by frame, decomposed into separate images, and the size of each image is adjusted to meet the input requirements of the model and reduce the amount of calculation. Gaussian filtering or median filtering is used to remove random noise in the image.
[0022] Furthermore, the oil leakage probability is output, including:
[0023] Using the sigmoid function of the output layer of the convolutional neural network model, the output is a probability distribution array;
[0024] The value of the element corresponding to the oil spill classification in the probability distribution array is taken as the oil spill probability.
[0025] A transformer oil leakage automatic alarm system, comprising a collection unit, a model training unit and an alarm unit;
[0026] The acquisition unit is used to acquire an oil leakage state image of the transformer; the oil leakage state image includes an oil leakage image and an oil non-leakage image; and data enhancement processing is performed on the oil leakage state image to obtain a processed image;
[0027] The model training unit is used to input the processed image into the deep learning network model to train the network model to obtain a trained model;
[0028] The alarm unit is used to collect a video stream of the transformer oil leakage status in real time according to a set interval, pre-process the video stream to obtain a processed frame image; input the processed frame image into a trained model, and output the oil leakage probability; if the oil leakage probability exceeds a preset threshold, oil leakage occurs and an alarm is issued.
[0029] Furthermore, the data enhancement processing includes rotation, flipping and scaling;
[0030] The rotation is used to rotate the image around the center point by a certain angle to obtain a new image;
[0031] The flipping is used to perform an axisymmetric transformation of the image along a certain axis, including horizontal flipping and vertical flipping;
[0032] The scaling is used to enlarge or reduce the image while maintaining the aspect ratio of the image.
[0033] Furthermore, the processed image is input into the deep learning network model for training the network model, which specifically includes:
[0034] Dividing the processed images into positive samples and negative samples; the positive samples are oil leakage images, and the negative samples are non-oil leakage images;
[0035] The divided images are used as training sets to train the convolutional neural network model, so that the network model can learn the oil leakage characteristics of the transformer; wherein the convolutional neural network model is set as a binary classifier; the oil leakage characteristics include the shape of the oil droplets, the dripping trajectory and the color change of the oil droplets; during the training process, the cross entropy loss function is used to optimize the model weights, and the optimization algorithm is used for back propagation.
[0036] Further, the video stream is preprocessed, specifically including:
[0037] The video stream is read frame by frame, decomposed into separate images, and the size of each image is adjusted to meet the input requirements of the model and reduce the amount of calculation. Gaussian filtering or median filtering is used to remove random noise in the image.
[0038] The beneficial effects of the present invention are as follows: a transformer oil leakage automatic alarm method and system disclosed in the present invention performs real-time monitoring and alarm based on computer vision and artificial intelligence, combines deep learning image classification technology, robotic process automation (RPA) and real-time video analysis, and automatically determines whether the transformer has oil leakage by collecting and processing videos from cameras around the transformer. It can automatically detect transformer oil leakage, accurately identify oil leakage, and prevent equipment damage and safety hazards caused by oil leakage. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:
[0040] Figure 1 It is a schematic flow chart of the automatic oil leakage alarm method of the present invention. DETAILED DESCRIPTION
[0041] The present invention is further described below in conjunction with the accompanying drawings, as shown in the drawings:
[0042] This embodiment discloses a transformer oil leakage automatic alarm method, comprising the following steps:
[0043] Collecting transformer oil leakage status images; the oil leakage status images include oil leakage images and non-oil leakage images;
[0044] Perform data enhancement processing on the oil spill state image to obtain a processed image;
[0045] Input the processed image into the deep learning network model to train the network model and obtain a trained model;
[0046] According to the set interval, a video stream of the transformer oil leakage status is collected in real time, and the video stream is preprocessed to obtain a processed frame image;
[0047] The processed frame image is input into the trained model and the oil leakage probability is output; if the oil leakage probability exceeds the preset threshold, oil leakage occurs and an alarm is issued.
[0048] In this embodiment, a large amount of pictures and video data of transformer oil leakage are collected to form a transformer oil leakage state image, wherein the oil leakage image includes an image of the oil leakage dripping process;
[0049] The dripping process includes the dynamic changes of oil leakage, which helps the model learn the temporal characteristics of oil leakage and improve recognition accuracy. By capturing the dripping behavior, the model can detect the early signs of oil leakage earlier, issue an alarm in advance, and prevent greater damage. In addition, the images in the dripping process vary, including different dripping speeds and dripping shapes. These data enrich the training set and enhance the generalization ability of the model. At the same time, by analyzing the dripping process, the model can more effectively distinguish between oil leakage and oil pollution or other similar phenomena in normal conditions, thereby reducing false alarms.
[0050] In this embodiment, the dataset formed by the oil spill status images is further enriched in diversity and robustness through data enhancement processing, such as rotation, flipping, scaling, etc.
[0051] The data enhancement processing includes rotation, flipping and scaling;
[0052] The rotation is used to rotate the image around the center point by a certain angle to obtain a new image; wherein the rotation angle can be 90 degrees, 180 degrees, 270 degrees or a random angle, so that the model can learn the features of the object in different directions.
