Composite insulator infrared temperature measurement correction method based on PSO-BP algorithm

Optimizing the BP neural network through the PSO-BP algorithm, the problem of the infrared temperature measurement accuracy of composite insulators is affected by distance and perspective angle, and higher temperature measurement accuracy and reliability are achieved, and real-time and automation of drone inspections are improved. , ensuring accurate evaluation of the insulator state and safety of the transmission line.

CN119043502APending Publication Date: 2024-11-29CHINA THREE GORGES UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411015432.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The infrared temperature measurement accuracy of composite insulators is significantly affected by distance and viewing angle. The existing calibration methods are complex and have limited applicability. It is difficult to maintain consistent measurement results in different working environments. Real-time and automation levels also need to be improved.

Method used

Using the composite insulator infrared temperature measurement correction method based on the PSO-BP algorithm, the weight and threshold of the BP neural network are optimized through the particle swarm optimization algorithm, and the PSO-BP neural network model is constructed to eliminate the influence of distance and perspective on infrared temperature measurement results, and improve the Temperature measurement accuracy and reliability.

Benefits of technology

Significantly improves the accuracy and reliability of infrared temperature measurement, ensures consistent measurement results in different environments, improves the real-time and automation level of drone inspections, ensures accurate evaluation of composite insulator status, and thus improves the safety of transmission lines. and operational reliability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119043502A_ABST
    Figure CN119043502A_ABST
Patent Text Reader

Abstract

The invention discloses a composite insulator infrared temperature measurement correction method based on a PSO-BP algorithm. The method comprises the following steps: building a composite insulator infrared temperature measurement experiment platform; obtaining infrared temperature measurement results of the infrared thermometer and the infrared thermal imager on the composite insulator at different distances and visual angles; making an infrared temperature measurement data set, and dividing the infrared temperature measurement data set into a training set and a verification set; a particle swarm optimization (PSO) is introduced to optimize the BP neural network, and a PSO-BP neural network model is constructed; and correcting an infrared temperature measurement result through a PSO-BP neural network model, and recording a temperature value to obtain a corrected infrared temperature measurement result. According to the method, the weight and the threshold value of the BP neural network are optimized by using the global search capability of the PSO, so that the precision and the reliability of infrared temperature measurement are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of composite insulator state assessment, and in particular to a composite insulator infrared temperature measurement correction method based on a PSO-BP algorithm. Background Art

[0002] The composite insulators of power transmission lines are subjected to the combined effects of electrical, mechanical and environmental factors for a long time. It is easy for acid corrosion and electrocorrosion to cause partial discharge heating inside the sheath, which leads to aging and deterioration of the core rod, and brittleness, thereby reducing the mechanical strength of the insulator. In severe working conditions, string breakage and line drop accidents may also occur, resulting in inestimable consequences. In recent years, infrared temperature measurement technology has been increasingly used in the inspection of power transmission lines. During the operation and maintenance of line insulators, drones equipped with infrared imagers are usually used for inspection. The distance of 3-10m is maintained during shooting, and the connection between the infrared lens and the insulator is at an angle of 30° to the conductor. According to the principle of infrared temperature measurement, the distance and shooting angle of view will affect the temperature measurement accuracy, so it is difficult to obtain the true temperature of the composite insulator very accurately. A slight measurement deviation of the temperature may cause misjudgment of the state of the composite insulator. Therefore, it is of great significance to study how to eliminate the influence of distance and angle of view on the infrared temperature measurement results and correct the temperature measurement accuracy for accurately evaluating the state of the insulator.

[0003] In recent years, some scholars at home and abroad have studied the correction of infrared temperature measurement accuracy, mainly focusing on improving temperature measurement accuracy and reliability. The methods used include using advanced infrared imaging technology, optimizing image processing algorithms, and improving temperature compensation models. Although some progress has been made, there are still some shortcomings. The main problems include: the temperature measurement accuracy is greatly affected by environmental factors, especially the changes in viewing angle and distance have a significant impact on the accuracy of temperature measurement; the existing calibration methods are highly complex and have limited applicability, and it is difficult to maintain consistent measurement results in different working environments; in addition, the real-time and automation levels need to be improved. These problems restrict the widespread application of infrared temperature measurement technology in composite insulator detection, and further research and optimization are urgently needed.

[0004] Reference [1]: Li Cao. Research on the relationship between the temperature measurement accuracy of infrared thermal imagers and the influence of the external environment [D]. Changchun University of Science and Technology, 2009. Record: Based on the basic principle of infrared thermal imager radiation measurement, various factors affecting the measurement error of thermal imagers were obtained, and the influence of emissivity and measurement distance on temperature measurement accuracy was analyzed through data, but the results of infrared thermal imagers were not corrected.

[0005] References [2]: PRMuniz, SPNcani, R. da S. Influence of Field of View of Thermal Imagers and Angle of View on Temperature Measurements by Infrared Thermovision[J].IEEE Sensors Journal, 2014, 14(3):729-733. This paper explores the influence of the field of view and viewing angle of a thermal imager on temperature measurement, and develops a model to correct the measurement error through multi-layer artificial neural network and regression analysis. However, this article only considers the field of view angle and does not consider other factors affecting infrared temperature measurement.

