Insulator detection digital filtering algorithm and system combined with Gaussian regression

Abstract

The invention discloses an insulator detection digital filtering algorithm and system combined with Gaussian regression, and aims to solve the problem of low accuracy caused by influence of various factors on insulator detection in the prior art. According to the algorithm, firstly, a voltage signal collected by a sensor is processed in real time through a digital filter, and electromagnetic field interference is eliminated; performing curve fitting on the distribution voltage value of the insulator string by using a Gaussian process regression (GPR) algorithm, and converting the distribution voltage value into a resistance value to approximate an actual value; and finally, predicting and evaluating data through the GPR model, and identifying a fault insulator. According to the invention, the accuracy and stability of insulator fault detection are improved, the method is suitable for real-time monitoring of insulators in an ultra-high voltage line, and safe operation of a power transmission line can be guaranteed; in addition, the algorithm also has an adaptive learning capability, model parameters can be continuously optimized according to actual operation data, and the detection precision is further improved. The modular design is convenient for integration into the existing monitoring system, and seamless joint and efficient operation and maintenance are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission lines, insulator detection and intelligent algorithms, and in particular to an insulator detection digital filtering algorithm and system combined with Gaussian regression. Background Art

[0002] At present, with the rapid development of the electric power industry, ultra-high voltage transmission lines play a vital role in power transmission. Ultra-high voltage transmission technology has become an important means to improve the efficiency and economic benefits of power transmission due to its long distance, large capacity, low loss, land saving and economy. However, the reliability problem of ultra-high voltage transmission lines has become increasingly prominent, which is affected by the power itself, the external environment and human factors.

[0003] In transmission lines, insulators are key insulation components, and the stability and reliability of their performance are directly related to the safe operation of the entire line. Porcelain insulators made of electric porcelain materials are widely used in power transmission and transformation projects of various voltage levels due to their good insulation performance, climate change resistance, heat resistance, and assembly flexibility. In particular, the large-scale application of porcelain insulators is indispensable in ultra-high voltage lines.

[0004] However, porcelain insulators will gradually deteriorate during long-term operation due to the combined effects of electric fields, mechanical loads, and external climatic conditions (such as sunlight, wind, rain, snow, fog, etc.). This degradation phenomenon will not only reduce the insulation performance of the insulators, but also pose a serious threat to the safe operation of the power grid. Therefore, the use of intelligent devices to detect zero-value insulators has become an important way to solve the technical difficulties in live operations on UHV lines at this stage.

[0005] Traditional insulator inspection methods are often based on fixed value judgment criteria, such as judging the performance status by measuring the resistance value or distributed voltage of the insulator. However, this method is often difficult to accurately judge the true status of the insulator when faced with changes in temperature and humidity in the operating environment and the influence of the degree of insulator pollution. In particular, when the voltage distribution of the insulator string is U-shaped or sawtooth-shaped, the traditional judgment criteria are even more difficult to apply.

[0006] In addition, existing insulator detection technologies also have problems such as inaccurate data processing and low detection efficiency. In ultra-high voltage lines, changing electromagnetic fields may interfere with sensors and other acquisition systems, resulting in unstable or distorted voltage signals. Such data inaccuracies will not only affect the insulator detection results, but also reduce the reliability of the entire detection system.

[0007] In summary, the existing insulator detection technology has obvious deficiencies and limitations when facing the complex environment and strict requirements of ultra-high voltage transmission lines. Therefore, there is an urgent need for a new insulator detection method that can accurately and efficiently judge the performance status of insulators and ensure the safe and stable operation of transmission lines. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide an insulator detection digital filtering algorithm and system combined with Gaussian regression to solve the accuracy problem of insulator fault detection in transmission lines, especially the technical problem that insulators in ultra-high voltage lines are gradually deteriorated by electric fields, mechanical loads and external climatic conditions, resulting in reduced insulation performance; and to overcome the technical limitations of the insulator detection method in the prior art that it is difficult to accurately judge the true state of the insulator due to the influence of temperature and humidity changes and the degree of insulator pollution.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: a digital filtering algorithm for insulator detection combined with Gaussian regression, comprising the following steps: Step 1: Use the IIR digital filter to process the voltage signal collected by the sensor in the ultra-high voltage line in real time to eliminate electromagnetic field interference; Step 2: Use the Gaussian process regression (GPR) algorithm to perform curve fitting on the distributed voltage value of the insulator string, obtain the predicted normal distribution voltage value, and compare the nonlinear distribution voltage value with the predicted curve; Step 3: The processed data is predicted and evaluated through the GPR model, converted into resistance values ​​based on the fitting error and curve ratio, and faulty insulators are identified.

[0010] In a preferred solution, the digital filter in Step 1 is a bandpass filter for filtering out power frequency interference and harmonic interference.

