A power transmission line single-end fault location method and related device

Abstract

This application provides a method and related equipment for single-end fault location in transmission lines, belonging to the field of fault location technology. The method includes: constructing a complex-domain transient analytical calculation model; generating batches of fault transient traveling wave time-domain waveform samples by scanning all parameters of the transmission line; establishing a fault panoramic sample library; performing continuous wavelet transforms on the time-domain waveform samples using high-time-domain resolution wavelet bases and high-frequency-domain resolution wavelet bases respectively, generating multiple grayscale time-frequency energy maps and mapping them to RGB three channels, fusing them into a three-channel multi-wavelet time-frequency fusion feature map; constructing a multi-task learning model based on a deep residual network, the model including a shared convolutional feature extraction layer and parallel classification and regression heads, training using the three-channel fusion feature map, outputting the fault occurrence section through the classification head, and outputting the precise distance of the fault point through the regression head. This application effectively improves the fault location accuracy in complex topology scenarios such as T-junction hybrid lines.

Description

Technical Field

[0001] This application relates to the field of fault location technology, and in particular to a method and related equipment for single-end fault location of transmission lines. Background Technology

[0002] Transmission line faults account for more than 70% of all power system faults. Quickly and accurately locating the fault point after it occurs is of great significance for shortening power outage time, reducing line inspection costs, and ensuring the safe and stable operation of the power grid.

[0003] However, traveling waves experience impedance abrupt changes at overhead line-cable connection points and T-junctions, resulting in complex refraction and reflection. High-frequency components attenuate more rapidly due to the skin effect and dielectric loss, leading to wavefront dispersion and a slower rise time. Directly extracting the wavefront timing is insufficient for complex topologies, necessitating a shift towards analysis methods based on the overall waveform characteristics.

[0004] In recent years, deep learning methods, represented by convolutional neural networks, have shown significant potential in the field of power system fault diagnosis. However, the large-scale application of deep networks faces the bottleneck of insufficient training data; in addition, generating a large number of samples using electromagnetic transient simulation software such as PSCAD is time-consuming and inefficient. Summary of the Invention

[0005] The main objective of this application is to propose a method, electronic device, storage medium, and program product for single-end fault location of transmission lines, which can efficiently generate sufficient training samples and build a matching deep learning model, while avoiding data leakage from interfering with model evaluation, thus becoming a key to promoting the practical application of intelligent fault location technology.

[0006] To achieve the above objectives, one aspect of this application proposes a method for single-end fault location in transmission lines, the method comprising: A complex-domain transient analytical calculation model was constructed, and fault transient traveling wave time-domain waveform samples were generated in batches by scanning all parameters of the transmission line to establish a fault panoramic sample library. For the time-domain waveform sample of the fault transient traveling wave, continuous wavelet transform is performed using high time-domain resolution wavelet basis and high-frequency resolution wavelet basis respectively to generate multiple gray-level time-frequency energy maps. The multiple gray-level time-frequency energy maps are mapped to the RGB three channels respectively and fused into a three-channel multi-wavelet time-frequency fusion feature map. A multi-task learning model based on a deep residual network is constructed. The multi-task learning model includes a shared convolutional feature extraction layer and parallel classification and regression heads. The multi-task learning model is trained using the three-channel multi-wavelet time-frequency fusion feature map. The classification head outputs the fault occurrence segment, and the regression head outputs the precise distance of the fault point.

[0007] In some embodiments, the construction of the complex-domain transient analytical calculation model, the batch generation of fault transient traveling wave time-domain waveform samples by scanning all parameters of the transmission line, and the establishment of a fault panoramic sample library include: Based on the tower geometry and conductor parameters of the transmission line, the frequency response parameters of the transmission line are obtained. A vector fitting algorithm using rational function approximation is then employed to extract the series impedance per unit length. and complex parallel admittance Complex domain representation:

[0008] In the formula, R ( s () represents the resistance per unit length of the line. L ( s () represents the inductance per unit length. G ( s )and C ( s ) represent the conductivity and susceptance per unit length, respectively, and s is the Laplace operator. s = c +j ω , c Here, j is the damping coefficient, and j is the imaginary unit in the complex domain. Based on the long-line equation, a two-port complex domain admittance matrix model of the transmission line is established:

[0009] In the formula, and These are the three-phase currents at the beginning and end of the transmission line, respectively. and These are the three-phase voltages at the beginning and end of the transmission line, respectively. and These are the self-admittance matrix and mutual admittance matrix of the line, respectively; Port equivalent modeling of the system impedance is performed, and the transformer is equivalently represented using a П-type equivalent circuit. The complex domain self-admittance sub-matrices, mutual admittance sub-matrices, system impedance admittance matrix, and transformer equivalent admittance matrix of each line are combined according to the admittance matrix construction rules to construct the system complex domain node admittance matrix Y after the fault. all ( s ), for Y all ( s Inverse the equation to obtain the complex domain nodal impedance matrix Z of the system. all ( s ); From the complex domain node impedance matrix Z of the system all ( s Extract the self-impedance Z corresponding to the fault node.ff ( s The positive-sequence impedance, negative-sequence impedance, and zero-sequence impedance of the fault node are obtained through symmetrical component transformation. A step fault excitation voltage source is applied to the fault node. A composite sequence network is constructed according to the boundary conditions of different fault types. The short-circuit current of each sequence complex domain at the fault port is calculated. The three-phase complex domain transient short-circuit current at the fault port is obtained by inverse transformation of the symmetrical components. According to the system complex domain node impedance matrix Z all ( s The three-phase complex transient short-circuit current at the fault port is calculated, and the three-phase complex transient voltage and three-phase complex transient current at the measurement point are converted into time-domain transient traveling wave waveforms through numerical inverse Laplace transform. A full parameter scan is performed on the fault closing angle, transition resistance, and fault distance to generate time-domain waveform samples of fault transient traveling waves under different fault factors in batches, and a fault panoramic sample library is established.

