Power grid fault comprehensive analysis method based on transient data deep mining

Abstract

The invention relates to the technical field of power systems, in particular to a transient data deep mining-based power grid fault comprehensive analysis method, which comprises the following steps of: obtaining power grid fault data; preprocessing the power grid fault data; obtaining a plurality of fault characteristic quantity data according to the preprocessed data; and performing fault type identification according to the fault characteristic quantity data to obtain a power grid fault comprehensive analysis result. According to the method, fault mode identification is carried out through the multiple fault characteristic quantities, the single characteristic quantity is concerned, and the multiple characteristic quantities are comprehensively analyzed and mined to determine the cause of the fault and the specific position of the fault, so that the fault analysis is more comprehensive and deep to assist in power grid fault processing and prevention.

Description

Technical Field

[0001] The present application relates to the technical field of power systems, and in particular to a comprehensive analysis method for power grid faults based on deep mining of transient data. Background Art

[0002] With the development of my country's economy, the demand and dependence on electric energy is constantly increasing. Transmission lines are the main lifeline of the power grid, which has an important impact on the security of the power grid. The requirements for safe operation of transmission lines are also constantly increasing. Transmission line fault location is an important equipment to ensure the safe operation of the line. In many years of production practice, power grid fault location has provided great help.

[0003] At present, the transmission line fault location based on transient traveling wave principle has become a standard configuration in domestic medium and high voltage transmission lines due to its high accuracy and reliable principle. However, the application of transient data in the existing traveling wave fault location system is mainly limited to fault location, and its application scope is relatively narrow, and it cannot provide auxiliary functions for power grid fault handling.

[0004] In response to the above problems, there is an urgent need for a method that can comprehensively analyze faults, which can not only determine the location of the fault point, but also provide information to identify the cause of the fault, etc., to assist in power grid fault handling and improve the level of power grid operation and management. Summary of the invention

[0005] In view of this, it is necessary to provide a comprehensive analysis method for power grid faults based on deep mining of transient data to solve the shortcomings of the existing technology that can only determine the location of the fault point but cannot provide a variety of information to assist in power grid fault handling and prevention.

[0006] A comprehensive analysis method for power grid faults based on deep mining of transient data, comprising:

[0007] Obtain power grid fault data;

[0008] Preprocessing the power grid fault data;

[0009] Obtaining a plurality of fault characteristic quantity data according to the preprocessed data;

[0010] The fault type is identified based on the fault characteristic data to obtain a comprehensive analysis result of the power grid fault.

[0011] Preferably, the fault type identification is performed according to the fault characteristic data, including:

[0012] Determine sample fault data, and determine the weight of each feature quantity in the sample fault data;

[0013] Training a classifier according to the weights of each feature quantity in the sample fault data;

[0014] The fault characteristic data is classified by a trained classifier to identify the fault type.

[0015] Preferably, training a classifier according to the weights of the feature quantities includes:

[0016] Train multiple weak classifiers;

[0017] After each training, the weight of the corresponding weak classifier is adjusted according to the error rate of the weak classifier; wherein, the weak classifier with a higher error rate is assigned a smaller weight;

[0018] Each time training is performed, the weight of the feature quantity that was misclassified in the previous round is increased for training; among them, the weight of the feature quantity that was correctly classified is reduced;

[0019] When the training results of the multiple weak classifiers meet a preset threshold, the training of the weak classifiers is completed.

[0020] Preferably, the trained weak classifiers are weighted and fused according to the weights to obtain a strong classifier;

[0021] The fault feature data is classified by the strong classifier to identify the fault type.

[0022] Preferably, the trained weak classifiers are weighted and fused according to the weights based on the AdaBoost algorithm to obtain a strong classifier, including:

[0023] The obtained strong classifier satisfies:

[0024]

[0025] Among them, L j represents a weak classifier, L j =(L1, L2, L3...L N );ω j represents the weight of the weak classifier, ω j =ω1,ω2......ω N ; T represents the number of iterations.

[0026] Preferably, the fault type is identified according to the fault characteristic data to obtain a comprehensive analysis result of the power grid fault, including:

[0027] According to the result of the strong classifier, a comprehensive analysis result of the power grid fault is obtained; the comprehensive analysis result of the power grid fault includes the cause of the fault.

[0028] Preferably, the method further includes determining the specific location where the fault occurs based on the result of the strong classifier and in combination with the topological structure of the power system.

