Defect diagnosis method based on power equipment infrared image temperature data distribution statistics

Abstract

The invention discloses a defect diagnosis method based on power equipment infrared image temperature data distribution statistics, and the method comprises the following steps: S1, dividing each power equipment infrared image collected historically into a defect image and a non-defect image, performing data processing on one or more component frame coordinates in the defect image and the non-defect image and corresponding temperature matrix data to generate a data set consisting of a defect statistical chart and a non-defect statistical chart; s2, training the YoV4 classification model through the data in the data set to obtain a statistical graph classification detection model; and S3, intercepting temperature matrix data of each component frame corresponding to the infrared image of the to-be-detected power equipment by using coordinates of each component frame of known similar power equipment to obtain a to-be-detected temperature data distribution statistical graph, and performing classification detection through the statistical graph classification detection model to obtain a classification detection result. The method has the characteristic of being capable of accurately identifying the defects of the power equipment.

Description

Technical Field

[0001] The invention relates to a method for judging defects of electric power equipment, in particular to a method for diagnosing defects based on distribution statistics of temperature data of infrared images of electric power equipment. Background Art

[0002] At present, among the fault diagnosis rules of various power equipment, infrared detection is a common means of detecting substation equipment. Defect diagnosis of infrared images is an important means of real-time control of the operating status of substation equipment. Defect diagnosis of infrared images mainly judges whether there is a fault in the power equipment through the temperature of each component of the power equipment, so it is particularly important to accurately judge whether the selected equipment components have a fault. However, since infrared images contain a variety of pseudo-colors, it increases the difficulty of manual identification, resulting in inaccurate diagnosis of power equipment. Therefore, it is urgent to develop a method that can accurately identify whether there are defects in infrared images of power equipment to meet usage needs. Summary of the invention

[0003] The purpose of the present invention is to provide a defect diagnosis method based on the distribution statistics of temperature data of infrared images of power equipment. The present invention has the characteristics of being able to accurately identify defects of power equipment.

[0004] The technical solution of the present invention is a defect diagnosis method based on the distribution statistics of infrared image temperature data of power equipment, comprising the following steps:

[0005] S1. Divide the infrared images of various power equipment collected historically into defect images and non-defect images, and process the coordinates of one or more component frames in the defect images and non-defect images and their corresponding temperature matrix data respectively to generate corresponding defect statistical graphs and non-defect statistical graphs, and form a data set with the obtained defect statistical graphs and non-defect statistical graphs;

[0006] The method of generating a statistical graph by data processing is as follows: marking each component of the power equipment, obtaining a component frame of the power equipment, finding the temperature matrix data of the component frame corresponding to the infrared image of the power equipment, flattening the temperature matrix data corresponding to each component frame in the infrared image of the power equipment, and concatenating them to obtain a one-dimensional array array_concatenate; after statistically processing the one-dimensional array array_concatenate, a statistical graph of temperature data is generated with each set temperature value within the temperature range of the power equipment as the horizontal coordinate and the number of times each set temperature value appears in all component frames as the vertical coordinate;

[0007] S2. Train the YoloV4 classification model using the data in the data set to obtain a statistical graph classification detection model;

[0008] S3. Use the known coordinates of each component frame of the same type of electrical equipment to intercept the temperature matrix data of each component frame corresponding to the infrared image of the electrical equipment to be detected, and use the data processing method in step S1 to process the data to obtain a distribution statistical graph of the temperature data to be detected. Use the statistical graph classification detection model to classify and detect the distribution statistical graph of the temperature data to be detected to obtain a classification detection result.

[0009] In the aforementioned defect diagnosis method based on the distribution statistics of temperature data of infrared images of power equipment, the data set in step S1 is randomly divided into a training set, a validation set and a test set in a ratio of 7:1:2 for training the YoloV4 classification model in step S2.