[0010] In some embodiments, the time-domain expression of the fault excitation voltage source is:

[0011] In the formula, Indicates the fault phase The fault excitation voltage source The instantaneous voltage amplitude at the fault point before the fault occurred. The fault closing angle at the fault point before the fault occurred. For system power frequency, The time when the fault occurred. This is a unit step signal.

[0012] In some embodiments, the formula for the numerical inverse Laplace transform is:

[0013] In the formula, It is a complex sequence. This is the numerical solution in the time domain after transformation. For the Hanning window function, , These represent the number of sampling points for the time-domain numerical solution and the complex-domain angular frequency, respectively.

[0014] In some embodiments, the high time-domain resolution wavelet basis is the Gauss1 wavelet basis and the Mexh wavelet basis, and the high-frequency domain resolution wavelet basis is the CMO1.5-1.0 wavelet basis; For the time-domain waveform samples of the fault transient traveling wave, continuous wavelet transform is performed using both high-time-domain resolution wavelet basis and high-frequency domain resolution wavelet basis to generate multiple gray-level time-frequency energy maps. These multiple gray-level time-frequency energy maps are then mapped onto the RGB three channels and fused into a three-channel multi-wavelet time-frequency fusion feature map, including: For the fault phase voltage waveform or fault phase current waveform of the same fault sample, continuous wavelet transform is performed using the gaus1 wavelet basis, mexh wavelet basis and cmor1.5-1.0 wavelet basis respectively, and each transform generates a grayscale time-frequency energy map. The grayscale time-frequency energy map generated by the gaus1 wavelet basis is assigned to the R channel, the grayscale time-frequency energy map generated by the mexh wavelet basis is assigned to the G channel, and the grayscale time-frequency energy map generated by the cmor1.5-1.0 wavelet basis is assigned to the B channel. These are then merged into a three-channel image, which serves as the three-channel multi-wavelet time-frequency fusion feature map.

[0015] In some embodiments, the classification head is structured as a fully connected layer with K-dimensional output, and the posterior probability of each segment is output via a Softmax function; the regression head is structured as a fully connected layer with 1-dimensional output, and linearly mapped to [0, ..., ... L The interval is given, and the predicted distance is output. L This represents the total length of the transmission line.

[0016] In some embodiments, the joint loss function of the multi-task learning model is:

[0017]

[0018] In the formula, Cross-entropy classification loss, The mean squared error regression loss, For the number of categories, The total number of training samples, and As preset task weights, For observation samples Category The predicted probability, For observation samples Real section labels, This represents the actual fault distance of the sample. This is the predicted fault distance output by the model.

[0019] In some embodiments, a deterministic partitioning strategy based on fault distance numerical features is used to construct training and validation sets, specifically by using a preset step size. Fault samples are generated, and the fault distance values ​​are uniformly represented as follows: ,in For integers, the partitioning rule is as follows: when When the number is even, the corresponding fault samples are assigned to the training set; when When the number is odd, the corresponding fault samples are assigned to the validation set.

[0020] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0021] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0022] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0023] The embodiments of this application include at least the following beneficial effects: This application provides a method, electronic device, storage medium, and program product for single-end fault location of transmission lines. This solution achieves full-parameter scanning and batch generation of fault transient traveling wave time-domain waveforms by constructing a complex-domain transient analytical calculation model, effectively solving the bottleneck problems of scarce training samples for deep learning fault location models and the long time consumption and low efficiency of traditional electromagnetic transient simulation. On this basis, continuous wavelet transforms are performed on the generated time-domain waveforms using high-time-domain resolution wavelet bases and high-frequency domain resolution wavelet bases, and multiple gray-scale time-frequency energy maps are mapped and fused into a three-channel multi-wavelet time-frequency fusion feature map. This breaks through the limitation that a single waveform or a single wavelet base cannot simultaneously take into account the time-domain wavefront abrupt change and the frequency-domain fine texture features due to the constraint of the Heisenberg uncertainty principle, and realizes the visual complementarity and integration of cross-resolution time-frequency information. Furthermore, a deep residual network multi-task learning model containing a shared convolutional feature extraction layer and parallel classification and regression heads is constructed and trained using the above-mentioned three-channel fusion feature map, so that fault segment identification and accurate location are jointly optimized and mutually promoted under the same framework. The three technical steps are linked and work together to form a complete end-to-end solution from data generation to feature representation to model training. This significantly improves the accuracy of fault segmentation and ranging in complex topology scenarios such as T-connected hybrid lines, and effectively promotes the practical application of intelligent fault location technology. Attached Figure Description

