GIS internal multi-source partial discharge signal positioning method

Abstract

The invention discloses a GIS internal multi-source partial discharge signal positioning method comprising the following steps: carrying out noise reduction processing on ultrahigh frequency signals collected by a plurality of sensors to obtain pure signals corresponding to the sensors; based on the pure signals corresponding to the sensors, the time difference of the partial discharge signals of the jth partial discharge source received by any two sensors is solved by using an initial energy increase cross-correlation method, and the multi-source time difference is obtained; according to the multi-source time difference, generating MN-1 spaces and calculating the fitness of each space, taking the positioning position in the space with the best fitness as the coordinate of the partial discharge source, i = 1, 2, 3,..., N, j = 1, 2, 3,..., M, N being the total number of sensors, and M being the total number of the partial discharge source; according to the invention, the positioning of the multi-source partial discharge signal can be realized.

Description

Technical Field

[0001] The invention relates to the technical field of GIS discharge detection, and in particular to a method for locating multi-source local discharge signals inside a GIS. Background Art

[0002] In high-voltage gas insulated switchgear (GIS), partial discharge (PD) is a common fault phenomenon that can cause equipment performance degradation, damage, and even accidents. Accurately and quickly locating multi-source partial discharge signals inside GIS is crucial for equipment health status monitoring and fault diagnosis.

[0003] In recent years, the local discharge ultrasonic positioning method based on the time difference of arrival (TDOA) has gradually attracted widespread attention due to its low complexity, no influence from strong electromagnetic environment and high accuracy. The positioning method based on the time difference of arrival is to obtain the sampling data of the same signal at different positions through multiple sensors, process the data at different positions, obtain the time difference information between each signal, and then construct and solve the positioning equation. By solving the positioning equation, the purpose of PD source positioning can be achieved. However, in the complex structural environment inside GIS, the propagation path of the signal is complex and easily affected by multipath effect and reflection interference, resulting in the TDOA method The accuracy of local discharge signal positioning inside GIS is not high.

[0004] Traditional time difference estimation algorithms include energy accumulation method, generalized cross correlation algorithm (Generalized CrossCorrelation, GCC), etc., and new algorithms such as wavelet convolution image method and time difference estimation based on source-filter model have also been proposed. In terms of positioning algorithms, the Chan algorithm and Fang algorithm of analytical expressions have been improved and applied, and intelligent algorithms such as particle swarm algorithm, genetic algorithm, artificial bee colony algorithm (Artificial bee colony algorithm, ABC) have also made certain progress.

[0005] The above studies mainly focus on single-source positioning technology. However, in actual power scenarios, multiple devices may generate local discharges at the same time or a single device may discharge at multiple locations. For example, inside a GIS, there may be multiple local discharge sources that discharge at the same time. Single-source positioning technology is difficult to accurately separate the time difference between signals from different sources, resulting in inaccurate positioning results.

[0006] Moreover, these methods are mainly based on ultra-high frequency (UHF) signals, whose signal width is in the nanosecond level. Local discharge occurs according to the power frequency law, resulting in the overlap of multi-source local discharge ultrasonic signals in the time domain, and the energy accumulation method and clustering algorithm are no longer applicable. In addition, the ultrasonic signal wave head rises slower than the UHF signal, and the energy accumulation method has a large error in extracting the time difference. In the face of complex noise environments, the positioning accuracy is easily affected by noise and interference, resulting in unstable positioning results. In addition, other existing time difference extraction algorithms also have large errors under low signal-to-noise ratios.

[0007] For example, in the related technology, the doctoral thesis "Research on Key Technologies for Local Discharge Location in Substations Based on Sensor Arrays, Wang Shudong, Hefei University of Technology" proposed to extract time difference through energy accumulation method based on local discharge signal denoising algorithm, as well as time difference extraction method based on fast S transform and singular value decomposition, etc.; However, in practical applications, due to the uncertainty of noise, the characteristics of on-site background noise may change with time and environmental conditions, which may cause the pre-set energy threshold or number of singular values ​​to be not applicable to all situations, thereby affecting the accuracy of time difference extraction; and the computational complexity of multi-resolution generalized S transform and fast S transform combined with singular value decomposition is high, requiring a large number of time-frequency transformations and matrix decomposition operations, which is computationally intensive and time-consuming, and may lead to performance bottlenecks for real-time processing, and may not be efficient enough for real-time monitoring and rapid response application scenarios; in addition, if there are multiple local discharge sources, their energy accumulation curves may interfere with each other, and this multi-signal interference makes it difficult to identify mutation points. In the doctoral dissertation "Research on Localization and Diagnosis Methods of Partial Discharge in Power Substations, Ning Shuguang, Hefei University of Technology", a transformer PD source location method based on TDOA and truncated singular value regularization was proposed. In this scheme, the linear transformation of the nonlinear equations using the spherical transformation method may introduce model errors. Although this transformation simplifies the calculation, it may not fully and accurately reflect the actual physical phenomena, thus affecting the positioning accuracy; secondly, the selection of regularization parameters in the TSVD method has a greater impact on the final solution. If the regularization parameters are not selected properly, it may lead to over-smoothing or under-smoothing, affecting the accuracy of the positioning results. Although the regularization parameter optimization method is given, the effectiveness and robustness of this method need to be further verified in practical applications; and because multi-source discharge signals may overlap and interfere with each other, it is difficult to distinguish the contributions of each discharge source from the received signal. The TDOA method relies on accurate time difference data. If multi-source signals interfere with each other, the extraction of time difference will become complicated and inaccurate. In terms of computational complexity: Although the spherical transformation and TSVD regularization strategy simplify the solution of the equation, there may still be a problem of high computational complexity. Especially when processing large-scale data, the computational efficiency and resource requirements of the optimization algorithm are still an issue that needs to be considered; and in the case of multi-source signals, the selection of TSVD regularization parameters becomes more complicated. Different discharge sources may require different regularization parameters. How to find an optimal parameter that is suitable for all discharge sources is a challenge. For ill-conditioned problems: Although the scheme uses the balance method to process the matrix A to weaken the degree of ill-conditioning of the time difference solution equation, in practical applications, the ill-conditioning of the matrix may not be completely eliminated and will still affect the solution results. The existence of ill-conditioning problems may lead to instability and sensitivity of the solution. Multi-source discharge signals require processing more data and more complex equations, and the computational burden of the TSVD method will increase significantly.Although TSVD has advantages in dealing with ill-conditioned problems, its computational efficiency and resource requirements may become bottlenecks when faced with complex multi-source signals. In addition, in the case of multi-source discharge, the cumulative effect of data noise and measurement errors will be more significant, which will have a greater impact on the positioning results. Dealing with these errors requires more complex and robust algorithms.

