Insulator partial discharge positioning method

Abstract

The invention relates to the technical field of partial discharge source positioning, in particular to an insulator partial discharge positioning method, which comprises the following steps of: acquiring positioning signals generated by insulator partial discharge by adopting a polygonal three-dimensional sensor array; decomposing and denoising the positioning signal to obtain a denoised time domain positioning signal; obtaining a frequency domain positioning signal based on the time domain positioning signal, and carrying out weighted cross-correlation processing to obtain signal time delay; any four sensors are selected to be combined, objective function solutions are established, and each solution corresponds to one positioning initial value; all the positioning initial values are clustered through a clustering algorithm, and the clustering center of the cluster with the most elements in the clustered clusters is the partial discharge position of the insulator. According to the method, errors possibly occurring in the positioning process are reduced, the reliability of overall positioning is improved, and the solution method can find balance between global search and local search by dynamically adjusting the search speed of particles.

Description

Technical Field

[0001] The present invention relates to the technical field of partial discharge source localization, and particularly to a method for localizing partial discharge of insulators. Background Art

[0002] During long-term operation, insulators may have defects such as aging, fouling or damage, which will lead to a decline in the insulation performance of insulators and generate partial discharge phenomena. As an early signal of insulator faults, partial discharge has extremely important warning effects. If not dealt with or repaired in time, partial discharge may further develop and ultimately lead to flashover or breakdown of insulators. Therefore, accurately locating the position where partial discharge occurs in insulators is of great significance for preventing insulator faults.

[0003] Currently, commonly used methods for localizing partial discharge of insulators include infrared thermal imaging localization, ultrasonic localization, etc. The infrared thermal imaging method uses an infrared thermal imager to scan the surface of the insulator and determines the position of the fault point by analyzing the generated thermal image. This method has a simple principle and is easy to implement, but in practical applications, it is easily affected by factors such as sunlight, rain, snow, and strong winds that can cause temperature changes on the surface of the insulator.

[0004] The ultrasonic localization method determines the position of the partial discharge source based on the time difference between ultrasonic signals generated by partial discharge reaching different sensors or the beam propagation direction. However, since the ultrasonic signal attenuates rapidly in the air and the wave velocity is unstable, the sensitivity of ultrasonic detection is greatly reduced, and it often needs to be used in combination with other technologies. Therefore, there is an urgent need for an effective method to determine the position where partial discharge occurs in insulators. Summary of the Invention

[0005] To solve the technical problems in the above-mentioned background art, the present invention provides a method for localizing partial discharge of insulators.

[0006] The technical solution of the present invention is as follows: A method for localizing partial discharge of insulators, and the specific method includes the following steps: S1. Use a polygonal three-dimensional sensor array to collect the localization signals generated by partial discharge of the insulator.

[0007] S2. Adaptively decompose the collected positioning signals from high to low frequencies to obtain the intrinsic mode function components at different frequencies. Calculate the correlation coefficients of each intrinsic mode function component, and select the correlation coefficients greater than the correlation coefficient threshold as high-quality coefficients. Perform wavelet packet decomposition on the intrinsic mode function components corresponding to the high-quality coefficients, process the wavelet packet coefficients obtained from the wavelet packet decomposition using an adjustable threshold function, and reconstruct to obtain the denoised high-quality intrinsic mode function components. Synthesize all the denoised high-quality intrinsic mode function components to obtain the denoised time-domain positioning signal.

[0008] S3. Based on the denoised time-domain positioning signal, obtain the frequency-domain positioning signal through Fourier transform respectively. Perform weighted cross-correlation processing on any two frequency-domain positioning signals to obtain the signal time delay between the two sensors corresponding to any two frequency-domain positioning signals in the polygonal three-dimensional sensor array.