[0011] In the preferred solution, the principle of the IIR digital filter filtering out power frequency interference and harmonic interference is as follows: Step 1.1: The system input-output relationship of the IIR digital filter is expressed by a differential equation: (1) In the formula, is the filter at time The output value of is the filter at time The input value of Indicates input signal In time The value of the current moment The input signal is delayed time unit; Indicates input signal In time The value of the current moment The input signal is delayed time unit; is the feedforward coefficient, which is multiplied by the past value of the input signal, from arrive , is the order of the feedforward coefficient; is the feedback factor, which is multiplied by the past value of the output signal, from arrive , is the order of the feedback coefficient; Step 1.2: Corresponding transfer function The expression is: (2) In the formula, The transfer function represents the input-output relationship of the filter. Domain expression; represents the feed-forward part of the filter; represents the feedback part of the filter; Indicates that the system delays the input signal response after time units; Step 1.3: Put the transfer function use Replace and define , , , , the actual frequency response of the IIR digital filter is expressed as follows: (3) In the formula, is the transfer function of the system; is the representation of the input matrix in the frequency domain; is the representation of the state matrix in the frequency domain; Transpose of the output frequency response vector; is the frequency response vector of the input; is the feedforward coefficient vector; is the feedback coefficient vector; The frequency response of the IIR digital filter is expressed as: (4) In the formula, is the amplitude-frequency characteristic function of the filter, which represents the frequency response of the filter at frequency; is the frequency response amplitude, which indicates the gain of the filter to the signal at the frequency; is the phase factor of the frequency response, which represents the effect of the filter on the signal phase at the frequency; Step 1.4: When the desired frequency response is The frequency response error of the IIR digital filter is defined as for: (5) In the formula, is the frequency response error, indicating the frequency The difference between the actual response of the filter and the expected response; The actual frequency response of the IIR filter is the filter in the complex exponential The value of the transfer function at ; The desired frequency response is the frequency response characteristic that is expected to be achieved when designing the filter; For frequency In the design of digital filters, The value range is or normalized to the Nyquist frequency ; This algorithm needs to remove the maximum and minimum values, and then find the average value of all remaining elements, so it is set here , to ensure that there are still at least 2 numbers to average after removing the maximum and minimum numbers, according to the sorting results of the bubble algorithm, remove the maximum and minimum numbers, and take the average of the remaining elements. The specific algorithm is: (6) In the formula, Represents each element in the data set N; is the total number of elements in the data set N; That is, it is the final average value obtained after removing the maximum and minimum values. Since the outliers are removed, the final average value obtained by this method is a value with higher credibility. Corresponding to the specific application environment, it is equivalent to removing some disturbances and abnormal signals.

[0012] In the preferred solution, the GPR algorithm in Step 2 uses a Gaussian kernel function for curve fitting to improve fitting accuracy. Compared with the commonly used BP and SVM prediction algorithms, the GPR algorithm has the advantages of fewer adjustable parameters, strong generalization ability, and easy implementation, and the prediction results have probabilistic significance.

[0013] In a preferred embodiment, the Gaussian process regression is a regression model based on the Gaussian kernel function (normal distribution), which is used to fit and model the pattern with noise in the data, and its expression is: (7) In the formula, is the Gaussian kernel function; is the independent variable of the Gaussian kernel function; is the amplitude coefficient; is the mean, i.e. the peak position; is the standard deviation, which is the width or spread of the distribution.

[0014] In the preferred scheme, the Gaussian process regression is a probability-based method that establishes a nonlinear relationship between input and output variables and provides a confidence interval for each predicted value. Its specific implementation process is to first assume the prior probability that the sample obeys the Gaussian process, then calculate the posterior probability in combination with Bayesian theory, and then obtain the final prediction model by calculating the hyperparameters.

[0015] In a preferred solution, the prediction and evaluation in Step 3 includes calculating the error between the predicted value and the measured value, and setting an error threshold. When the error exceeds the threshold, it is determined that the insulator is faulty.

[0016] In the preferred scheme, the specific steps of the GPR model predicting and evaluating the processed data are: after the signal is processed by a digital filter, it is fitted into a Gaussian distribution, and the electrical or physical signal of the porcelain insulator is collected by a sensor; the noise is removed by a digital filter to retain useful signal characteristics; Gaussian regression fitting is performed on the filtered signal to evaluate the fitting error and the change of the fitting curve; and whether there is a significant anomaly based on the fitting result, thereby identifying damage or other faults of the zero-value porcelain insulator.

[0017] The preferred solution also includes a step of visually displaying the processed data so as to intuitively observe the working status and fault conditions of the insulator.

[0018] An insulator detection system using an insulator detection digital filtering algorithm combined with Gaussian regression as described above, comprising a sensor, a data processing unit and a fault identification unit, wherein the sensor is used to collect voltage signals in an ultra-high voltage line, the data processing unit is used to perform the digital filtering and GPR algorithm processing, and the fault identification unit is used to identify faulty insulators based on the processing results.