[0010] In the aforementioned defect diagnosis method based on the distribution statistics of temperature data of infrared images of power equipment, the specific process of YoloV4 classification model training in step S2 is as follows: the YoloV4 classification model uses mobilenet0.25 as the backbone feature extraction network and SGD as the optimizer to perform end-to-end network training; the YoloV4 classification model training adopts the transfer learning method, with a total of 800 iterations set. In the first 300 iterations, the backbone network parameters are frozen, and the batch_size is set to 64. In the next 500 iterations, the weight value of the entire network is trained, and the batch_size is set to 32; the learning rate adopts the cosine decay method, the initial value is set to 1e-2, and the minimum learning rate is set to 1e-4, and finally the statistical graph classification detection model is obtained.

[0011] In the aforementioned defect diagnosis method based on the distribution statistics of temperature data of infrared images of power equipment, in step S3, when processing the temperature matrix data of each component frame, the obtained array array_concatenate is first used to make a preliminary judgment on the defect situation of the equipment to be detected. If all temperature values ​​in the array array_concatenate are less than -10°C, it can be directly determined that the components of the power equipment to be detected are free of defects without the need for analysis and judgment by the statistical chart classification detection model. On the contrary, if all temperature values ​​in the array array_concatenate are greater than 100 degrees, the power equipment to be detected is a defective device. If the array arr ay_concatenate does not meet the above two conditions, the statistical chart to be detected is then used, and then the statistical chart classification detection model is used to classify and detect the distribution statistics of the temperature data to be detected, so as to determine whether the power equipment to be detected has defects.

[0012] Compared with the prior art, the present invention uses the infrared images of power equipment collected historically to classify defective images and non-defective images, and then obtains a temperature data distribution statistical graph that can reflect the status of the equipment by performing specific data processing on the components and temperature matrix data of the corresponding equipment in the above image; based on the temperature data distribution statistical graph of the infrared image of the power equipment, the YoloV4 deep learning method is used to classify the temperature data distribution statistical graph into defective and non-defective, thereby determining whether the components of the power equipment are defective. The method of the present application can effectively improve the accuracy of infrared image recognition of power equipment.

[0013] The applicant tested the defective and non-defective data of 23 common power equipment in substations in the same experimental environment. The precision and recall of the model are as follows:

[0014] Accuracy Recall defect 98.2％ 98.3％ Non-defective 98.5％ 98.4％ BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is an infrared image of defective power equipment;

[0016] Figure 2 It is a defect statistics chart;

[0017] Figure 3 It is an infrared image of a non-defective device;

[0018] Figure 4 Yes and No Defect Statistic Chart;

[0019] Figure 5 This is a diagram that uses a defective isolating switch as an example to illustrate the process of generating a temperature statistical diagram from a temperature matrix and component frame coordinates. DETAILED DESCRIPTION

[0020] The present invention is further described below in conjunction with the accompanying drawings and embodiments, but they are not intended to limit the present invention.

[0021] Embodiment: A defect diagnosis method based on distribution statistics of temperature data of infrared images of electric power equipment comprises the following steps:

[0022] S1. Divide the infrared images of various power equipment collected historically into defect images and non-defect images, and process the coordinates of one or more component frames in the defect images and non-defect images and their corresponding temperature matrix data respectively to generate corresponding defect statistical graphs and non-defect statistical graphs, and form a data set with the obtained defect statistical graphs and non-defect statistical graphs;

[0023] The method of generating a statistical graph by data processing is as follows: marking each component of the power equipment, obtaining a component frame of the power equipment, finding the temperature matrix data of the component frame corresponding to the infrared image of the power equipment, flattening the temperature matrix data corresponding to each component frame in the infrared image of the power equipment, and concatenating them to obtain a one-dimensional array array_concatenate; after statistically processing the one-dimensional array array_concatenate, a statistical graph of temperature data is generated with each set temperature value within the temperature range of the power equipment as the horizontal coordinate and the number of times each set temperature value appears in all component frames as the vertical coordinate;

