Transformer oil leakage detection method based on two-dimensional neural network algorithm

Abstract

The invention discloses a transformer oil leakage detection method based on a two-dimensional neural network algorithm, and belongs to the technical field of power equipment monitoring and fault diagnosis, and the method comprises the steps: 1, carrying out the image collection and classification of the oil leakage condition of a transformer, and building an annotation data set; and 2, building a data transmission protocol by using an Internet of Things transmission chip, and stably transmitting cloud platform data. And step 3, according to the transformer data set label, establishing a convolution processing image model by using a two-dimensional neural network algorithm. And step 4, utilizing a multi-scale convolution module, and using convolution kernels of different sizes to extract the characteristics of the oil leakage condition of the transformer. According to the transformer oil leakage detection method provided by the invention, the oil leakage condition of the transformer can be found in time, modules and protocols of Internet of Things transmission are considered, and the oil leakage condition of the transformer is detected on the premise of ensuring stable transmission.

Description

Technical Field

[0001] This invention belongs to the field of power equipment monitoring and fault diagnosis technology, specifically relating to a transformer oil leakage detection method based on a two-dimensional neural network algorithm. Background Technology

[0002] Transformers play a crucial role in power systems, responsible for the transmission and distribution of electrical energy. However, with the increasing operating time of transformers, oil leakage has gradually become one of their common faults. Oil leakage not only affects the reliable operation of the transformer itself but also poses a serious threat to the safe and stable operation of the power system. Transformers are one of the core pieces of equipment in a power system, and their normal operation relies on a tight sealing structure to ensure the stability of the internal insulating medium. However, due to factors such as long-term operation and harsh environments, transformer oil leakage has gradually become apparent. Oil leakage is not merely a loss of oil; more importantly, it poses a potential threat to the insulation condition and overall operational safety of the transformer. The impacts of oil leakage on transformers and power systems are as follows: Leakage of internal oil in the transformer damages the sealing condition, potentially allowing external factors such as air and moisture to intrude, accelerating the aging of the internal insulation materials and reducing their insulation performance. Oil leakage lowers the oil level in the transformer's oil conservator, causing inaccurate oil level gauge readings. Maintenance personnel find it difficult to accurately monitor the oil level, affecting their judgment of the transformer's condition. When the oil level drops, live joints, switches, and other components may operate in an oil-free, insulated state, increasing the risk of breakdown, short circuits, and burnout. If oil leakage is not addressed promptly, it can reduce the safety of transformer equipment and even cause explosions, posing a significant threat to the stable operation of the power system. Monitoring and prevention of oil leakage include improving sealing technology, regular inspection and maintenance, and introducing online monitoring systems. Continuously improving transformer sealing technology and using high-efficiency sealing materials enhances the transformer's sealing performance and reduces the risk of oil leakage. Establishing a regular transformer inspection and maintenance mechanism, including checking the sealing structure and replacing aging components, can effectively reduce the incidence of oil leakage. Introducing advanced online monitoring systems can monitor transformer oil temperature, oil level, and other parameters in real time, promptly detecting anomalies and issuing early warnings, thus improving the efficiency of oil leakage detection.

[0003] Transformer oil leakage is a common fault in power systems, seriously affecting the safe and reliable operation of transformers and the entire power system. Introducing advanced online monitoring systems can effectively prevent and mitigate oil leakage, improving the stability and security of the power system. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a transformer oil leakage detection method based on a two-dimensional neural network algorithm, thus solving the problems mentioned in the background art.

[0005] The objective of this invention is achieved as follows: a transformer oil leakage detection method based on a two-dimensional neural network algorithm, comprising the following steps:

[0006] S1. Collect and classify images of transformer oil leakage to create a labeled dataset;

[0007] S2. Use IoT transmission chips to build a data transmission protocol to stably transmit cloud platform data;

[0008] S3. Based on the transformer dataset annotations, use a two-dimensional neural network algorithm to establish a convolutional image processing model;

[0009] S4. Using a multi-scale convolution module, extract features of transformer oil leakage using convolution kernels of different sizes.

[0010] Furthermore, the image collection for transformer oil leakage includes classifying the captured images of transformer conditions according to different weather and climate conditions, and building a sample comparison database:

[0011]

[0012] Among them, B i C represents the set of bounding boxes for the i-th image. ij Describe the category of the j-th annotation box. This indicates the position and size of the j-th annotation box.