[0029] Preferably, the fault type is identified according to the fault characteristic data to obtain a comprehensive analysis result of the power grid fault, including:

[0030] Standardizing each characteristic quantity in the fault characteristic quantity data;

[0031] Determining the weight of each of the feature quantities;

[0032] Calculating a fault score of a power grid fault according to the characteristic quantities and their weights;

[0033] According to the result of the fault scoring, determining the fault type corresponding to the currently observed feature quantity combination;

[0034] The fault score satisfies: L = ω1·I peak +ω2·T fault +ω3·R rate +ω4·R drop +ω n ·N m

[0035] Wherein, L represents the fault score, ω1, ω2, ω3, ω4, ω n Represents the weight of each feature; I peak is the peak value of the fault current; T fault is the fault duration; R rate is the rate of change of resistance; R drop is the voltage drop; N m Indicates other feature quantities.

[0036] Preferably, a plurality of fault characteristic quantity data are obtained according to the preprocessed data, including:

[0037] Obtaining the fault characteristic quantity data by performing frequency domain analysis on the preprocessed data;

[0038] The fault characteristic quantity data include: the amplitude of the steady-state quantity, the amplitude of the transient quantity, the amplitude and change rate of the transition resistance value, the phase-to-phase coupling characteristic quantity, the energy spectrum distribution and the transient waveform.

[0039] Compared with the prior art, the beneficial effect of the present application is that fault type identification is performed based on fault characteristic data, not only focusing on a single characteristic, but also determining the cause of the fault and the specific location of the fault through comprehensive analysis and mining of multiple characteristic quantities, making the fault analysis more comprehensive and in-depth, so as to assist in power grid fault handling and prevention and improve the level of power grid operation and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1A flow chart of a comprehensive power grid fault analysis method based on deep mining of transient data provided in an embodiment of the present application.

[0041] Figure 2 A flowchart of a comprehensive power grid fault analysis method based on deep mining of transient data is provided in another embodiment of the present application.

[0042] Figure 3 Schematic diagram of the basic principles of feature extraction and classification.

[0043] Figure 4 Schematic diagram of the process of obtaining a strong classifier.

[0044] Figure 5 It is a structural diagram of the transient feature quantity classification model based on Adaboost classifier. DETAILED DESCRIPTION

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] like Figure 1 As shown, Figure 1 1 is a flow chart of a method for comprehensive analysis of power grid faults based on deep mining of transient data provided by an embodiment of the present application. The method for comprehensive analysis of power grid faults based on deep mining of transient data comprises the following steps:

[0047] S101: Obtain power grid fault data.

[0048] The power grid fault data is a waveform characteristic signal when a line fails, which contains broadband data of steady-state and transient quantities, as well as sampling noise, surrounding environmental interference, etc. Therefore, in order to provide guarantee for subsequent processing, after obtaining the power grid fault data, the original signal can be pre-processed by bandpass filtering method to eliminate the burrs in the original signal curve, remove high-frequency noise components, and eliminate the influence of environmental noise and electromagnetic interference of surrounding equipment.

[0049] S102: Preprocessing the power grid fault data.

[0050] Preprocessing the power grid fault data may include extracting steady-state quantities (signals), transient quantities, fault phases, etc. from the fault data.

[0051] The steady-state quantity signal is composed of power frequency signals and is a stable signal, so the feature extraction method for fault steady state is relatively simple. The commonly used filtering algorithm in power system adopts the high-precision and real-time moving data window Fourier filtering improved algorithm to extract the steady-state quantity of traveling wave.

[0052] Transient quantities usually refer to the rapid changes in system response within a short period of time after a fault occurs, such as transient voltage, transient current, frequency change, etc. The method to extract the transient component of the signal is "frequency domain filtering", which is to use Fourier transform and then perform filtering, then inverse transform the filtered signal back to the time domain, and subtract this inverse transformed signal from the original signal.

[0053] The fault phase can be used to determine which phase or phases the fault occurred in. The fault phase coefficient is usually used to describe the relative changes between phases when a fault occurs in the power system. In the symmetrical component method, the fault phase coefficient (or fault coefficient) can represent the multiple relationship between the voltage or current of a phase relative to another phase after the fault. The voltage and current data at the time of the fault are selected to facilitate the analysis of the fault phase.

[0054] S103: Obtain a plurality of fault characteristic quantity data according to the preprocessed data.

[0055] In the present application, fault characteristic data is extracted by performing frequency domain analysis on the fault data. The fault characteristic data may include: the amplitude of the steady-state quantity, the amplitude of the transient quantity, the amplitude and change rate of the transition resistance value, the phase-to-phase coupling characteristic quantity and energy spectrum distribution and transient waveform, etc.

[0056] Assuming that the fault types include lightning faults, wildfire faults, external connection faults and icing faults, for each fault type, indicators that can characterize its characteristics can be selected, such as the amplitude of the fault current and voltage, the fault duration, the resistance change rate, etc.