[0006] Reference [3]: Wang Liming, Fu Kaiwei, Mei Hongwei, et al. Influence of ambient humidity on infrared temperature measurement of composite insulators [J]. High Voltage Technology, 2019, 45(06): 1955-1961. The correlation between the temperature difference measurement between the high-voltage and low-voltage ends of the composite insulator and the ambient relative humidity was studied, and the recommended humidity for using infrared temperature measurement technology was obtained, but the influence of other factors on infrared temperature measurement was not considered.

[0007] Reference [4]: ​​Wang Haijuan, Hu Zhenqi, Xia Qing, et al. Study on the influence of distance and humidity on the surface temperature detection of coal gangue mountain [J]. Infrared Technology, 2015, 37(07): 618-623. Record: The surface temperature of coal gangue mountain was detected by infrared thermal imager, and the distance-temperature curve and humidity-temperature curve were fitted, and a correction model was established. However, the two models were not combined, which made it more troublesome to apply.

[0008] An infrared temperature measurement correction algorithm based on the data regression prediction of BP neural network optimized by particle swarm optimization algorithm (PSO-BP) is proposed. Based on the original BP algorithm, the algorithm uses the global search capability of PSO to optimize the weights and thresholds of the BP neural network, thereby improving the training efficiency and performance of the network. Therefore, this method can be applied to infrared temperature measurement correction. Summary of the invention

[0009] Aiming at the problem of infrared temperature measurement correction of composite insulators, the present invention provides a composite insulator infrared temperature measurement correction method based on PSO-BP algorithm, aiming to solve the problems existing in the prior art that the temperature measurement accuracy is significantly affected by distance and viewing angle, the calibration method is complicated and has limited applicability. By combining the particle swarm optimization algorithm PSO and the BP neural network, the present invention uses the global search capability of PSO to optimize the weights and thresholds of the BP neural network, thereby improving the accuracy and reliability of infrared temperature measurement.

[0010] The technical solution adopted by the present invention is:

[0011] The infrared temperature measurement correction method of composite insulator based on PSO-BP algorithm includes the following steps:

[0012] Step 1: Build a composite insulator infrared temperature measurement experimental platform;

[0013] Step 2: Obtain infrared temperature measurement results of the composite insulator using an infrared thermometer and an infrared thermal imager at different distances and viewing angles;

[0014] Step 3: Create an infrared temperature measurement data set and divide the infrared temperature measurement data set into a training set and a validation set;

[0015] Step 4: Particle swarm algorithm PSO is used to optimize BP neural network and construct PSO-BP neural network model;

[0016] Step 5: Correct the infrared temperature measurement results through the PSO-BP neural network model, and record the temperature value to obtain the corrected result of infrared temperature measurement.

[0017] In the step 1, in order to explore the influence of distance and viewing angle on infrared temperature measurement and obtain an infrared temperature measurement data set, the present invention builds an infrared temperature measurement experimental platform for composite insulators, conducts experiments on 110kV composite insulators, and studies the relationship between the measured value and the true value of the infrared thermal imager at different measurement distances and viewing angles.

[0018] The composite insulator infrared temperature measurement experimental platform includes a booster, a suspension vehicle, a composite insulator, a fiber optic thermometer, an infrared thermal imager, and an infrared thermometer;

[0019] The boost device mainly includes a high-voltage power supply system and a high-voltage measurement system;

[0020] The high-voltage power supply system includes a voltage regulator, a step-up transformer (150kV, 50kVA), and a protection resistor (10kΩ, 10kW);

[0021] The high voltage measurement system includes a voltmeter (voltage division ratio 1:1000);

[0022] The suspension vehicle is used to suspend 110kV composite insulators. The ground wire of the pressurized equipment is connected to the top suspension insulator of the suspension vehicle.

[0023] The model of the infrared thermal imager is FLUKET Ti100 3, which is used for the infrared temperature of composite insulators.

[0024] Fiber optic thermometer, used to measure the actual temperature of composite insulators;

[0025] The 110kV composite insulator is connected to the primary side of the step-up transformer through a protective resistor, the secondary side of the step-up transformer is connected to the voltage regulator, and a voltmeter is connected in parallel to the secondary side of the step-up transformer;

[0026] The model of the infrared thermometer is FLUKE 566 laser thermometer, which is also used to perform infrared temperature measurement on composite insulators and compare the temperature measurement results with those of the infrared thermal imager.

[0027] In order to eliminate the interference of other factors such as temperature, humidity and wind speed on the test, the infrared temperature measurement experiment of composite insulators was carried out in the laboratory, so the wind speed was 0m / s and the time for infrared temperature measurement was short, so the temperature and humidity could be considered constant.

