CN-121978457-A - Power distribution network fault distance measurement and transition resistance estimation method based on physical information neural network

CN121978457ACN 121978457 ACN121978457 ACN 121978457ACN-121978457-A

Abstract

The application discloses a power distribution network fault location and transition resistance estimation method based on a physical information neural network, which relates to the technical field of power distribution network fault diagnosis and artificial intelligent deep learning and comprises the following steps of collecting real-time voltage and current data after power distribution network faults through measuring devices arranged on all lines; the method comprises the steps of inputting voltage and current data into a pre-trained physical information neural network model to obtain a fault distance estimated value and a transition resistance estimated value, wherein the physical information neural network model is constructed based on a feedforward neural network model, is obtained through training a loss function of a fusion data driving loss item and a physical constraint loss item, and is used for establishing a nonlinear mapping relation between the voltage and current data, the fault distance and the transition resistance. The application has the advantages of improving the accuracy and generalization capability of fault location and reducing the dependence on a large number of fault samples.

Inventors

WANG XIAOJUN
LIU CHANGYU
LIU ZHAO
SI FANGYUAN
ZHANG DAHAI
LUO GUOMIN

Assignees

北京交通大学

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (10)

1. The power distribution network fault location and transition resistance estimation method based on the physical information neural network is characterized by comprising the following steps of: s1, acquiring real-time voltage and current data after power distribution network faults through measuring devices arranged on all lines; S2, inputting the voltage and current data into a pre-trained physical information neural network model to obtain a fault distance estimated value and a transition resistance estimated value; the physical information neural network model is constructed based on a feedforward neural network model, is obtained through training a loss function of a fusion data driving loss term and a physical constraint loss term, and is used for establishing a nonlinear mapping relation between voltage and current data, a fault distance and a transition resistance.
2. The method according to claim 1, wherein the training method of the physical information neural network model comprises: Acquiring voltage and current sample data containing various fault conditions and corresponding fault distance and transition resistance labels; Constructing a feedforward neural network model; The method comprises the steps of constructing a loss function, wherein the loss function comprises a data driving loss term and a physical constraint loss term, the data driving loss term is determined based on deviation between an estimated value output by a feedforward neural network model and a corresponding label value, and the physical constraint loss term is used for embedding a physical rule as a constraint into a training process of the feedforward neural network model; And training the feedforward neural network model by using the voltage and current sample data, and updating model parameters by optimizing the loss function to obtain a trained physical information neural network model.
3. The method of claim 2, wherein the method of acquiring voltage and current sample data comprises: building a simulation model of the power distribution network; Simulating various fault working conditions by changing fault feeder lines, fault types, fault distances, transition resistances and fault initial phase angle parameters; And collecting voltage and current waveform data under each fault working condition, and recording corresponding fault distance and transition resistance as labels.
4. The method according to claim 1, characterized in that before said S1, the method further comprises: monitoring the amplitude change of the zero sequence current of the power distribution network; and when the zero sequence current amplitude difference of the adjacent time window exceeds a set threshold value, judging that the fault occurs and executing the S1.
5. The method of claim 2, wherein the feedforward neural network model includes a plurality of hidden layers, the number of neurons of the hidden layers being configured [256,512,512,256], and Dropout and layer normalization are introduced during training to improve the numerical stability and generalization ability of the model.
6. The method of claim 2, wherein the physical constraint loss term comprises at least one of a fault differential equation constraint term, a boundary condition constraint term, and a line parameter regularization constraint term.
7. The method of claim 6, wherein the constructing of the fault differential equation constraint term comprises: Acquiring measurement data in a preset time window after a fault occurs, wherein the measurement data comprises a discrete time sequence of voltage of a fault phase, current of the fault phase, zero sequence voltage and zero sequence current; according to the measurement data and the preset sampling frequency, calculating the voltage of the fault phase, the current of the fault phase, the zero sequence voltage, the first time derivative and the second time derivative of the zero sequence current; Substituting the measurement data, the calculated first-order time derivative and second-order time derivative, and the fault distance estimated value and the transition resistance estimated value output by the feedforward neural network model into a preset fault differential equation, and calculating to obtain the estimated voltage of a fault phase; calculating a residual term based on the estimated voltage and the actually measured voltage of the fault phase, and averaging the squares of the residual term in the whole preset time window to obtain a constraint term of a fault differential equation: ; Wherein, the Constraint terms of a differential equation of a fault; the measured value of the phase A voltage at the time t; an estimated value of the phase A voltage at the time t; is the total number of sampling points within the time window.
8. The method of claim 7, wherein the boundary condition constraint term is used to impose a physical range constraint on the fault distance estimate and the transition resistance estimate output by the feedforward neural network model; The boundary condition constraint terms have the following expression: ; Wherein, the Constraint items for boundary conditions; Is a linear rectification function; is the total length of the line; A fault distance estimation value; is the estimated value of the transition resistance.
9. The method of claim 8, wherein the line parameter regularization constraint term is used to constrain each of the learnable line parameters in the feedforward neural network model to keep a deviation of each of the learnable line parameters from a corresponding reference value within a preset error radius; The expression of the line parameter regularization constraint term is as follows: Wherein, the Regularizing constraint items for the line parameters; the i-th learnable line parameter in the feedforward neural network model; the reference value of the ith line parameter, n is the total number of the learnable line parameters, and e is the error radius coefficient.
10. The method of claim 9, wherein the penalty function is expressed as a weighted sum of a data-driven penalty term and a physically constrained penalty term, expressed as: Wherein, the For the data-driven loss term, In order to physically constrain the loss term(s), Is a weight factor; Constraint terms of a differential equation of a fault; constraint items for boundary conditions; Constraint terms are regularized for the line parameters.