[0057] Furthermore, fault type identification based on fault feature data can be achieved through a classifier, for example:

[0058] First, the sample fault data is determined, and the weight of each feature in the sample fault data is determined; the classifier is trained according to the weight of each feature in the sample fault data. The weight of each feature in the sample fault data can be determined by the analytic hierarchy process (AHP) to reflect its importance in fault identification.

[0059] Then, the fault feature data is classified through the trained classifier to identify the fault type.

[0060] Furthermore, multiple weak classifiers may be trained; the trained weak classifiers may be weightedly fused according to weights to obtain a strong classifier; and the fault feature data may be classified by the strong classifier to identify the fault type.

[0061] When training weak classifiers, the weight of the corresponding weak classifier is adjusted according to the error rate of the weak classifier after each training; the weak classifier with a higher error rate is given a smaller weight. At the same time, each training can increase the weight of the feature quantity that was misclassified in the previous round for training; the weight of the feature quantity that was correctly classified is reduced; those feature quantities (samples) that were misclassified in the previous round of classification will be given a larger weight, so that the subsequent weak classifiers will pay more attention to these samples that are difficult to classify correctly.

[0062] When the training results of the multiple weak classifiers meet a preset threshold, the training of the weak classifiers is completed. The preset threshold may be a preset number of training rounds, and the training is completed when the number of training rounds of the weak classifiers meets the preset threshold.

[0063] Finally, during fusion, the trained weak classifiers can be weighted and fused according to the weights based on the AdaBoost algorithm to obtain a strong classifier.

[0064] Furthermore, the obtained strong classifier satisfies:

[0065]

[0066] Among them, L j represents a weak classifier, L j =(L1, L2, L3...L N );ω j represents the weight of the weak classifier, ω j =ω1,ω2......ω N ; T represents the number of iterations.

[0067] Furthermore, the fault feature data is classified by a strong classifier to identify the fault type, including:

[0068] Different power grid faults are classified; fault characteristic data corresponding to each fault type are extracted; the changing trend of fault characteristic data under different fault conditions is determined to complete the characteristic description of the fault; based on the characteristic description of the fault, the fault type is identified through a strong classifier.

[0069] Specifically, first of all, according to the different manifestations and causes of power grid faults, the faults are divided into different types, including but not limited to: single-phase grounding fault, phase-to-phase short circuit fault, three-phase short circuit fault, line break fault, lightning fault, forest fire fault, external hanging fault, icing fault, etc.

[0070] For each fault type, extract its unique fault characteristic quantity data, including but not limited to: Steady-state quantity: such as the fundamental component of voltage and current. Transient quantity: such as the voltage and current changes at the moment of fault occurrence. Transition resistance value: the transition resistance at the fault point. Phase-to-phase coupling characteristic quantity: such as the amplitude ratio, phase difference or energy ratio of the two-phase current or voltage. Energy spectrum distribution: the energy distribution of the signal, reflecting the energy size of different frequency components. Transient waveform characteristics: the mutation point in the transient waveform, such as the peak value of voltage or current, the width of the waveform, etc.

[0071] Determine the changing trend of fault characteristic quantity data under different fault conditions, and complete the characteristic quantity description of the fault: For each fault type, analyze the changing trend of its fault characteristic quantity data, for example: Single-phase grounding fault: The transition resistance value is small, the transient current waveform characteristics are obvious, and the high-frequency components in the energy spectrum increase. Phase-to-phase short-circuit fault: The transient current waveform has a peak, and the energy spectrum distribution shows high-frequency components. Three-phase short-circuit fault: The transient current is extremely large and the energy spectrum distribution is uniform. Line break fault: The voltage drops and the energy spectrum distribution is abnormal. Lightning fault: The transient current waveform rises steeply, and the high-frequency components of the energy spectrum distribution are significant. Wildfire fault: The transition resistance value changes, and the energy spectrum distribution changes. External connection fault: The transient current waveform changes, and the energy spectrum distribution changes. Ice fault: The transition resistance value changes, and the energy spectrum distribution changes. Complete the characteristic quantity description of the fault Based on the above change trend, complete the characteristic quantity description for each fault type to form a set of characteristic quantities that describe the fault type.

[0072] Based on the characteristic description of the fault, the fault type is identified through a strong classifier. The extracted fault characteristic data is input into the trained strong classifier to identify the fault type. The specific steps are as follows:

[0073] Feature standardization: Standardize the extracted fault feature data to ensure that different feature quantities are compared at the same level.

[0074] Application of feature quantity weights: Perform weighted processing on the standardized feature quantity data according to the previously determined feature quantity weights.