[0024] Figure 1This is a flowchart of a single-end fault location method for transmission lines provided in an embodiment of this application; Figure 2 This is a composite sequence network topology diagram of different fault types in the fault network in the embodiments of this application; Figure 3 This is a schematic diagram of a simulation model of a 110kV AC transmission line system built to verify the validity of this application; Figure 4 This is a fault segmentation result diagram of the model in the embodiment of this application on the test sample; Figure 5 This is a scatter plot of the absolute regression error of the model as a function of fault distance in the embodiments of this application; Figure 6 This is a cumulative probability distribution diagram of the absolute error of the fault in the embodiments of this application; Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0027] In recent years, deep learning methods, represented by Convolutional Neural Networks (CNNs), have shown significant potential in the field of power system fault diagnosis. The first existing technology proposes a method based on deep mining of traveling wave panoramic features, using CNNs to process time-domain waveforms, demonstrating the superiority of deep learning in resisting transition resistance and noise. The second existing technology achieves near-end fault identification by recognizing time-domain waveform features. The third existing technology uses artificial neural networks (ANNs) to handle flexible DC grid faults, verifying the classification ability of neural networks under complex topologies. However, most of the above existing studies use relatively simple early neural network models. For example, the third existing technology uses a traditional shallow ANN, and the first existing technology uses a variant of the classic LeNet-5 model proposed in 1998. Faced with the complex traveling wave reflection textures in T-connected hybrid lines and the weak features under high-resistance faults, these early shallow network models have obvious limitations, such as limited feature extraction capabilities, encountering the gradient vanishing problem, and insufficient generalization ability when facing strong noise or parameter fluctuations.

[0028] To address the limitations of shallow networks, the fourth existing technical solution proposes a deep residual network (ResNet). By introducing "skip connections," a network with up to 152 layers was successfully trained, significantly improving the model's ability to represent complex features. However, the large-scale application of deep networks still faces the bottleneck of insufficient training data. Methods using electromagnetic transient simulation software such as PSCAD to generate large numbers of samples are time-consuming and inefficient, failing to meet the demands of deep learning models for massive training data. Therefore, exploring efficient data generation methods and establishing matching deep learning models are crucial for promoting the practical application of intelligent fault location technology.

[0029] Furthermore, in standard deep learning practices, datasets are typically divided into training and validation sets using random or hierarchical random partitioning. However, for the specific task of transmission line fault location, random partitioning presents a hidden but serious problem: data leakage. If the distance values ​​of two fault samples are extremely close or identical, and these two samples are randomly assigned to the training and validation sets, the model has actually seen samples very similar to the validation samples during the training phase, thus artificially inflating its performance on the validation set. This problem has not yet received sufficient attention or effective solutions in existing research.

[0030] In view of this, this application provides a method, electronic device, storage medium, and program product for single-end fault location of transmission lines based on complex domain transient analysis, three-channel multi-wavelet time-frequency fusion, and deep residual network. The solution includes the following core contents: I. Constructing a complex-domain transient analytical calculation model and generating a comprehensive fault sample library. Using a vector fitting algorithm with rational function approximation, the frequency response of the transmission line is extracted, and the complex-domain impedance and parallel admittance per unit length are calculated. Based on long-line equations, a system complex-domain nodal admittance matrix including the system impedance equivalent network and the transformer's П-type equivalent circuit is established. A step fault excitation voltage source is applied to the fault node, and a composite sequence network is constructed according to the boundary conditions of different fault types. The three-phase complex-domain transient voltage and current sequences at the measurement points are solved, and converted into time-domain transient traveling wave waveforms through numerical inverse Laplace transform. An offline fault sample library can be built in batches through full-parameter scanning.

[0031] II. Generation of Time-Frequency Complementary Three-Channel Multi-Wavelet Fusion Feature Map. For the extracted transient traveling wave time-domain waveform, based on the Heisenberg uncertainty principle constraint characteristics of continuous wavelet transform in time-frequency analysis, high time-domain resolution bases (gaus1, mexh) and high-frequency domain resolution bases (cmor1.5-1.0) are selected to extract time-frequency features from the time-domain waveform of the same fault sample. The three generated independent grayscale time-frequency energy maps are mapped to the RGB three channels respectively and fused into a single time-frequency panoramic three-channel image containing cross-resolution complementary information, which serves as the input feature of the deep residual network.

[0032] III. Parallel Multi-Task Deep Residual Network Training Based on Spatial Deterministic Partitioning. A deep learning network based on the standard ResNet-50 is established for the ranging task. Parallel "classification heads" and "regression heads" are constructed after the convolutional feature extraction layer to output the precise distances to the fault location and fault segment, respectively. A deterministic alternating partitioning strategy based on the numerical spatial features of the fault distance is adopted to construct training and validation sets, reducing data leakage caused by random partitioning. Automatic mixed-precision training and Warmup cosine annealing scheduling strategies are introduced to perform stable training using a joint equal-weight loss function, ultimately achieving high-precision single-end fault ranging of transmission lines.

[0033] This application provides a method for single-end fault location in transmission lines, relating to the field of fault location technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle-mounted terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the single-end fault location method for transmission lines, but is not limited to the above forms.

