Power transmission line fault location method based on ICEEMDAN and improved YOLOv8

Abstract

The application discloses a power transmission line fault ranging method based on ICEEMDAN and improved YOLOv8, respectively utilizes the improved YOLOv8 network to construct a fault identification model and a fault positioning model, utilizes the Kelvin-Bell transformation method to perform phase-mode transformation on three-phase voltage transient signals of a fault line, selects a line-mode component as a research object of a fault traveling wave, adopts an ICEEMDAN algorithm to decompose the line-mode component, uses an NTEO algorithm to strengthen modal components of the line-mode component, highlights fault signal characteristics, and obtains a fault traveling wave energy spectrum diagram and an amplified traveling wave energy spectrum diagram; the fault identification model is used to determine an initial fault time range, and then the fault positioning model is used to obtain a fault time point, i.e., accurate arrival times of fault traveling wave fronts at two ends of the line, according to the amplified traveling wave energy spectrum diagram.The application improves the accuracy of line fault ranging, is not affected by transition resistance, fault distance and transmission lines, has high reliability and good practicability.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of power system fault location, and particularly relates to a power transmission line fault distance measurement method based on ICEEMDAN and improved YOLOv8. BACKGROUND

[0002] The rapid growth of distribution networks and the increase in load demand make the distribution network play a crucial role in urban development. However, the distribution network is prone to line faults due to various events such as adverse weather conditions, instantaneous contact with animals and plants, infrastructure aging, etc. These line faults not only result in high maintenance costs, but also cause productivity reduction and significant economic losses. Therefore, an accurate and rapid line fault location method is of great significance for the safe and stable operation of the distribution system.

[0003] In real life, many principles and algorithms for power grid line fault location have been proposed according to actual needs, among which the traveling wave method is widely used. However, the key to applying the traveling wave method is to accurately determine the traveling wave front, which will directly affect the accuracy of the traveling wave fault location. In addition, most improved traveling wave methods also require manual selection of the initial traveling wave front to ensure data reliability.

[0004] Currently, the traveling wave arrival time determination method uses wavelet modulus maximum value to detect the abrupt point of the traveling wave signal, determines the time when the fault traveling wave arrives at the monitoring point, and realizes line fault location, which has achieved good results. However, the wavelet transform method needs to select a suitable wavelet basis function and decomposition scale, and has poor adaptability.

[0005] The fault location method based on EMD decomposition has phenomena such as modal aliasing and end effect, which will have a certain impact on the determination of the traveling wave arrival time, and the transformation calculation amount is large, making it difficult to achieve satisfactory results. ICEEMDAN adds noise after decomposition and calculates the mean value based on EMD, which not only preserves the effective information of the original signal, but also greatly reduces the redundant components brought by the added white noise, further improving the decomposition effect. SUMMARY

[0006] The purpose of the present application is to solve the above problems, and to provide a power transmission line fault distance measurement method based on ICEEMDAN and improved YOLOv8. The key data of the line fault signal is converted into a picture, which is visualized and has high positioning accuracy. The improved YOLOv8 is used to construct a fault recognition model and a fault location model, solving the problem of manual selection of the initial traveling wave front in the traveling wave method.

[0007] In order to achieve the above object, the technical scheme provided by the application is a power transmission line fault ranging method based on ICEEMDAN and improved YOLOv8, which respectively uses an improved YOLOv8 network to construct a fault identification model and a fault positioning model, the fault identification model is used to identify and determine whether a fault occurs and an initial fault time range according to a fault voltage waveform diagram, and the fault positioning model is used to obtain a reduced fault time range or a fault time point according to a fault traveling wave energy spectrum diagram.

[0008] The power transmission line fault ranging method based on ICEEMDAN and improved YOLOv8 comprises the following steps:

[0009] Step 1: Collecting three-phase voltage transient signals at both ends of the line after a fault occurs to obtain a fault voltage waveform diagram, and using the Kelvin-Bell transformation method to perform phase-to-mode transformation on the three-phase voltage transient signals and selecting line-mode components as fault traveling wave research objects;

[0010] Step 2: Using the ICEEMDAN algorithm to decompose the line-mode components, using the NTEO algorithm to strengthen the modal components of the line-mode components, and highlighting the fault signal features to obtain a fault traveling wave energy spectrum diagram and an amplified traveling wave energy spectrum diagram;