[0024] Specifically, mark each component of the defective power equipment to obtain the component frame of the defective equipment, find the temperature matrix data collected by the infrared temperature sensor corresponding to the infrared image of the power equipment, intercept the temperature data in each component frame of the defective equipment, flatten the temperature data in each component frame, and splice them. Use the temperature values ​​in the component frame of the equipment as the horizontal axis and the number of each temperature value as the vertical axis to generate a statistical graph of the temperature data. Do the same processing on the non-defective data. At this time, a data set consisting of defect statistical graphs and non-defect statistical graphs will be generated. The sorted data set is randomly divided into a training set, a validation set, and a test set at a ratio of 7:1:2. The defective equipment and its corresponding defect statistical graph are shown in the figure. Figure 1 and Figure 2 As shown, non-defective equipment and its corresponding non-defective statistics are shown in Figure 3 and Figure 4 shown. Figure 1 and Figure 3 R01, R02 and R03 in the figure all represent three component boxes.

[0025] Distribution law of statistical chart: There will be two peaks on the statistical chart. The first peak is formed by the background temperature value, and the second peak is formed by the temperature value of the component in the component box. After the second peak, the curve of the defect statistical chart will form a small peak or tailing phenomenon. The small peak or tailing phenomenon depends on the size and number of the defect temperature values. If the temperature of the defective component is concentrated, a small peak will be formed. If the temperature of the defective component is discrete, the statistical chart will have a tailing phenomenon.

[0026] The following takes a defective isolating switch as an example to illustrate the process of generating a temperature statistical graph from a temperature matrix and component frame coordinates. The entire implementation process is as follows: Figure 5As shown, R01, R02 and R03 are three component frames of the disconnector. First, the temperature data of the corresponding component frames are intercepted from the temperature matrix of the disconnector according to the coordinates of R01, R02 and R03. The temperature data at this time is three two-dimensional arrays R01_array, R02_array and R03_array. Each two-dimensional array is flattened to convert it into three one-dimensional arrays. The three one-dimensional arrays are concatenated in sequence according to the order of R01, R02 and R03 to generate a new one-dimensional array array_concatenate. Next, the data in the concatenated new one-dimensional array needs to be processed to convert the temperature less than -10 degrees and greater than 10 The temperature value of 0 degrees is set to -10 and 100 respectively. The number of times the elements in the one-dimensional array after statistical processing appear in the interval [-10,100] with a step of 0.1 for each temperature value will be generated. At this time, a two-dimensional array array containing 1100 elements will be generated. The structure of each element of the array is (15.5,60), which means that the temperature 15.5 appears 60 times in the component frame. The 1100 elements of the array array are used to draw a curve graph. At this time, the horizontal axis of the curve graph is the temperature value in the interval [-10,100] with a step of 0.1, and the vertical axis is the number of times each temperature value appears in all component frames. The generated curve graph is the defect statistics graph corresponding to the three component frames of the disconnector, as shown in the figure below. Figure 2 shown.

[0027] Preferably, the best use state of the present invention is that the component frame does not contain other devices as much as possible.

[0028] S2. Train the YoloV4 classification model using the data in the data set to obtain a statistical graph classification detection model;

[0029] S3. Use the known coordinates of each component frame of the same type of electrical equipment to intercept the temperature matrix data of each component frame corresponding to the infrared image of the electrical equipment to be detected, and use the data processing method in step S1 to process the data to obtain a distribution statistical graph of the temperature data to be detected. Use the statistical graph classification detection model to classify and detect the distribution statistical graph of the temperature data to be detected to obtain a classification detection result.

[0030] The data set in step S1 is randomly divided into a training set, a validation set, and a test set in a ratio of 7:1:2 for training the YoloV4 classification model in step S2.

[0031] The specific process of YoloV4 classification model training in step S2 is as follows: the YoloV4 classification model uses mobilenet0.25 as the backbone feature extraction network and SGD as the optimizer to perform end-to-end network training; the YoloV4 classification model training adopts the transfer learning method, with a total of 800 iterations set. In the first 300 iterations, the backbone network parameters are frozen, and the batch_size is set to 64. The weight values ​​of the entire network are trained in the next 500 iterations, and the batch_size is set to 32; the learning rate adopts the cosine decay method, the initial value is set to 1e-2, and the minimum learning rate is set to 1e-4, and finally the statistical graph classification detection model is obtained.