[0034] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0035] like Figure 1 As shown, this embodiment provides a method for single-end fault location in transmission lines, including the following steps: Step S1: Construct a complex-domain transient analytical calculation model, scan all parameters of the transmission line to generate batch fault transient traveling wave time-domain waveform samples, and establish a fault panoramic sample library; Step S2: For the fault transient traveling wave time-domain waveform sample, continuous wavelet transform is performed using high time-domain resolution wavelet basis and high-frequency domain resolution wavelet basis respectively to generate multiple gray-level time-frequency energy maps. The multiple gray-level time-frequency energy maps are mapped to the RGB three channels respectively and fused into a three-channel multi-wavelet time-frequency fusion feature map. Step S3: Construct a multi-task learning model based on a deep residual network. The multi-task learning model includes a shared convolutional feature extraction layer and parallel classification and regression heads. The multi-task learning model is trained using the three-channel multi-wavelet time-frequency fusion feature map. The classification head outputs the fault occurrence segment, and the regression head outputs the precise distance of the fault point.

[0036] The solutions of the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific application examples.

[0037] (1) Construction of complex transient analytical computation model and sample generation This embodiment describes in detail the construction process of the complex-domain transient analytical calculation model, which is used to generate time-domain waveform samples of fault transient traveling waves in batches.

[0038] 1.1) Obtaining and vector fitting of line frequency response parameters Based on the actual tower geometry and related conductor parameters of the transmission line, the frequency response parameters of the transmission line can be obtained through the detailed parameter modeling module of the PSCAD / EMTDC electromagnetic transient simulation program. Then, based on descending matching, a vector fitting algorithm using rational function approximation is used to obtain the series impedance per unit length. Z (s) and complex parallel admittance Y Complex representation of (s):

[0039] In the formula, R ( s () represents the resistance per unit length of the line. L ( s () represents the inductance per unit length. G ( s )and C ( s These represent the conductivity and susceptance per unit length, respectively. s This is the Laplace operator. s = c +j ω , c The damping coefficient has a value of [value missing]. c = 2lnN / (N·dt), where j is the imaginary unit of the complex field. ω =(2m+1)π / (Ndt), where N=T / dt, T is the analytical calculation time of this model, dt is the analytical calculation step size, and m=0,1,……N-1.

[0040] 1.2) Establishment of the complex domain admittance matrix model of a two-port transmission line Based on the long-line equation, the two-port complex domain admittance matrix model of the transmission line can be further obtained:

[0041] In the formula, I1(s) and I2(s) are the three-phase currents at the beginning and end of the transmission line, respectively, and U1(s) and U2(s) are the three-phase voltages at the beginning and end of the transmission line, respectively. y s (s) and y m (s) represent the self-admittance and mutual admittance matrices of the line, respectively, both with a dimension of 3×3, and the corresponding calculation formulas are as follows:

[0042] In the formula, l This is the length of the transmission line segment. T Here is the Clarke transformation matrix. T and T -1 as follows:

[0043] 1.3) Equivalent modeling of system components Port equivalent modeling of the system impedance is performed, assuming the system impedance is... Z s The two-port complex domain admittance matrix model of the system impedance is:

[0044] The transformer is equivalently evaluated using a type-P equivalent circuit. Assume nodes... i When connected to a transformer, if the primary side of the transformer uses a star-grounded connection, the equivalent impedance of the load connected to the transformer, referred to the primary side via the transformer, is represented by the three-phase admittance matrix Y, which has positive and negative zero sequences. T,i ( s )for:

[0045] In the formula, S is the symmetric component transformation matrix, S -1 It is the inverse transformation matrix of the symmetric components. Z TL1,i (s), Z TL2,i (s) and Z TL0,i (s) represent the positive, negative, and zero-sequence impedances transferred to the primary side of the transformer, respectively. Z L,i (s) represents the equivalent impedance of the load. Z T,i (s) represents the transformer impedance. k i This represents the transformer's turns ratio.

[0046] If the primary side of the transformer uses a star-connected ungrounded configuration, the transformer does not have an equivalent zero-sequence network. Y TL,i ( s )for:

[0047] 1.4) Construction of the system complex domain nodal admittance matrix Assuming the system had n nodes before the failure, and the failure point became the new node after the failure, a system admittance matrix of dimension 3(n+1)×3(n+1) is constructed. The complex domain self-admittance submatrix of each line can be obtained according to the above steps. y s ( s ) and line mutual admittance sub-matrix y m ( s The line self-admittance sub-matrix, mutual admittance sub-matrix, system impedance admittance matrix, and transformer equivalent admittance matrix, according to the rules for constructing admittance matrices, together constitute the complex domain node admittance matrix Y of the system after a fault. all ( s The inverse of this matrix yields the complex domain nodal impedance matrix Z of the system after the fault. all ( s From the system complex domain nodal impedance matrix Z) all ( s In the process, extract the self-impedance Z corresponding to the fault node. ff ( s This can be transformed into an order impedance matrix using symmetric component transformation. In the formula, the operator... a = e j120° , a 2 = e j240° .

[0048]

[0049] In the formula, Z Fff(1) ( s () represents the positive sequence impedance corresponding to the self-impedance of the fault node. Z Fff(2) ( s ) is the negative sequence impedance. Z Fff(0) ( s ) is the zero-sequence impedance.

[0050] 1.5) Fault excitation application and complex short-circuit current calculation A fault excitation voltage source is applied to the faulty node. The time-domain expression of the fault excitation voltage source is as follows:

[0051] In the formula, Δ u f-φ ( t ) indicates the faulty phase φ The fault excitation voltage source U f The instantaneous voltage amplitude at the fault point before the fault occurred. θ f-φ The fault closing angle at the fault point before the fault occurred. α For system power frequency, t 0 represents the time when the fault occurred. ε ( t The given equation represents a unit step signal. Applying a Laplace transform to both sides of the equation, followed by a symmetric component transform, yields its positive-sequence component Δ. u f(1) ( s ), negative order component Δ u f(2) ( s ) and zero-sequence component Δ u f(0) (s).