[0009] S4. Arbitrarily select four sensors in the polygonal three-dimensional sensor array for combination. Establish the objective function corresponding to each combination based on the signal time delay and the distance from the positioning signal to the sensor. Randomly initialize a group of particles, where each particle represents a potential solution of the objective function. Update the velocity and position of the particles by combining the particle swarm algorithm and the gravitational search algorithm. Repeat the update step and calculate the objective function value after each update. After the preset number of iterations, output the global optimal solution, which is the solution of the objective function, and each solution of the objective function corresponds to a positioning initial value.

[0010] S5. Use the clustering algorithm to cluster all the positioning initial values. The clustering center of the cluster with the most elements after clustering is the local discharge position of the insulator.

[0011] When updating the velocity of the particles by combining the particle swarm algorithm and the gravitational search algorithm, the specific update formula is: , where, and represent the velocities of the i-th particle at the (u + 1)-th and u-th iterations respectively, u is the current iteration number, represents the inertia weight, and are learning factors, and are random variables between [0, 1], is the historical best position of the i-th particle, is the global best position among all the particles, is the gravitational factor, represents the position of the i-th particle at the u-th iteration, is the acceleration of the i-th particle at the u-th iteration.

[0012] Gravitational factor The calculation formula is as follows; , where, is the minimum gravitational factor, is the maximum gravitational factor, is the preset number of iterations.

[0013] Inertia weight The specific calculation method is; , where, represents the maximum inertia weight, represents the minimum inertia weight, is the preset number of iterations, and u is the current number of iterations.

[0014] When updating the position of particles by combining the particle swarm algorithm and the gravitational search algorithm, the specific update formula is: , where, and respectively represent the positions of the i-th particle at the (u + 1)-th and u-th iterations, represents the velocity of the i-th particle at the (u + 1)-th iteration, and u is the current number of iterations.

[0015] Randomly select four sensors , , , to form a combination, and use as the reference sensor to establish the objective function as: , In the formula, is the speed of the positioning signal propagating in the air, and the value is , is the time difference between the time when the -th sensor receives the positioning signal and the time when the reference sensor receives the positioning signal, that is, the time delay, represents the distance from the power source to the -th sensor. When h takes 1, 2, 3, 4, , , and , , , , where, the distance from the power source to the The distance calculation formula for a sensor is as follows: , where x, y, and z are the coordinate values of the power source on the x, y, and z axes respectively, and x h , y h , z h are the coordinate values of the h-th sensor on the x, y, and z axes respectively.

[0016] In S3, weighted cross-correlation processing is performed based on any two frequency-domain positioning signals. The specific weighting function is; , where is the weighting function, is the angular frequency, and respectively represent the auto-power spectral densities of the two frequency-domain positioning signals, m and n respectively represent the serial numbers of the sensors, is the weighting coefficient, and .

[0017] The polygonal three-dimensional sensor array is a cone array, a sphere array, or a cylinder array, and the number N of sensors in the polygonal three-dimensional sensor array satisfies N ≥ 5.

[0018] The method for calculating the correlation coefficient in S2 is; , where represents the correlation coefficient of the i-th intrinsic mode function component, L is the length of the positioning signal, is the mean value of the positioning signal, is the mean value of the i-th intrinsic mode function component, represents the value of the positioning signal at time, represents the value of the i-th intrinsic mode function component at time, represents the time.

[0019] The number of clusters in S5 is selected according to the elbow method.

[0020] Compared with the prior art, the present invention has the following beneficial effects: Through the design of the N-element polygonal three-dimensional sensor array, signal focusing can be achieved in the spatial domain, thereby increasing the intensity of the received signal and correspondingly reducing the influence of background noise. Through the collaborative work of multiple sensors, the positioning accuracy can be effectively improved. When one sensor fails or is interfered with, other sensors can still provide effective information, thereby reducing the errors that may occur during the positioning process and enhancing the overall positioning reliability.

[0021] Perform adaptive decomposition on the positioning signal so that the intrinsic mode function components at different frequencies can be analyzed independently. By screening the high-quality intrinsic mode function components with larger correlation coefficients for wavelet packet decomposition, the main feature parts in the signal can be extracted more accurately, reducing redundant information. The adjustable threshold function can be flexibly adjusted according to the actual signal characteristics during the denoising process, significantly improving the denoising effect.