[0032] In step S3, when processing the temperature matrix data of each component frame, the obtained array array_concatenate is first used to make a preliminary judgment on the defect of the equipment to be detected. If all temperature values ​​in the array array_concatenate are less than -10°C, it can be directly determined that the components of the power equipment to be detected are free of defects without the need for analysis and judgment by the statistical chart classification detection model. On the contrary, if all temperature values ​​in the array array_concatenate are greater than 100 degrees, the power equipment to be detected is a defective device. If the array arr ay_concatenate does not meet the above two conditions, the statistical chart to be detected is then used. The statistical chart classification detection model is used to perform classification detection on the distribution statistical chart of the temperature data to be detected, and it can be determined whether the power equipment to be detected has defects.

Claims

1. A defect diagnosis method based on the distribution statistics of infrared image temperature data of power equipment, characterized in that: The following steps are involved: S1. Divide the infrared images of various power equipment collected historically into defect images and non-defect images, and process the coordinates of one or more component frames in the defect images and non-defect images and their corresponding temperature matrix data respectively to generate corresponding defect statistical graphs and non-defect statistical graphs, and form a data set with the obtained defect statistical graphs and non-defect statistical graphs; The method of generating a statistical graph by data processing is as follows: marking each component of the power equipment, obtaining a component frame of the power equipment, finding the temperature matrix data of the component frame corresponding to the infrared image of the power equipment, flattening the temperature matrix data corresponding to each component frame in the infrared image of the power equipment, and concatenating them to obtain a one-dimensional array array_concatenate; after statistically processing the one-dimensional array array_concatenate, a statistical graph of temperature data is generated with each set temperature value within the temperature range of the power equipment as the horizontal coordinate and the number of times each set temperature value appears in all component frames as the vertical coordinate; S2. Train the YoloV4 classification model using the data in the data set to obtain a statistical graph classification detection model; S3. Use the known coordinates of each component frame of the same type of electrical equipment to intercept the temperature matrix data of each component frame corresponding to the infrared image of the electrical equipment to be detected, and use the data processing method in step S1 to process the data to obtain a distribution statistical graph of the temperature data to be detected. Use the statistical graph classification detection model to classify and detect the distribution statistical graph of the temperature data to be detected to obtain a classification detection result.

2. The defect diagnosis method based on the distribution statistics of infrared image temperature data of power equipment according to claim 1 is characterized in that: The data set in step S1 is randomly divided into a training set, a validation set, and a test set in a ratio of 7:1:2 for training the YoloV4 classification model in step S2.

3. The defect diagnosis method based on the distribution statistics of infrared image temperature data of power equipment according to claim 1 is characterized in that: The specific process of YoloV4 classification model training in step S2 is as follows: the YoloV4 classification model uses mobilenet0.25 as the backbone feature extraction network and SGD as the optimizer to perform end-to-end network training; the YoloV4 classification model training adopts the transfer learning method, with a total of 800 iterations set. In the first 300 iterations, the backbone network parameters are frozen, and the batch_size is set to 64. The weight values ​​of the entire network are trained in the next 500 iterations, and the batch_size is set to 32; the learning rate adopts the cosine decay method, the initial value is set to 1e-2, and the minimum learning rate is set to 1e-4, and finally the statistical graph classification detection model is obtained.

4. The defect diagnosis method based on the distribution statistics of infrared image temperature data of power equipment according to claim 1 is characterized in that: In step S3, when processing the temperature matrix data of each component frame, the obtained array array_concatenate is first used to make a preliminary judgment on the defect of the equipment to be detected. If all temperature values ​​in the array array_concatenate are less than -10°C, it is possible to directly determine that the components of the power equipment to be detected are free of defects without the need for analysis and judgment by the statistical chart classification detection model. On the contrary, if all temperature values ​​in the array array_concatenate are greater than 100 degrees, the power equipment to be detected is a defective device. If the array array_concatenate does not meet the above two conditions, the statistical chart to be detected is used, and then the statistical chart classification detection model is used to classify and detect the distribution statistical chart of the temperature data to be detected, so as to determine whether the power equipment to be detected has defects.

Images