Description

Power distribution network fault distance measurement and transition resistance estimation method based on physical information neural network Technical Field The invention relates to the technical field of power distribution network fault diagnosis and artificial intelligence deep learning, in particular to a power distribution network fault distance measurement and transition resistance estimation method based on a physical information neural network. Background Locating line fault points quickly and accurately is a core requirement for power distribution network fault handling. In actual power distribution network operation, because line faults frequently occur and fault points are hidden, fault troubleshooting is difficult, and power supply reliability is further affected. Therefore, the rapid and accurate positioning of the fault point has important significance for recovering power supply of the system and guaranteeing the reliability of the distribution network. The existing fault locating method can be divided into a physical driving method and an artificial intelligence method. The physical driving method mainly comprises a traveling wave method, a signal injection method, an impedance method and the like. The traveling wave method utilizes the propagation time difference of the wave head to determine the fault distance by detecting the transient traveling wave signal when the fault occurs, and has the advantages of high positioning precision, high response speed and the like. However, the traveling wave method relies on high-frequency/high-speed sampling and synchronization accuracy, and the implementation cost is relatively high, so that the practical use in a conventional power distribution network still has a certain difficulty. The signal injection method identifies the fault location by injecting a specific frequency signal into the system and analyzing the reflected wave or impedance change, but the method also faces the problems of high equipment cost, signal attenuation and the like. The impedance method is based on steady-state power frequency measurement, calculates equivalent impedance of a fault point by measuring terminal voltage and current, and reversely deduces the fault distance. The method is simple to implement and high in calculation speed, and is the most widely applied fault positioning mode in the current power distribution network. However, the presence of the transition resistance may cause the voltage drop to be distributed over the ground, especially when the transition resistance is large, the fault location result tends to be systematically small. Therefore, in order to improve the ranging accuracy, some scholars introduce a joint solution mechanism of the fault distance and the transition resistance in the modeling process so as to realize accurate depiction of fault path voltage distribution. In addition, the impedance method is prone to multiple position estimation in a power distribution system having multiple branch lines, resulting in inaccurate positioning. With the rapid popularization of machine learning and artificial intelligence, a power distribution network fault positioning method based on data driving is rapidly developed. The main idea is to model fault location as regression problem, and realize end-to-end fault distance estimation by establishing nonlinear mapping relation between measurement information and fault distance. Preliminary applications of the artificial neural network ANN, the support vector machine SVM and other artificial intelligent models show feasibility and applicability of the data driving scheme in the field of fault location of the power distribution network. The method can automatically extract the characteristics and avoid explicit parameter modeling, but has obvious limitations that on one hand, the sample label of the fault distance cannot be completely covered, the dependence on the number of samples is too high, and samples of different fault types which can be obtained in an actual power distribution network are very few, on the other hand, the pure data driving method lacks electrical mechanism constraint, the result is difficult to ensure to conform to the physical rule, and the interpretation of the predicted result and the reliability of the actual application are insufficient. In an actual power distribution network environment, the fault types are various, including single-phase earth fault, two-phase short circuit fault, two-phase earth fault, three-phase short circuit fault and the like, and the change of the transition resistance and the fault distance under each fault type can obviously influence the ranging result. In particular in the case of high-resistance ground faults, the transition resistance can reach tens of ohms or even higher, resulting in a significant increase in the range error of conventional impedance methods. Meanwhile, the power distribution network is complex in structure and has multiple bran