A Method and System for Predicting Single-Phase Ground Fault Sections in Distribution Lines Based on Transient Signals

Abstract

This invention discloses a method and system for predicting single-phase grounding fault sections in distribution lines based on transient signals. Wavelet transform is used to perform multi-scale analysis on transient current or voltage signals, extracting wavelet coefficients and their energy that reflect fault characteristics. These are then input as feature values ​​into a pre-trained neural network model for distributed training, predicting the fault section. This invention significantly improves the accuracy of fault location, reduces false alarms and missed alarms, greatly shortens fault location time, and improves the fault response speed of the power system, helping to reduce power outage time and economic losses.

Description

Technical Field

[0001] This invention relates to the field of power distribution line fault technology, specifically to a method and system for predicting single-phase grounding fault sections in power distribution lines based on transient signals. Background Technology

[0002] A distribution network refers to a power grid that receives electrical energy from the transmission network or regional power plants and distributes it locally or in stages according to voltage to various users through distribution facilities. It consists of overhead lines, cables, poles, distribution transformers, disconnect switches, reactive power compensators, and some auxiliary facilities, and plays an important role in distributing electrical energy within the power grid.

[0003] Single-phase grounding faults in distribution lines are a common type of fault in power systems, and their rapid and accurate location is crucial for ensuring the continuity and stability of power supply. Traditional fault segment prediction methods mainly rely on steady-state signal analysis, fault recording devices, and traveling wave methods. However, these methods often suffer from insufficient location accuracy, slow response speed, and poor adaptability when faced with complex network structures, transient fault characteristics, and environmental factors, leading to prolonged fault investigation time and affecting the quality of power supply services. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for predicting single-phase grounding fault sections in power distribution lines based on transient signals. This invention significantly improves the accuracy of fault location and reduces false alarms and missed alarms by leveraging the powerful nonlinear fitting capabilities of deep learning models combined with the refined features provided by wavelet analysis. This invention demonstrates good accuracy in predicting fault sections even when facing complex network structures, transient fault characteristics, and the influence of environmental factors.

[0005] To achieve this objective, the present invention provides a method for predicting single-phase grounding fault sections in distribution lines based on transient signals, which includes the following steps:

[0006] Collect transient current or transient voltage signals generated in power distribution lines when a single-phase ground fault occurs;

[0007] Wavelet transform is used to perform multi-scale analysis on the acquired transient current signal or transient voltage signal to obtain the wavelet coefficients of the transient current signal or transient voltage signal at each wavelet transform scale and at the time axis or spatial axis position;

[0008] Calculate the energy of wavelet coefficients of transient current or transient voltage signals at each scale and at each time or spatial axis position, and determine and select the largest energy among all wavelet transform scales as the main characteristic value of transient current or transient voltage signals;

[0009] The main feature values ​​of transient current or transient voltage signals are input into multiple trained neural network models for distributed training to obtain multiple corresponding fault segment prediction results. These multiple fault segment prediction results are then weighted and averaged for fusion. ， The fused fault section prediction results are obtained.

[0010] A single-phase grounding fault section prediction system for distribution lines based on transient signals includes a signal acquisition module, a wavelet analysis module, a feature calculation module, and multiple neural network modules.

[0011] The signal acquisition module is used to acquire transient current signals or transient voltage signals generated in power distribution lines when a single-phase ground fault occurs.

[0012] The wavelet analysis module is used to perform wavelet transform multi-scale analysis on the transient current signal or transient voltage signal acquired by the signal acquisition module, and obtain the wavelet coefficients of the transient current signal or transient voltage signal at each wavelet transform scale and time axis or spatial axis position.

[0013] The feature calculation module is used to calculate the energy of the wavelet coefficients of the transient current signal or transient voltage signal obtained by the wavelet analysis module at each scale and time axis or spatial axis position at each wavelet transform scale, and to determine and select the largest energy among all wavelet transform scales as the main feature value of the transient current signal or transient voltage signal.