[0011] Step 3: Using the fault voltage waveform diagram obtained in step 1 and the fault traveling wave energy spectrum diagram and the amplified traveling wave energy spectrum diagram obtained in step 2 to establish a traveling wave detection data set;

[0012] Step 4: Constructing a fault identification model and a fault positioning model respectively and training using the traveling wave detection data set;

[0013] Step 5: For a newly occurring single-phase line-to-ground fault, reading voltage transient signals at both ends of the line, obtaining a fault voltage waveform diagram and inputting the fault voltage waveform diagram into the fault identification model, and using the fault identification model to determine whether there is a line fault and to determine an initial fault time range;

[0014] Step 6: According to the initial fault time range, obtaining line-mode components of three-phase voltage transient signals at a corresponding time stage, obtaining a fault traveling wave energy spectrum diagram through ICEEMDAN decomposition and NTEO algorithm strengthening of step 2, and inputting the fault traveling wave energy spectrum diagram into the fault positioning model to obtain a reduced fault time range;

[0015] Step 7: According to the reduced fault time range, obtaining line-mode components of three-phase voltage transient signals at a corresponding time stage, obtaining an amplified traveling wave energy spectrum diagram through ICEEMDAN decomposition and NTEO algorithm strengthening of step 2, and using the fault positioning model to obtain a fault time point, i.e., an accurate arrival time of a fault traveling wave head at both ends of the line, according to the amplified traveling wave energy spectrum diagram;

[0016] Step 8: According to the accurate arrival time of the fault wave head at both ends of the line obtained in step 6, the fault point position is calculated.

[0017] Further, in step 2, the ICEEMDAN is used to decompose the fault traveling wave voltage line mode component, including:

[0018] 1) Based on the original signal s, a new signal sequence is generated after adding the auxiliary noise of empirical mode decomposition:

[0019] ; (1)

[0020] Wherein, represents the added i-th group of zero-mean Gaussian white noise; represents the signal-to-noise ratio of the initial added noise; represents the i-th intrinsic mode function component of signal empirical mode decomposition; represents the new sequence formed after adding the i-th group of auxiliary noise;

[0021] 2) The local mean of the signal is calculated , and the overall mean of the signal is calculated to obtain the first-order residual component ,

[0022] ; (2)

[0023] ; (3)

[0024] Wherein, represents the local mean of the signal, represents the overall mean of the signal;

[0025] 3) The first-order intrinsic mode function component of the signal is obtained by subtracting the original signal from the first-order residual component ,

[0026] ; (4)

[0027] 4) Based on the first-order residual component , a plurality of auxiliary noises are added to calculate the mean to obtain a new second-order residual component , and the difference is calculated to obtain the second-order intrinsic mode function component of the signal ,

[0028] ; (5)

[0029] ; (6)

[0030] In the formula, represents the signal-to-noise ratio of the second added noise;

[0031] 5) repeat the calculation to get the kth stage residual component and the kth stage intrinsic modal function component ,

[0032] ; (7)

[0033] ; (8)

[0034] wherein , respectively represent the k-1th, kth stage residual component; represents the signal-to-noise ratio of the kth added noise;

[0035] 6) continue iteration until the resulting residual signal can no longer be decomposed, satisfying the decomposition termination condition, to obtain all intrinsic modal function components and residual components.

[0036] Further, in step 2, the NTEO algorithm comprises:

[0037] For a continuous-time signal , the TEO energy operator is expressed as:

[0038] ; (9)

[0039] wherein , respectively represent the first and second derivatives of ;

[0040] And for a discrete-time signal , the TEO energy operator is expressed as:

[0041] ; (10)

[0042] A resolution parameter is introduced, and the NTEO energy operator is expressed as:

[0043] ; (11)

[0044] Increasing the interval of the sampled signal points and thus reducing the interference of the continuous noise, the sensitivity to the signal frequency is improved, thereby enhancing the noise immunity.