[0019] The insulator detection digital filtering algorithm and system combined with Gaussian regression provided by the present invention have the following beneficial effects: 1. The present invention solves the problem of accuracy in insulator fault detection in transmission lines, especially the technical problem that insulators in ultra-high voltage lines gradually deteriorate due to the influence of electric fields, mechanical loads and external climatic conditions, resulting in reduced insulation performance; 2. The present invention introduces a Gaussian regression algorithm. Compared with the traditional method based on the judgment standard of fixed values, it does not consider the influence of the operating environment and the degree of insulator pollution. It can more accurately fit the distributed voltage value of the insulator string and adapt to various complex nonlinear relationships. 3. The present invention pre-processes the collected signals through a digital filter, effectively eliminating electromagnetic field interference and noise, and improving the accuracy and reliability of the data; 4. The present invention adopts dynamic reference voltage distribution generation, and generates a new reference voltage distribution as a judgment standard each time a measurement is made, which is not affected by the number of insulators and the operating environment, and avoids deviations between voltage distributions or rough voltage distributions; 5. The present invention uses Gaussian regression algorithm and digital filter to improve the accuracy of insulator fault detection compared with traditional methods, which are difficult to accurately judge the fault condition of the insulator when facing the influence of the operating environment and the degree of insulator pollution; 6. The present invention effectively solves the problem of data processing complexity and stability caused by the voltage signal collected in the ultra-high voltage line being easily disturbed by the electromagnetic field, resulting in data instability or distortion, through the preprocessing of the digital filter; 7. The present invention improves the accuracy and stability of insulator fault detection through the combined application of Gaussian regression algorithm and digital filter, more accurately identifies faulty insulators, and reduces the risk of misjudgment and missed judgment; 8. The present invention is not affected by the temperature and humidity of the operating environment and the degree of insulator pollution, can adapt to complex operating environments, and can work stably in various complex environments; 9. The present invention reduces manual inspection and maintenance costs by improving detection efficiency and accuracy, reduces the risk of power outages caused by insulator failures, ensures the safety of personnel during inspections, and improves the economic efficiency of transmission line maintenance; 10. The design of the robot for live-line operation of the present invention eliminates the potential safety hazard of manual high-altitude inspection. The automated detection greatly reduces the risk of manual contact with high-voltage lines, thus ensuring the safety of personnel during the inspection process. 11. The present invention effectively eliminates noise and abnormal signals through IIR digital filter and median average filtering algorithm, improves the resistance measurement accuracy of insulators, can accurately detect low-value or zero-value insulators, and ensures the safe operation of transmission lines; 12. Compared with traditional regression models (such as linear regression), which require data to conform to a certain distribution form (such as a linear relationship), the Gaussian process regression of the present invention is a parameter-free model that can adapt to various complex nonlinear relationships through the flexibility of the kernel function; 13. Compared with the deep learning model, the Gaussian process regression of the present invention does not require a large amount of training data, but can still make effective predictions and provide a suitable uncertainty measure to help reduce the cost of data collection; 14. The present invention reduces manual inspection and maintenance costs by improving detection efficiency and accuracy, reduces the risk of power outages caused by insulator failures, and improves the overall maintenance economy of transmission lines; 15. The Gaussian process regression method adopted in the present invention can achieve high-precision fault prediction under finite samples. Compared with traditional statistical methods, it significantly improves the intelligence level of insulator fault diagnosis and provides strong technical support for the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further described below in conjunction with the accompanying drawings and implementation examples: Figure 1 This is a diagram of the voltage fluctuation distribution under actual measurement of the present invention; Figure 2 It is a schematic diagram of Gaussian regression of the present invention; Figure 3 It is a flow chart of the Gaussian regression processing algorithm of the present invention; Figure 4 It is a schematic diagram of the actual curve of the present invention and the curve after algorithm fitting; Figure 5 A schematic diagram of converting the predicted distributed voltage result of the present invention into insulation resistance through a ratio; Figure 6 It is the overall flow chart of the detection system of the present invention. DETAILED DESCRIPTION

[0021] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments: Example 1 An insulator detection digital filtering algorithm combined with Gaussian regression includes the following steps: Step 1: Use an IIR digital filter (Infinite Impulse Response) to process the voltage signal collected by the sensor in the ultra-high voltage line in real time to eliminate electromagnetic field interference; Step 2: Use the Gaussian process regression (GPR) algorithm to perform curve fitting on the distributed voltage value of the insulator string, obtain the predicted normal distribution voltage value, and compare the nonlinear distribution voltage value with the predicted curve; Step 3: The processed data is predicted and evaluated through the GPR model, converted into resistance values ​​based on the fitting error and curve ratio, and faulty insulators are identified.

[0022] In this embodiment, the digital filter in Step 1 is a bandpass filter, which is used to filter out power frequency interference and harmonic interference.