[0014] Multiple neural network modules are used. Each neural network module is used to receive the main feature values ​​of the transient current signal or transient voltage signal calculated by the feature calculation module, and to perform distributed training using the trained neural network model in the corresponding neural network module. Each neural network module obtains a corresponding fault section prediction result.

[0015] The beneficial effects of this invention are:

[0016] A method and system for predicting single-phase grounding fault sections in distribution lines based on transient signals are disclosed. This method utilizes wavelet transform to perform multi-scale analysis on transient current or voltage signals, extracting wavelet coefficients and their energy that reflect fault characteristics. This approach effectively captures the transient characteristics of the fault and, compared to traditional steady-state signal-based analysis methods, more accurately identifies the high-frequency fluctuations unique to faults. A deep learning model, such as a deep feedforward neural network, is constructed to learn and predict the feature values ​​obtained from wavelet analysis, outputting the fault section. Through the powerful nonlinear fitting capability of the deep learning model, combined with the refined features provided by wavelet analysis, the accuracy of fault location is significantly improved, reducing false alarms and missed alarms.

[0017] The system adopts a distributed processing architecture, in which multiple intelligent terminals process transient signals in parallel. The central processing unit performs weighted averaging and fusion of the prediction results from multiple terminals. The intelligent terminals independently complete data processing and prediction, which reduces the burden on the central processing unit. The central processing unit only needs to fuse the prediction results, which simplifies the system structure, significantly shortens the fault location time, improves the fault response speed of the power system, and helps to reduce power outage time and economic losses.

[0018] The central processing unit dynamically adjusts the weights of the smart terminals based on their historical prediction accuracy, thereby achieving adaptive fusion of prediction results. This constructs a fusion mechanism that can dynamically adjust the weights according to the prediction performance of the terminals, ensuring that the fusion results can fully utilize the advantages of each terminal. This enhances the system's adaptability and prediction stability to complex fault scenarios. Even when the prediction performance of some terminals declines, the overall fault location performance can be ensured by adjusting the weights. Attached Figure Description

[0019] Figure 1 This is a block diagram of the method for predicting single-phase grounding fault sections in power distribution lines based on transient signals, as presented in this invention.

[0020] Figure 2 This is a system structure diagram of the present invention. Detailed Implementation

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0022] like Figure 1 The present invention discloses a method for predicting single-phase ground fault sections in distribution lines based on transient signals, which includes the following steps:

[0023] S1 collects transient current or transient voltage signals generated in the power distribution line when a single-phase ground fault occurs;

[0024] S2, Wavelet transform is used to perform multi-scale analysis on the transient current signal or transient voltage signal collected in step S1 to obtain the wavelet coefficients of the transient current signal or transient voltage signal at each wavelet transform scale and time axis or spatial axis position;

[0025] S3, calculate the energy of the wavelet coefficients of the transient current signal or transient voltage signal obtained in step S2 at each scale and time axis or spatial axis position at each wavelet transform scale, determine and select the energy at the largest wavelet transform scale as the main characteristic value of the transient current signal or transient voltage signal;

[0026] S4. Input the main feature values ​​of the transient current signal or transient voltage signal obtained in step S3 into the trained multilayer feedforward neural network model to predict the fault section.

[0027] S5, the trained multi-layer feedforward neural network model is trained in a distributed manner to obtain multiple fault segment prediction results, and the multiple fault segment prediction results are weighted and averaged and fused. ， The fused fault section prediction results are obtained.

[0028] A method for predicting single-phase ground fault sections in power distribution lines based on transient signals, wherein step S1, acquiring transient current signals or transient voltage signals generated in the power distribution line when a single-phase ground fault occurs, is implemented as follows:

[0029] High-precision transient signal sensors, such as current transformers (CTs) or voltage transformers (PTs), are installed in power distribution lines to monitor and record transient current or voltage signals generated during a single-phase ground fault in real time. These sensors should have a fast response capability to capture the transient process of the fault. The acquired transient current or voltage signals are stored in digital form for subsequent computer processing.