[0052] Based on the boundary conditions for different fault types, the following can be constructed: Figure 2 The composite sequence network shown can be used to calculate the complex short-circuit current under different fault types. Among them, R f This is the transition resistance during a fault.

[0053] 1) Single-phase ground fault:

[0054] 2) Two-phase ground fault: Using the superposition theorem, calculate the positive-sequence component Δ. u f(1) ( s ), negative order component Δ u f(2) ( s ) and zero-sequence component Δ u f(0) The sequence short-circuit currents under the action of (s) are then added together:

[0055]

[0056]

[0057]

[0058] 3) Two-phase fault:

[0059] 4) Three-phase fault:

[0060] After obtaining the sequence complex domain short-circuit currents at the fault port, the three-phase complex domain transient short-circuit currents at the fault port can be obtained by inverse transformation of the symmetrical components:

[0061] 1.6) Calculation and time-domain conversion of transient voltage and current at the measurement point The system's complex domain nodal impedance matrix Z all ( s The three-phase complex transient short-circuit current at the fault port obtained from the above formula is used to obtain the three-phase complex transient voltage at the measurement point using the following formula:

[0062] In the formula, U x1 ( s ) and U x2 ( s Z represents the three-phase complex transient short-circuit voltage at the beginning and end nodes of the line, respectively. fx1 ( s () is the impedance matrix Z from the complex domain nodes all ( s The three mutual impedance matrices Z between the fault point f and the line head node x1 extracted from the data are shown in the figure. fx2 ( s Similarly, the three-phase complex transient current at the beginning of the line can be further obtained from this:

[0063] The obtained three-phase transient voltages and currents are both complex domain sequences, which need to be converted into time domain sequences through inverse numerical Laplace transform. The formula for inverse numerical Laplace transform is as follows:

[0064] In the formula, F ( s () is a complex field sequence. f ( k Δ t Then it is F ( s The numerical solution in the time domain after transformation, where σ is the Hanning window function, k , m These are the number of sampling points for the time-domain numerical solution and the complex-domain angular frequency, respectively, and their values ​​are... k , m =0,1,…, N -1.

[0065] The complex-domain analytical calculation model established in this module can generate transient traveling wave waveforms of three-phase fault voltage and three-phase fault current at measurement points consistent with actual fault conditions. The aim is to establish a fault sample library for training deep residual networks. (2) Generation of three-channel multi-wavelet time-frequency fusion feature map This embodiment describes in detail the method for generating a three-channel multi-wavelet time-frequency fusion feature map. Continuous Wavelet Transform (CWT) is a linear time-frequency analysis method for a given mother wavelet. ψ ( t ),Signal x ( t The definition of ) is:

[0066] in a As a scale factor, b This is the translation factor. Specifically, the wavelet basis function... ψ a,b ( t ) at scale a The time domain support width Δ t a and frequency domain support width Δ ω a Satisfy Δ t a Δ ω a ≥1 / 2, and the product of the two is equal only when the wavelet is a Gabor function.

[0067] The time-frequency resolution of continuous wavelet transform is constrained by the Heisenberg uncertainty principle; time-domain resolution and frequency-domain resolution cannot be simultaneously optimized. Different wavelet genes employ different Heisenberg constraint allocation strategies, resulting in systematic differences in the information distribution of the generated time-frequency maps. High time-domain resolution bases (such as mexh and gaus1) preferentially encode fault information as the precise location of the wavefront on the time axis, while high frequency-domain resolution bases (such as cmor and morl) preferentially encode fault information as subtle changes in spectral texture. Due to the inherent incompatibility of the Heisenberg constraint, the two encoding strategies capture different aspects of the same physical phenomenon.

[0068] Based on the above principles, this embodiment proposes a three-channel multi-wavelet time-frequency fusion feature strategy. The CWT time-frequency images generated by three wavelet bases are stacked into an RGB three-channel image, which is used as input to the ResNet-50 residual network. Different wavelet bases have complementary perspectives when characterizing the same physical fault. When the three channels are jointly input, the amount of information obtained by the network is greater than that of any single channel. Specifically, for the fault phase voltage or current waveform of the same fault sample, continuous wavelet transform is performed using three mother wavelets, each transform generating a 224×224 grayscale time-frequency energy map. The three grayscale images are assigned R, G, and B channels respectively, and merged into a 224×224×3 three-channel image. The three channels are simultaneously accessed by the first layer convolutional kernel of the CNN, allowing cross-resolution information synthesis to be completed in a shallow layer.

[0069] The three wavelet bases ultimately determined in this embodiment are gaus1 (high time-domain resolution base), mexh (high time-domain resolution base), and cmor1.5-1.0 (high-frequency domain resolution base). Specifically, the gray-scale time-frequency energy map generated by the gaus1 wavelet base is assigned to the R channel, the gray-scale time-frequency energy map generated by the mexh wavelet base is assigned to the G channel, and the gray-scale time-frequency energy map generated by the cmor1.5-1.0 wavelet base is assigned to the B channel.