[0013] Furthermore, the IoT transmission chip establishes a data transmission protocol that includes using a 4G IoT module for TCP / IP protocol data transmission, and the stability judgment formula is as follows:

[0014]

[0015] Data transmission efficiency improvement index:

[0016]

[0017] This index comprehensively evaluates and considers the impact of effective bandwidth, protocol overhead, connection reliability, and latency on data transmission efficiency.

[0018] Furthermore, the establishment of the convolutional image processing model includes using a two-dimensional neural network algorithm to detect targets based on images labeled with database samples; the convolution operation obtains a feature map by weighted summation through sliding convolution kernels on the input image.

[0019] Y(i,j)=∑ m ∑ n X(i+m,j+n)·W(m,n)+b

[0020] Where Y(i,j) is the pixel value of the output feature map, X(i+m,j+n) is the pixel value of the input image, W(m,n) is the weight of the convolution kernel, and b is the bias term;

[0021] Adding pooling operations reduces the spatial size of the feature map and lowers the computational cost. Max pooling is used for calculation, and the expression is:

[0022] Q(i,j)=max m,n X(i·s+m,j·s+n)

[0023] Where Q(i,j) is the pixel value of the feature map after pooling, X(i·s+m,j·s+n) is the pixel value within the pooling window, and s is the pooling window stride; the feature map of the target is obtained through the processing of the two-dimensional neural network, laying a good data foundation for classification.

[0024] Furthermore, the multi-scale convolution module includes convolution kernels of various sizes to effectively handle features of different scales:

[0025] Y1 = f1(W1*X + b1)

[0026] Y3 = f1(W3*X + b3)

[0027] Y5 = f1(W5*X + b5)

[0028] Where Y represents the feature maps generated by convolutional kernels of different sizes, W is the weight of each convolutional kernel, b is the corresponding bias, and * indicates the convolution operation; finally, multi-scale output is performed:

[0029] Y out =concatenate(Y1,Y3,Y5)

[0030] The feature map generated in the multi-scale convolution module is Y. out The multi-scale convolution module can extract features at different scales, which helps the model to better detect transformer oil leakage.

[0031] The beneficial effects of this invention are as follows: Considering the modules and protocols for IoT transmission, this invention detects transformer oil leakage while ensuring stable transmission. Oil leakage monitoring: Several cameras are deployed at common oil leakage points on the transformer to collect video data around the transformer, obtaining real-time oil leakage information. The acquired video information is transmitted to the system control terminal via a 4G transmission module. A two-dimensional convolutional neural network is used to process the obtained image information. The convolutional layer performs convolution operations on the input image by sliding a small matrix called a convolution kernel or filter, extracting specific feature information. In the two-dimensional convolutional layer, the convolution kernel is a two-dimensional matrix that performs element-wise multiplication and summation with the input image to obtain the output feature map. The convolutional layer can view the convolution kernel as a set of learnable parameters that automatically adjust with the input training data. Thus, during training, the convolutional layer can learn different convolution kernels to extract different features from the input image. During testing, errors caused by factors such as weather, lighting, and season are input into the system as correction parameters. After extensive data training, an algorithm for comparing and judging transformer oil leakage images is established to ultimately determine whether the transformer has experienced oil leakage. Data training of the detection system achieves a detection accuracy rate of 96.5%. By continuously monitoring and timely transmitting temperature data, the terminal temperature early warning system can quickly respond to potential problems, improving the safety and reliability of the equipment. Attached Figure Description

[0032] Figure 1 This is a flowchart of the process of the present invention. Detailed Implementation

[0033] The present invention will now be described in further detail with reference to the accompanying drawings. It should be noted that all directional terms such as up, down, front, back, left, and right appearing in the present invention are not intended to limit the present invention, but are only used to more clearly explain and interpret the present invention.

[0034] like Figure 1 As shown, this invention discloses a transformer oil leakage detection method based on a two-dimensional neural network algorithm, comprising:

[0035] Step 1: Collect and classify images of transformer oil leakage to create a labeled dataset.

[0036] Step 2: Use IoT transmission chips to build a data transmission protocol to stably transmit data from the cloud platform.

[0037] Step 3: Based on the transformer dataset annotations, use a two-dimensional neural network algorithm to build a convolutional image processing model.