[0028] In the step 2, infrared temperature measurement results of the composite insulator at different distances and viewing angles are obtained using an infrared thermometer and an infrared thermal imager. The specific test steps are as follows:

[0029] 2.1: First, suspend the composite insulator as required: Select the first composite insulator for suspension, and suspend the composite insulator vertically on the composite suspension vehicle in the test room according to the requirements of GB / T4585-2004 "Artificial pollution test for high-voltage insulators for AC systems", clamp the top of the composite insulator to ensure stability. And measure the vertical distance from the composite insulator to the ground to ensure that the distance between any part of the composite insulator and the ground object is more than 0.7m;

[0030] 2.2: Install the fiber optic thermometer: After the insulator is hung, install the fiber optic thermometer outside the safety fence, and stick the three probes CH1, CH2, and CH3 of the fiber optic thermometer on the surface of the composite insulator through insulating and thermally conductive double-sided tape. Probe CH4 is directly placed in the air to measure the ambient temperature.

[0031] 2.3: Apply voltage to the high voltage end at the bottom of the composite insulator:

[0032] Use copper wire to connect the live wire end of the step-up transformer to the bottom of the composite insulator, and connect the ground wire end to the top of the composite insulator. In order to simulate the actual operation, the voltage is set to 1.1 times the operating margin; before pressurization, confirm that the safety guardrail and grounding rod are correctly placed. After confirmation, pressurize the composite insulator. The applied voltage type is AC voltage, the pressurization size is 100 / ×1.1≈70kV, and the pressurization time is 120min;

[0033] 2.4: Temperature measurement of composite insulators: After the pressurization is completed, the actual temperature of the composite insulator is measured using a fiber optic thermometer, and the infrared temperature of the composite insulator is obtained through an infrared thermometer and an infrared thermal imager; according to the requirements of the infrared inspection of drones, the distance and viewing angle of the infrared thermometer and infrared thermal imager are changed for measurement. The measuring distances are set to 2m, 3m, 4m and 5m respectively, and the shooting angle is adjusted at each distance. The shooting angle is set to -30° to 30°, with each 10° as an interval. At each temperature measurement position, three temperature measurements are recorded, and the average of the three temperature measurements is taken as the measured temperature to ensure the reliability of the data. Record the temperature of the infrared thermometer, infrared thermal imager and fiber optic thermometer.

[0034] 2.5: Replace the insulator and retest: After the record is completed, control the voltage regulator to reduce the voltage, and close the voltage regulator by opening the switch, hang the grounding rod, remove the tested composite insulator and replace it with the second composite insulator, repeat test steps 2..1 to 2.4, and test the remaining insulators in turn.

[0035] In the step three, an infrared temperature measurement data set is prepared, and the infrared temperature measurement data set is divided into a training set and a validation set in a ratio of 8:2.

[0036] In the step 4, the PSO-BP neural network model parameters are configured, deep learning training is performed, the training effect is tested, and the trained PSO-BP neural network model is obtained; the details are as follows:

[0037] In order to optimize and solve the problem that the traditional BP algorithm is easy to fall into the local optimal solution and the training speed is relatively slow, the present invention uses the particle swarm algorithm PSO to optimize the BP neural network. The particle swarm algorithm PSO is an intelligent optimization algorithm that simulates the social behavior of animal groups. Each particle is regarded as a solution. By tracking the relationship between the fitness value F(i), individual extreme value PiBest, and group extreme value PgBest of each particle, its own speed and position are updated. The speed update formula is shown in formula (1):

[0038]

[0039] In formula (1), v i (t+1) is the velocity of the particle at t+1; w is the inertia weight; v i (t) is the velocity of the particle at time t; c1, c2 are acceleration constants; r1 and r2 are random numbers between 0 and 1; x i (t) is the position of the particle at time t.

[0040] The position update formula is shown in formula (2):

[0041] x i (t+1)=xi (t)+v i (t+1) (2)

[0042] In the formula, x i (t+1) is the position of the particle at time t+1.

[0043] The back propagation algorithm is an algorithm used to train neural networks. It adjusts the network weights to minimize the output error. The weight adjustment formula is:

[0044]

[0045] In the formula, w ij (t+1) and w ij (t) represents the weights of connected neurons at t+1 and t respectively; η is the learning rate; E is the total error.

[0046] The particle swarm algorithm PSO is used to optimize the BP neural network, which overcomes the shortcomings of the traditional BP neural network that relies on initial weights and is prone to falling into local optimal solutions, and improves the overall performance of the network. The training steps are as follows:

[0047] 1) Determine the structure of BP neural network:

[0048] Determine the number of network layers, number of neurons and activation function of the BP neural network; the activation value of the neural network is:

[0049] a l =σ(W l a l-1 +b l ) (4)

[0050] In the formula, a l is the activation value of the lth layer, σ is the activation function, W l is the weight matrix of the lth layer, a l-1 is the activation value of the l-1th layer, b l is the bias vector of the lth layer.

[0051] 2) Initialization parameters:

[0052] The initial weights and biases of the BP neural network are randomly generated, and the PSO particle swarm is initialized, and the initial position and speed of each particle are set; the weight calculation formula is:

[0053]

[0054] Where W ij is the weight between the i-th neuron and the j-th neuron in the l-th layer, φ is a normal distribution, n (l-1) is the number of neurons in the l-1th layer.