[0008] In the patent application document with publication number CN114355114A, it is proposed to accurately detect the arrival time of the sensor according to the pulse waveform, and distinguish the discharge types of multiple discharge pulses in combination with the PRPD spectrum, so that multiple pulses of different types can be located and identified at the same time; however, this scheme does not have a filtering and noise reduction module, which makes the arrival time detected according to the waveform have deviations, which has a great impact on the accuracy and robustness of subsequent detections; secondly, when the arrival time difference is obtained, the arrival time is only subtracted and the rising edge trigger mechanism is used to adjust the obtained time difference according to the correlation parameter, which is easy to cause confusion about the source of the time difference signal.

[0009] In summary, traditional methods usually require complex signal processing algorithms and a large amount of calculations, which are computationally complex and time-consuming, and are not conducive to real-time monitoring and fault diagnosis. New methods are urgently needed to solve the problem of rapid positioning of multi-source discharge signals. Summary of the invention

[0010] The technical problem to be solved by the present invention is how to solve the problem of locating multi-source partial discharge signals inside GIS.

[0011] The present invention solves the above technical problems by the following technical means:

[0012] The present invention proposes a method for locating multi-source partial discharge signals inside a GIS, the method comprising:

[0013] Perform noise reduction on the UHF signals collected by several sensors to obtain the pure signals corresponding to each sensor;

[0014] Based on the pure signals corresponding to each sensor, the time difference between any two sensors receiving the partial discharge signal of the jth partial discharge source is calculated using the initial energy increase cross-correlation method to obtain the multi-source time difference;

[0015] Generate M according to the multi-source time difference N-1 spaces and calculate the fitness of each space, and take the positioning position in the space with the best fitness as the coordinate of the local discharge source, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., M, N is the total number of sensors, and M is the total number of local discharge sources.

[0016] Furthermore, the noise reduction process is performed on the ultra-high frequency signals collected by the plurality of sensors to obtain the pure signals corresponding to the sensors, including:

[0017] Performing singular value decomposition denoising processing on the ultra-high frequency signal to obtain a reconstructed filtered signal;

[0018] The reconstructed filtered signal is subjected to dual-tree complex wavelet transform noise reduction processing, and the processed wavelet coefficients are subjected to dual-tree complex wavelet inverse transform to obtain pure signals corresponding to each sensor.

[0019] Furthermore, the performing singular value decomposition denoising processing on the ultra-high frequency signal to obtain a reconstructed filtered signal includes:

[0020] Converting the UHF signal into a Hannkel matrix, and performing singular value decomposition on the Hankel matrix to obtain a singular entropy increment F1 and an energy value F2;

[0021] The singular entropy increment F1 and the energy value F2 are subjected to binary classification of the singular values ​​of the corresponding feature quantities by using K-means clustering, the index position of the singular value corresponding to the narrowband interference is determined and the corresponding value of the index position is set to 0;

[0022] The reconstructed filtered signal with narrowband interference removed is obtained through inverse singular value decomposition.

[0023] Furthermore, the reconstructed filtered signal is subjected to dual-tree complex wavelet transform noise reduction processing, and the processed wavelet coefficients are subjected to dual-tree complex wavelet inverse transform to obtain a pure signal corresponding to each sensor, including:

[0024] The reconstructed filtered signal is subjected to dual-tree complex wavelet transform denoising, and the lth wavelet coefficient λ on the kth scale is k，l The optimal threshold ψ is obtained using the SURE-whale optimization algorithm;

[0025] Based on the optimal threshold ψ, performing secondary threshold denoising processing on the wavelet coefficients of the reconstructed filtered signal to obtain processed wavelet coefficients;

[0026] The processed wavelet coefficients are inversely transformed using dual-tree complex wavelet transform to obtain the pure signals corresponding to each sensor.

[0027] Furthermore, the reconstructed filtered signal is subjected to dual-tree complex wavelet transform denoising, and the lth wavelet coefficient λ on the kth scale is k，l The optimal threshold ψ is obtained using the SURE-whale optimization algorithm, including:

[0028] The variance of the wavelet coefficients at each scale is calculated, and the initial threshold corresponding to the wavelet coefficients at each scale is calculated using the SURE algorithm:

[0029]

[0030] Where: nk 、M k are the length of the wavelet coefficients at scale k and the median absolute deviation of the wavelet coefficients used to estimate the noise level;

[0031] Randomly generate the initial whale population position, each whale represents a set of thresholds of different scales corresponding to the dual-tree complex wavelet transform, take the initial threshold as the initial optimal solution, and set the maximum number of iterations and the number of individuals;

[0032] According to the initial threshold of each individual whale, the wavelet coefficients obtained by decomposition are threshold processed, and the joint fitness of the processed signals is calculated;

[0033] Update the whale's position, simulate the whale's spiral rise by calculating the distance from the optimal position, and update the position:

[0034]

[0035] In the formula, μ represents the step size, ξ(b) represents the gradient value of the bth iteration, and ψ k (b) represents the threshold of the b-th iteration, ψ k (b+1) represents the threshold value of the b+1th iteration;

[0036] When the maximum number of iterations is reached or the convergence condition is met, the optimal threshold ψ is obtained.