[0023] Furthermore, the principle of the IIR digital filter to filter out power frequency interference and harmonic interference is as follows: Step 1.1: The system input-output relationship of the IIR digital filter is expressed by a differential equation: (1) In the formula, is the filter at time The output value of is the filter at time The input value of Indicates input signal In time The value of the current moment The input signal is delayed time unit; Indicates input signal In time The value of the current moment The input signal is delayed time unit; is the feedforward coefficient, which is multiplied by the past value of the input signal, from arrive , is the order of the feedforward coefficient; is the feedback factor, which is multiplied by the past value of the output signal, from arrive , is the order of the feedback coefficient; Step 1.2: Corresponding transfer function The expression is: (2) In the formula, The transfer function represents the input-output relationship of the filter. Domain expression; represents the feed-forward part of the filter; represents the feedback part of the filter; Indicates that the system delays the input signal response after time units; Step 1.3: Put the transfer function use Replace and define , , , , the actual frequency response of the IIR digital filter is expressed as follows: (3) In the formula, is the transfer function of the system; is the representation of the input matrix in the frequency domain; is the representation of the state matrix in the frequency domain; Transpose of the output frequency response vector; is the frequency response vector of the input; is the feedforward coefficient vector; is the feedback coefficient vector; The frequency response of the IIR digital filter is expressed as: (4) In the formula, is the amplitude-frequency characteristic function of the filter, which represents the frequency response of the filter at frequency; is the frequency response amplitude, which indicates the gain of the filter to the signal at the frequency; is the phase factor of the frequency response, which represents the effect of the filter on the signal phase at the frequency; Step 1.4: When the desired frequency response is The frequency response error of the IIR digital filter is defined as for: (5) In the formula, is the frequency response error, indicating the frequency The difference between the actual response of the filter and the expected response; The actual frequency response of the IIR filter is the filter in the complex exponential The value of the transfer function at ; The desired frequency response is the frequency response characteristic that is expected to be achieved when designing the filter; For frequency In the design of digital filters, The value range is or normalized to the Nyquist frequency ; This algorithm needs to remove the maximum and minimum values, and then find the average value of all remaining elements, so it is set here (Ensure that there are still at least 2 numbers to average after removing the maximum and minimum numbers). According to the sorting result of the bubble algorithm, remove the maximum and minimum numbers, and take the average of the remaining elements. The specific algorithm is: (6) In the formula, Represents each element in the data set N; is the total number of elements in the data set N; That is, it is the final average value obtained after removing the maximum and minimum values. Since the outliers are removed, the final average value obtained by this method is a value with higher credibility. Corresponding to the specific application environment, it is equivalent to removing some disturbances and abnormal signals.

[0024] Furthermore, the GPR algorithm in Step 2 uses a Gaussian kernel function for curve fitting to improve fitting accuracy. Compared with the commonly used BP (Back Propagation) and SVM (Support Vector Machine) prediction algorithms, the GPR algorithm has the advantages of fewer adjustable parameters, strong generalization ability, and easy implementation, and the prediction results have probabilistic significance.

[0025] Furthermore, the Gaussian process regression is a regression model based on the Gaussian kernel function (normal distribution), which is used to fit and model the pattern with noise in the data, and its expression is: (7) In the formula, is the Gaussian kernel function; is the independent variable of the Gaussian kernel function; is the amplitude coefficient; is the mean, i.e. the peak position; is the standard deviation, which is the width or spread of the distribution.

[0026] Furthermore, the Gaussian process regression is a probability-based method that establishes a nonlinear relationship between input and output variables and provides a confidence interval for each predicted value. Its specific implementation process is to first assume the prior probability that the sample obeys the Gaussian process, then calculate the posterior probability in combination with Bayesian theory, and then obtain the final prediction model by calculating the hyperparameters.

[0027] Furthermore, the prediction and evaluation in Step 3 include calculating the error between the predicted value and the measured value, and setting an error threshold. When the error exceeds the threshold, it is determined that the insulator is faulty.

[0028] Furthermore, the specific steps of the GPR model for predicting and evaluating the processed data are: after the signal is processed by a digital filter, it is fitted into a Gaussian distribution, and the electrical or physical signal of the porcelain insulator is collected by a sensor; the noise is removed by a digital filter to retain useful signal characteristics; Gaussian regression fitting is performed on the filtered signal to evaluate the fitting error and the change of the fitting curve; and whether there is a significant anomaly based on the fitting result, thereby identifying damage or other faults of the zero-value porcelain insulator.

[0029] Furthermore, the method also includes a step of visually displaying the processed data so as to intuitively observe the working status and fault conditions of the insulator.

[0030] Example 2 In another preferred embodiment, based on the above-mentioned embodiment 1, an insulator detection system that uses the insulator detection digital filtering algorithm combined with Gaussian regression described in embodiment 1 includes a sensor, a data processing unit and a fault identification unit, wherein the sensor is used to collect voltage signals in the ultra-high voltage line, the data processing unit is used to perform the digital filtering and GPR algorithm processing, and the fault identification unit is used to identify faulty insulators based on the processing results.

[0031] In practical applications, the system can monitor the working status of insulators in real time, detect and alarm faulty insulators in a timely manner, and provide strong guarantees for the safe operation of ultra-high voltage lines.

[0032] Example 3 In another preferred embodiment, based on the above embodiments 1, 2, and 3, the technical solution of a digital filtering algorithm for insulator detection combined with Gaussian regression of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] 1. Main features of zero-value insulator string: Porcelain suspension insulators are mainly composed of porcelain parts, iron caps, steel feet and zinc sleeves. The adhesives include cement, quartz stone and porcelain sand. The porcelain parts have the largest insulation resistivity and are the main material components for electrical insulation in the insulator. Its circuit equivalent diagram can be regarded as a large capacitor, which is composed of body resistance and surface resistance in parallel.

[0034] Low zero-value insulators are mainly caused by factors such as manufacturing process, working environment, and physical and chemical reactions. In actual operation, insulators in transmission lines are affected by various factors (natural environment, mechanical stress of the line, etc.), resulting in insulators with insulation degradation in the entire insulator string, which will further reduce the reserved margin of the insulator string and even greatly reduce the insulation performance of the entire line, thereby further accelerating the rate of low zero-value insulator formation and causing serious damage to the line.