[0030] The purpose of step S1 is that when a single-phase ground fault occurs in the power distribution line, the system needs to use transient signal sensors to capture the generated transient current or voltage signals in a timely manner. This design ensures a fast response and records these transient information, providing raw data for subsequent analysis.

[0031] A method for predicting single-phase grounding fault sections in distribution lines based on transient signals, wherein the wavelet transform in step S2 is implemented as follows:

[0032] (1) Select wavelet basis functions suitable for transient signal analysis, such as Morlet wavelet, Daubechies wavelet or other wavelets with time-frequency localization characteristics;

[0033] (2) Set the scale sequence of wavelet transform to ensure that it can cover the frequency domain range of transient signals;

[0034] The scaling sequence of wavelet transform is used to ensure that the entire frequency domain of the transient signal is covered, including all frequency components contained in the signal. There is no fixed numerical range for the specific scaling sequence setting, because it depends on the characteristics of the actual transient signal, the wavelet basis functions used, and the specific needs of the analysis.

[0035] Methods to ensure coverage of the frequency domain of transient signals include: ① analyzing the spectral characteristics of the transient signals of power distribution lines to understand the range of their main frequency components; ② selecting the basis functions for time-frequency localization characteristics; ③ setting the scale parameter a (scale factor) of the wavelet transform according to the spectral distribution of the signal and the selected wavelet basis functions, gradually increasing it from a small value to cover a wide range from high frequency to low frequency; ④ calculating the results after the wavelet transform to verify whether it can effectively capture all the key frequency components of the signal, and adjusting the scale sequence until all key frequency bands are included.

[0036] (3) Perform discrete wavelet transform on the transient signal acquired in step one to obtain wavelet coefficients at each scale and location.

[0037] A method for predicting single-phase grounding fault sections in distribution lines based on transient signals is proposed. In step S2, wavelet transform is used to perform multi-scale analysis on the transient current signal or transient voltage signal acquired in step one to obtain the wavelet coefficients of the transient current signal or transient voltage signal, expressed by the following formula:

[0038]

[0039] Where x(t) represents the transient current signal or transient voltage signal, ψ * (t) represents the complex conjugate of the wavelet basis function, a is the scaling parameter, and b is the translation parameter.

[0040] The specific scale selection is determined by the characteristics of the transient signal and the required frequency range for analysis. Generally, values ​​for the scale parameter 'a' are chosen to cover the entire or key frequency bands of the signal. These scale values ​​can be continuous or discrete, and are determined based on the actual application scenario and signal characteristics. For example, using a scale sequence that gradually increases from small to large allows analysis from high-frequency details to low-frequency trends. There is no fixed value for the specific scale selection; it is determined by the signal analysis requirements and the characteristics of wavelet transform.

[0041] A method for predicting single-phase ground fault sections in distribution lines based on transient signals is proposed. This method calculates the energy of wavelet coefficients at various scales and time or spatial axes for the transient current or voltage signals obtained in step S2. The formula is as follows:

[0042] Ej=l∑∣X[j,l]∣ 2 (2)

[0043] Where Ej is the energy at the j-th scale, and X[j,l] are the wavelet coefficients at the corresponding scale and position;

[0044] The purpose of step S2 is to perform wavelet transform analysis. Wavelet transform is used to perform multi-scale analysis on the acquired transient signal, decompose the signal at different scales, and extract fault features. This design can effectively separate the high-frequency fluctuation information unique to the fault, help distinguish fault modes, and improve the accuracy and pertinence of the analysis.

[0045] In this invention, real-time meteorological data (such as temperature, humidity, and wind speed), line load status, and equipment health status are introduced as auxiliary features during the fault diagnosis process. These environmental and status parameters can be acquired in real time through Internet of Things (IoT) technology and input into the neural network model along with transient signal features, enabling the prediction model to more accurately reflect the impact of environmental factors on fault performance and location.