[0075] Classification prediction: Input the weighted feature data into the strong classifier to obtain the classification result, that is, the prediction of the fault type.

[0076] S104: Identify the fault type according to the fault characteristic data to obtain a comprehensive analysis result of the power grid fault.

[0077] Analysis of classification results: Based on the output of the strong classifier, the classification results are interpreted to determine the specific fault type. In addition, based on the output of the strong classifier, the specific location of the fault can be determined in combination with the topological structure of the power system.

[0078] According to the strong classifier, based on the fault feature data and combined with the fault feature description, the fault type can be determined, and then the cause and development process can be analyzed according to the changes of each feature under this type.

[0079] In addition, in some implementations, after obtaining a plurality of fault characteristic quantity data according to the preprocessed data, this implementation further provides a method for determining the fault type, including:

[0080] Determine the weight of each feature quantity;

[0081] According to the characteristic quantities and their weights, the fault score of each power grid fault is calculated; the fault score satisfies:

[0082] L=ω1·I peak +ω2·T fault +ω3·R rate +ω4·R drop +ω n ·N m

[0083] Where L represents the fault score, ω1, ω2, ω3, ω4, ω n Represents the weight of each feature; I peak is the peak value of the fault current; T fault is the fault duration; R rate is the rate of change of resistance; R drop is the voltage drop; N m Indicates other feature quantities.

[0084] A standard fault score is preset for each fault type. If the fault score calculated by the currently observed feature quantity combination is the same as the preset one, the fault type can be preliminarily determined. This embodiment is suitable for rapid fault screening and provides preliminary guidance for the screening process. If more complex models and algorithms are required in practical applications, strong classifiers can be used for judgment to ensure the accuracy and reliability of fault identification.

[0085] In addition, the determination result of the fault type of this embodiment may be compared with the result based on the strong classifier, etc., so that the results of the two solutions can be mutually verified to further improve the accuracy of identification.

[0086] The following is a further introduction to the comprehensive analysis method of power grid faults based on deep mining of transient data of the present application in conjunction with a specific embodiment. Figure 2 As shown, the comprehensive power grid fault analysis method based on transient data deep mining provided in this embodiment includes the following steps:

[0087] S201: Preprocessing power grid fault data.

[0088] The collected power grid fault data includes fault steady-state signal data, fault transient signal data, harmonic signal data, etc., and contains interference from sampling noise, surrounding environment, etc. In order to eliminate the influence of environmental noise and electromagnetic interference from surrounding equipment, the original signal is first preprocessed by bandpass filtering to eliminate the burrs in the original signal curve and remove high-frequency noise components, providing guarantee for subsequent feature extraction.

[0089] After removing the noise, the steady-state quantity, transient quantity, fault quantity, etc. are extracted. For example:

[0090] Extract steady-state quantities:

[0091] The fault steady-state quantity signal is composed of power frequency signal, which is a stable signal. The filtering algorithm can be used to extract the characteristics of the fault steady state. The steady-state quantity extraction of traveling waves can adopt the high-precision and real-time moving data window Fourier filtering improved algorithm.

[0092] First, perform Fourier transform on the original signal x(t) to obtain its frequency domain representation X(f)

[0093] X(f)=F{x(t)}

[0094] Frequency domain filtering: Apply a bandpass filter function H(f) in the frequency domain. The bandpass filter can remove both low-frequency and high-frequency components in the signal and retain only the components within a specific frequency range.

[0095] Y(f)=X(f)*H(f)

[0096] Where * represents the convolution operation, X(f) is the frequency domain representation of the signal, H(f) is the frequency domain representation of the filter, and Y(f) is the frequency domain signal after filtering.

[0097] Inverse Fourier transform: Perform an inverse Fourier transform on the filtered frequency domain signal Y(f) to obtain the time domain signal y(t).

[0098] y(t)=F -1 {Y(f)}

[0099] Extract transient quantities:

[0100] Transient quantities usually refer to the rapid changes in system response within a short period of time after a fault occurs, such as transient voltage, transient current, frequency change, etc. The method to extract the transient component of the signal is "frequency domain filtering", which is to use Fourier transform and then perform filtering, then inverse transform the filtered signal back to the time domain, and subtract this inverse transformed signal from the original signal.

[0101] The signal y(t) obtained after steady-state filtering is subtracted from the original signal x(t) by differential operation to extract the transient signal z(t), that is:

[0102] z(t)=x(t)-y(t)

[0103] Extract the fault phase:

[0104] By extracting the fault phase in the traveling wave, it is possible to determine which phase or phases the fault occurred in. The fault phase coefficient is usually used to describe the relative changes between phases when a fault occurs in the power system. In the symmetrical component method, the fault phase coefficient (or fault coefficient) can represent the multiple relationship of the voltage or current of a phase relative to another phase after the fault. The following are the general steps for calculating the fault phase coefficient:

[0105] 1) Data acquisition: Collect the transient data of three-phase voltage and current at the time of fault occurrence.