[0070] (3) Construction and training of deep residual network model This embodiment describes in detail the construction of the deep residual network model, the training / validation set partitioning strategy, and the training protocol.

[0071] 3.1) Training and validation set partitioning based on fault distance spatial features In standard deep learning practices, datasets are typically divided into training and validation / test sets using random or hierarchical random partitioning. However, for the specific task of power transmission line fault localization, random partitioning presents a hidden but serious problem: data leakage. Data leakage refers to the unexpected overlap of information between the training and validation sets, leading to a systematic overestimation of the model's performance on the validation set. In the context of fault localization, if the distance between two fault samples is extremely close or identical, and these two samples are randomly assigned to the training and validation sets, the model has already seen samples very similar to the validation samples during the training phase, resulting in a falsely inflated performance on the validation set.

[0072] To mitigate the impact of data leakage, this patent proposes a deterministic partitioning strategy based on the numerical characteristics of fault distance. Specifically, samples are generated in 0.5km increments, and the fault distance value can be uniformly represented as... ,in k Integer ( k= 0, 1, 2, ..., n). The partitioning rules are as follows:

[0073] Taking the verification embodiment of this example, The fault distance is generated in increments of 0.5 km. This deterministic partitioning strategy physically isolates the training and validation samples in space, ensuring that there is no cross-contamination between neighboring samples in the fault distance dimension between the training and validation sets.

[0074] 3.2) General Training Protocol Automatic Mixed Precision Training: Utilizing NVIDIA's automatic mixed precision training framework, some operations are dynamically downgraded from FP32 to FP16 during training, reducing memory usage by approximately 40% and increasing training speed by approximately 1.5-2 times without sacrificing model accuracy.

[0075] Warmup + Cosine Annealing Learning Rate Scheduling: A linear warmup strategy is used in the initial training phase. In the first Twarmup iterations, the learning rate increases linearly from 0 to a preset maximum value to avoid excessively large gradient updates in the initial phase from disrupting the randomly initialized weights. After warmup, the learning rate decays according to the cosine annealing strategy.

[0076] The implementation details and hyperparameter settings of the above strategies are summarized in Table 1.

[0077] Table 1 Training Hyperparameter Settings

[0078] 3.3) Deep Residual Network Structure This embodiment uses the standard ResNet-50 as a fixed reference model. Fault localization requires solving two tasks: the fault location segment (classification task) and the precise distance to the fault point (regression task). To meet this requirement, this embodiment replaces the 1000-dimensional vector of the output layer with two parallel output heads. The classification head has a fully connected layer structure—K-dimensional output, which outputs the posterior probability of each segment via Softmax. The regression head has a fully connected layer structure—1-dimensional output, which is linearly mapped to [0, ... L The output heads predict the distance. Both output heads share all the convolutional feature extraction layers of ResNet-50, separating only at the fully connected layers.

[0079] 3.4) Joint Loss Function Multi-task output requires setting a joint loss function. Let the classification loss be the cross-entropy loss. L cls The regression loss is the mean squared error loss. L reg :

[0080] The joint loss function is a weighted sum of the two:

[0081] In the formula, Cross-entropy classification loss, The mean squared error regression loss, For the number of categories, The total number of training samples, and As preset task weights, For observation samples Category The predicted probability, For observation samples Real section labels, This represents the actual fault distance of the sample. This represents the predicted fault distance output by the model. In this embodiment, we take... In other words, classification and regression tasks are weighted equally. This choice is based on the following considerations: In the early stages of training, the classification loss is usually greater than the regression loss, and the gradient signal of the classification task plays a dominant role in the early stages of training, which helps the network to first establish a coarse-grained segment discrimination ability; as training progresses, the classification loss decreases rapidly, and the gradient contribution of the regression loss gradually increases, and the network turns to fine-tuning the distance prediction accuracy. This natural gradient balancing mechanism keeps the simple equal-weighted training process stable.

[0082] (4) Example illustration To verify the effectiveness of the proposed single-end fault location method for transmission lines based on complex domain transient analysis, three-channel multi-wavelet time-frequency fusion, and deep residual network, a simulation model was built in PSCAD / EMTDC electromagnetic transient simulation software. Figure 3 The simulation model of the 110kV AC transmission line system is shown. The time window is T=5ms, dt is 10us, and the simulation recording step size is consistent with the analytical calculation, i.e., Δ... t The distance is 10µs. Fault points f are set on lines MK1, K1K2, K2T, TP, and TO. The fault distance is calculated from point M (the measurement point). Among them, lines SM, MK1, K2T, TP, and TO are all overhead transmission lines, while K1K2 is a cable line. The fault section is mainly identified by the following three physical sections: Seg0 is the main section, from the measurement point M to the T-junction (0-60km), which includes 20km of overhead line, 20km of cable, and 20km of overhead line in sequence; Seg1 is a T-junction branch (60-80km), with a total length of 20km of overhead line; Seg2 is another T-junction branch (60-72km), with a total length of 12km of overhead line.

[0083] Taking the most common and widely covered single-phase ground fault in transmission lines as an example, 2196 test samples were generated randomly on each section of the line to test the fault location performance of the model under the real statistical distribution.

[0084] Sample generation involves three core variables: fault closing angle, transition resistance, and fault distance, all of which are fully scanned within their respective ranges at fixed steps. The specific parameter ranges and the number of possible values ​​are summarized in Table 2.