[0046] A method for predicting single-phase grounding fault sections in distribution lines based on transient signals, wherein in step S3, the specific method for determining and selecting the energy at several maximum scales as the main characteristic values ​​of the transient current signal or transient voltage signal is as follows:

[0047] Calculate and sort the scale energies: First, calculate the energy at each scale according to formula (2). Then, sort the energies at each scale in descending order and select the largest energy values ​​Ej from the sorted energy sequence. max These are the main characteristic values ​​of transient current signals or transient voltage signals.

[0048] The purpose of step S3 is to calculate the eigenvalues. By calculating the energy of the wavelet coefficients obtained in step 2 at various scales, these eigenvalues ​​are used as the eigenvalues ​​of the transient signal. This reflects the signal's intensity distribution and frequency composition, especially the energy concentration in the frequency band where the fault characteristics are located. This design can extract the information that best characterizes the fault features, providing discriminative input for subsequent machine learning models.

[0049] A method for predicting single-phase grounding fault sections in distribution lines based on transient signals is provided. In step S4, the neural network model adopts a multi-layer feedforward neural network, including at least one hidden layer and one output layer, wherein the output layer is used to predict the fault section. The training dataset of the neural network model contains transient signal feature values ​​of various types of single-phase grounding faults and corresponding fault section labels. The training of the neural network model adopts the backpropagation algorithm, and the optimization objective is to minimize the prediction error.

[0050] A method for predicting single-phase ground fault sections in distribution lines based on transient signals, wherein in step S4, the formula is expressed as:

[0051] y = f(W2.σ(W1.Ej) max +b1)+b2), (3)

[0052] Where W1 is the weight matrix of the hidden layer, W2 is the weight matrix of the output layer, b1 and b2 are the corresponding bias terms, b1 is the bias term of the hidden layer weight matrix, b2 is the bias term of the output layer weight matrix, σ is the activation function, f is the activation function of the output layer, and y is the predicted fault segment.

[0053] Step S4 aims to predict the faulty section using a neural network. The feature values ​​calculated in step 3 are input into a trained multi-layer feedforward neural network model. The model learns the correlation between fault features and the faulty section to predict the specific location of the fault. Through distributed training and weighted fusion of prediction results, the positioning speed and accuracy are improved, false alarms and false negatives are reduced, and the system's adaptability to complex scenarios is optimized. This design combines the efficient feature extraction of wavelet analysis with the accurate prediction advantages of deep learning models, achieving fast and accurate faulty section prediction.

[0054] A method for predicting single-phase grounding fault sections in distribution lines based on transient signals, wherein in step S5, in the distributed processing module, the central processing unit dynamically adjusts the weight wi of each intelligent terminal according to the historical prediction accuracy, and the central processing unit performs weighted average fusion of multiple prediction results to obtain the fused prediction result;

[0055] When performing weighted average fusion, the specific formula can be expressed as:

[0056]

[0057] in, This represents the predicted result after fusion, where N is the number of smart terminals participating in the fusion, and w i It is the weight of the i-th smart terminal, y i It is the prediction result of the i-th smart terminal;

[0058] Prediction result y i For a single-phase grounding fault prediction system for power distribution lines, the system typically includes location information for the faulty section, such as distance measurements from the fault point, fault type identification, or more specific geographical location information. The specific information depends on the system's design and application requirements, but the core objective is to determine the precise location of the fault for rapid response and repair. Therefore, each smart terminal's prediction result may be one or a set of values, representing a quantitative estimate or classification label of the faulty section. By weighted fusion of these prediction results, the system aims to obtain a more accurate and reliable fault location output.

[0059] This invention constructs a complete fault diagnosis and decision support system, including functions such as fault alarm, fault level assessment, fault cause inference, and fault repair suggestions. For example, based on the predicted fault section, combined with historical fault data, equipment information, and operation and maintenance knowledge base, possible fault causes (such as insulation aging, external damage, bird damage, etc.) are inferred, and targeted fault handling suggestions and preventive measures are provided.