[0045] Preferably, in step 3, the traveling wave detection data set comprises:

[0046] 1) a fault voltage waveform diagram, i.e. a three-phase voltage transient signal diagram after the fault occurs;

[0047] 2) the fault traveling wave energy spectrum diagram, the fault traveling wave line module component is obtained from the fault three-phase voltage transient signal in the fault occurrence time range through phase module transformation, and then the first order intrinsic mode function component is obtained by decomposing the line module component through the ICEEMDAN algorithm, and the NTEO energy operator is used to strengthen the first order intrinsic mode function component, so that the fault traveling wave characteristic energy spectrum diagram is obtained;

[0048] 3) the amplified traveling wave energy spectrum diagram, the fault traveling wave line module component is obtained from the fault three-phase voltage transient signal in the reduced fault traveling wave time range through phase module transformation, and then the first order intrinsic mode function component is obtained by decomposing the line module component through the ICEEMDAN algorithm, and the NTEO energy operator is used to strengthen the first order intrinsic mode function component, so that the amplified fault traveling wave characteristic energy spectrum diagram is obtained.

[0049] Preferably, the construction process of the improved YOLOv8 network specifically comprises:

[0050] a) the GhostConv module in the GhostNet network is integrated into the BottleNeck module to obtain a GhostBottleNeck module;

[0051] b) the GhostBottleNeck module is used to replace the BottleNeck layer of the C3 module of YOLOv5 to obtain a C3Ghost module;

[0052] c) a ScalSeq module is constructed according to the principle of the scale sequence feature fusion module in the ASF-YOLO network;

[0053] d) all C2f modules in the backbone network Backbone and the neck network Neck in YOLOv8 are replaced by C3Ghost modules, and the Concat module and the Upsample module in the neck network Neck are replaced by ScalSeq modules, so that the improved YOLOv8 network is obtained.

[0054] Preferably, the C3Ghost module comprises two convolution layers Conv in parallel, wherein one convolution layer Conv is connected with the convolution layer Conv at the end through a plurality of GhostBottleNeck modules, a splicing layer Concat and the convolution layer Conv at the end in parallel, and the other convolution layer Conv in parallel is directly connected with the convolution layer Conv at the end.

[0055] Preferably, the GhostBottleNeck module comprises a parallel ghost convolution layer GhostConv and a depthwise separable convolution layer DWConv connected to the end fusion layer Add through a convolution layer Conv, and the ghost convolution layer GhostConv is connected to the end fusion layer Add through another depthwise separable convolution layer DWConv, another ghost convolution layer GhostConv.

[0056] Further, in step 8, the calculation of the line fault point position comprises:

[0057] 1) According to the parameters of the transmission line, the wave speed of the zero-mode or line-mode component corresponding to the fault traveling wave of the line is calculated , and the calculation formula is:

[0058] ; (12)

[0059] In the formula, , respectively, the inductance and capacitance of the unit length line under the same phase sequence;

[0060] 2) According to the double-end traveling wave distance measurement principle, the position of the fault on the line is F point, and the time when the fault occurs is , the fault traveling wave propagates from the F point to the M and N points at the left and right ends of the line, wherein M is the left end point of the line, and N is the right end point of the line;

[0061] The time when the initial wave head of the fault traveling wave reaches the M point is recorded as , and the time when it reaches the N point is recorded as , then:

[0062] ; (13)

[0063] ; (14)

[0064] ; (15)

[0065] In the formula, represents the distance from the F point to the M point;

[0066] 3) For the mixed line of overhead line and cable line, the cable line is equivalent to the corresponding length of overhead line, and the wave speed of the traveling wave in the overhead line and the cable line is represented by , respectively, and the calculation formula for converting the wave speed of different lines is:

[0067] ; (16)

[0068] In the formula to normalize the cable equivalent overhead line length, to represent the cable length in the distribution network.

[0069] The fault location system of the power transmission line fault location method comprises:

[0070] The traveling wave generation module is used for simulating and obtaining the three-phase fault transient voltage signal under the 10MHz sampling rate after the line fault occurs, and outputs the fault voltage waveform diagram.

[0071] The fault voltage waveform detection module is used for processing the fault voltage waveform diagram to obtain the fault voltage waveform detection diagram with a bounding box, the fault target probability and the fault initial time range of the traveling wave main body.

[0072] The phase-mode transformation module is used for converting the three-phase fault transient voltage signal in the fault initial time range into the corresponding line-mode component through the Kelvin-Bell transformation matrix.

[0073] The ICCEMDAN algorithm module is used for decomposing the fault feature of the line-mode component in the given time range through the ICCEMDAN algorithm to obtain the corresponding first-order intrinsic mode function component.