[0038] Step 4: Use a multi-scale convolution module to extract features of transformer oil leakage using convolution kernels of different sizes.

[0039] A further improvement of this invention is that the transformer sample status database is generated by using a two-dimensional neural network to accurately classify and identify the transformer status, effectively preventing and mitigating oil leakage problems and improving the stability and safety of the power system.

[0040] The specific process of step one above is as follows:

[0041] Based on different weather and climate conditions, the transformer images were categorized, and a sample comparison database was built:

[0042]

[0043] Among them, B i C represents the set of bounding boxes for the i-th image. ij Describe the category of the j-th annotation box. This indicates the position and size of the j-th annotation box.

[0044] The specific process of step two is as follows:

[0045] The stability assessment formula for using a 4G IoT module for TCP / IP protocol data transmission is as follows:

[0046]

[0047] Data transmission efficiency improvement index:

[0048]

[0049] This index comprehensively evaluates and considers the impact of effective bandwidth, protocol overhead, connection reliability, and latency on data transmission efficiency.

[0050] The specific process of step three is as follows:

[0051] Based on images labeled with samples from the database, a two-dimensional neural network algorithm is used for target detection. The convolution operation involves weighted summation of the features obtained through sliding convolution kernels across the input image.

[0052] Y(i,j)=∑ m ∑ n X(i+m,j+n)·W(m,n)+b

[0053] Where Y(i,j) is the pixel value of the output feature map, X(i+m,j+n) is the pixel value of the input image, W(m,n) is the weight of the convolution kernel, and b is the bias term.

[0054] Adding pooling operations reduces the spatial size of the feature map and lowers the computational cost; max pooling is used for computation.

[0055] Q(i,j)=max m,n X(i·s+m,j·s+n)

[0056] Where Q(i,j) is the pixel value of the feature map after pooling, X(i·s+m,j·s+n) is the pixel value within the pooling window, and s is the pooling window stride.

[0057] The feature map of the target can be obtained through the processing of two-dimensional neural networks, which lays a good data foundation for classification.

[0058] The specific process of step four is as follows:

[0059] Define convolutional kernels of various sizes to effectively handle features of different scales:

[0060] Y1 = f1(W1*X + b1)

[0061] Y3 = f1(W3*X + b3)

[0062] Y5 = f1(W5*X + b5)

[0063] Where Y represents the feature maps generated by convolutional kernels of different sizes, W is the weight of each convolutional kernel, and b is the corresponding bias. * indicates the convolution operation. Finally, multi-scale output is performed:

[0064] Y out =concatenate(Y1,Y3,Y5)

[0065] Finally, the feature map generated in the multi-scale convolution module is Y. out Multi-scale convolutional modules can extract features at different scales, which helps the model to better detect transformer oil leakage.

[0066] Oil leakage monitoring: Several cameras are installed at common oil leakage points of the transformer to collect video data around the transformer, obtain real-time oil leakage information of the transformer, and transmit the obtained video information to the system control terminal through a 4G transmission module.

[0067] Wireless Transmission Module: Utilizing the ME909S-821 IoT module, this powerful module can choose between always-on or ultra-low-power operating modes. In always-on mode, the 4G IoT module maintains a continuous network connection, uploading real-time video data of the transformer's surroundings. The complete 4G IoT module connects to the sensor, operating on the TCP / IP protocol. By maintaining a continuous connection with the server, it ensures real-time data transmission even when always online. Temperature data is monitored and collected in real-time via a cloud platform. The ME909S-821 IoT module is bound to the sensor, communicating using the TCP / IP protocol. The terminal temperature warning system receives, parses, and processes the data from the cloud platform. The system judges based on preset temperature thresholds; if an abnormal temperature is detected, a temperature warning is triggered. Through continuous monitoring and timely transmission of temperature data, the cloud platform enables the terminal temperature warning system to respond quickly to potential problems, improving equipment safety and reliability.

[0068] Database Establishment: Initially, image information on transformer oil leakage sites was collected. High-resolution cameras were used to capture images of the leakage areas, ensuring clear visualization of potential anomalies. Next, the collected images were meticulously annotated, including areas of normal and abnormal transformer conditions. Image samples were categorized according to weather conditions to ensure the system can detect oil leaks under varying climatic conditions. A sample comparison database was generated for later image comparison. Later, image processing techniques were introduced to classify and determine the differences between the current operating images and normal and abnormal samples in the database.