[0106] 2) Waveform synchronization: Ensure that the waveform data of all phases are collected synchronously, that is, the samples at the same moment correspond to the same time point.

[0107] 3) Fault location: Determining the exact moment when a fault occurs can usually be achieved by looking for sudden changes in the current or voltage waveform.

[0108] 4) Signal filtering: Filter the collected signal to remove high-frequency noise and unnecessary frequency components.

[0109] 5) Data selection before and after the fault: Select the voltage and current data before and after the fault occurs to facilitate the analysis of the changes caused by the fault.

[0110] Calculate the amplitude before and after the fault: Calculate the amplitude of each phase voltage and current before and after the fault.

[0111] S202: Extracting power grid fault characteristic data.

[0112] Feature extraction and selection: Since the amount of data obtained from images and waveforms is huge, in order to effectively realize classification and identification, the original data must be transformed. For each type of fault, select indicators that can characterize its characteristics to obtain the features that best reflect the essence of classification. The feature quantity data extracted from the disturbance data in this embodiment include: steady-state quantity amplitude, transient quantity amplitude, transition resistance amplitude and change rate, phase-to-phase coupling characteristic quantity, energy spectrum distribution, transient waveform and other aspects, to obtain feature quantity descriptions of different power grid faults, so as to study the occurrence and evolution of major faults, analyze the change trend of each feature quantity during the fault process, and complete the feature quantity description of typical faults. This helps staff to identify and respond to different types of power grid faults more quickly. This embodiment shows the following four types of fault classification: 1) lightning strike fault; 2) mountain fire fault (initial high resistance fault, gradually decreasing); 3) external hook-up; 4) ice-covered fault. The relationship between the classification of power grid faults and the five types of power grid fault feature quantities is shown in Table 1 below:

[0113] Table 1: Relationship between the classification of power grid faults and the five power grid fault characteristic quantities

[0114]

[0115]

[0116]

[0117] Extracting feature data may include:

[0118] 1) The amplitude of the steady-state quantity and the amplitude of the transient quantity during the fault.

[0119] The time domain signal is initially decomposed into sinusoidal components of different frequencies using Fourier transform, and then the time domain extreme points of each component are extracted as characteristic parameters. The amplitudes of steady-state and transient quantities during faults can effectively reflect various types of faults in the power grid system. The steady-state component is the long-term behavior of the system response, while the transient component is the short-term behavior of the system response, which will decay to zero over time.

[0120] For system responses with multiple frequency components, the total amplitude is usually the vector sum of the amplitudes of the individual frequency components. Mathematically, this can be achieved by Fourier transforming the time domain signal into the frequency domain and then summing the amplitudes of the individual frequency components.

[0121] Using the extracted steady-state quantity y(t) and transient quantity z(t), the following are the steps to calculate the transient quantity amplitude during a fault using Fourier transform:

[0122] Given a transient frequency domain signal Z(f), its Fourier transform is expressed as, where f is the frequency. Each point Z(f) in the spectrum k ) can be expressed as:

[0123]

[0124] Among them, A k is the frequency component f k The amplitude, φ k is the frequency component f k The phase of j is an imaginary unit, satisfying j 2 =-1.

[0125] Use the Inverse Fourier Transform (IFT) to convert the frequency domain components back to the time domain signal. For discrete signals, use the Inverse Fast Fourier Transform (IFFT):

[0126]

[0127] Among them, f k is the frequency of the kth component.

[0128] Calculate the time domain signal: Convert to real and imaginary form:

[0129] z(t)=∑ k A k cos(2πf k t+φ k )+j∑ k A k sin(2πf k t+φ k )

[0130] Usually only the real part is considered because it represents the actual physical signal.

[0131] Compute the square of a time-domain signal:

[0132] Calculate the time interval [t1, t2] of the transient magnitude. Calculate the square of z(t) in the time interval [t1, t2]:

[0133] z 2 (t)=(∑ k A k cos(2πf k t+φ k )) 2

[0134] Integrate to find the average:

[0135] In the time interval [t1, t2], z 2 (t) is integrated and the average value is calculated:

[0136]

[0137] Calculate the effective value:

[0138] Calculate the effective value (RMS) of the amplitude of z(t) in the time interval [t1, t2]:

[0139]

[0140] To calculate the amplitude of the steady-state quantity, use the same steps as above.