[0055] 3) Use particle swarm optimization (PSO) to optimize weights and biases:

[0056] Calculate the fitness F(i) of each particle and update the individual extreme value P iBest and the group extreme value P gBest ; Update its own speed and position by tracking the relationship between each particle's fitness value F(i), individual extreme value PiBest, and group extreme value PgBest. The speed update formula is shown in formula (6):

[0057]

[0058] In the formula, v i (t+1) is the velocity of the particle at t+1; w is the inertia weight; v i (t) is the velocity of the particle at time t; c1, c2 are acceleration constants; r1 and r2 are random numbers between 0 and 1; x i (t) is the position of the particle at time t.

[0059] The position update formula is shown in formula (7):

[0060] x i (t+1)=x i (t)+v i (t+1) (7);

[0061] In the formula, x i (t+1) is the position of the particle at time t+1.

[0062] 4) Initialize BP neural network:

[0063] After obtaining the optimal solution of the PSO algorithm, set this solution as the initial weight and bias of the BP neural network:

[0064]

[0065] Where W ij best is the optimal weight obtained through PSO optimization.

[0066] 5) Training BP neural network:

[0067] Input distance x1, viewing angle x2, infrared temperature value x3, calculate the neuron outputs of the hidden layer and the output layer, get the corrected infrared temperature value y', and calculate the error between the output value and the target value, adjust the weight and bias, and iterate continuously until the preset value is reached. Output the optimized PSO-BP neural network model, and the loss function J is:

[0068]

[0069] Where n is the number of samples; y i is the actual value of the ith sample, is the predicted value of the ith sample.

[0070] In the step 5, the PSO-BP neural network model is loaded, the infrared temperature measurement data to be corrected is read, and the infrared temperature measurement correction value is output.

[0071] It also includes step 6: input the infrared temperature measurement data to be predicted, check the PSO-BP neural network model correction results, and compare them with other models. The details are as follows:

[0072] After the training is completed, the temperature measurement test is carried out again, the infrared temperature measurement results are corrected through the correction models of different methods, and the accurate temperature values ​​are recorded to obtain the corrected results of infrared temperature measurement.

[0073] The commonly used statistical regression models, including linear, nonlinear and exponential function fitting, and commonly used machine learning models, such as extreme learning machine (ELM), radial basis function neural network (RBF) and partial least squares regression (PLS), are used. At the same time, deep learning models, such as BP neural network optimized by genetic algorithm (GA-BP) and standard BP neural network, are combined to correct the infrared temperature measurement data respectively. The infrared temperature measurement data after statistical correction are compared and analyzed to compare the correction accuracy of the PSO-BP infrared temperature measurement correction model proposed in the present invention and other models.

[0074] The present invention provides a composite insulator infrared temperature measurement correction method based on PSO-BP algorithm, which has the following beneficial effects:

[0075] 1) This method can not only effectively eliminate the influence of distance and viewing angle on the infrared temperature measurement results, but also maintain consistent measurement results in different working environments, significantly improve the real-time and automation level of UAV infrared inspection, ensure accurate assessment of the status of composite insulators, and thus improve the safety and operational reliability of transmission lines.

[0076] 2) The present invention particularly relates to a composite insulator infrared temperature measurement distance and viewing angle compensation calibration method based on the PSO-BP algorithm. This method can improve the accuracy of drone infrared inspection and make the insulator fault level judgment more accurate.

[0077] 3) The present invention first performs infrared temperature measurement on the composite insulator in the test to obtain an infrared temperature measurement data set. Then the data set is trained to establish a correction model, and finally the correction effect of the algorithms proposed by the present invention is verified through experiments. The experiment shows that before correction, the average temperature difference between the infrared thermometer and the thermal imager is 1.65°C and 1.32°C, respectively, and the average error is 17.06% and 9.07%. After correction by the PSO-BP algorithm, the average temperature difference is reduced to 0.12°C and 0.08°C, and the average error is reduced to 1.36% and 0.97%.

[0078] 4) The correction results show that the PSO-BP-based composite insulator infrared temperature measurement correction method proposed in the present invention has a significant correction effect, effectively improves the infrared temperature measurement accuracy of composite insulators, and this method is a comprehensive correction method, which can add other influencing factors for correction.

[0079] 5) This method can be applied to the field of UAV infrared inspection of power transmission lines, which can greatly improve the accuracy of UAV infrared temperature measurement and the accuracy of insulator status determination, and provide strong technical support for UAV regular inspection of composite insulators and maintenance of power transmission line safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0081] Figure 1 The overall flow chart for insulator temperature measurement correction.

[0082] Figure 2 This is the wiring schematic diagram of the infrared temperature measurement test;

[0083] Among them: 1-110kV composite insulator, 2-fiber optic thermometer, 3-infrared thermal imager, 4-voltage regulator, 5-step-up transformer 5, 6-protective resistor, 7-voltmeter, 8-laptop computer, 9-switch.