[0037] Furthermore, based on the pure signals corresponding to each sensor, the time difference between any two sensors receiving the partial discharge signal of the j-th partial discharge source is obtained by using the initial energy increase cross-correlation method to obtain the multi-source time difference, including:

[0038] Based on the pure signal corresponding to each sensor, the initial local energy capacity change rate of the pure signal corresponding to each sensor is calculated;

[0039] Starting from the start time t1, repeatedly decrease the sampling time t1 s Determine t2, where t1 and t2 satisfy:

[0040] E[w](t1)≥B

[0041] E[w](t2)≤b

[0042] 2E[w](t2)-E[w](t2+t s )>E[w](t2-t s )

[0043] Where B and b are the upper and lower thresholds corresponding to the initial local energy capacity change rate E[w] respectively;

[0044] Calculate the estimated value of the signal arrival time at the sensor τ(w)=t2+T / 4 according to time t2, where T / 4 is the width of the window function that compensates for the peak shift effect of E[w](t);

[0045] According to the estimated values ​​of the arrival time of the signals corresponding to any two sensors, the measured value of the arrival time difference of the signals of the two sensors is calculated.

[0046] Furthermore, the calculation formula for the measured value of the signal arrival time difference of the two sensors is:

[0047] t α，β =τ α，β +argmax(w α (τ α，β +t)*w β (τ α，β +t)), t∈[-ε,ε]

[0048] In the formula, τ α，β is the difference between the estimated time of arrival of the PD signal of the jth PD source at sensor α and sensor β, t α，β is the measured value of the arrival time difference of the signals from sensor α and sensor β, ε is the size of the cross-correlation window, argmax is the parameter operator for finding the optimal solution of the function, and w α (τ α，β +t)、w β (τ α，β +t) are the received frequency domain signals of the arrival time difference offset of sensor α and sensor β, t is the time series of the signal received by the sensor, and * is the convolution operation.

[0049] Furthermore, the upper threshold and lower threshold corresponding to the initial local energy capacity change rate E[w] are respectively:

[0050]

[0051]

[0052] Where t is the time series of the signal received by the sensor.

[0053] Further, the generating M according to the multi-source time difference N-1 spaces and calculate the fitness of each space, and use the location in the space with the best fitness as the coordinates of the local discharge source, including:

[0054] Generate M based on the multi-source time difference calculated for N sensors and M local discharge sources N-1 space, in which is recorded as real space, and the space formed by the other multi-source time differences is recorded as simulated space, where It represents the time difference between the Nth sensor and the 1st sensor receiving the signal of the jth local discharge source. The subscript of j from 1 to N-1 represents the time difference separation scale of the jth local discharge source relative to the same source time difference N sensors. If the values ​​are all equal, the space formed is real space;

[0055] In M N-1 The space forms an appropriate amount of custom particles, and the particle p can be expressed as (x p ,y p , z p ,s p ), (x p ,y p , z p ) represents the coordinate of particle p in the sth space, s p Indicates that the space where particle p is located is the sth space, s = 1, 2, 3, ..., M N-1 ;

[0056] Calculate M separately N-1 The fitness corresponding to the space is calculated, and the positioning position in the space with the best fitness is taken as the coordinate of the local discharge source.

[0057] Further, the calculation of M N-1 The fitness corresponding to each space is calculated, and the location in the space with the best fitness is used as the coordinate of the local discharge source, including:

[0058] Calculate M separately N-1 The fitness of the space is Among them, t i1，j It represents the time difference between the i-th sensor and the first sensor receiving the signal from the j-th local discharge source, t i1，s It represents the time difference between the i-th sensor and the first sensor receiving the signal from the s-th local discharge source;

[0059] The M spaces with the best fitness are selected to approximate the real space, and the positioning positions in the M spaces are used as the coordinates of the local discharge source.

[0060] In addition, the present invention also proposes a GIS internal multi-source partial discharge signal positioning system, the system comprising:

[0061] A noise reduction processing module is used to perform noise reduction processing on the ultra-high frequency signals collected by several sensors to obtain the pure signals corresponding to each sensor;

[0062] A time difference calculation module is used to calculate the time difference between any two sensors receiving the partial discharge signal of the jth partial discharge source by using the initial energy increase cross-correlation method based on the pure signal corresponding to each sensor, and obtain the multi-source time difference;

[0063] A positioning module is used to generate M according to the multi-source time difference N-1 spaces and calculate the fitness of each space, and take the positioning position in the space with the best fitness as the coordinate of the local discharge source, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., M, N is the total number of sensors, and M is the total number of local discharge sources.

[0064] The advantages of the present invention are:

[0065] (1) The present invention generates M based on the multi-source time difference calculated from N sensors and M local discharge sources. N-1 spaces, and then the fitness of each space is calculated, and the positioning position in the space with the best fitness is used as the coordinate of the local discharge source. The present invention can accurately separate and combine the time differences of homologous signals in multi-source time differences by introducing the screening of fitness and particle space dimensions, thereby improving the accuracy of time difference extraction, thereby effectively separating and locating multiple local discharge sources, simplifying the process of multi-source positioning, and realizing efficient positioning of multi-source local discharge signals.

[0066] (2) The present invention first performs singular value decomposition (SVD) denoising processing on the collected UHF signal, effectively separates the part containing periodic narrowband interference in the SVD through K-means clustering, and then performs dual-tree complex wavelet transform for denoising processing. The optimal adaptive threshold is obtained by using the SURE-whale optimization algorithm for each layer of wavelet coefficients after the dual-tree wavelet transform, thereby improving the precision of the denoising process. The present invention adopts a secondary signal denoising method that combines SVD and adaptive wavelet threshold, which is suitable for adaptive denoising processing of UHF signals of multiple local discharge sources collected in a complex noise environment.