[0060] In addition to transient current or voltage signals, other types of fault-related signals, such as partial discharge signals, acoustic emission signals, and optical radiation signals, can be collected and fused. These multi-source signals can often reflect fault characteristics from different perspectives. By combining wavelet transform and other signal processing techniques, the fusion of multi-modal features can be achieved, improving the comprehensiveness and reliability of fault identification.

[0061] like Figure 2 As shown, a single-phase ground fault section prediction system for distribution lines based on transient signals includes a signal acquisition module, a wavelet analysis module, a feature calculation module, and a neural network module.

[0062] The signal acquisition module is used to acquire transient current signals or transient voltage signals generated in power distribution lines when a single-phase ground fault occurs.

[0063] The wavelet analysis module is used to perform wavelet transform multi-scale analysis on the transient current signal or transient voltage signal acquired by the signal acquisition module to obtain the wavelet coefficients of the transient current signal or transient voltage signal at each scale and time axis or spatial axis position;

[0064] The feature calculation module is used to calculate the energy of the wavelet coefficients of the transient current signal or transient voltage signal obtained by the wavelet analysis module at each scale and time axis or spatial axis position, and to determine and select the largest energy at several scales as the main feature values ​​of the transient current signal or transient voltage signal.

[0065] The neural network module includes at least one hidden layer and an output layer, which is used to receive the main feature values ​​of the transient current signal or transient voltage signal calculated by the feature calculation module and predict the fault section.

[0066] A single-phase grounding fault section prediction system for power distribution lines based on transient signals is provided. The system further includes a distributed processing module, which comprises multiple intelligent terminals and a central processing unit. The multiple intelligent terminals process transient current signals or transient voltage signals respectively and send their respective prediction results to the central processing unit. The central processing unit performs a weighted average fusion of the multiple prediction results to obtain a fused prediction result.

[0067] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A method for predicting single-phase grounding fault sections in distribution lines based on transient signals, characterized in that, Includes the following steps: Collect transient current or transient voltage signals generated in power distribution lines when a single-phase ground fault occurs; Wavelet transform is used to perform multi-scale analysis on the acquired transient current signal or transient voltage signal to obtain the wavelet coefficients of the transient current signal or transient voltage signal at each scale and at the time axis or spatial axis position; Calculate the energy of wavelet coefficients of transient current or transient voltage signals at each scale and at each position on the time or spatial axis, and determine and select the largest energy at all scales as the main characteristic value of the transient current or transient voltage signal; The main feature values ​​of transient current signals or transient voltage signals are input into multiple trained neural network models for distributed training to obtain multiple corresponding fault segment prediction results. The multiple fault segment prediction results are then weighted and averaged to obtain the fused fault segment prediction result. During the distributed training process, the weights w used for fusing prediction results are dynamically adjusted based on the historical prediction accuracy of each trained neural network model. i The prediction results of each trained neural network model are sorted according to their weights w. i A weighted average fusion is performed to obtain the fused prediction result. The specific formula for weighted average fusion is as follows: in, This represents the predicted fault segment after fusion, where N is the number of trained neural network models participating in the fusion. It is the first The weights of a trained neural network model It is the first The fault segment prediction results of a trained neural network model; Prediction results This includes location information of the faulty section; the specific location information depends on the specific design and application requirements of the system, with the core objective being to determine the precise location of the fault; the prediction result of each smart terminal is one or a set of values, representing a quantitative estimate or classification label of the faulty section. By weighted fusion of these prediction results, the system obtains accurate and reliable fault location output.

2. The method for predicting single-phase grounding fault sections in distribution lines based on transient signals according to claim 1, characterized in that: The specific method of wavelet transform is as follows: Select wavelet basis functions suitable for the analysis of transient current or transient voltage signals; Set the wavelet transform scaling sequence to ensure that it can cover the frequency domain range of the transient signal; Discrete wavelet transform is performed on the acquired transient current or voltage signals to obtain wavelet coefficients at various scales and locations.