[0141] It is also possible to select a specific calculation method for the amplitude of the steady-state quantity and the transient quantity according to actual conditions such as the fault type and specific parameters of the system at the time of the fault.

[0142] 2) Transition resistance value R(t) and rate of change.

[0143] The transition resistance value is the additional resistance caused by factors such as arcing, equipment characteristics or grounding conditions at the fault point. The transition resistance change rate usually refers to the change in resistance value over time due to the thermal effect of equipment or materials near the fault point in the power system when a fault occurs. This resistance value and change rate can be indirectly estimated by measuring the fault current and voltage. Accurate estimation of transition resistance helps power grid dispatchers to deeply understand the nature and potential causes of the fault, and provides important reference for subsequent fault diagnosis and repair work by construction personnel. The following are some basic steps to extract transition resistance:

[0144] Measuring fault current and voltage: First, it is necessary to measure the current and voltage at the fault point when a fault occurs. These measurements can be completed by the protection devices and measuring equipment installed on the power system.

[0145] Calculate instantaneous power: The instantaneous power P(t) can be obtained through the instantaneous voltage V(t) and instantaneous current I(t):

[0146] P(t)=V(t)×I(t)

[0147] Estimation of transition resistance: The transition resistance R(t) can be estimated from the instantaneous power and instantaneous voltage using Ohm's law:

[0148]

[0149] Calculate the rate of change of resistance: The rate of change of the transition resistance can be obtained by differentiating R(t). This can be achieved by numerical differentiation methods, such as using the two-point difference formula:

[0150]

[0151] Where Δt is the time interval.

[0152] Analyze the changing trend: By observing the changing trend of the transition resistance, we can understand the increase or decrease of the transition resistance over time.

[0153] 3) Phase coupling characteristic

[0154] Phase-to-phase coupling reveals the interaction and influence between different phases in the power grid. Under different power grid fault conditions, the strength and characteristics of phase-to-phase coupling will change significantly. Phase-to-phase coupling characteristics are usually used to describe the degree of coupling between phases in multiphase power systems, especially in electromagnetic coupling analysis of transmission lines. The following are the general steps for calculating phase-to-phase coupling characteristics:

[0155] Define coupling coefficient: Define a phase coupling coefficient C ij , where i and j represent two different phases (e.g., phase A and phase B). The coefficients can be based on one of the following relationships:

[0156] Amplitude coupling: Based on the ratio or difference of the two-phase current or voltage amplitudes.

[0157] Phase coupling: Based on the absolute value or correlation of the phase difference between the two-phase current or voltage.

[0158] Energy coupling: based on the energy ratio of two-phase current or voltage signals within a certain period of time.

[0159] For example, you can use one of the following formulas:

[0160]

[0161] Among them, V i and V j are the voltage or current signals of the i-th phase and the j-th phase respectively, and the operator <·,·> represents the inner product of the two signals.

[0162] Analyze the coupling coefficient: The calculated coupling coefficient can be used to determine the phase characteristics:

[0163] High coupling coefficient: indicates that there is a strong coupling relationship between the two phases, which may be a normal working state or a specific type of fault (such as a two-phase short circuit).

[0164] Low coupling coefficient: indicates that the coupling between the two phases is weak, which may be due to a fault in another phase (such as a single-phase grounding fault).

[0165] Threshold setting: Set the threshold of the coupling coefficient to distinguish different working states or fault types.

[0166] 4) Energy spectrum distribution:

[0167] By performing Fourier transform on a signal, we can obtain the spectrum of the signal and then calculate the energy of each frequency component. The energy spectrum is usually expressed as a function of frequency. Through a detailed analysis of the energy spectrum, we can gain insight into the energy flow and conversion of the power grid during a fault, thereby more accurately determining the type and location of the fault.

[0168] Since the active component of the fault current accounts for a small proportion and the integral function is prone to accumulate some fixed errors, the improvement of its detection sensitivity is limited. The following are the general steps for calculating the energy spectrum:

[0169] Determine the energy spectral density function: Assume a function that describes the energy distribution, usually denoted as E total , where f represents the frequency.

[0170] Calculate the total energy: Calculate the integral of the energy spectral density function within the range of interest to obtain the total energy E total :

[0171]

[0172] Where f1 and f2 are the lower and upper limits of the energy spectrum.

[0173] Normalization factor: Divide the total energy by the length of the integration interval (frequency range) to obtain the normalization factor N:

[0174]

[0175] Where f1 and f2 are the lower and upper limits of the frequency.

[0176] Normalized energy spectral density: Divide the original energy spectral density function by the normalization factor N to obtain the normalized energy spectral density E'(f):

[0177]

[0178] Verify normalization: Verify whether the integral of the normalized energy spectral density is equal to 1.