[0067] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flow chart of a method for locating multi-source partial discharge signals inside a GIS proposed in one embodiment of the present invention;

[0069] Figure 2 It is a principle block diagram of locating multi-source partial discharge signals inside GIS in one embodiment of the present invention;

[0070] Figure 3 It is a structural schematic diagram of a multi-source partial discharge signal locating system inside a GIS proposed in one embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] like Figure 1 to Figure 2 As shown, an embodiment of the present invention proposes a method for locating multi-source partial discharge signals inside a GIS, the method comprising the following steps:

[0073] S10, performing noise reduction processing on the ultra-high frequency signals collected by a plurality of sensors to obtain a pure signal corresponding to each sensor;

[0074] It should be noted that the sensor arranged in this embodiment is a UHF (Ultra High Frequency) sensor.

[0075] S20, based on the pure signals corresponding to each sensor, using the initial energy increase cross-correlation method to calculate the time difference between any two sensors receiving the partial discharge signal of the j-th partial discharge source, and obtain the multi-source time difference;

[0076] S30, generating M according to the multi-source time difference N-1 spaces and calculate the fitness of each space, and take the positioning position in the space with the best fitness as the coordinate of the local discharge source, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., M, N is the total number of sensors, and M is the total number of local discharge sources.

[0077] In this embodiment, the UHF signals of M partial discharge sources collected by N sensors are first subjected to noise reduction processing, providing a pure signal for the subsequent efficient positioning of multi-source partial discharge signals. The multi-source time difference is calculated based on the pure signal corresponding to each sensor, and M N-1 spaces, and then calculate the fitness of each space, and take the positioning position in the space with the best fitness as the coordinates of the local discharge source; by introducing the screening of fitness and particle space dimensions, the time difference of the same source signal can be accurately separated and combined in the multi-source time difference, which improves the accuracy of time difference extraction, thereby effectively separating and locating multiple local discharge sources, simplifying the process of multi-source positioning, and realizing efficient positioning of multi-source local discharge signals.

[0078] As a further preferred technical solution, the step S10: performing noise reduction processing on the ultra-high frequency signals collected by a plurality of sensors to obtain a pure signal corresponding to each sensor includes the following steps:

[0079] S11, performing singular value decomposition denoising processing on the ultra-high frequency signal to obtain a reconstructed filtered signal;

[0080] S12, performing dual-tree complex wavelet transform noise reduction processing on the reconstructed filtered signal, and performing dual-tree complex wavelet inverse transform on the processed wavelet coefficients to obtain pure signals corresponding to each sensor.

[0081] This embodiment uses a secondary signal denoising method that combines singular value decomposition and adaptive wavelet threshold, which is suitable for denoising ultra-high frequency signals collected in a complex noise environment.

[0082] As a further preferred technical solution, the step S11: performing singular value decomposition denoising processing on the UHF signal to obtain a reconstructed filtered signal, specifically comprises the following steps:

[0083] S111, converting the UHF signal into a Hankel matrix, and performing singular value decomposition on the Hankel matrix to obtain a singular entropy increment F1 and an energy value F2;

[0084] Specifically, this embodiment determines the window length L for constructing the Hankel matrix. Usually, L should satisfy (1<L<N), where N is the signal sampling value. The selection of the window length affects the size and characteristics of the Hankel matrix. Constructing the Hankel matrix: Construct the Hankel matrix H according to the sampled signal (x[n]) and the window length L. The definition of the Hankel matrix is ​​that the elements of each row or column are arranged in time series order, and the elements along the diagonal are the same.

[0085] Then perform SVD decomposition on the Hankel matrix H, which can be expressed as H = U∑V T , where U and V are orthogonal matrices, and ∑ is a diagonal matrix whose diagonal elements are the singular values ​​of H.

[0086] S112, using K-means clustering to perform binary classification of the singular values ​​of the corresponding feature quantities on the singular entropy increment F1 and the energy value F2, determine the index position of the singular value corresponding to the narrowband interference and set the corresponding value of the index position to 0;

[0087] It should be noted that the process of performing K-means clustering on the singular entropy increment F1 and the energy value F2 in this embodiment is as follows:

[0088] 1) Randomly initialize the centroids of two clusters; 2) For each data point in the data set, assign it to the category represented by the nearest centroid; 3) Update the centroid of each cluster to make it the average of all points in the category; 4) Repeat steps 2) and 3) until the centroid no longer changes or meets a certain stopping condition, such as the number of iterations reaches a predetermined value or the change of the centroid is less than a certain threshold; 5) Finally, the singular values ​​of the corresponding feature quantities are divided into two categories: normal signal features and narrowband interference features.

[0089] S113. Obtain a reconstructed filtered signal with narrowband interference removed by inverse singular value decomposition.

[0090] It should be noted that the UHF signal collected in this embodiment is first subjected to singular value decomposition denoising processing, and the part containing periodic narrowband interference in the singular value decomposition is effectively separated through K-means clustering to obtain a reconstructed filtered signal; by utilizing the Hankel structure, matrix compression or screening can be performed in advance, so that the actual SVD calculation is only concentrated on the part representing the main information (i.e., the eigenvalue corresponding to the larger singular value), thereby reducing the amount of calculation compared to simple singular value decomposition for denoising.

[0091] As a further preferred technical solution, the step S12: performing dual-tree complex wavelet transform denoising processing on the reconstructed filtered signal, and performing dual-tree complex wavelet inverse transform on the processed wavelet coefficients to obtain a pure signal corresponding to each sensor, specifically includes the following steps:

[0092] S121, perform dual-tree complex wavelet transform noise reduction processing on the reconstructed filtered signal, and transform the lth wavelet coefficient λ on the kth scale k，l The optimal threshold ψ is obtained using the SURE-whale optimization algorithm;

[0093] Specifically, the calculation process of the optimal threshold is:

[0094] 1) Initial SURE threshold estimation: Calculate the variance of the wavelet coefficients at each scale and use the SURE algorithm to calculate the initial threshold:

[0095]

[0096] Among them, n k 、M k are the length of the wavelet coefficients on scale k and the median absolute deviation (MAD) of the wavelet coefficients used to estimate the noise level, and 0.6745 is to make the estimate unbiased under Gaussian noise.