[0035] The voltage drop of the insulator string of UHV transmission lines generally presents a U-shaped curve. When there are zero-value insulators in the insulator string, the voltage distribution on each insulator changes, resulting in an increase in the withstand voltage of each insulator. If the body resistance of the zero-value insulator is zero, the surface voltage drop is also zero, which is equivalent to a conductor and directly affects the leakage current in the insulator string. However, due to the capacitance differences of a single insulator string on the tower, interference and noise between the three phases of AC, the voltage distribution may also be offset. In addition, the temperature and humidity of the operating environment and the degree of contamination of the insulators cause the U-shaped distribution to appear sawtooth, making it almost impossible to apply these judgment criteria, such as the attached Figure 1 shown.

[0036] 2. Digital filtering to collect data: In ultra-high voltage lines, changing electromagnetic fields may interfere with sensors and other collection systems, thereby affecting the accuracy and continuous stability of the data collection process, causing the collected voltage signal to be unstable or severely distorted. For this reason, the accuracy of the data is ensured through the digital filter process, which greatly improves the reliability of this algorithm in insulator fault detection.

[0037] For IIR digital filters, the difference equation representing the input-output relationship of the system can be expressed as follows: (1) The corresponding transfer function expression is: (2) In formula (2), use Replace and define , , , , the actual frequency response of the IIR digital filter is expressed as follows: (3) In the formula, is the transfer function of the system; is the representation of the input matrix in the frequency domain; is the representation of the state matrix in the frequency domain; Transpose of the output frequency response vector; is the frequency response vector of the input; is the feedforward coefficient vector; is the feedback coefficient vector; The frequency response of the IIR digital filter can also be expressed as: (4) In the formula, is the amplitude-frequency characteristic function of the filter, which represents the frequency response of the filter at frequency; is the frequency response amplitude, which indicates the gain of the filter to the signal at the frequency; is the phase factor of the frequency response, which represents the effect of the filter on the signal phase at the frequency; When the desired frequency response is The frequency response error of the IIR digital filter is defined as for: (5) This algorithm needs to remove the maximum and minimum values, and then find the average of all remaining elements. Therefore, it is set here (to ensure that there are still at least 2 numbers to average after removing the maximum and minimum numbers). According to the sorting results of the bubble algorithm, the maximum and minimum numbers are removed, and the average of the remaining elements is taken. The specific algorithm is: (6) In the formula, That is, it is the final average value obtained after removing the maximum and minimum values. Since the outliers are removed, the final average value obtained by this method is a value with higher credibility. Corresponding to the specific application environment, it is equivalent to removing some disturbances and abnormal signals.

[0038] 3. Gaussian regression algorithm processing: Gaussian process regression is a non-parametric model proposed by Williams et al., and its essence is kernel function learning with probabilistic meaning. At first, when neural networks were popular, Neal et al. found that when using neural networks to solve practical problems, the number of hidden nodes should not be limited. When the number of hidden nodes is set to infinity, the Gaussian prior distribution of the network weights approaches a Gaussian process, so the parameter research of the neural network is gradually turned to the covariance matrix in the more direct Gaussian process. The idea of ​​​​Gaussian process regression when making predictions is: Simply put, Gaussian process regression is a kind of "regression". The so-called regression is to fit a law to seemingly isolated things. Unlike traditional regression methods, Gaussian process regression can fit nonlinear laws, and the kernel function plays a key role here.

[0039] Introducing Gaussian process regression into insulator detection helps model the changing trend of insulation voltage distribution and thus identify outliers. Gaussian regression is a regression model based on Gaussian function (normal distribution), which is usually used to fit and model patterns with noise in data. Its common form is: (7) Compared with commonly used prediction algorithms such as BP and SVM, GPR has the advantages of fewer adjustable parameters, strong generalization ability, easy implementation, and the prediction results have probabilistic significance.

[0040] As attached Figure 2As shown in the figure, the curve is the mean function curve of the prediction model. The circles above represent known data points. Each point corresponds to the mean and variance of the respective Gaussian distribution. The area above and below the curve represents the range of two standard deviations of the Gaussian distribution. Then, the conditional probability mean and variance expressions of the prediction points are obtained using Bayesian theory. In short, Gaussian process regression first assumes the prior probability that the sample obeys the Gaussian process, then calculates the posterior probability in combination with Bayesian theory, and then obtains the final prediction model by calculating the hyperparameters. The process is shown in the attached figure. Figure 3 shown.

[0041] 4. Model prediction: After the signal is processed by the digital filter, the data can be fitted into the form of Gaussian distribution, and the electrical or physical signal of the porcelain insulator can be collected through a suitable sensor. The noise is removed by the digital filter to retain the useful signal characteristics. Gaussian regression fitting is performed on the filtered signal to evaluate the fitting error and the change of the fitting curve. According to the fitting results, whether there is a significant anomaly is detected, so as to identify the damage or other faults of the zero-value porcelain insulator.