[0022] The weighting function takes into account the signals of multiple channels, so it can effectively utilize the correlation between multi-channel signals and enhance the effect of signal processing. By adjusting the weighting coefficient, the main components in the signal can be highlighted, especially in the case of low signal-to-noise ratio, the ability of signal delay estimation can be improved.

[0023] By dynamically adjusting the search speed of the particles, the flexibility of the algorithm in global search and local search can be improved, enabling the algorithm to widely explore the solution space in the initial stage of the search and concentrating on the fine search for the optimal solution in the later stage, enhancing the convergence speed and global optimization ability of the search algorithm.

[0024] By using the clustering algorithm to cluster multiple groups of initial positioning values obtained, the outliers can be effectively identified and excluded from the positioning results. By selecting the clustering center of the cluster with the most elements in the cluster as the final result of this positioning, the accuracy of positioning can be improved. Description of the Drawings

[0025] In the drawings: Figure 1 is the arrangement method of the five-element cone sensor array; Figure 2 is the positioning signal collected by the sensor; Figure 3 is the effect diagram after denoising; Figure 4 is the schematic diagram of signal delay; Figure 5 is the effect diagram of signal delay estimation. Detailed Implementation Manner

[0026] Example 1 The technical solution of the present invention is as follows: A method for local discharge positioning of insulators, the specific method includes the following steps: S1. Use a polygonal three-dimensional sensor array to collect the positioning signal generated by the local discharge of the insulator.

[0027] The polygonal three-dimensional sensor array is a cone array, a sphere array or a cylinder array, and the number N of sensors in the polygonal three-dimensional sensor array is N≥5.

[0028] In this embodiment, a five-element cone sensor array is used to collect the positioning signals generated by the partial discharge of the insulator. The arrangement of the five-element cone sensor array is as follows Figure 1 shown.

[0029] The sensor antenna selects a spherical antenna with a diameter of 8 cm, which can effectively detect the partial discharge positioning signals in the frequency band between 100 MHz and 1000 MHz. Compared with ultrasonic signals, the signals in this detection frequency band have good frequency response characteristics and propagation characteristics.

[0030] Each sensor is electrically connected to an oscilloscope through a feeder. The sampling rate of the oscilloscope is 10 GSa / s, and the positioning signals collected by the sensor are understood in combination with Figure 2 this.

[0031] Through the collaborative work of multiple sensors, the positioning accuracy can be effectively improved. When one sensor fails or is interfered with, other sensors can still provide effective information, thereby reducing the errors that may occur during the positioning process and enhancing the overall positioning reliability.

[0032] S2. The collected positioning signals are adaptively decomposed from high to low frequencies to obtain the intrinsic mode function components at different frequencies. Calculate the correlation coefficient of each intrinsic mode function component, and select the correlation coefficients greater than the correlation coefficient threshold as high-quality coefficients. Perform wavelet packet decomposition on the intrinsic mode function components corresponding to the high-quality coefficients, process the wavelet packet coefficients obtained by the wavelet packet decomposition using an adjustable threshold function, and reconstruct to obtain the denoised high-quality intrinsic mode function components. Synthesize all the denoised high-quality intrinsic mode function components to obtain the denoised time-domain positioning signal.

[0033] The collected positioning signals are adaptively decomposed from high to low frequencies to obtain the intrinsic mode function components at different frequencies. Calculate the correlation coefficient of each intrinsic mode function component. The calculation method of the correlation coefficient is as follows; , where represents the correlation coefficient of the i-th intrinsic mode function component, L is the length of the positioning signal, is the mean value of the positioning signal, is the mean value of the i-th intrinsic mode function component, represents the value of the positioning signal at time, represents the value of the i-th intrinsic mode function component at time, represents the time.

[0034] The closer the correlation coefficient of the intrinsic mode function component is to 1, the stronger the correlation between this intrinsic mode function component and the original data.