[0097] 2) Initialization: Randomly generate the initial whale population position (threshold), each whale represents a set of positions (thresholds) of different scales corresponding to the dual-tree complex wavelet transform, take the SURE threshold as the initial optimal solution, and set parameters such as the maximum number of iterations and the number of individuals.

[0098] 3) Fitness evaluation: According to the position of each individual (threshold ψk), the wavelet coefficients obtained by decomposition are thresholded and the joint fitness of the processed signal is calculated (the weighted combination of SURE risk and signal quality indicators such as mean square error, signal-to-noise ratio, etc.).

[0099] 4) Update position: Update the whale's position, simulate the behavior of surrounding prey by calculating the distance to the best position; simulate the whale's spiral rise and update the position

[0100]

[0101] In the formula, μ represents the step size, ξ(b) represents the gradient value of the b-th iteration, and according to a certain probability, a global search is performed to avoid falling into the local optimum.

[0102] 5) Fitness comparison: Compare the fitness value of each individual and update the optimal position.

[0103] 6) Iteration: Repeat steps 3) to 5) until the maximum number of iterations is reached or a certain convergence condition is met.

[0104] 7) Output result: The optimal threshold ψ obtained in the end is the threshold corresponding to the individual with the highest fitness. Adaptive optimal threshold noise reduction can be achieved for different local discharge sources.

[0105] It should be noted that, in this embodiment, the global search mechanism of the whale optimization algorithm (WOA) is combined with the SURE algorithm for initial threshold estimation, which can better explore the parameter space and find a better solution by minimizing Stein's unbiased risk estimate, thereby improving the noise reduction effect; by dynamically adjusting weights and multi-objective optimization strategies, the system can adapt to different signal characteristics and noise conditions according to the actual noise reduction effect and problem requirements. For complex UHF signal noise reduction problems, this method has better adaptability and robustness, and can handle various noise components and signal characteristics.

[0106] S122, based on the optimal threshold ψ, performing secondary threshold denoising processing on the wavelet coefficients of the reconstructed filtered signal to obtain processed wavelet coefficients;

[0107] Specifically, the formula for performing secondary threshold noise reduction processing on the signal using the optimal threshold is expressed as:

[0108]

[0109] In the formula, It is represented as a threshold function, and ξ represents the parameter that determines the shape of the threshold function.

[0110] S123, performing a dual-tree complex wavelet inverse transform on the processed wavelet coefficients to obtain a pure signal corresponding to each sensor.

[0111] It should be noted that this embodiment performs dual-tree complex wavelet transform on the reconstructed filtered signal for denoising, and uses the whale optimization algorithm to obtain the optimal adaptive threshold for each layer of wavelet coefficients after the dual-tree wavelet transform, thereby improving the precision of the denoising process.

[0112] This embodiment uses a secondary noise reduction method when processing the collected multi-source partial discharge signals. The first noise reduction is to perform singular value decomposition noise reduction on the signal in the form of Hankel matrix to filter out the narrowband interference of the signal. The second noise reduction is to perform dual-tree complex wavelet transform on the signal and perform optimal threshold noise reduction on the wavelet coefficients. In the second noise reduction, the whale optimization algorithm is used to achieve the solution of adaptive optimal thresholds for different discharge sources, so as to maintain a good noise reduction effect for different discharge sources.

[0113] As a further preferred technical solution, the step S20: based on the pure signals corresponding to each sensor, using the initial energy increase cross-correlation method to calculate the time difference between any two sensors receiving the partial discharge signal of the j-th partial discharge source, to obtain the multi-source time difference, specifically includes the following steps:

[0114] S21, based on the pure signal corresponding to each sensor, calculating the initial local energy capacity change rate of the pure signal corresponding to each sensor;

[0115] Specifically, for the pure signal w(t) of a single sensor, find its initial local energy capacity change rate E[w]:

[0116]

[0117]

[0118] In the formula, h T (t) is set to a Hann window function with a width of T = 4ns, t is the time value of the pure signal w(t), * is the convolution operator, and x≥0 indicates that the rate of change after differentiation is non-negatively rectified.

[0119] S22, starting from the starting time t1, repeatedly decrease the sampling time t1 s Determine t2, where t1 and t2 satisfy:

[0120] E[w](t1)≥B

[0121] E[w](t2)≤b

[0122] 2E[w](t2)-E[w](t2+t s )>E[w](t2-t s )

[0123] Where B and b are the upper and lower thresholds corresponding to the initial local energy capacity change rate E[w], respectively:

[0124]

[0125]

[0126] Where t is the time series of the signal received by the UHF sensor.

[0127] S23. Calculate the estimated value of the signal arrival time at the sensor τ(w)=t2+T / 4 according to time t2, where T / 4 is the width of the window function for compensating the effect of the peak shift of E[w](t).

[0128] S24. Calculate a measured value of a difference in arrival time of the signals of any two sensors based on the estimated values ​​of the arrival time of the signals of the two sensors.

[0129] This embodiment performs two noise reduction processes after receiving the signal to remove narrowband interference and white noise in the signal as much as possible. Secondly, when obtaining the arrival time difference, the initial energy increase cross-correlation method is used to obtain the multi-source time difference signal according to the initial local energy capacity change rate. Subsequently, the fitness of the time difference generation space is obtained to approximate the defined real space, thereby realizing the separation of the same source time difference.

[0130] As a further preferred technical solution, the calculation formula for the measured value of the signal arrival time difference of the two sensors is:

[0131] t α，β =τ α，β +argmax(w α (τ α，β +t)*w β (τ α，β +t)), t∈[-ε,ε]

[0132] In the formula, τ α，β is the difference between the estimated time of arrival of the PD signal of the jth PD source at sensor α and sensor β, t α，β is the measured value of the arrival time difference of the signals from sensor α and sensor β, ε is the size of the cross-correlation window, argmax is the parameter operator for finding the optimal solution of the function, and w α (τ α，β +t)、w β (τα，β +t) are the received frequency domain signals of the arrival time difference offset of UHF sensor α and UHF sensor β, t is the time series of the signal received by the UHF sensor, and * is the convolution operation.