[0074] The NTEO algorithm module is used for strengthening the fault feature of the first-order intrinsic mode function component in the given time range through the NTEO algorithm to obtain the corresponding energy spectrum data, the fault traveling wave energy spectrum diagram or the amplified traveling wave energy spectrum diagram.

[0075] The traveling wave energy spectrum detection module is used for obtaining the fault traveling wave energy spectrum detection diagram with a bounding box, the fault target probability and the fault reduced time range of the traveling wave main body by respectively passing the line-mode component in the fault initial time range through the ICCEMDAN algorithm module and the NTEO algorithm module; and obtaining the amplified traveling wave energy spectrum detection diagram with a bounding box, the fault target probability and the accurate arrival time of the traveling wave head by screening the energy spectrum data in the fault reduced time range.

[0076] The line fault point calculation module is used for integrating the accurate arrival time of the traveling wave head at both ends of the line, i.e. the time corresponding to the maximum point in the energy spectrum diagram, to calculate the accurate fault point position of the line.

[0077] Compared with the prior art, the beneficial effects of the present application include:

[0078] 1) The application utilizes the Kelvin transformation method to perform phase-mode transformation on the three-phase voltage transient signal of the fault line, selects the line-mode component as the research object of the fault traveling wave, decomposes the line-mode component by using the ICEEMDAN algorithm, uses the NTEO algorithm to strengthen the modal component of the line-mode component, highlights the fault signal characteristics, obtains the fault traveling wave energy spectrum diagram and the amplified traveling wave energy spectrum diagram, determines the initial fault time range by using the fault identification model, and then obtains the fault time point, i.e., the accurate arrival time of the fault traveling wave head at both ends of the line according to the amplified traveling wave energy spectrum diagram, thereby improving the accuracy of the line fault distance measurement, and the method is not affected by the transition resistance, fault distance and transmission line, has high reliability and good practicability.

[0079] 2) The application uses the ICEEMDAN algorithm to decompose the line-mode component of the fault traveling wave signal, and then uses the NTEO algorithm to strengthen the fault traveling wave characteristics, so that the obtained initial wave head time of the fault traveling wave has extremely high accuracy.

[0080] 3) The application improves the YOLOv8 model for the traveling wave detection task, the improved YOLOv8 model has the characteristics of light weight, can automatically read the fault traveling wave information according to the picture form of the traveling wave key data, and accurately selects the initial wave head of the fault traveling wave. BRIEF DESCRIPTION OF DRAWINGS

[0081] The application will be further described below in combination with the drawings and examples.

[0082] Figure 1 It is a schematic diagram of the power transmission line fault distance measurement method of the embodiment of the application.

[0083] Figure 2 It is a simulation line diagram of the embodiment of the application.

[0084] Figure 3 It is a fault voltage waveform diagram of the embodiment of the application.

[0085] Figure 4 It is a fault traveling wave energy spectrum detection diagram of the embodiment of the application.

[0086] Figure 5 It is an amplified traveling wave energy spectrum detection diagram of the embodiment of the application.

[0087] Figure 6 It is a double-end traveling wave method principle diagram of the embodiment of the application.

[0088] Figure 7 It is a structure schematic diagram of the improved YOLOv8 of the embodiment of the application.

[0089] Figure 8 It is a structure schematic diagram of the C3Ghost module of the embodiment of the application.

[0090] Figure 9 Schematic diagram of the ScalSeq module of an embodiment of the application. DETAILED DESCRIPTION

[0091] In order to make the technical problems, technical solutions and beneficial effects of the present application clearer, the present application will be further described in detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application and not to limit the present application.

[0092] As shown in Figure 1 , the power line fault ranging method based on ICEEMDAN and improved YOLOv8 includes the following steps:

[0093] Step one: Collect a large number of three-phase voltage transient signals at the M and N ends of the line after the fault, collect the fault voltage waveform diagram, use the Kelvin-Bell transformation to perform phase-to-mode transformation on the three-phase voltage transient signals, obtain the line mode component and the ground mode component of the voltage, and select the line mode component as the fault traveling wave research object;

[0094] Use PSCAD to simulate five power lines, as shown in Figure 2 , including pure overhead lines, pure cable lines and hybrid lines. Set one of the lines to have a single-phase ground fault at a sampling rate of 10 MHz, and repeat the simulation many times under different fault occurrence times, different grounding resistances, and different fault occurrence positions, and obtain the voltage transient signal saving file at both ends of the line each time.