[0035] In this embodiment, the correlation coefficient threshold is set to 0.5, that is, the intrinsic mode function components with a correlation coefficient greater than 0.5 are considered high-quality intrinsic mode function components.

[0036] Perform wavelet packet decomposition on the selected high-quality intrinsic mode function components. The wavelet basis function selected for decomposition is the "db45" wavelet basis, and the decomposition level is selected as 3 layers; Use an adjustable threshold function for the wavelet packet coefficients obtained by decomposition for quantization processing. The adjustable threshold function has the following expression: , where is the adjustment factor, is the B-th wavelet packet coefficient at scale A obtained by wavelet packet decomposition, is the new wavelet packet coefficient obtained after processing with the threshold function, is the fixed threshold, and the mathematical expression is , is the variance of the noise, is the length of the positioning signal.

[0037] In this embodiment takes the value of 5.

[0038] Perform inverse wavelet packet transform on the obtained new wavelet packet coefficients to reconstruct the denoised high-quality intrinsic mode function components; Reconstruct all the denoised high-quality intrinsic mode function components to obtain the denoised time-domain positioning signal, Figure 3 which is the effect diagram after denoising by the method of the present invention.

[0039] S3. Based on the denoised time-domain positioning signal, obtain the frequency-domain positioning signal through Fourier transform respectively. Based on any two frequency-domain positioning signals, perform weighted cross-correlation processing to obtain the signal time delay between any two sensors corresponding to the two frequency-domain positioning signals in the polygon three-dimensional sensor array, as Figure 4 shown.

[0040] Suppose the two denoised time-domain positioning signals are respectively , , where t is the position of the positioning signal in the time domain, and m and n respectively represent the serial numbers of the sensors.

[0041] Use Fourier transform to transform the time-domain positioning signal into a frequency-domain positioning signal. The specific steps are as follows: , , Among them, represents the frequency-domain localization signal obtained after Fourier transform, represents the frequency-domain localization signal obtained after Fourier transform, is the imaginary unit, satisfying , is the angular frequency.

[0042] Then, the cross-power spectral density between the two frequency-domain localization signals and can be expressed as: , denotes taking the conjugate complex number of, is the cross-power spectral density.

[0043] Perform a weighting process on the obtained cross-power spectral density, and the specific weighting function is; , Among them, is the weighting function, is the angular frequency, and respectively represent the auto-power spectral densities of the two frequency-domain localization signals, m and n respectively represent the serial numbers of the sensors, is the weighting coefficient, and .

[0044] In this embodiment takes the value of 0.85, and the obtained generalized cross-power spectral density is .

[0045] Perform an inverse Fourier transform on the obtained generalized cross-power spectral density to obtain the generalized cross-correlation function in the time domain as: , is the time difference between the signals received by the two sensors, that is, the signal time delay. By performing peak detection on , the time delay value of the localization signal arriving between the two sensors can be obtained.

[0046] In a digital communication system, the time delay estimation method based on cross-correlation can estimate the time delay between the reference signal and the received signal, which is divided into integer-level and fractional-level time delay estimations. In order to improve the accuracy of the time delay estimation of the present invention, the cross-power spectral density of the signal is processed by means of frequency-domain weighting, combined with Figure 5Understand the signal delay estimation results.

[0047] Similarly, the time delay of the positioning signal received between any other two sensors can be obtained.

[0048] S4. Arbitrarily select four sensors from the polygon three-dimensional sensor array, establish an objective function corresponding to each combination based on the signal delay and the distance from the positioning signal to the sensor, randomly initialize a group of particles, where each particle represents a potential solution of the objective function, and use a combination of the particle swarm algorithm and the gravitational search algorithm to update the velocity and position of the particles. Repeat the update step and calculate the objective function value after each update. After the preset number of iterations, output the global optimal solution, which is the solution of the objective function, and each solution of the objective function corresponds to a positioning initial value.