[0042] Attached Figure 4 The solid line in the figure represents the voltage distribution measured when the faulty insulator is located at position 9 of the transmission line insulator string. After the curve is approximated by the Gaussian regression method, the result is shown as the dotted line in the figure. However, the faulty insulator at position 9 causes the approximation curve to have a small value centered at position 9, which deviates from the voltage distribution of the solid insulator. Therefore, in order to solve these problems related to the approximation value, any measurement data that deviates greatly from the reference value should be excluded when performing curve fitting.

[0043] Through the previous steps, the ratio of the measured actual distribution voltage to the predicted distribution voltage can be obtained. The calculated ratio is then converted into an estimated insulation value of the insulator under low humidity and low pollution conditions. Figure 5 As shown, in this conversion process, the ratio of the distribution voltage to the insulation resistance obtained experimentally is used. If the ratio is 100%, the insulation resistance value is of the order of magnitude M ohms. Finally, the insulation resistance value of each insulator can be obtained by the ratio of the distribution voltage to the predicted distribution voltage. Under low humidity and low pollution conditions, the Gaussian regression model can more accurately estimate the resistance value of the insulator. For the generated data, the model will give the corresponding predicted value and provide a confidence interval for the prediction. If the predicted value of a data point is outside the range of the confidence interval, it can be considered an outlier, which may represent a zero or low value insulator. Finally, if the estimated resistance value is lower than the preset threshold, the insulator will be judged as faulty.

[0044] Application of Gaussian regression model: After data preprocessing, Gaussian regression can be used to model the relationship between data. Gaussian process regression is a probability-based method that can establish nonlinear relationships between input and output variables and provide confidence intervals for each predicted value. The process is shown in the attached figure. Figure 6 .

[0045] Example 4 In another preferred embodiment, based on the above embodiments 1 and 23, a digital filtering algorithm for insulator detection combined with Gaussian regression of the present invention will be described in detail in combination with specific embodiments.

[0046] First, voltage signals are collected in real time through sensors installed on the UHV lines. These signals contain the distributed voltage information of the insulator strings, but are also mixed with noise such as power frequency interference and harmonic interference.

[0047] Next, the collected signal is processed in real time using a digital filter. In this embodiment, the digital filter is selected as a bandpass filter, which can filter out power frequency interference and harmonic interference and retain the distributed voltage signal of the insulator string.

[0048] Then, the processed signal is input into the Gaussian process regression (GPR) algorithm. The GPR algorithm uses the Gaussian kernel function to perform curve fitting on the distributed voltage value of the insulator string. Through the fitting of the GPR algorithm, the voltage distribution curve of the insulator string under prediction can be obtained. The resistance value of each insulator is calculated according to the fitting error and the curve ratio, and the faulty insulator is then identified.

[0049] Finally, based on the prediction and evaluation results of the GPR model, the faulty insulator is identified; specifically, by calculating the error between the predicted value and the measured value and setting an error threshold, when the error exceeds the threshold, the insulator is determined to be faulty.

[0050] In order to further verify the effect of the present invention, the following experiment was conducted: multiple insulators were set on a simulated ultra-high voltage line, some of which were set to a faulty state; the algorithm of the present invention was used to detect the insulators, and all faulty insulators were successfully identified with a low false alarm rate.

[0051] Example 5 In another preferred embodiment, based on the above embodiments 1, 2, 3, and 4, this embodiment optimizes the GPR model in the algorithm; specifically, in the fitting process of the GPR algorithm, a regularization term is introduced to avoid the occurrence of overfitting. At the same time, the parameters of the Gaussian kernel function are adjusted to improve the fitting accuracy and generalization ability.

[0052] Through experimental verification, the accuracy of the optimized GPR model in insulator detection has been further improved, and it is more robust to noise.

[0053] In the preferred scheme, in the preferred scheme, the GPR algorithm in Step 2 uses a Gaussian kernel function for curve fitting to improve fitting accuracy. Compared with the commonly used BP and SVM prediction algorithms, the GPR algorithm has the advantages of fewer adjustable parameters, strong generalization ability, and easy implementation, and the prediction results have probabilistic significance; the above settings not only simplify the model adjustment process, but also significantly improve the stability and reliability of the prediction; in practical applications, the scheme has demonstrated excellent performance, especially when processing nonlinear and high-dimensional data, its prediction accuracy and efficiency are better than traditional methods.

[0054] In the preferred scheme, the Gaussian process regression is a regression model based on the Gaussian kernel function (normal distribution), which is used to fit and model noisy patterns in data; the above setting can effectively process complex data relationships in high-dimensional space, and achieve high-precision prediction and uncertainty estimation through nonlinear mapping of input data, and is suitable for regression analysis problems in various fields.

[0055] In the preferred scheme, the Gaussian process regression is a probability-based method that establishes a nonlinear relationship between input and output variables and provides a confidence interval for each predicted value. Its specific implementation process is to first assume the prior probability that the sample obeys the Gaussian process, then calculate the posterior probability in combination with Bayesian theory, and then obtain the final prediction model by calculating the hyperparameters; the above settings can effectively handle the nonlinear regression problem of small sample data sets, while taking into account the uncertainty of the data, and improving the accuracy and reliability of the prediction results; in addition, by adjusting the hyperparameters, the model performance can be further optimized to meet the needs of different application scenarios.