[0084] Figure 3 For dirty insulator diagram;

[0085] Figure 4 Take a pose picture for an infrared thermometer.

[0086] Figure 5(a) shows the temperature measurement result of the infrared thermometer.

[0087] Figure 5(b) shows the temperature measurement result of the infrared thermal imager.

[0088] Figure 6 This is a schematic diagram of the BP neural network structure.

[0089] Figure 7 Modification flow chart of PSO-BP neural network model.

[0090] Figure 8(a) is a graph of the model iteration error.

[0091] Figure 8(b) Error curve of the training process.

[0092] Figure 8(c) is a comparison of the correction effects.

[0093] Figure 9(a) shows the correction result of the traditional fitting model of infrared thermometer;

[0094] Figure 9(b) shows the correction result of the infrared thermal imager learning model;

[0095] Figure 9(c) shows the correction result of the deep learning model of the infrared thermal imager.

[0096] Figure 10(a) shows the correction result of the traditional fitting model of infrared thermal imager;

[0097] Figure 10(b) shows the correction result of the infrared thermal imager machine learning model;

[0098] Figure 10(c) shows the correction result of the deep learning model of the infrared thermal imager.

[0099] Fig.11 This is the infrared temperature measurement correction diagram. DETAILED DESCRIPTION

[0100] Infrared temperature measurement of composite insulators is an important means to determine their status. In order to reduce the influence of infrared lens shooting distance and viewing angle on the infrared temperature measurement accuracy of composite insulators of transmission lines. The present invention proposes a back propagation artificial neural network prediction model (PSO-BP neural network model) based on particle swarm optimization algorithm for infrared temperature correction of composite insulators.

[0101] Firstly, a high-voltage temperature measurement test was conducted on FXBW4-110 / 70 110kV composite insulators under normal and contaminated conditions. The surface temperature of the composite insulator was collected by infrared temperature measurement and optical fiber temperature measurement. The infrared temperature measurement value, distance and viewing angle were used as the input layer parameters of the PSO-BP neural network model, and the optical fiber temperature measurement value was used as the output layer parameter. The initial weight and threshold of the BP neural network were optimized by the PSO algorithm. Then, the infrared temperature of the composite insulator was corrected by traditional statistical regression methods such as linear, nonlinear and exponential, and machine learning models such as ELM, RBF and PLS. The results show that the PSO-BP neural network model is superior to traditional statistical regression methods, machine learning models and deep learning algorithms in correcting infrared temperature measurement errors. The average errors of infrared thermometers and thermal imagers were reduced from 17.06% and 9.07% to 1.36% and 0.97% respectively.

[0102] Example:

[0103] The present invention uses a WT-YH-400 optical fiber temperature measuring device to measure the temperature of the surface of the composite insulator. The temperature measurement accuracy is ±0.5°C. The temperature value obtained by the optical fiber temperature measuring device can be used as an accurate value to judge the measurement accuracy of other temperature measurement methods, and as a data set input for subsequent infrared temperature measurement correction methods. The present invention uses two types of infrared temperature measuring instruments, namely infrared thermometers and infrared thermal imagers. Infrared thermometers mainly use lasers to measure point temperature and have no image display function, while infrared thermal imagers can generate infrared images to display the specific temperature distribution of the measured object. The two types of infrared temperature measurement results will be corrected later. The specific parameters of the three temperature measuring instruments are shown in Table 1.

[0104] Table 1 Infrared temperature measuring instrument parameters

[0105]

[0106] The samples used in the test were three 110kV FXBW4-110 / 70 composite insulators with equalizing rings, one of which was used to simulate the operation of normal insulators, and the other two were used to simulate the operation of contaminated insulators. In accordance with the requirements of GB / T4585-2004, tap water with trisodium phosphate or other cleaning agents was heated to 50°C, and the surfaces of the three normal insulators were cleaned and rinsed with tap water. Then, two composite insulators with surface contamination were artificially constructed using the solid coating method. 40g of loess, 1000g of tap water and an appropriate amount of NaCl were mixed, and the contaminated liquid was covered on the two cleaned and dried insulators by pouring, and then placed in the environment to dry naturally for 24 hours. Photos of contaminated insulators are shown in the figure. Figure 3 shown.

[0107] According to the above test steps, the infrared temperature measurement results of several commonly used types of composite insulators at different distances and viewing angles were obtained by infrared thermometer and infrared thermal imager.

[0108] Specific results such as Figure 4 As shown in Figure 5(a) and Figure 5(b), the temperature measurement results show that as the temperature measurement distance of the infrared thermometer and the infrared thermal imager increases, the temperature difference between the measured temperature and the accurate temperature basically shows a trend of increasing, and the temperature measurement accuracy will also be affected by changing the pitch angle of the temperature measurement. The larger the pitch angle, the greater the temperature measurement error. The maximum temperature difference of the infrared thermometer is 1.6℃. The infrared thermal imager is 1.1℃. It can be seen from the infrared temperature measurement principle and experiments that distance and viewing angle are important factors affecting the accuracy of infrared temperature measurement. In order to ensure the effectiveness of the infrared temperature measurement value, a certain temperature compensation method must be used to correct the influence of these factors on the infrared temperature measurement.