[0049] For an N - element polygon three - dimensional sensor array, combining any four elements (with one element as a reference), there are combination methods, and each combination method corresponds to an objective function. Searching and solving the obtained objective functions can obtain positioning initial values.

[0050] In this embodiment, for a five - element cone array, assuming the coordinates of the insulator partial discharge source are , and the coordinates of the five positioning sensors are , , , , .

[0051] Arbitrarily select four sensors , , , to form a combination, and use as the reference sensor. The established objective function is: , In the formula, is the propagation speed of the positioning signal in the air, with a value of , is the time difference between the moment when the th sensor receives the positioning signal and the moment when the reference sensor receives the positioning signal, that is, the time delay, represents the distance from the discharge source to the th sensor. When h takes 1, 2, 3, 4, , , and , , can be obtained. 。

[0052] Among them, the distance calculation formula from the power source to the th sensor is: , where x, y, and z are the coordinate values of the power source on the x, y, and z axes respectively, and x h , y h , z h are the coordinate values of the hth sensor on the x, y, and z axes respectively.

[0053] Each particle represents a potential solution to the objective function. Calculate the objective function value of each particle, and call the objective function value of each particle the fitness value of each particle.

[0054] Update the individual optimal solution and the global optimal solution of each particle according to the objective function value.

[0055] When updating the velocity of particles by combining the particle swarm algorithm and the gravitational search algorithm, the specific update formula is: , where and represent the velocities of the Ith particle at the (u + 1)th and uth iterations respectively, u is the current iteration number, represents the inertia weight, and are learning factors, and are random variables between [0, 1], is the historical best position of the Ith particle, is the global best position among all particles, is the gravitational factor, represents the position of the Ith particle at the uth iteration, is the acceleration of the Ith particle at the uth iteration.

[0056] The calculation formula of the gravitational factor is as follows; , where is the minimum gravitational factor, is the maximum gravitational factor, is the preset number of iterations.

[0057] Compared with traditional numerical solution methods, dynamically adjusting the search speed of particles through the gravitational factor can find a balance between global search and local search, enabling the algorithm to widely explore the solution space in the initial stage and concentrate on the fine search for the optimal solution in the later stage, thus improving the convergence speed and global optimization ability.

[0058] Inertia weight The specific calculation method is as follows: , where, represents the maximum inertia weight, represents the minimum inertia weight, is the preset number of iterations, and u is the current number of iterations.

[0059] The inertia weight can be dynamically adjusted according to the needs of the algorithm. In the initial stage of iteration, a larger inertia weight is helpful for global search; in the later stage of iteration, a smaller inertia weight is beneficial for local search.

[0060] This dynamic adjustment strategy enables the algorithm to flexibly adjust the search strategy according to the requirements of different stages, thereby improving the overall performance of the algorithm.

[0061] When updating the position of particles by combining the particle swarm algorithm and the gravitational search algorithm, the specific update formula is: , where, and respectively represent the positions of the i-th particle at the (u + 1)-th and u-th iterations, represents the velocity of the i-th particle at the (u + 1)-th iteration, and u is the current number of iterations.

[0062] Repeat the above steps of updating particle velocity and position, and simultaneously update the fitness value of each particle until the preset number of iterations is reached, then output the global optimal solution, that is, obtain a set of initial positioning values . Similarly, solve the objective functions obtained under the other four combination methods to respectively obtain the initial positioning values , , , .

[0063] The solution method realizes the alternating iteration of the algorithm by mixing the search strategies of the particle swarm optimization algorithm and the gravitational search algorithm.

[0064] This alternating iteration method can make full use of the advantages of the two algorithms, avoid falling into local optimal solutions, and improve the convergence of the algorithm.

[0065] S5. Use the clustering algorithm to cluster all the initial positioning values. Among the clusters after clustering, the clustering center of the cluster with the most elements is the partial discharge position of the insulator.

[0066] In this embodiment, the five groups of initial positioning values solved when N = 5 , , , , are clustered using the clustering algorithm.