[0095] Step two: decompose the traveling wave voltage line mode component using the ICEEMDAN algorithm, strengthen the prominent features of the traveling wave signal using the NTEO algorithm, collect the fault traveling wave energy spectrum diagram and the amplified traveling wave energy spectrum diagram, and establish a traveling wave detection dataset;

[0096] Read each obtained simulation file, perform phase-to-mode transformation on the three-phase transient voltage signals, and decompose the traveling wave voltage line mode component using the ICEEMDAN algorithm. The specific method is as follows:

[0097] 1) Based on the original signal s, add auxiliary noise for EMD decomposition to generate a new signal sequence:

[0098] (1)

[0099] wherein, represents the i-th set of added zero-mean Gaussian white noise; represents the signal-to-noise ratio of the noise; represents the i-th intrinsic mode function component of a signal after EMD decomposition; represents a new sequence formed after adding i sets of auxiliary noise.

[0100] 2) Obtain the local mean of the signal , and obtain the first order residual component by taking the overall mean of the signal :

[0101] (2)

[0102] (3)

[0103] wherein, represents the local mean of the signal, represents the overall mean of the signal, represents the kth stage residual component of the original signal.

[0104] 3) Obtain the first order intrinsic modal function component of the signal by subtracting the original signal from the first order residual component , that is,

[0105] (4)

[0106] 4) Based on the first order residual component , add multiple sets of auxiliary noise to obtain a new second order residual component by taking the mean , and obtain the second order intrinsic modal function component of the signal by subtracting the first order residual component :

[0107] (5)

[0108] (6)

[0109] 5) Repeat the calculation of the kth stage residual component and the kth stage IMF component:

[0110] (7)

[0111] (8)

[0112] 6) Continue iteration until the residual signal obtained cannot be further decomposed, and the decomposition termination condition is met, to obtain all intrinsic modal function components and residual components.

[0113] The first order intrinsic modal function component obtained by decomposing the fault traveling wave is strengthened by using the NTEO algorithm to highlight the characteristics of the traveling wave signal. The specific method is:

[0114] For a continuous time signal , the TEO energy operator is expressed as:

[0115] (9)

[0116] In the formula , respectively represent the first and second derivatives of ;

[0117] For discrete-time signals , the TEO energy operator is expressed as:

[0118] (10)

[0119] Introducing a resolution parameter , the NTEO energy operator is expressed as:

[0120] (11)

[0121] Increasing the interval of sampling signal points and thus reducing the interference of continuous noise, the sensitivity to signal frequency is improved, thereby enhancing the anti-noise performance of the algorithm.

[0122] Step three: establishing a traveling wave detection data set;

[0123] Collecting the fault traveling wave energy spectrum obtained by the NTEO algorithm and the amplified traveling wave energy spectrum, as well as the fault voltage waveform obtained before, to form a traveling wave detection data set.

[0124] Step four: building an improved YOLOv8 network and training using the traveling wave detection data set;

[0125] The improved YOLOv8 model is shown in Figure 7 , and the construction process of the improved YOLOv8 network specifically includes:

[0126] a) Integrate the GhostConv module in the GhostNet network into the BottleNeck module to obtain the GhostBottleNeck module;

[0127] b) Replace the BottleNeck layer of the C3 module of YOLOv5 with the GhostBottleNeck module to obtain the C3Ghost module, and the structure is shown in Figure 8 ;

[0128] c) According to the principle of the scale sequence feature fusion module in the ASF-YOLO network, construct the ScalSeq module, and the structure is shown in Figure 9 ;

[0129] d) Replace all C2f modules in the backbone network Backbone and neck network Neck in YOLOv8 with C3Ghost modules, and replace the Concat module and Upsample module in the neck network Neck with ScalSeq modules to obtain the improved YOLOv8 network.

[0130] The training of the improved YOLOv8 model requires labeling of specific targets in the data set pictures. The Labelimg tool can be used for labeling the traveling wave category and the digital category to obtain a labeled traveling wave detection data set. The data set is divided into training, verification, and test sets in a ratio of 7:2:1.