[0109] In order to verify the effectiveness of the proposed temperature measurement correction algorithm, the present invention will re-conduct a verification test and use other commonly used correction algorithms to compare with the algorithm of the present invention.

[0110] First, an infrared temperature rise test is carried out to obtain a data set for infrared temperature measurement correction. Then, the data set is trained through different correction algorithms such as statistical regression, machine learning, and deep learning. After the training is completed, the correction model is exported. At the same time, the insulator temperature rise test is carried out again, and the infrared temperature measurement image of the test insulator is taken, and the infrared temperature measurement data as well as the distance and viewing angle are recorded. The recorded data is input into the trained correction model for infrared temperature measurement correction and compared with the accurate temperature measured in the verification test. In addition to using the PSO-BP neural network model for correction, the most commonly used data correction methods are also used, including statistical regression correction methods with linear, nonlinear and exponential fitting corrections, and machine learning correction methods of ELM, PBF and PLS. The infrared thermometer and infrared thermal imager are corrected for temperature measurement, and the correction effects of different algorithms on the two infrared temperature measuring instruments are compared.

[0111] The commonly used statistical regression models currently include linear, nonlinear, exponential and logarithmic function fitting. However, since the input viewing angle value has negative values, the logarithmic and power function fitting is not considered in the present invention. The present invention uses the linear, nonlinear and exponential function fitting methods to perform fitting analysis on the infrared temperature measurement data set. The fitting results are shown in Table 2. 2 represents the correlation coefficient of the fitting curve. The closer it is to 1, the stronger the model explanatory ability and the better the fitting effect. r1 is the corrected temperature value of the infrared thermal imager, d is the distance, θ is the viewing angle, T m2 is the temperature value measured by the infrared thermal imager, T r2 is the corrected temperature value of the infrared thermometer, T m2 It is the temperature value measured by infrared thermometer.

[0112] Table 2 Fitting correction results

[0113]

[0114]

[0115] The model training and verification of machine learning and deep learning are carried out in Windows 11 system, the CPU is Intel(R) Core(TM) i5-13490F 4.80GHz, the GPU is NVIDIAGeForce RTX 3050, and the programming software is MatlabR2021a. The specific correction steps of the deep learning correction algorithm are as follows:

[0116] 1) First, make a data set: pre-process the data set obtained in the experiment, including data cleaning, missing value processing, etc., and then divide the data set, dividing the infrared temperature measurement data obtained in the experiment into a training set and a validation set according to 8:2;

[0117] 2) Set training parameters. Set the number of hidden layer nodes, maximum number of iterations, and error threshold for deep learning and machine learning;

[0118] 3) Model training: Use different algorithms to train the data set and export the trained model;

[0119] 4) Training effect test: obtain training measurement indicators, including root mean square error and accuracy, etc., to check the training effect. The error change curve of the PSO-BP algorithm training process used in the present invention and the comparison of the correction effect of the validation set are shown in Figures 8(a), 8(b), and 8(c). It can be seen from Figures 8(a) and 8(b) that with the increase in the number of training times, the training error becomes smaller and smaller, and the model accuracy is also better; and it can be seen from Figure 8(c) that the difference between the result value after algorithm correction and the true value is not large, and the correction result is more accurate.

[0120] After the training is completed, the temperature measurement test is carried out again, the infrared temperature measurement results are corrected by the correction models of different methods, and the accurate temperature values ​​are recorded. The corrected results of infrared temperature measurement are shown in Figure 9(a), Figure 9(b), and Figure 9(c).

[0121] The specific correction data are shown in Table 3:

[0122] Table 3 Comparison of correction results

[0123]

[0124] From the correction results, it can be seen that before correction, the temperature difference between the infrared thermometer and the infrared thermal imager is large. After correction by different algorithms, the temperature difference of infrared temperature measurement is improved by different amounts. Therefore, the correction effects of different algorithms on different temperature measuring instruments are also different. As shown in Figure 9(a) and Figure 10(a), the linear, nonlinear and exponential statistical regression correction methods have certain correction effects on infrared temperature measurement, but the overall correction effect is poor, especially for some extreme value points. The statistical regression correction has weak generalization ability when processing more complex data.

[0125] As shown in Figure 10(a) and Figure 10(b), for machine learning models, the ELM model performs relatively smoothly at most test points, but is not as good as the model based on the optimization algorithm in handling extreme values. Although the RBF model and PLS model have good correction effects at individual points, their overall performance is unstable. The model with the best correction effect is the deep learning model.