[0067] The number of clusters in S5 is selected according to the elbow method.

[0068] The specific operation of S5 is as follows: By calculating the sum of squared errors within clusters under different numbers of clusters and plotting the relationship between the number of clusters and the sum of squared errors within clusters, find the "elbow" position in the graph (i.e., the point where the rate of decrease in the sum of squared errors within clusters slows down) as the number of clusters K.

[0069] In this embodiment, the value of the number of clusters K is 2, that is, randomly select 2 data points as the initial clustering centers.

[0070] For each data point, calculate its distance from all the clustering centers and assign it to the cluster where the nearest clustering center is located.

[0071] For each cluster, select a new clustering center such that the total distance from the new clustering center to all other data points within the cluster is minimized. This step is called distance iteration.

[0072] Repeat the assignment and distance iteration steps until the clustering centers no longer change or reach the preset number of iterations.

[0073] Select the clustering center of the cluster with the most elements as the final result of the positioning.

[0074] By clustering the multiple groups of initial positioning values solved through the clustering algorithm, outliers can be effectively identified and excluded from the positioning results. By selecting the clustering center of the cluster with the most elements as the final result of this positioning, the positioning accuracy can be improved.

[0075] Through the design of the N - element polygon three - dimensional sensor array, signal focusing can be achieved in the spatial domain, thereby increasing the received signal strength and correspondingly reducing the influence of background noise. Through the collaborative work of multiple sensors, the positioning accuracy can be effectively improved. When one sensor fails or is interfered with, other sensors can still provide effective information, thereby reducing the errors that may occur during the positioning process and enhancing the overall positioning reliability.

[0076] The positioning signal is adaptively decomposed so that the intrinsic mode function components at different frequencies can be independently analyzed. By screening the high-quality intrinsic mode function components with larger correlation coefficients for wavelet packet decomposition, the main feature parts in the signal can be extracted more accurately, reducing redundant information. The adjustable threshold function can be flexibly adjusted according to the actual signal characteristics during the denoising process, significantly improving the denoising effect.

[0077] The weighting function takes into account the signals of multiple channels, so it can effectively utilize the correlation between multi-channel signals and enhance the effect of signal processing. By adjusting the weighting coefficient, the main components in the signal can be highlighted, especially in the case of low signal-to-noise ratio, the ability of signal delay estimation can be improved.

[0078] By dynamically adjusting the search speed of the particles, the flexibility of the algorithm in global search and local search can be improved, enabling the algorithm to widely explore the solution space in the initial stage of the search and focusing on the fine search for the optimal solution in the later stage, thus enhancing the convergence speed and global optimization ability of the search algorithm.

[0079] By clustering the multiple groups of initial positioning values obtained by the solution through the clustering algorithm, outliers can be effectively identified and excluded from the positioning results. By selecting the clustering center of the cluster with the largest number of elements in the cluster as the final result of this positioning, the accuracy of positioning can be improved.

Claims

1. A method for locating partial discharge of an insulator, characterized in that: The specific method includes the following steps: S1. Using a polygonal stereo sensor array to collect positioning signals generated by partial discharge of insulators; S2. Adaptively decompose the collected positioning signal from high to low frequency to obtain the intrinsic modal function components at different frequencies, calculate the correlation coefficient of each intrinsic modal function component, screen out the correlation coefficient greater than the correlation coefficient threshold as the high-quality coefficient, perform wavelet packet decomposition on the intrinsic modal function component corresponding to the high-quality coefficient, process the wavelet packet coefficient obtained by the wavelet packet decomposition using an adjustable threshold function, reconstruct the denoised high-quality intrinsic modal function component, synthesize all the denoised high-quality intrinsic modal function components, and obtain the denoised time domain positioning signal; S3, obtaining frequency domain positioning signals by Fourier transform based on the denoised time domain positioning signals, performing weighted cross-correlation processing based on any two frequency domain positioning signals, and obtaining the signal delay between two sensors corresponding to any two frequency domain positioning signals in the polygonal stereo sensor array; S4. Randomly select four sensors from the polygonal stereo sensor array for combination, establish the objective function corresponding to each combination based on the signal delay and the distance from the positioning signal to the sensor, randomly initialize a group of particles, each particle represents a potential solution of the objective function, and use the particle swarm algorithm and the gravitational search algorithm to update the speed and position of the particles. Repeat the updating steps and calculate the objective function value after each update. After a preset number of iterations, output the global optimal solution, which is the solution of the objective function. Each solution of the objective function corresponds to an initial positioning value. S5. Cluster all the initial positioning values ​​using a clustering algorithm. The cluster center of the cluster with the most elements in the cluster is the partial discharge position of the insulator.