[0085] Table 2. Scan range of sample parameters for complex domain analysis

[0086] A total of 17,005 samples were generated, all of which were generated in batches based on the complex domain analytical calculation model of Module 1.

[0087] Furthermore, the transient traveling waves of transmission line faults are inevitably affected by thermal noise, electromagnetic coupling interference, and analog-to-digital conversion quantization errors in actual measurements. As a first-order approximation model, Gaussian white noise is often used for unified modeling in engineering. This example sets three signal-to-noise ratio (SNR) levels, generating corresponding copies for each of the 17,005 original samples, ultimately constructing an augmented dataset of approximately 51,015 samples: samples without added noise serve as the model performance benchmark; the 40 dB level simulates typical experimental environments; and the 30 dB level simulates harsh operating conditions such as strong electromagnetic interference or channel degradation.

[0088] Figure 4 The fault segmentation results for all test samples were analyzed, with 2182 samples accurately segmented and 14 samples incorrectly segmented, resulting in an accuracy rate of 99.36%. Figure 5 The scatter plot shows the absolute error of the model regression as a function of fault distance. The mean absolute error for all test samples is 0.1892 km. To make the ranging results more intuitive, Figure 6 The cumulative error probability distribution of all test samples is shown. It can be seen that 90% of the test samples have an absolute error within 0.37km, 95% of the test samples have an absolute error within 0.54km, and 99% of the test samples have an absolute error within 1.2km, which verifies the effectiveness of the method proposed in this invention patent.

[0089] In summary, compared with the prior art, this application has at least the following advantages and beneficial effects: 1) An innovative complex-domain analytical data generation mechanism significantly improves sample construction efficiency. This application directly generates fault time-domain waveforms under specific fault factors, realizing the batch offline generation of tens of thousands of high-quality fault samples under full-parameter scanning, thus eliminating the dependence on traditional complex simulation models.

[0090] 2) A multi-wavelet basis, multi-channel time-frequency fusion strategy achieves complementary advantages in features. Traditional single waveform or single wavelet basis methods are constrained by the Heisenberg uncertainty principle and cannot simultaneously take into account temporal wavefront abrupt changes and fine-grained texture features in the frequency domain. This application jointly employs high temporal resolution wavelet basis (gaus1, mexh) and high-frequency resolution wavelet basis (cmor1.5-1.0), stacking the generated time-frequency images into an RGB three-channel image and inputting it into ResNet-50 to achieve cross-resolution information integration, forming complementary visual features and greatly enhancing the model's ability to mine traveling wave features.

[0091] 3) Deterministic spatial feature partitioning strategy to suppress data leakage risk. Traditional random partitioning results in fault samples with adjacent distances existing in both the training and validation sets, artificially inflating the model's performance on the validation set. This application proposes a deterministic parity-interval partitioning strategy based on the numerical spatial features of fault distance, which physically isolates training and validation samples in space, reducing the risk of data leakage caused by random partitioning.

[0092] 4) Parallel multi-task learning architecture of deep residual network, taking into account both segment identification and distance measurement. This application uses ResNet-50 as the backbone network and constructs parallel "classification head" and "regression head" after the convolutional feature extraction layer to output the fault occurrence segment and the accurate distance of the fault point, respectively, realizing the integration of fault segment identification and accurate distance measurement.

[0093] 5) Superior fault location performance. Experimental results show that the fault location accuracy of this application reaches 99.36% in the T-connection mixed line scenario, the average absolute error of all test samples is 0.1892km, and the absolute error of 90% of the test samples is within 0.37km, which verifies the effectiveness of the method of this application.

[0094] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0095] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0096] Please see Figure 7 , Figure 7 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 701 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 702 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702 and is called and executed by the processor 701 using the methods described in the embodiments of this application. The input / output interface 703 is used to implement information input and output; The communication interface 704 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 705 transmits information between various components of the device (e.g., processor 701, memory 702, input / output interface 703, and communication interface 704); The processor 701, memory 702, input / output interface 703, and communication interface 704 are connected to each other within the device via bus 705.

[0097] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0098] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0099] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0100] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0101] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0102] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0103] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0105] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0106] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0107] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0109] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0110] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0111] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0112] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A single-ended fault location method for a power transmission line, characterized in that, The method includes the following steps: A complex-domain transient analytical calculation model was constructed, and fault transient traveling wave time-domain waveform samples were generated in batches by scanning all parameters of the transmission line to establish a fault panoramic sample library. For the time-domain waveform sample of the fault transient traveling wave, continuous wavelet transform is performed using high time-domain resolution wavelet basis and high-frequency resolution wavelet basis respectively to generate multiple gray-level time-frequency energy maps. The multiple gray-level time-frequency energy maps are mapped to the RGB three channels respectively and fused into a three-channel multi-wavelet time-frequency fusion feature map. A multi-task learning model based on a deep residual network is constructed. The multi-task learning model includes a shared convolutional feature extraction layer and parallel classification and regression heads. The multi-task learning model is trained using the three-channel multi-wavelet time-frequency fusion feature map. The classification head outputs the fault occurrence segment, and the regression head outputs the precise distance of the fault point.