[0131] An operating environment is established, the traveling wave detection data set is imported, and the corresponding training parameters are set. The improved YOLOv8 model is trained to obtain an optimized improved YOLOv8 model for traveling wave detection.

[0132] Step five: a new single-phase ground fault occurs. The three-phase voltage transient signals at the M and N ends of the line after the fault are collected, the fault voltage waveform diagram is output, and the trained improved YOLOv8 model is called to determine the initial fault time range.

[0133] Five transmission lines are simulated using PSCAD, as shown in Figure 2 . The five transmission lines include pure overhead lines, pure cable lines, and hybrid lines. A single-phase ground fault occurs on one of the lines at a sampling rate of 10 MHz. The fault occurrence time, ground resistance, and fault location are different from those in the traveling wave detection data set. The voltage transient signals at the two ends of the line are saved and the three-phase fault voltage waveform diagram is output, as shown in Figure 3 .

[0134] The three-phase fault voltage waveform diagram is input into the trained improved YOLOv8 model. The fault voltage waveform detection diagram, the probability of the detection target, and the detection frame position and size of the detection target are output. The initial fault time range is obtained by integrating the information.

[0135] Step six: according to the initial fault time range, the corresponding three-phase voltage transient signals are read. After phase-to-mode transformation, the ICEEMDAN algorithm is used to decompose the traveling wave voltage line mode component. The NTEO algorithm is used to strengthen the prominent features of the traveling wave signal to obtain the fault traveling wave energy spectrum diagram. The improved YOLOv8 model is used to obtain a reduced fault time range.

[0136] From the above fault occurrence time range, the voltage transient signal files at the two ends of the line are read. First, the phase-to-mode transformation is performed to obtain the traveling wave voltage line mode component. Then, the ICEEMDAN algorithm is used to decompose the traveling wave voltage line mode component to obtain the first-order intrinsic mode function component. Finally, the NTEO algorithm is used to strengthen the prominent features of the first-order intrinsic mode function component to make the fault traveling wave features more concrete. The fault traveling wave energy spectrum diagram is output.

[0137] The fault traveling wave energy spectrum diagram is input into the trained improved YOLOv8 model. The fault traveling wave energy spectrum detection diagram is output, as shown in Figure 4As shown, and the probability of detecting the target and the detection frame position and size of the frame-selected detection target, the integrated information obtains the reduced fault time range.

[0138] Step seven: According to the reduced fault time range, read the traveling wave characteristic signal obtained by ICEEMDAN and NTEO decomposition at the corresponding stage, output the amplified traveling wave energy spectrum graph, and use the improved YOLOv8 model trained to obtain the accurate arrival time of the initial fault traveling wave head at both ends of the line.

[0139] According to the reduced fault time range obtained above, read the voltage transient signal file at both ends of the line, select the first-order intrinsic mode function component in the time range decomposed by ICEEMDAN, and then use NTEO algorithm to strengthen the characteristic of the traveling wave signal to obtain the amplified traveling wave energy spectrum graph.

[0140] Input the amplified traveling wave energy spectrum graph into the improved YOLOv8 model trained, and output the amplified traveling wave energy spectrum detection graph as shown. Figure 5 , and the probability of detecting the target and the detection frame position and size of the frame-selected detection target, the integrated information obtains the accurate arrival time of the initial fault traveling wave head at both ends of the line.

[0141] Step eight: According to the accurate arrival time of the initial fault traveling wave head at both ends of the line, accurately calculate the fault point position.

[0142] The principle of double-end traveling wave method is shown in Figure 6 After obtaining the accurate arrival time of the initial fault traveling wave head at both ends of the line, the fault point position is calculated according to the principle, and the calculation process is as follows:

[0143] 1) According to the parameters of the transmission line, calculate the wave speed of the zero-mode or line-mode component corresponding to the fault traveling wave of the line , and the calculation formula is:

[0144] (12)

[0145] In the formula, , are the inductance and capacitance of the unit length line under the same phase sequence, respectively;

[0146] 2) According to the principle of double-end traveling wave distance measurement, assume that the position of the fault on the line is F point, and the time when the fault occurs is , the fault traveling wave propagates from F point to M and N points at both ends of the line, where M is the left end point of the line and N is the right end point of the line.