2. A method for locating partial discharge of an insulator according to claim 1, characterized in that: When the particle swarm algorithm and the gravitational search algorithm are combined to update the particle velocity, the specific update formula is: ， in, and They represent the speed of the I-th particle at the u+1th and uth iterations, respectively, where u is the current iteration number. represents the inertia weight, and is the learning factor, and is a random variable between [0, 1], is the best historical position of the Ith particle, is the global best position among all particles, is the gravitational factor, represents the position of the I-th particle at the u-th iteration, is the acceleration of the I-th particle at the u-th iteration.

3. The method for locating partial discharge of an insulator according to claim 2, characterized in that: The gravitational factor The calculation formula is as follows; ， in, is the minimum gravitational factor, is the maximum gravitational factor, is the preset number of iterations.

4. The method for locating partial discharge of an insulator according to claim 2, characterized in that: The inertia weight The specific calculation method is: ， in, represents the maximum inertia weight, represents the minimum inertia weight, is the preset number of iterations, and u is the current number of iterations.

5. The method for locating partial discharge of an insulator according to claim 1, characterized in that: When the particle position is updated by combining the particle swarm algorithm and the gravitational search algorithm, the specific update formula is: ， in, and Respectively represent the position of the I-th particle at the u+1-th and u-th iterations, It represents the velocity of the Ith particle at the u+1th iteration, where u is the current iteration number.

6. The method for locating partial discharge of an insulator according to claim 1, characterized in that: Any four sensors , , , Form a combination and As a reference sensor, the objective function established for: ， In the formula, is the speed at which the positioning signal propagates in the air, and its value is , For the The time difference between the time when the first sensor receives the positioning signal and the time when the reference sensor receives the positioning signal is called the delay. Indicates that the power is discharged to the The distance between the sensors, when h is 1, 2, 3, 4, we get , , and , , , ; Among them, discharge the power to The distance calculation formula for each sensor is: ， Where x, y, and z are the coordinates of the discharge source on the x, y, and z axes, respectively. h ,y h 、z h are the coordinate values ​​of the hth sensor on the x, y, and z axes respectively.

7. The method for locating partial discharge of an insulator according to claim 1, characterized in that: The weighted cross-correlation processing described in S3 is performed based on any two frequency domain positioning signals, and the specific weighting function is: , in, is the weighting function, is the angular frequency, and They represent the auto-power spectral density of the two frequency domain positioning signals, m and n represent the serial numbers of the sensors, is the weighting coefficient, and .

8. The method for locating partial discharge of an insulator according to claim 1, characterized in that: The polygonal stereo sensor array is a cone array, a sphere array or a column array, and the number of sensors in the polygonal stereo sensor array is N≧5.

9. The method for locating partial discharge of an insulator according to claim 1, characterized in that: The correlation coefficient calculation method described in S2 is: ， in, represents the correlation coefficient of the i-th intrinsic mode function component, L is the length of the positioning signal, is the mean of the positioning signal, is the mean of the i-th intrinsic mode function component, Indicates that the positioning signal is The value of the moment, Indicates that the i-th intrinsic mode function component is The value of the moment, Indicates time.

10. The method for locating partial discharge of an insulator according to claim 1, characterized in that: The number of clusters described in S5 was chosen according to the elbow rule.

Images