2. The method according to claim 1, characterized in that, The construction of the complex-domain transient analytical calculation model involves batch generating fault transient traveling wave time-domain waveform samples through full parameter scanning of transmission lines, establishing a comprehensive fault sample library, including: Based on the tower geometry and conductor parameters of the transmission line, the frequency response parameters of the transmission line are obtained. A vector fitting algorithm using rational function approximation is then employed to extract the series impedance per unit length. and complex parallel admittance Complex domain representation: In the formula, R ( s () represents the resistance per unit length of the line. L ( s () represents the inductance per unit length. G ( s )and C ( s ) represent the conductivity and susceptance per unit length, respectively, and s is the Laplace operator. s = c +j ω , c Here, j is the damping coefficient, and j is the imaginary unit in the complex domain. Based on the long-line equation, a two-port complex domain admittance matrix model of the transmission line is established: In the formula, and These are the three-phase currents at the beginning and end of the transmission line, respectively. and These are the three-phase voltages at the beginning and end of the transmission line, respectively. and These are the self-admittance matrix and mutual admittance matrix of the line, respectively; Port equivalent modeling of the system impedance is performed, and the transformer is equivalently represented using a П-type equivalent circuit. The complex domain self-admittance sub-matrices, mutual admittance sub-matrices, system impedance admittance matrix, and transformer equivalent admittance matrix of each line are combined according to the admittance matrix construction rules to construct the system complex domain node admittance matrix Y after the fault. all ( s ), for Y all ( s Inverse the equation to obtain the complex domain nodal impedance matrix Z of the system. all ( s ); From the complex domain node impedance matrix Z of the system all ( s Extract the self-impedance Z corresponding to the fault node. ff ( s The positive-sequence impedance, negative-sequence impedance, and zero-sequence impedance of the fault node are obtained through symmetrical component transformation. A step fault excitation voltage source is applied to the fault node. A composite sequence network is constructed according to the boundary conditions of different fault types. The short-circuit current of each sequence complex domain at the fault port is calculated. The three-phase complex domain transient short-circuit current at the fault port is obtained by inverse transformation of the symmetrical components. According to the system complex domain node impedance matrix Z all ( s The three-phase complex transient short-circuit current at the fault port is calculated, and the three-phase complex transient voltage and three-phase complex transient current at the measurement point are converted into time-domain transient traveling wave waveforms through numerical inverse Laplace transform. A full parameter scan is performed on the fault closing angle, transition resistance, and fault distance to generate time-domain waveform samples of fault transient traveling waves under different fault factors in batches, and a fault panoramic sample library is established.

3. The method according to claim 2, characterized in that, The time-domain expression of the fault excitation voltage source is: In the formula, Indicates the fault phase The fault excitation voltage source The instantaneous voltage amplitude at the fault point before the fault occurred. The fault closing angle at the fault point before the fault occurred. For system power frequency, The time when the fault occurred. This is a unit step signal.

4. The method according to claim 2, characterized in that, The formula for the numerical inverse Laplace transform is: In the formula, It is a complex sequence. This is the numerical solution in the time domain after transformation. For the Hanning window function, , These represent the number of sampling points for the time-domain numerical solution and the complex-domain angular frequency, respectively.

5. The method according to claim 1, characterized in that, The high time-domain resolution wavelet basis is the gaus1 wavelet basis and the mexh wavelet basis, and the high-frequency domain resolution wavelet basis is the cmor1.5-1.0 wavelet basis; For the time-domain waveform samples of the fault transient traveling wave, continuous wavelet transform is performed using both high-time-domain resolution wavelet basis and high-frequency domain resolution wavelet basis to generate multiple gray-level time-frequency energy maps. These multiple gray-level time-frequency energy maps are then mapped onto the RGB three channels and fused into a three-channel multi-wavelet time-frequency fusion feature map, including: For the fault phase voltage waveform or fault phase current waveform of the same fault sample, continuous wavelet transform is performed using the gaus1 wavelet basis, mexh wavelet basis and cmor1.5-1.0 wavelet basis respectively, and each transform generates a grayscale time-frequency energy map. The grayscale time-frequency energy map generated by the gaus1 wavelet basis is assigned to the R channel, the grayscale time-frequency energy map generated by the mexh wavelet basis is assigned to the G channel, and the grayscale time-frequency energy map generated by the cmor1.5-1.0 wavelet basis is assigned to the B channel. These are then merged into a three-channel image, which serves as the three-channel multi-wavelet time-frequency fusion feature map.

6. The method according to claim 1, characterized in that, The classification head consists of a fully connected layer with a K-dimensional output, which outputs the posterior probability of each segment via a Softmax function; the regression head consists of a fully connected layer with a 1-dimensional output, which is linearly mapped to [0, ..., ...] after passing through a Sigmoid function. L The interval is given, and the predicted distance is output. L This represents the total length of the transmission line.

7. The method according to claim 1, characterized in that, The joint loss function of the multi-task learning model is: In the formula, Cross-entropy classification loss, The mean squared error regression loss, For the number of categories, The total number of training samples, and As preset task weights, For observation samples Category The predicted probability, For observation samples Real section labels, This represents the actual fault distance of the sample. This is the predicted fault distance output by the model.

8. The method according to claim 1, characterized in that, A deterministic partitioning strategy based on numerical features of fault distance is used to construct training and validation sets, specifically: using a preset step size... Fault samples are generated, and the fault distance values ​​are uniformly represented as follows: ,in For integers, the partitioning rule is as follows: when When the number is even, the corresponding fault samples are assigned to the training set; when When the number is odd, the corresponding fault samples are assigned to the validation set.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 8.

Images