[0053] The flipping is used to transform the image along a certain axis, including horizontal flipping and vertical flipping. Horizontal flipping: flipping the image left and right along the vertical center axis. This operation can simulate the mirror effect, so that the model can learn the characteristics of horizontal mirroring without deformation. Vertical flipping: flipping the image up and down along the horizontal center axis. This operation can simulate the shape of objects under different perspectives in some cases.
[0054] The scaling is used to enlarge or reduce the image while maintaining the aspect ratio of the image. This can simulate the effects of shooting at different distances and make the model more robust to changes in the size of the object.
[0055] Through the above data enhancement processing, the diversity of the data set can be increased, the generalization ability of the model can be improved, and thus its performance in practical applications can be improved.
[0056] In this embodiment, the processed image is input into the deep learning network model to train the network model, which specifically includes:
[0057] The processed images are divided into positive samples and negative samples; the positive samples are oil leakage images, and the negative samples are no oil leakage images; the purpose of dividing positive and negative samples is to allow the model to better learn and identify the characteristics of oil leakage. Positive samples provide examples of oil leakage, and negative samples allow the model to understand the appearance of the transformer in a normal state, thereby improving recognition accuracy and reducing false positives and false negatives.
[0058] The divided images are used as training sets to train the convolutional neural network model, so that the network model learns the oil leakage characteristics of the transformer; wherein the convolutional neural network model is set as a binary classifier for identifying oil leakage and non-leakage, and the model structure includes several convolutional layers, pooling layers, fully connected layers, etc.; the oil leakage characteristics include the shape of the oil droplets, the drop trajectory, and the color change of the oil droplets; during the training process, the cross entropy loss function is used to optimize the model weights, and the optimization algorithm such as Adam or SGD is used for back propagation. The final model can effectively distinguish between oil leakage and non-leakage images.
[0059] In this embodiment, by integrating with the RPA process robot, real-time video data from cameras around the transformer is collected regularly, and the RPA robot automatically obtains the video stream captured by the camera at fixed time periods. The fixed time period can be set according to actual working conditions.
[0060] By collecting video streams at intervals in real time, RPA robots can automatically process video streams within a fixed time period to ensure timely response to changes. Video streams provide continuous frame information, and the model can better capture dynamic changes and time series features. Video streams can provide more contextual information, helping the model to more accurately understand object behavior and changes, and improve the accuracy of detection and recognition.
[0061] Preprocessing the video stream specifically includes:
[0062] The video stream is read frame by frame and decomposed into individual images, usually extracted at a fixed frame rate for subsequent analysis and processing;
[0063] Adjust the size of each image to fit the model's input requirements and reduce computational effort, using interpolation techniques such as bilinear interpolation for image scaling.
[0064] Use Gaussian filtering or median filtering to remove random noise from the image to improve image quality and ensure that the model can recognize features more accurately.
[0065] In this embodiment, the processed frame image is input into the trained model, and the oil leakage probability is output, which specifically includes:
[0066] Using the sigmoid function of the output layer of the convolutional neural network model, a probability distribution array is output; each element of this array corresponds to the predicted probability of a category, and the sum of all elements is 1; if the model is used to distinguish between two different types of oil spills, the returned array contains 2 elements, each element represents the predicted probability of the corresponding category; the value of the element corresponding to the oil spill classification in the probability distribution array is taken as the oil spill probability.
[0067] If the probability of oil leakage exceeds the preset threshold, oil leakage occurs and an alarm is issued. The preset threshold can be set to 0.5; once the model detects an oil leak, the alarm mechanism will be triggered immediately to notify the relevant maintenance personnel. The alarm information will be sent through a variety of channels, such as SMS, email or APP push, to ensure that the oil leak situation is delivered to the responsible personnel as soon as possible. In addition, logs can be recorded and historical data can be stored to facilitate subsequent analysis and tracing.
[0068] The present invention not only has the function of oil leakage detection, but also can realize system self-improvement through continuous data collection and model optimization. Each oil leakage detection result is recorded and fed back to the model as new data. By continuously introducing new samples and scenarios, the model can be continuously updated to adapt to new environments and different types of transformer oil leakage patterns.
[0069] The present invention also relates to an automatic transformer oil leakage alarm system, which corresponds to the automatic transformer oil leakage alarm method of the above embodiment and can be understood as a system for implementing the above alarm method. The system includes a collection unit, a model training unit and an alarm unit;
[0070] The acquisition unit is used to acquire an oil leakage state image of the transformer; the oil leakage state image includes an oil leakage image and an oil non-leakage image; and data enhancement processing is performed on the oil leakage state image to obtain a processed image;
[0071] The model training unit is used to input the processed image into the deep learning network model to train the network model to obtain a trained model;
[0072] The alarm unit is used to collect a video stream of the transformer oil leakage status in real time according to a set interval, pre-process the video stream to obtain a processed frame image; input the processed frame image into a trained model, and output the oil leakage probability; if the oil leakage probability exceeds a preset threshold, oil leakage occurs and an alarm is issued.