3. The method for predicting single-phase grounding fault sections in distribution lines based on transient signals according to claim 2, characterized in that: The specific method for using wavelet transform to perform multi-scale analysis on the acquired transient current or voltage signals to obtain the wavelet coefficients X(a,b) of the transient current or voltage signals at various scales and on the time or spatial axes is as follows: X(a,b)= (1) Where x(t) represents the transient current signal or transient voltage signal, ψ ∗ (t) represents the complex conjugate of the wavelet basis function, a is the scaling parameter, and b is the translation parameter.

4. The method for predicting single-phase grounding fault sections in distribution lines based on transient signals according to claim 3, characterized in that: The specific method for calculating the energy of wavelet coefficients at various scales and positions along the time or spatial axes of transient current or voltage signals is as follows: Ej=l∑∣X[j,l]∣ 2 (2) Where Ej is the energy at the j-th scale, and X[j,l] are the wavelet coefficients at the corresponding scale and position.

5. The method for predicting single-phase grounding fault sections in distribution lines based on transient signals according to claim 4, characterized in that: The specific method for determining and selecting the largest energy across all scales as the principal characteristic value of the transient current signal or transient voltage signal is as follows: Calculate the energy at each scale according to formula (2), and select the largest energy value Ej among all scales. max , which serves as the main characteristic value of transient current or transient voltage signals.

6. The method for predicting single-phase grounding fault sections in distribution lines based on transient signals according to claim 5, characterized in that: The neural network model employs a multi-layer feedforward neural network, including at least one hidden layer and one output layer. The output layer is used to predict the fault section. The training dataset of the neural network model contains the main feature values ​​of transient current signals or transient voltage signals of various types of single-phase ground faults and the corresponding fault section labels. The training of the neural network model adopts the backpropagation algorithm, and the optimization objective is to minimize the prediction error.

7. The method for predicting single-phase grounding fault sections in distribution lines based on transient signals according to claim 6, characterized in that: The formula for predicting faulty sections in a neural network model is expressed as follows: y=f(W2.σ(W1.Ej max +b1)+b2), (3) Where W1 is the weight matrix of the hidden layer, W2 is the weight matrix of the output layer, b1 and b2 are the corresponding bias terms, b1 is the bias term of the hidden layer weight matrix, b2 is the bias term of the output layer weight matrix, σ is the activation function, f is the activation function of the output layer, and y is the predicted fault segment.

8. A system for predicting single-phase grounding fault sections in distribution lines based on transient signals, according to the method for predicting single-phase grounding fault sections in distribution lines based on transient signals as described in claim 1, is characterized in that, It includes a signal acquisition module, a wavelet analysis module, a feature calculation module, and multiple neural network modules: The signal acquisition module is used to acquire transient current signals or transient voltage signals generated in power distribution lines when a single-phase ground fault occurs. The wavelet analysis module is used to perform wavelet transform multi-scale analysis on the transient current signal or transient voltage signal acquired by the signal acquisition module, and obtain the wavelet coefficients of the transient current signal or transient voltage signal at each scale and time axis or spatial axis position. The feature calculation module is used to calculate the energy of the wavelet coefficients of the transient current signal or transient voltage signal obtained by the wavelet analysis module at each scale and time axis or spatial axis position, and to determine and select the largest energy at all scales as the main feature value of the transient current signal or transient voltage signal. Multiple neural network modules are used. Each neural network module is used to receive the main feature values ​​of the transient current signal or transient voltage signal calculated by the feature calculation module, and to perform distributed training using the trained neural network model in the corresponding neural network module. Each neural network module obtains a corresponding fault section prediction result.

9. A single-phase grounding fault section prediction system for distribution lines based on transient signals according to claim 8, characterized in that: It also includes a distributed processing module; The distributed processing module performs a weighted average fusion of the prediction results for all fault sections to obtain the fused prediction results for the fault sections.

Images