[0133] The calculation method of the signal arrival time difference measurement value of the two sensors adopted in this embodiment has a small amount of calculation and is suitable for application in real-time monitoring and positioning of discharge signals.

[0134] As a further preferred technical solution, the step S30: generating M according to the multi-source time difference N-1 The space is divided into two parts, and the fitness of each space is calculated. The positioning position in the space with the best fitness is used as the coordinate of the local discharge source, including the following steps:

[0135] S31, generating M according to the multi-source time difference calculated for N sensors and M local discharge sources N-1 space, in which Recorded as real space, the rest of the space formed by multi-source time difference are not all equal and are all integers between 1 and M) is recorded as the simulative space, It represents the time difference between the Nth sensor and the 1st sensor receiving the signal of the jth local discharge source. The subscript of j from 1 to N-1 represents the time difference separation scale of the jth local discharge source relative to the same source time difference N sensors. If the values ​​are all equal, the space formed is real space;

[0136] S32, in M N-1 The space forms an appropriate amount of custom particles, and the particle p can be expressed as (x p ,y p , z p ,s p ), (x p ,y p , z p ) represents the coordinate of particle p in the sth space, s p Indicates that the space where particle p is located is the sth space, s = 1, 2, 3, ..., M N-1 ;

[0137] It should be noted that since each particle needs to evaluate and update the function, too many particles will increase the computational overhead. It is necessary to consider the available computing resources and choose an appropriate number of particles. Generally speaking, 20-50 particles are usually a good choice. For more complex problems, the number of particles can be appropriately increased to about 100.

[0138] It should be noted that by adding the particle space dimension as the fourth dimension on the basis of the three-dimensional coordinates, the space formed by the time difference data is used to perform homologous time difference separation. The definition of real space is the time difference formed by the same local discharge source. The fitness of the space is subsequently calculated, and the optimal space is taken to be close to the real space, thereby achieving the effect of homologous time difference separation.

[0139] S33, calculate M respectively N-1 The fitness corresponding to the space is calculated, and the positioning position in the space with the best fitness is taken as the coordinate of the local discharge source.

[0140] Specifically, the step S33: respectively calculate M N-1 The fitness corresponding to the space is calculated, and the location in the space with the best fitness is used as the coordinate of the local discharge source, which includes:

[0141] Calculate M separately N-1 The fitness of the space is Among them, t i1，j It represents the time difference between the i-th sensor and the first sensor receiving the signal from the j-th local discharge source, t i1，s It represents the time difference between the i-th sensor and the first sensor receiving the signal from the s-th local discharge source;

[0142] The M spaces with the best fitness are selected to approximate the real space, and the positioning positions in the M spaces are used as the coordinates of the local discharge source.

[0143] This embodiment uses the increase in the local energy content of the UHF sensor detection signal to extract the arrival time difference. This method does not require complex transformation calculations. The time point when the sensor receives the signal corresponds to the lowest point of the first peak found in the energy function. It saves a lot of time for the subsequent use of multi-source time differences to generate real space and pseudo space, including the subsequent spatial fitness calculation does not need to consume a lot of time and computing resources. At the same time, some low-frequency noise attached to the UHF signal does not affect the calculation of the increase trend of the local energy content, and a certain noise reduction function can be achieved. Therefore, this embodiment can achieve rapid positioning of multi-source discharge signals.

[0144] It should be noted that this embodiment can realize the positioning of multi-source partial discharge signals inside the GIS, and can also realize the positioning of partial discharge signals generated by multiple GIS simultaneously. Multiple GIS insulation devices are also interconnected through busbars, cables or other connectors. This embodiment can be used separately on each GIS device, and partial discharge positioning can also be performed on the overall equipment.

[0145] In addition, if Figure 3As shown, the second embodiment of the present invention further proposes a multi-source partial discharge signal positioning system inside a GIS, the system comprising:

[0146] The noise reduction processing module 10 is used to perform noise reduction processing on the ultra-high frequency signals collected by a plurality of sensors to obtain a pure signal corresponding to each sensor;

[0147] The time difference calculation module 20 is used to calculate the time difference between any two sensors receiving the partial discharge signal of the j-th partial discharge source by using the initial energy increase cross-correlation method based on the pure signals corresponding to each sensor, and obtain the multi-source time difference;

[0148] A positioning module 30 is used to generate M according to the multi-source time difference. N-1 spaces and calculate the fitness of each space, and take the positioning position in the space with the best fitness as the coordinate of the local discharge source, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., M, N is the total number of sensors, and M is the total number of local discharge sources.

[0149] As a further preferred technical solution, the noise reduction processing module 10 includes:

[0150] A primary noise reduction unit, used for performing singular value decomposition noise reduction processing on the ultra-high frequency signal to obtain a reconstructed filtered signal;

[0151] The secondary noise reduction unit is used to perform dual-tree complex wavelet transform noise reduction processing on the reconstructed filtered signal, and perform dual-tree complex wavelet inverse transform on the processed wavelet coefficients to obtain pure signals corresponding to each sensor.

[0152] The primary noise reduction unit is specifically used to perform the following steps:

[0153] Converting the UHF signal into a Hankel matrix, and performing singular value decomposition on the Hankel matrix to obtain a singular entropy increment F1 and an energy value F2;

[0154] The singular entropy increment F1 and the energy value F2 are subjected to binary classification of the singular values ​​of the corresponding feature quantities by using K-means clustering, the index position of the singular value corresponding to the narrowband interference is determined and the corresponding value of the index position is set to 0;

[0155] The reconstructed filtered signal with narrowband interference removed is obtained through inverse singular value decomposition.