[0147] The time when the initial wave head of the fault traveling wave arrives at M point is recorded as , and the time when it arrives at N point is recorded as , then:

[0148] (13)

[0149] (14)

[0150] (15)

[0151] 3) For the mixed line of overhead line and cable line, the cable line is equivalent to the corresponding length of overhead line, and the wave velocity of traveling wave in overhead line and cable line is represented by , respectively. The calculation formula of different line wave velocity conversion is:

[0152] (16)

[0153] In the formula, is the equivalent overhead line length of normalized cable, and represents the cable length in distribution network.

Claims

1. A transmission line fault location method based on ICEEMDAN and improved YOLOv8, characterized by: The improved YOLOv8 network is used to construct a fault identification model and a fault location model. The fault identification model is used to identify whether a fault has occurred and determine the initial fault time range based on the fault voltage waveform. The fault location model is used to obtain a narrowed fault time range or a fault time point according to the fault traveling wave energy spectrum diagram; The method comprises the following steps: Step 1: Collect the three-phase voltage transient signals at both ends of the line after the fault occurs to obtain the fault voltage waveform. Use the Karenberg transform method to perform phase-mode transformation on the three-phase voltage transient signals, and select the line mode component as the fault traveling wave research object; Step 2: Use the ICEEMDAN algorithm to decompose the line mode components, and use the NTEO algorithm to enhance the modal components of the line mode components, highlight the fault signal characteristics, and obtain the fault traveling wave energy spectrum and the amplified traveling wave energy spectrum; Step 3: Using the fault voltage waveform obtained in step 1 and the fault traveling wave energy spectrum and amplified traveling wave energy spectrum obtained in step 2, a traveling wave detection data set is established; Step 4: Build a fault identification model and a fault location model respectively, and train them using the traveling wave detection dataset; Step 5: For a newly occurring single-phase line ground fault, read the voltage transient signals at both ends of the line, obtain the fault voltage waveform, and input it into the fault identification model. The fault identification model is used to determine whether there is a line fault and determine the initial fault time range. Step 6: Based on the initial fault time range, obtain the line mode components of the three-phase voltage transient signal for the corresponding time period. After decomposition by ICEEMDAN and enhancement by the NTEO algorithm in step 2, obtain the fault traveling wave energy spectrum. This is input into the fault location model to obtain the narrowed fault time range. Step 7: Based on the narrowed fault time range, obtain the line mode components of the three-phase voltage transient signal for the corresponding time period. After decomposition using the ICEEMDAN algorithm and enhancement using the NTEO algorithm in step 2, obtain an amplified traveling wave energy spectrum. Use the fault location model to determine the fault time point, i.e., the exact arrival time of the fault traveling wave front at both ends of the line, based on the amplified traveling wave energy spectrum. Step 8: Based on the accurate arrival time of the fault traveling wave front at both ends of the line obtained in step 7, calculate the fault point location.

2. The transmission line fault location method according to claim 1, characterized in that: In step 2, ICEEMDAN is used to decompose the line mode components of the fault traveling wave voltage, including: 1) Based on the original signal s, a new signal sequence is generated by adding auxiliary noise of empirical mode decomposition: ；(1) in, represents the added i-th group of zero-mean Gaussian white noise; represents the signal-to-noise ratio of the initially added noise; represents the i-th intrinsic mode function component of the empirical mode decomposition of the signal; represents the new sequence formed after adding the i-th group of auxiliary noise; 2) Find the local mean of the signal , and then calculate the overall average of the signal to obtain the first-order residual component , ；(2) ；(3) in, represents the local mean of the signal, Indicates the overall mean of the signal; 3) The first-order intrinsic mode function component of the signal is obtained by taking the difference between the original signal and the first-order residual component , ；(4) 4) Take the first-order residual component Based on this, we add multiple groups of auxiliary noise and calculate the mean to get the new second-order residual component. , and then calculate the difference to get the second-order intrinsic mode function component of the signal , ；(5) ；(6) In the formula represents the signal-to-noise ratio of the second noise addition; 5) Repeat the calculation to obtain the residual component of the kth stage and the kth stage intrinsic mode function component , ；(7) ；(8) In the formula 、 Represent the residual components of the k-1th and kth stages respectively; represents the signal-to-noise ratio of the k-th noise addition; 6) Continue iterating until the residual signal can no longer be decomposed and the decomposition termination condition is met, and all intrinsic mode function components and residual components are obtained.