[0179] 5) Transient waveform (rising edge and pulse width):

[0180] The transient waveform directly reflects the instantaneous changes in the voltage and current waveforms when the power grid fails. The multi-scale analysis capability of wavelet transform is used to identify the mutation points of the signal, which may correspond to the rising edge of the waveform. In-depth study of transient waveforms can reveal the instantaneous dynamic characteristics of faults and provide direct and powerful support for fault diagnosis and repair.

[0181] The peak of a waveform usually refers to the highest point in the waveform graph, which represents the maximum amplitude of the signal. The peak characteristics of each waveform can be analyzed through its mathematical expression, and the position and height of the peak can be determined by finding the extreme point by taking the derivative, especially when looking for the maximum value (peak) or minimum value (valley) of the signal. The following are the detailed steps and related formulas:

[0182] Signal definition:

[0183] Assume there is a continuous time signal x(t), where t is the time variable.

[0184] Find the first-order derivative:

[0185] Derivative the signal x(t) to find the derivative The first derivative represents the rate of change of a signal.

[0186] Find extreme points:

[0187] The extreme point is the point where the derivative is zero, that is, x'(t) = 0. Solve this equation to find the value of t.

[0188] Determine the extreme value type:

[0189] For each point where the derivative is zero, we need to determine whether it is a maximum or a minimum. This can be done by calculating the second-order derivative x”(t).

[0190] If x”(t)>0, then the point is a minimum point.

[0191] If x”(t)<0, then the point is a maximum point.

[0192] Calculate the peak value:

[0193] From all the maximum points, select the point with the largest x(t) value as the location of the peak. Once the location of the peak t is determined peak , we can use t peak Substitute the original signal x(t) to calculate the peak value:

[0194] x peak =x(t peak )

[0195] 6) Equivalent pulse width of transient waveform.

[0196] The extraction of equivalent pulse width of transient waveform is a key point in studying the occurrence or recovery of faults. The equivalent pulse width usually refers to the duration of the main part of the transient waveform containing energy. The following are some basic steps and methods to extract the equivalent pulse width of transient waveform:

[0197] Determine the transient waveform:

[0198] First, the transient waveform part of the system response needs to be determined. This is usually done by observing the time history plot of the system response. The transient waveform is the part of the system response that decays to zero over time.

[0199] Energy calculation:

[0200] The energy of a transient waveform is calculated using the following formula:

[0201]

[0202] Where E is energy, V(t) is the voltage waveform, t0 is the time when the transient waveform starts, and t f is the time when the transient waveform ends.

[0203] Determine the cycle:

[0204] For periodic signals, energy can be calculated by integrating over one cycle. However, for transient waveforms, they are usually not periodic. If a transient waveform exhibits periodicity over a certain period of time, the period can be considered as a cycle to calculate the energy. If the transient waveform is not periodic, then the concept of period does not apply and the energy calculation should be performed over the entire duration of the transient waveform.

[0205] Consider attenuation:

[0206] When calculating energy, it is necessary to consider the attenuation characteristics of the transient waveform. If the attenuation of the transient waveform is exponential, the exponential decay model can be used to estimate t f .

[0207] Equivalent pulse width calculation:

[0208] Define an equivalent pulse width T eq , so that most (e.g. 90% or 95%) of the energy is contained within this pulse width. The equivalent pulse width can be calculated as follows:

[0209]

[0210] Where E eq The equivalent pulse width T eq By adjusting T eq The value of E eq Close to some percentage of E (e.g. 90% or 95%).

[0211] S203: Train weak classifiers.

[0212] like Figure 3 A schematic diagram of the basic principle of feature extraction and classification is shown in Figure 1. A multi-index decision model can be constructed, which uses various fault features and corresponding weights to identify the fault type. For each fault type, an indicator that can characterize its characteristics is selected, such as the amplitude of the fault current and voltage, the fault duration, the resistance change rate, etc., and multiple weak classifiers are trained.

[0213] S204: Obtain a strong classifier.

[0214] like Figure 4 The flowchart of obtaining a strong classifier is shown in FIG. 1 . The analytic hierarchy process (AHP) method can be used to determine the weight of each feature quantity to reflect its importance in fault identification. In addition, since different feature quantities may have different dimensions and value ranges, they need to be standardized for comparison and synthesis.