[0126] As can be seen from Figures 9(c) and 10(c), in the deep learning model, the correction effects of the PSO-BP and GA-BP models are significantly better than those of the BP model. Specifically, after the temperature measurement accuracy of the infrared thermometer was corrected by PSO-BP and GA-BP, the average errors were 1.36%, 1.47% and 0.97%, 0.99%, respectively, which are higher than the traditional BP model. The correction effect of GA-BP is more stable, but the correction effect is slightly inferior at some test points with large temperature differences. The PSO-BP model has higher correction accuracy at most test points. Whether it is an infrared thermometer or an infrared thermal imager, the accuracy of infrared temperature measurement correction using deep learning is significantly higher than the other two types of methods. Therefore, among all the correction methods, the PSO-BP neural network model proposed in the present invention performs the best and is more suitable for high-precision correction applications of infrared temperature measurement. The specific correction effects are as follows: Fig.11 As shown. Fig.11 It can be seen that the accuracy of the temperature measurement results after correction by the PSO-BP algorithm is greatly improved compared with the uncorrected values, which shows that the algorithm is suitable for infrared temperature measurement correction.

Claims

1. The infrared temperature measurement correction method of composite insulator based on PSO-BP algorithm is characterized by The following steps are involved: Step 1: Build a composite insulator infrared temperature measurement experimental platform; Step 2: Obtain infrared temperature measurement results of the composite insulator using an infrared thermometer and an infrared thermal imager at different distances and viewing angles; Step 3: Create an infrared temperature measurement data set and divide the infrared temperature measurement data set into a training set and a validation set; Step 4: Particle swarm algorithm PSO is used to optimize BP neural network and construct PSO-BP neural network model; Step 5: Correct the infrared temperature measurement results through the PSO-BP neural network model, and record the temperature value to obtain the corrected result of infrared temperature measurement.

2. The infrared temperature measurement correction method for composite insulators based on the PSO-BP algorithm according to claim 1 is characterized in that: In the step 1, a composite insulator infrared temperature measurement experimental platform is built to test a 110 kV composite insulator. The composite insulator infrared temperature measurement experimental platform includes a booster, a suspension vehicle, a composite insulator, an optical fiber temperature meter (2), and an infrared thermal imager (3); an infrared temperature meter; The boost device comprises a high-voltage power supply system and a high-voltage measurement system; the high-voltage power supply system comprises a voltage regulator (4), a boost transformer (5), and a protective resistor (6); the high-voltage measurement system comprises a voltmeter (7); A suspension vehicle is used to suspend the composite insulator (1), and the ground wire of the pressurized equipment is connected to the top suspension insulator of the suspension vehicle; An infrared thermal imager (3) for measuring the infrared temperature of the composite insulator (1); An optical fiber thermometer (2) for measuring the actual temperature of the composite insulator (1); The composite insulator (1) is connected to the primary side of the step-up transformer (5) through a protective resistor (6), the secondary side of the step-up transformer (5) is connected to a voltage regulator (4), and a voltmeter (7) is connected in parallel to the secondary side of the step-up transformer (5); The infrared thermometer also performs infrared temperature measurement on the composite insulator (1), and compares the temperature measurement result with the infrared thermal imager (3); In order to eliminate the interference of other factors such as temperature, humidity and wind speed on the test, the infrared temperature measurement experiment of composite insulators was carried out in the laboratory.

3. The infrared temperature measurement correction method for composite insulators based on the PSO-BP algorithm according to claim 1 is characterized in that: In the step 2, infrared temperature measurement results of the composite insulator at different distances and viewing angles are obtained using an infrared thermometer and an infrared thermal imager. The specific test steps are as follows: 2.1: First, hang the composite insulator as required: select the first composite insulator for hanging, and hang the composite insulator vertically on the composite hanging vehicle in the test room as required, clamp the top of the composite insulator to ensure stability; and Measure the vertical distance from the composite insulator to the ground to ensure that there is a certain distance between any part of the composite insulator and the ground object; 2.2: Install the optical fiber thermometer (2): After the insulator is hung, install the optical fiber thermometer (2) outside the safety fence, and stick the three probes CH1, CH2, and CH3 of the optical fiber thermometer (2) on the surface of the composite insulator through insulating and heat-conductive double-sided adhesive. Probe CH4 is directly placed in the air to measure the ambient temperature; 2.3: Apply voltage to the high voltage end at the bottom of the composite insulator: Use a copper wire to connect the live wire end of the step-up transformer (5) to the bottom end of the composite insulator, and connect the ground wire end to the top end of the composite insulator; in order to simulate the actual operation, the voltage is set with an operation margin of 1.1 times; before pressurizing, confirm that the safety guardrail and the grounding rod are correctly placed; after confirming that they are correct, pressurize the composite insulator; 2.4: Temperature measurement of composite insulators: After the pressurization is completed, the actual temperature of the composite insulator is measured using the optical fiber thermometer (2), and the infrared temperature of the composite insulator is obtained through the infrared thermometer and the infrared thermal imager (3); according to the requirements of the infrared inspection of the drone, the distance and viewing angle of the infrared thermometer and the infrared thermal imager (3) are changed for measurement, and the measuring distance is set to 2m, 3m, 4m and 5m respectively, and the shooting angle is adjusted at each distance, and the shooting angle is set to -30° to 30°, with each 10° as an interval; at each temperature measurement position, three temperature measurements are recorded, and the average of the three temperature measurements is taken as the measured temperature to ensure the reliability of the data; the temperature of the infrared thermometer, the infrared thermal imager (3) and the optical fiber thermometer (2) is recorded; 2.5: Replace the insulator and retest: After the record is completed, control the voltage regulator to reduce the voltage, and close the voltage regulator by opening the switch, hang the grounding rod, remove the tested composite insulator and replace it with the second composite insulator, repeat test steps 2..1 to 2.4, and test the remaining insulators in turn.