[0156] The secondary noise reduction unit is specifically used to perform the following steps:

[0157] The reconstructed filtered signal is subjected to dual-tree complex wavelet transform denoising, and the lth wavelet coefficient λ on the kth scale is k，l The whale optimization algorithm is used to obtain the optimal threshold mountain;

[0158] Based on the optimal threshold value mountain, performing secondary threshold noise reduction processing on the wavelet coefficients of the reconstructed filtered signal to obtain processed wavelet coefficients;

[0159] The processed wavelet coefficients are inversely transformed using dual-tree complex wavelet transform to obtain the pure signals corresponding to each sensor.

[0160] As a further preferred technical solution, the time difference calculation module 20 includes:

[0161] An initial local energy capacity change rate calculation unit, used to calculate the initial local energy capacity change rate of the pure signal corresponding to each sensor based on the pure signal corresponding to each sensor;

[0162] The time determination unit is used to repeatedly decrease the sampling time t1 from the starting time t1 s Determine t2, where t1 and t2 satisfy:

[0163] E[w](t1)≥B

[0164] E[w](t2)≤b

[0165] 2E[w](t2)-E[w](t2+t s )>E[w](t2-t s )

[0166] Where s and b are the upper and lower thresholds corresponding to the initial local energy capacity change rate E[w] respectively;

[0167] An arrival time estimation unit, used to calculate an estimated value of the signal arrival time at the sensor τ(w)=t2+T / 4 according to time t2, wherein T / 4 is a window function width for compensating for the effect of peak shift of E[w](t);

[0168] The arrival time difference measurement unit is used to calculate the arrival time difference measurement value of the signals of the two sensors according to the estimated arrival time values ​​of the signals corresponding to any two sensors.

[0169] As a further preferred technical solution, the calculation formula for the measured value of the signal arrival time difference of the two sensors is:

[0170] t α，β =τ α，β +argmax(w α (τ α，β +t)*w β (τ α，β +t)), t∈[-ε,ε]

[0171] In the formula, τ α，βis the difference between the estimated time of arrival of the PD signal of the jth PD source at sensor α and sensor β, t α，β is the measured value of the arrival time difference of the signals from sensor α and sensor β, ε is the size of the cross-correlation window, argmax is the parameter operator for finding the optimal solution of the function, and w α (τ α，β +t)、w β (τ α，β +t) are the received frequency domain signals of the UHF sensor α and β band arrival time difference offsets, t is the time series of the signal received by the UHF sensor, and * is the convolution operation.

[0172] As a further preferred technical solution, the positioning module 30 includes:

[0173] A spatial generation unit is used to generate M local discharge sources according to the multi-source time differences calculated for N sensors and M local discharge sources. N-1 space, in which is recorded as real space, and the space formed by the other multi-source time differences is recorded as simulated space. It represents the time difference between the Nth sensor and the 1st sensor receiving the signal of the jth local discharge source. The subscript of j from 1 to N-1 represents the time difference separation scale of the jth local discharge source relative to the same source time difference N sensors. If the values ​​are all equal, the space formed is real space;

[0174] Particle generation unit, used in M N-1 The space forms an appropriate amount of custom particles, and the particle p can be expressed as (x p ,y p , z p ,s p ), (x p ,y p , z p ) represents the coordinate of particle p in the sth space, s p Indicates that the space where particle p is located is the sth space, s = 1, 2, 3, ..., M N-1 ;

[0175] Coordinate determination unit, used to calculate M N-1 The fitness corresponding to the space is calculated, and the positioning position in the space with the best fitness is taken as the coordinate of the local discharge source.

[0176] Furthermore, the calculation formula of the fitness corresponding to the space is Among them, t i1，j It represents the time difference between the i-th sensor and the first sensor receiving the signal from the j-th local discharge source, t i1，sIt represents the time difference between the i-th sensor and the first sensor when they receive the signal from the s-th local discharge source.

[0177] It should be noted that other embodiments or implementation methods of the GIS internal multi-source partial discharge signal locating system described in the present invention can refer to the above-mentioned method embodiments, which will not be repeated here.

[0178] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0179] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0180] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for locating multi-source partial discharge signals inside a GIS, characterized in that: The method comprises: Perform noise reduction on the UHF signals collected by several sensors to obtain the pure signals corresponding to each sensor; Based on the pure signals corresponding to each sensor, the time difference between any two sensors receiving the partial discharge signal of the jth partial discharge source is calculated using the initial energy increase cross-correlation method to obtain the multi-source time difference; Generate M according to the multi-source time difference N-1 spaces and calculate the fitness of each space, and take the positioning position in the space with the best fitness as the coordinate of the local discharge source, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., M, N is the total number of sensors, and M is the total number of local discharge sources.

2. The method for locating multi-source partial discharge signals inside a GIS as claimed in claim 1, characterized in that: The noise reduction process is performed on the ultra-high frequency signals collected by the plurality of sensors to obtain the pure signals corresponding to the sensors, including: Performing singular value decomposition denoising processing on the ultra-high frequency signal to obtain a reconstructed filtered signal; The reconstructed filtered signal is subjected to dual-tree complex wavelet transform noise reduction processing, and the processed wavelet coefficients are subjected to dual-tree complex wavelet inverse transform to obtain pure signals corresponding to each sensor.

3. The method for locating multi-source partial discharge signals inside a GIS as claimed in claim 2, characterized in that: The step of performing singular value decomposition denoising processing on the ultra-high frequency signal to obtain a reconstructed filtered signal includes: Converting the UHF signal into a Hankel matrix, and performing singular value decomposition on the Hankel matrix to obtain a singular entropy increment F1 and an energy value F2; The singular entropy increment F1 and the energy value F2 are subjected to binary classification of the singular values ​​of the corresponding feature quantities by using K-means clustering, the index position of the singular value corresponding to the narrowband interference is determined and the corresponding value of the index position is set to 0; The reconstructed filtered signal with narrowband interference removed is obtained through inverse singular value decomposition.