[0056] In the preferred solution, the prediction and evaluation in Step 3 include calculating the error between the predicted value and the measured value, and setting an error threshold. When the error exceeds the threshold, it is determined that the insulator is faulty. The above settings can realize real-time monitoring and accurate early warning of the insulator status, and improve the efficiency and accuracy of fault detection. At the same time, the system can automatically record fault information to provide data support for subsequent analysis and maintenance.

[0057] In the preferred scheme, the specific steps of the GPR model predicting and evaluating the processed data are: after the signal is processed by a digital filter, it is fitted into a Gaussian distribution, and the electrical or physical signal of the porcelain insulator is collected by a sensor; the noise is removed by a digital filter to retain useful signal characteristics; Gaussian regression fitting is performed on the filtered signal to evaluate the fitting error and the change of the fitting curve; based on the fitting results, it is detected whether there are significant anomalies, so as to identify damage or other faults of the zero-value porcelain insulator; the above settings can also monitor the state changes of the porcelain insulator in real time to ensure the safe operation of the power grid; if an abnormality is detected, the alarm system is immediately triggered to notify the maintenance personnel to deal with it in time, effectively preventing power accidents caused by porcelain insulator failures and improving the reliability and stability of power grid operation.

[0058] The preferred solution also includes a step of visually displaying the processed data so as to intuitively observe the working status and fault conditions of the insulators. The above settings not only improve the operation and maintenance efficiency, but also ensure that insulator faults can be discovered and handled in a timely manner through real-time data monitoring and alarm systems, effectively ensuring the stable operation and safety of the power system.

[0059] In summary, the insulator detection digital filtering algorithm combined with Gaussian regression proposed in the present invention solves the accuracy problem of insulator fault detection in transmission lines, especially the technical problem that insulators in ultra-high voltage lines are gradually deteriorated by electric fields, mechanical loads and external climatic conditions, resulting in reduced insulation performance; it overcomes the technical limitation that the insulator detection method in the prior art is affected by temperature and humidity changes and the degree of insulator pollution, making it difficult to accurately judge the true state of the insulator; the present invention innovatively introduces the Gaussian regression algorithm, fits the distributed voltage value of the insulator string into a curve, and converts it into a resistance value to approximate the actual value. This method can more accurately reflect the true state of the insulator and improve the accuracy of the detection. The accuracy of detection is improved; the application of digital filters effectively eliminates these interferences and noises, improves the accuracy and continuous stability of data, and the application of this digital filtering technology in insulator detection is novel and helps to improve the reliability and accuracy of detection; the technical solution of the present invention generates a new reference voltage distribution as a judgment standard each time a measurement is made, which is not affected by these factors, thereby improving the accuracy and stability of detection. This method for generating a dynamic reference voltage distribution is novel and helps to improve the flexibility and adaptability of insulator detection; Gaussian regression algorithms are usually used to fit and model patterns with noise in data, and the present invention creatively applies it to insulator detection, and The distributed voltage curve of the insulator string is combined and the uncertainty of the prediction is quantified to identify potential faults; this method not only improves the accuracy of detection, but also can adapt to various complex nonlinear relationships, such as U-shaped or sawtooth curves, demonstrating the unique advantages of the Gaussian regression algorithm in the field of insulator detection; the present invention combines the digital filter with the Gaussian regression algorithm to form a new insulator detection method, the digital filter is used to pre-process the collected voltage signal to eliminate interference and noise; and the Gaussian regression algorithm is used to model and fit the processed signal to identify outliers; this combination not only improves the accuracy and stability of the data, but also enhances the reliability and precision of the detection, demonstrating creative technical integration ... The introduction of technical means such as Gaussian regression algorithm and digital filter successfully solves the technical problem that the traditional insulator detection method is difficult to accurately judge the fault condition of the insulator in the face of complex operating environments, and can work stably in various complex environments, and is not affected by the temperature and humidity of the operating environment and the degree of insulator pollution; the present invention applies digital filtering technology to insulator signal acquisition and processing, effectively improves the accuracy and stability of the signal, introduces Gaussian regression algorithm, adapts to the complex changes of insulator voltage distribution, and improves the accuracy and adaptability of fault detection; and then through algorithm optimization and model prediction, it realizes real-time monitoring of insulator performance and fault warning, providing a strong guarantee for the safe operation of transmission lines;In addition, the method also has the ability of self-learning, can continuously optimize the detection model, and adapt to long-term changes such as insulator aging, ensuring the continued advancement and practicality of the detection technology, and setting a new benchmark for the intelligent transformation of the power industry. ;

Claims

1. A digital filtering algorithm for insulator detection combined with Gaussian regression, characterized in that: The method comprises the following steps: Step 1: Use the IIR digital filter to process the voltage signal collected by the sensor in the ultra-high voltage line in real time to eliminate electromagnetic field interference; Step 2: Use the Gaussian process regression (GPR) algorithm to perform curve fitting on the distributed voltage value of the insulator string, obtain the predicted normal distribution voltage value, and compare the nonlinear distribution voltage value with the predicted curve; Step 3: The processed data is predicted and evaluated through the GPR model, converted into resistance values ​​based on the fitting error and curve ratio, and faulty insulators are identified.

2. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 1 is characterized in that: The digital filter in Step 1 is a bandpass filter, which is used to filter out power frequency interference and harmonic interference.

3. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 2 is characterized in that: The principle of the IIR digital filter to filter out power frequency interference and harmonic interference is as follows: Step 1.1: The system input-output relationship of the IIR digital filter is expressed by a differential equation: （1）； In the formula, is the filter at time The output value of is the filter at time The input value of Indicates input signal In time The value of the current moment The input signal is delayed time unit; Indicates input signal In time The value of the current moment The input signal is delayed time unit; is the feedforward coefficient, which is multiplied by the past value of the input signal, from arrive , is the order of the feedforward coefficient; is the feedback factor, which is multiplied by the past value of the output signal, from arrive , is the order of the feedback coefficient; Step 1.2: Corresponding transfer function The expression is: （2）； In the formula, The transfer function represents the input-output relationship of the filter. Domain expression; represents the feed-forward part of the filter; represents the feedback part of the filter; Indicates that the system delays the input signal response after time units; Step 1.3: Put the transfer function use Replace and define , , , , the actual frequency response of the IIR digital filter is expressed as follows: （3）； In the formula, is the transfer function of the system; is the representation of the input matrix in the frequency domain; is the representation of the state matrix in the frequency domain; Transpose of the output frequency response vector; is the frequency response vector of the input; is the feedforward coefficient vector; is the feedback coefficient vector; The frequency response of the IIR digital filter is expressed as: （4）； In the formula, is the amplitude-frequency characteristic function of the filter, which represents the frequency response of the filter at frequency; is the frequency response amplitude, which indicates the gain of the filter to the signal at the frequency; is the phase factor of the frequency response, which represents the effect of the filter on the signal phase at the frequency; Step 1.4: When the desired frequency response is The frequency response error of the IIR digital filter is defined as for: （5）； In the formula, is the frequency response error, indicating the frequency The difference between the actual response of the filter and the expected response; The actual frequency response of the IIR filter is the filter in the complex exponential The value of the transfer function at ; The desired frequency response is the frequency response characteristic that is expected to be achieved when designing the filter; For frequency In the design of digital filters, The value range is or normalized to the Nyquist frequency ; The algorithm needs to remove the maximum and minimum values, and then find the average of all remaining elements. According to the sorting results of the bubble algorithm, the maximum and minimum numbers are removed, and the average of the remaining elements is taken. The specific algorithm is: （6）； In the formula, Represents each element in the data set N; is the total number of elements in the data set N; That is, it is the final average value obtained after removing the maximum and minimum values. Since the outliers are removed, the final average value obtained by this method is a value with higher credibility. Corresponding to the specific application environment, it is equivalent to removing some disturbances and abnormal signals.

4. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 3 is characterized in that: The GPR algorithm in Step 2 uses a Gaussian kernel function for curve fitting to improve fitting accuracy. Compared with the commonly used BP and SVM prediction algorithms, the GPR algorithm has the advantages of fewer adjustable parameters, strong generalization ability, and easy implementation, and the prediction results have probabilistic significance.

5. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 4, characterized in that: The Gaussian process regression is a regression model based on the Gaussian kernel function, which is used to fit and model the pattern with noise in the data, and its expression is: （7）； In the formula, is the Gaussian kernel function; is the independent variable of the Gaussian kernel function; is the amplitude coefficient; is the mean, i.e. the peak position; is the standard deviation, which is the width or spread of the distribution.

6. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 5 is characterized in that: The Gaussian process regression is a probability-based method that establishes a nonlinear relationship between input and output variables and provides a confidence interval for each predicted value. Its specific implementation process is to first assume the prior probability that the sample obeys the Gaussian process, then calculate the posterior probability in combination with Bayesian theory, and then obtain the final prediction model by calculating the hyperparameters.

7. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 6 is characterized in that: The prediction and evaluation in Step 3 include calculating the error between the predicted value and the measured value, and setting an error threshold. When the error exceeds the threshold, it is determined that the insulator is faulty.

8. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 7 is characterized in that: The specific steps of the GPR model for predicting and evaluating the processed data are: after the signal is processed by a digital filter, it is fitted into a Gaussian distribution form, and the electrical or physical signal of the porcelain insulator is collected by a sensor; the noise is removed by a digital filter to retain useful signal characteristics; Gaussian regression fitting is performed on the filtered signal to evaluate the fitting error and the change of the fitting curve; and whether there is a significant anomaly based on the fitting result, thereby identifying damage or other faults of the zero-value porcelain insulator.

9. The insulator detection digital filtering algorithm combined with Gaussian regression according to claim 8, characterized in that: The method also includes a step of visually displaying the processed data so as to directly observe the working status and fault conditions of the insulator.

10. An insulator detection system using the insulator detection digital filtering algorithm combined with Gaussian regression as claimed in any one of claims 1 to 9, characterized in that: It includes a sensor, a data processing unit and a fault identification unit. The sensor is used to collect voltage signals in an ultra-high voltage line. The data processing unit is used to execute digital filtering and GPR algorithm processing. The fault identification unit is used to identify faulty insulators according to the processing results.

Images