4. The infrared temperature measurement correction method for composite insulators based on the PSO-BP algorithm according to claim 1 is characterized in that: In the step 4, the PSO-BP neural network model parameters are configured, deep learning training is performed, the training effect is tested, and the trained PSO-BP neural network model is obtained; the details are as follows: The particle swarm algorithm PSO is used to optimize the BP neural network. The particle swarm algorithm PSO is an intelligent optimization algorithm that simulates the social behavior of animal groups. Each particle is regarded as a solution. It updates its speed and position by tracking the relationship between each particle's fitness value F(i), individual extreme value PiBest, and group extreme value PgBest. The speed update formula is shown in formula (1): In formula (1), v i (t+1) is the velocity of the particle at t+1; w is the inertia weight; v i (t) is the velocity of the particle at time t; c1, c2 are acceleration constants; r1 and r2 are random numbers between 0 and 1; x i (t) is the position of the particle at time t; The position update formula is shown in formula (2): x i (t+1)=x i (t)+v i (t+1) (2) In the formula, x i (t+1) is the position of the particle at time t+1; The back propagation algorithm minimizes the output error by adjusting the network weights. The weight adjustment formula is: In the formula, w ij (t+1) and w ij (t) represents the weights of connected neurons at t+1 and t respectively; η is the learning rate; E is the total error; The particle swarm algorithm PSO is used to optimize the BP neural network. The training steps are as follows: 1) Determine the structure of BP neural network: Determine the number of network layers, number of neurons and activation function of the BP neural network; the activation value of the neural network is: a l =σ(W l a l-1 +b l ) (4) In the formula, a l is the activation value of the lth layer, σ is the activation function, W l is the weight matrix of the lth layer, a l-1 is the activation value of the l-1th layer, b l is the bias vector of the lth layer; 2) Initialization parameters: The initial weights and biases of the BP neural network are randomly generated, and the PSO particle swarm is initialized, and the initial position and speed of each particle are set; the weight calculation formula is: Where W ij is the weight between the i-th neuron and the j-th neuron in the l-th layer, φ is a normal distribution, n (l-1) is the number of neurons in the l-1th layer; 3) Use particle swarm optimization (PSO) to optimize weights and biases: Calculate the fitness F(i) of each particle and update the individual extreme value P iBest and the group extreme value P gBest ; Update its own speed and position by tracking the relationship between each particle's fitness value F(i), individual extreme value PiBest, and group extreme value PgBest. The speed update formula is shown in formula (6): In the formula, v i (t+1) is the velocity of the particle at t+1; w is the inertia weight; v i (t) is the velocity of the particle at time t; c1, c2 are acceleration constants; r1 and r2 are random numbers between 0 and 1; x i (t) is the position of the particle at time t; The position update formula is shown in formula (7): x i (t+1)=x i (t)+v i (t+1) (7); In the formula, x i (t+1) is the position of the particle at time t+1; 4) Initialize BP neural network: After obtaining the optimal solution of the PSO algorithm, set this solution as the initial weight and bias of the BP neural network: Where W ij best is the optimal weight obtained through PSO optimization; 5) Training BP neural network: Input distance x1, viewing angle x2, infrared temperature value x3, calculate the neuron outputs of the hidden layer and the output layer, get the corrected infrared temperature value y', and calculate the error between the output value and the target value, adjust the weight and bias, and iterate continuously until the preset value is reached. Output the optimized PSO-BP neural network model, and the loss function J is: Where n is the number of samples; y i is the actual value of the ith sample, is the predicted value of the ith sample.

5. The infrared temperature measurement correction method for composite insulators based on the PSO-BP algorithm according to claim 1 is characterized in that: It also includes step 6: input the infrared temperature measurement data to be predicted, check the PSO-BP neural network model correction results, and compare with other models; the details are as follows: After the training is completed, the temperature measurement test is carried out again, the infrared temperature measurement results are corrected through the correction models of different methods, and the accurate temperature values ​​are recorded to obtain the corrected results of infrared temperature measurement; The infrared temperature measurement data are corrected by using the current statistical regression models, including linear, nonlinear and exponential function fitting, and machine learning models: extreme learning machine (ELM), radial basis function neural network (RBF) and partial least squares regression (PLS), combined with deep learning models: BP neural network optimized by genetic algorithm (GA-BP) and standard BP neural network. The infrared temperature measurement data after statistical correction are compared and analyzed to compare the correction accuracy of the PSO-BP infrared temperature measurement correction model proposed by the correction method and other models.

Citation Information

Cited By

  • Multi-dimensional parameter calibration method for infrared temperature measurement of composite insulator

    CN119756606A