4. The method for locating multi-source partial discharge signals inside a GIS as claimed in claim 2, characterized in that: The reconstructed filtered signal is subjected to dual-tree complex wavelet transform noise reduction processing, and the processed wavelet coefficients are subjected to dual-tree complex wavelet inverse transform to obtain a pure signal corresponding to each sensor, including: The reconstructed filtered signal is subjected to dual-tree complex wavelet transform denoising, and the lth wavelet coefficient λ on the kth scale is k，l The optimal threshold ψ is obtained using the SURE-whale optimization algorithm; Based on the optimal threshold ψ, performing secondary threshold denoising processing on the wavelet coefficients of the reconstructed filtered signal to obtain processed wavelet coefficients; The processed wavelet coefficients are inversely transformed using dual-tree complex wavelet transform to obtain the pure signals corresponding to each sensor.

5. The method for locating multi-source partial discharge signals inside a GIS as claimed in claim 4, characterized in that: The reconstructed filtered signal is subjected to dual-tree complex wavelet transform denoising, and the lth wavelet coefficient λ on the kth scale is k，l The SURE-Whale optimization algorithm is used to obtain the optimal threshold mountain, including: The variance of the wavelet coefficients at each scale is calculated, and the initial threshold corresponding to the wavelet coefficients at each scale is calculated using the SURE algorithm: Where: n k 、M k are the length of the wavelet coefficients at scale k and the median absolute deviation of the wavelet coefficients used to estimate the noise level; Randomly generate the initial whale population position, each whale represents a set of thresholds of different scales corresponding to the dual-tree complex wavelet transform, take the initial threshold as the initial optimal solution, and set the maximum number of iterations and the number of individuals; According to the initial threshold of each individual whale, the wavelet coefficients obtained by decomposition are threshold processed, and the joint fitness of the processed signals is calculated; Update the whale's position, simulate the whale's spiral rise by calculating the distance from the optimal position, and update the position: Where μ represents the step size, ξ(b) represents the gradient value of the b-th iteration, ψk(b) represents the threshold of the b-th iteration, and ψk(b+1) represents the threshold of the b+1-th iteration; When the maximum number of iterations is reached or the convergence condition is met, the optimal threshold value is obtained.

6. The method for locating multi-source partial discharge signals inside a GIS according to claim 1, characterized in that: The method of using the initial energy increase cross-correlation method based on the pure signals corresponding to each sensor to obtain the time difference between any two sensors receiving the partial discharge signal of the jth partial discharge source to obtain the multi-source time difference includes: Based on the pure signal corresponding to each sensor, the initial local energy capacity change rate of the pure signal corresponding to each sensor is calculated; Starting from the start time t1, repeatedly decrease the sampling time t1 s Determine t2, where t1 and t2 satisfy: E[w](t1)≥B E[w](t2)≤b 2E[w](t2)-E[w](t2+t s )>E[w](t2-t s ) Where B and b are the upper and lower thresholds corresponding to the initial local energy capacity change rate E[w] respectively; Calculate the estimated value of the signal arrival time at the sensor τ(w)=t2+T / 4 according to time t2, where T / 4 is the width of the window function that compensates for the peak shift effect of E[w](t); According to the estimated values ​​of the arrival time of the signals corresponding to any two sensors, the measured value of the arrival time difference of the signals of the two sensors is calculated.

7. The method for locating multi-source partial discharge signals inside a GIS as claimed in claim 6, characterized in that: The calculation formula for the measured value of the signal arrival time difference of the two sensors is: t α，β =t α，β +argmax(w α (t α，β +t)*w β (t α，β +t)),t∈[-ε,ε] In the formula, τ α，β is the difference between the estimated time of arrival of the PD signal of the jth PD source at sensor α and sensor β, t α，β is the measured value of the arrival time difference of the signals from sensor α and sensor β, ε is the size of the cross-correlation window, argmax is the parameter operator for finding the optimal solution of the function, and w a (τ α，β +t)、w β (τ α，β +t) are the received frequency domain signals of the UHF sensor α and β band arrival time difference offsets, t is the time series of the signal received by the UHF sensor, and * is the convolution operation.

8. The method for locating multi-source partial discharge signals inside a GIS as claimed in claim 6, characterized in that: The upper and lower thresholds corresponding to the initial local energy capacity change rate E[w] are: Where t is the time series of the signal received by the UHF sensor.

9. The method for locating multi-source partial discharge signals inside a GIS according to claim 1, characterized in that: The method generates M according to the multi-source time difference N-1 spaces and calculate the fitness of each space, and use the location in the space with the best fitness as the coordinates of the local discharge source, including: Generate M based on the multi-source time difference calculated for N sensors and M local discharge sources N-1 space, in which is recorded as real space, and the space formed by the other multi-source time differences is recorded as simulated space. It represents the time difference between the Nth sensor and the 1st sensor receiving the signal of the jth local discharge source. The subscript of j from 1 to N-1 represents the time difference separation scale of the jth local discharge source relative to the N sensors with the same time difference. If the values ​​are all equal, the space formed is real space. In M N-1 The space forms an appropriate amount of custom particles, and the particle p can be expressed as (x p ,y p , z p ,s p ), (x p ,y p , z p ) represents the coordinate of particle p in the sth space, s p Indicates that the space where particle p is located is the sth space, s = 1, 2, 3, ..., M N -1 ; Calculate M separately N-1 The fitness corresponding to the space is calculated, and the positioning position in the space with the best fitness is taken as the coordinate of the local discharge source.

10. The method for locating multi-source partial discharge signals inside a GIS according to claim 8, characterized in that: The calculation of M N-1 The fitness corresponding to each space is calculated, and the location in the space with the best fitness is used as the coordinate of the local discharge source, including: Calculate M separately N-1 The fitness of the space is Among them, t i1，j It represents the time difference between the i-th sensor and the first sensor receiving the signal from the j-th local discharge source, t i1，s It represents the time difference between the i-th sensor and the first sensor receiving the signal from the s-th local discharge source; The M spaces with the best fitness are selected to approximate the real space, and the positioning positions in the M spaces are used as the coordinates of the local discharge source.

Images