[0069] Video processing: The video information received by the terminal is parsed into images at a certain frame rate. This patent uses a two-dimensional convolutional neural network to process the obtained image information. The convolutional layer performs convolution operations on the input image by sliding a small matrix called a convolution kernel or filter to extract specific feature information. In the two-dimensional convolutional layer, the convolution kernel is a two-dimensional matrix that performs element-wise multiplication and summation with the input image to obtain the output feature map. The convolutional layer can be viewed as a set of learnable parameters that automatically adjust with the input training data. In this way, during training, the convolutional layer can learn different convolution kernels to extract different features of the input image. During detection, the detection errors caused by factors such as weather, lighting, and season are used as correction parameters input into the system. After training with a large amount of data, a transformer oil leakage image comparison and discrimination algorithm is established to ultimately determine whether the transformer has oil leakage. By training the detection system with data, the detection accuracy can reach 96.5%. This invention proposes a method for detecting transformer oil leakage, which can detect transformer oil leakage in a timely manner. It also takes into account the modules and protocols of Internet of Things transmission, and detects transformer oil leakage while ensuring stable transmission.

[0070] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting transformer oil leakage based on a two-dimensional neural network algorithm, characterized in that, Includes the following steps: S1. Collect and classify images of transformer oil leakage to create a labeled dataset; S2. Use IoT transmission chips to build a data transmission protocol to stably transmit cloud platform data; S3. Based on the transformer dataset annotations, use a two-dimensional neural network algorithm to establish a convolutional image processing model; S4. Using a multi-scale convolution module, extract features of transformer oil leakage using convolution kernels of different sizes.

2. The transformer oil leakage detection method based on a two-dimensional neural network algorithm according to claim 1, characterized in that: The image collection for transformer oil leakage includes classifying the images of transformers based on different weather and climate conditions, and building a sample comparison database: Among them, B i C represents the set of bounding boxes for the i-th image. ij Describe the category of the j-th annotation box. This indicates the position and size of the j-th annotation box.

3. The transformer oil leakage detection method based on a two-dimensional neural network algorithm according to claim 1, characterized in that: The IoT transmission chip establishes a data transmission protocol that includes TCP / IP protocol data transmission using a 4G IoT module. The stability judgment formula is as follows: Data transmission efficiency improvement index: This index comprehensively evaluates and considers the impact of effective bandwidth, protocol overhead, connection reliability, and latency on data transmission efficiency.

4. The transformer oil leakage detection method based on a two-dimensional neural network algorithm according to claim 1, characterized in that: The establishment of the convolutional image processing model includes using a two-dimensional neural network algorithm to detect targets based on images labeled with database samples; the convolution operation obtains a feature map by weighted summation through sliding convolution kernels on the input image. Y(i,j)=∑ m ∑ n X(i+m,j+n)·W(m,n)+b Where Y(i,j) is the pixel value of the output feature map, X(i+m,j+n) is the pixel value of the input image, W(m,n) is the weight of the convolution kernel, and b is the bias term; Adding pooling operations reduces the spatial size of the feature map and lowers the computational cost. Max pooling is used for calculation, and the expression is: Q(i,j)=max m,n X(i·s+m,j·s+n) Where Q(i,j) is the pixel value of the feature map after pooling, X(i·s+m,j·s+n) is the pixel value within the pooling window, and s is the pooling window stride; the feature map of the target is obtained through the processing of the two-dimensional neural network, laying a good data foundation for classification.

5. The transformer oil leakage detection method based on a two-dimensional neural network algorithm according to claim 1, characterized in that: The multi-scale convolution module includes convolution kernels of various sizes to effectively handle features of different scales: Y1 = f1(W1*X + b1) Y3 = f1(W3*X + b3) Y5 = f1(W5*X + b5) Where Y represents the feature maps generated by convolutional kernels of different sizes, W is the weight of each convolutional kernel, b is the corresponding bias, and * indicates the convolution operation; finally, multi-scale output is performed: Y out =concatenate(Y1,Y3,Y5) The feature map generated in the multi-scale convolution module is Y. out The multi-scale convolution module can extract features at different scales, which helps the model to better detect transformer oil leakage.

Images