[0073] The present invention provides an objective and scientific transformer oil leakage automatic alarm method and system. By collecting and enhancing oil leakage status image data and using a deep learning network model for efficient training, accurate identification of transformer oil leakage behavior is achieved. The method is able to process video streams in real time, quickly output oil leakage probability, and automatically issue an alarm when oil leakage is detected, effectively improving the transformer's fault response speed and intelligent monitoring level.
[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.
Claims
1. A transformer oil leakage automatic alarm method, characterized in that: include: Collecting transformer oil leakage status images; the oil leakage status images include oil leakage images and non-oil leakage images; Perform data enhancement processing on the oil spill state image to obtain a processed image; Input the processed image into the deep learning network model to train the network model and obtain a trained model; According to the set interval, a video stream of the transformer oil leakage status is collected in real time, and the video stream is preprocessed to obtain a processed frame image; Input the processed frame image into the trained model and output the oil leakage probability; If the oil leakage probability exceeds the preset threshold, oil leakage occurs and an alarm is issued.
2. The transformer oil leakage automatic alarm method according to claim 1 is characterized in that: The oil leakage image includes an image of the oil leakage droplet falling process.
3. The transformer oil leakage automatic alarm method according to claim 1 is characterized in that: The data enhancement processing includes rotation, flipping and scaling; The rotation is used to rotate the image around the center point by a certain angle to obtain a new image; The flipping is used to perform an axisymmetric transformation of the image along a certain axis, including horizontal flipping and vertical flipping; The scaling is used to enlarge or reduce the image while maintaining the aspect ratio of the image.
4. The transformer oil leakage automatic alarm method according to claim 1 is characterized in that: The processed image is input into the deep learning network model for network model training, including: Dividing the processed images into positive samples and negative samples; the positive samples are oil leakage images, and the negative samples are non-oil leakage images; The divided images are used as training sets to train the convolutional neural network model, so that the network model can learn the oil leakage characteristics of the transformer; wherein the convolutional neural network model is set as a binary classifier; the oil leakage characteristics include the shape of the oil droplets, the dripping trajectory and the color change of the oil droplets; during the training process, the cross entropy loss function is used to optimize the model weights, and the optimization algorithm is used for back propagation.
5. The transformer oil leakage automatic alarm method according to claim 1 is characterized in that: Preprocessing the video stream specifically includes: The video stream is read frame by frame, decomposed into separate images, and the size of each image is adjusted to meet the input requirements of the model and reduce the amount of calculation. Gaussian filtering or median filtering is used to remove random noise in the image.
6. The transformer oil leakage automatic alarm method according to claim 4 is characterized in that: Output oil leakage probability, including: Using the sigmoid function of the output layer of the convolutional neural network model, the output is a probability distribution array; The value of the element corresponding to the oil spill classification in the probability distribution array is taken as the oil spill probability.
7. A transformer oil leakage automatic alarm system, characterized in that: It includes a collection unit, a model training unit and an alarm unit; The acquisition unit is used to acquire an oil leakage state image of the transformer; the oil leakage state image includes an oil leakage image and an oil non-leakage image; Perform data enhancement processing on the oil spill state image to obtain a processed image; The model training unit is used to input the processed image into the deep learning network model to train the network model to obtain a trained model; The alarm unit is used to collect a video stream of the transformer oil leakage status in real time according to a set interval, pre-process the video stream to obtain a processed frame image; input the processed frame image into a trained model, and output the oil leakage probability; If the oil leakage probability exceeds the preset threshold, oil leakage occurs and an alarm is issued.
8. The transformer oil leakage automatic alarm system according to claim 7 is characterized in that: The data enhancement processing includes rotation, flipping and scaling; The rotation is used to rotate the image around the center point by a certain angle to obtain a new image; The flipping is used to perform an axisymmetric transformation of the image along a certain axis, including horizontal flipping and vertical flipping; The scaling is used to enlarge or reduce the image while maintaining the aspect ratio of the image.
9. The transformer oil leakage automatic alarm system according to claim 7, characterized in that: The processed image is input into the deep learning network model for network model training, including: Dividing the processed images into positive samples and negative samples; the positive samples are oil leakage images, and the negative samples are non-oil leakage images; The divided images are used as training sets to train the convolutional neural network model, so that the network model can learn the oil leakage characteristics of the transformer; wherein the convolutional neural network model is set as a binary classifier; the oil leakage characteristics include the shape of the oil droplets, the dripping trajectory and the color change of the oil droplets; during the training process, the cross entropy loss function is used to optimize the model weights, and the optimization algorithm is used for back propagation.
10. The transformer oil leakage automatic alarm system according to claim 7, characterized in that: Preprocessing the video stream specifically includes: The video stream is read frame by frame, decomposed into separate images, and the size of each image is adjusted to meet the input requirements of the model and reduce the amount of calculation. Gaussian filtering or median filtering is used to remove random noise in the image.