3. The transmission line fault location method according to claim 2, characterized in that: In step 2, the NTEO algorithm includes: For continuous-time signals , TEO energy operator Expressed as: ；(9) In the formula 、 Respectively The first and second derivatives of For discrete-time signals , the TEO energy operator is expressed as: ；(10) Introducing resolution parameters , the NTEO energy operator is expressed as: ；(11) Increasing the interval between sampling signal points can reduce the interference of continuous noise, improve the sensitivity to signal frequency, and thus enhance noise resistance.

4. The transmission line fault location method according to claim 3, characterized in that: In step 3, the traveling wave detection dataset includes: 1) Fault voltage waveform, i.e. three-phase voltage transient signal diagram after the fault occurs; 2) Fault traveling wave energy spectrum: The fault three-phase voltage transient signal within the fault occurrence time range is transformed into the fault traveling wave line mode component through phase mode transformation. The line mode component is then decomposed using the ICEEMDAN algorithm to obtain the first-order intrinsic mode function component. The first-order intrinsic mode function component is enhanced using the NTEO energy operator to obtain the fault traveling wave characteristic energy spectrum; 3) Amplified traveling wave energy spectrum: The fault three-phase voltage transient signal within the reduced fault traveling wave time range is transformed into the fault traveling wave line mode component by phase mode transformation. The line mode component is then decomposed using the ICEEMDAN algorithm to obtain the first-order intrinsic mode function component. The first-order intrinsic mode function component is enhanced using the NTEO energy operator to obtain the amplified fault traveling wave characteristic energy spectrum.

5. The power transmission line fault location method according to claim 1, 2, 3 or 4, characterized in that: The construction process of the improved YOLOv8 network specifically includes: a) Integrate the GhostConv module in the GhostNet network into the BottleNeck module to obtain the GhostBottleNeck module; b) Use the GhostBottleNeck module to replace the BottleNeck layer of the C3 module of YOLOv8 to obtain the C3Ghost module; c) Based on the principle of the scale sequence feature fusion module in the ASF-YOLO network, the ScalSeq module is constructed; d) Replace all the C2f modules in the backbone network Backbone and the neck network Neck in YOLOv8 with C3Ghost modules, and replace the Concat module and Upsample module in the neck network Neck with ScalSeq modules to obtain an improved YOLOv8 network.

6. The power transmission line fault location method according to claim 5, characterized in that: The C3Ghost module includes two parallel convolutional layers Conv, one of which is connected to the terminal convolutional layer Conv via multiple GhostBottleNeck modules and a concatenation layer Concat, and the other parallel convolutional layer Conv is directly connected to the terminal convolutional layer Conv.

7. The power transmission line fault location method according to claim 6, characterized in that: The GhostBottleNeck module includes a parallel ghost convolution layer GhostConv and a depth-wise separable convolution layer DWConv. The depth-wise separable convolution layer DWConv is connected to the terminal fusion layer Add via the convolution layer Conv. The ghost convolution layer GhostConv is connected to the terminal fusion layer Add via another depth-wise separable convolution layer DWConv and another ghost convolution layer GhostConv.

8. The power transmission line fault location method according to claim 7, characterized in that: In step 8, the calculation to obtain the position of the line fault point includes: 1) Calculate the wave velocity of the zero mode or line mode component corresponding to the fault traveling wave of the line according to the parameters of the transmission line , the calculation formula is: ；（12） In the formula 、 are the inductance and capacitance of the unit length line under the same phase sequence respectively; 2) According to the principle of double-terminal traveling wave ranging, the position of the fault on the line is set as point F, and the time of the fault is recorded as , the fault traveling wave propagates from point F to points M and N at the left and right ends of the line, where M is the left end point of the line and N is the right end point of the line; The moment when the initial wave head of the fault traveling wave reaches point M is recorded as The time of reaching point N is recorded as , then: ；（13） ；（14） ；（15） In the formula Indicates the distance from point F to point M; 3) For the mixed lines of overhead lines and cable lines, normalization is performed, and the cable lines are equivalent to overhead lines of corresponding lengths. The speeds of traveling waves in overhead lines and cable lines are expressed as 、 The calculation formula for converting wave speeds of different lines is: ；（16） In the formula is the normalized equivalent overhead line length of the cable, Indicates the length of cables in the distribution network.

Images