[0215] Constructing decision-making models: Figure 5The structural diagram of the transient feature quantity classification model based on the Adaboost classifier shown in the figure can be used to construct a multi-index decision-making based on the Adaboost algorithm using the selected indicators and weights. The AdaBoost algorithm is an enhanced algorithm that improves the performance of weak classifiers through iteration and fusion weighting. First, all samples are supplemented with a sampling weight (usually the initial weight is set to a uniform distribution), and multiple weak classifiers are trained on this sample. After the initial classification of the sample, the error rate of each classifier is obtained. In the next classification, the classifier with a larger error rate is assigned a smaller weight; for the sample that was misclassified last time, its sampling weight is increased, and the weight is re-assigned according to its error rate. In this way, it is iterated in sequence, and finally a strong classifier of weighted fusion of multiple weak classifiers is obtained.

[0216] AdaBoost is a basic weak classifier that is automatically weighted and fused to form a strong classifier. Given n labeled samples {(x1,y1),(x2,y2),...,(x n ,y n )}, where x i is the sample feature vector, y i (y i ∈{+1,-1}) is the corresponding label, ω(i) is the weight of n classifiers, and the initial classifier weight Initial sample weight The number of iterations is i. Finally, based on the established model and the weights assigned to different features, the model is trained.

[0217] Input the initial feature parameters into the weighted sum model to obtain multiple initial weak classifiers L j :

[0218] L j =L(x,y,o)

[0219] Calculate each weak classifier L j The classification error E err :

[0220]

[0221] Adjust the weight coefficient ω of the classifier according to its classification error j :

[0222] ω j =(1 / 2)ln[(1-E err ) / E err ]

[0223] According to the weight coefficient, redistribute the weight of each sample and adjust the sample distribution:

[0224]

[0225] c j is the generalization coefficient. If the sample distribution is the same as the previous one, the loop will be exited, otherwise the loop will continue.

[0226] These weak classifiers are then combined into a strong classifier:

[0227]

[0228] Where the weak classifier L j =(L1,L2,...,L N ), weight ω j =(ω1,ω2,...,ω N ).

[0229] S205: Fault identification is performed using the trained strong classifier.

[0230] According to the output of the strong classifier, the specific fault type is determined. After the specific fault type is determined, the classification result is explained by the strong classifier. For example, as shown in Table 1, assuming it is a lightning fault, the cause of the fault may be a sudden voltage rise caused by lightning, resulting in overvoltage; its development process may be a ground fault, with a sharp increase in transient current and a wave tail time shorter than 20 microseconds.

[0231] In addition, the specific fault type can be determined by interpreting the classification results through a strong classifier, and the specific location of the fault can be determined by combining the topological structure of the power system.

Claims

1. A comprehensive analysis method for power grid faults based on deep mining of transient data, characterized in that: include: Obtain power grid fault data; Preprocessing the power grid fault data; Obtaining a plurality of fault characteristic quantity data according to the preprocessed data; The fault type is identified based on the fault characteristic data to obtain a comprehensive analysis result of the power grid fault.

2. The method according to claim 1, characterized in that Fault type identification is performed according to the fault characteristic data, including: Determine sample fault data, and determine the weight of each feature quantity in the sample fault data; Training a classifier according to the weights of each feature quantity in the sample fault data; The fault characteristic data is classified by a trained classifier to identify the fault type.

3. The method according to claim 2, characterized in that Training a classifier according to the weights of the feature quantities includes: Train multiple weak classifiers; After each training, the weight of the corresponding weak classifier is adjusted according to the error rate of the weak classifier; wherein, the weak classifier with a higher error rate is assigned a smaller weight; Each time training is performed, the weight of the feature quantity that was misclassified in the previous round is increased for training; among them, the weight of the feature quantity that was correctly classified is reduced; When the training results of the multiple weak classifiers meet a preset threshold, the training of the weak classifiers is completed.

4. The method according to claim 3, characterized in that include: Perform weighted fusion on the trained weak classifiers according to the weights to obtain a strong classifier; The fault feature data is classified by the strong classifier to identify the fault type.

5. The method according to claim 4, characterized in that include: Based on the AdaBoost algorithm, the trained weak classifiers are weighted and fused according to weights to obtain a strong classifier; the fault feature data is classified by the strong classifier to identify the fault type.

6. The method according to claim 5, characterized in that Based on the AdaBoost algorithm, the trained weak classifiers are weighted and fused according to the weights to obtain a strong classifier, including: The obtained strong classifier satisfies: Among them, L j represents a weak classifier, L j =(L 1， L2, L 3…… L N );ω j represents the weight of the weak classifier, ω j =ω1,ω2……ω N ; T represents the number of iterations.

7. The method according to any one of claims 1 to 6, characterized in that: Fault type identification is performed according to the fault characteristic data to obtain a comprehensive analysis result of the power grid fault, including: According to the result of the strong classifier, a comprehensive analysis result of the power grid fault is obtained; the comprehensive analysis result of the power grid fault includes the cause of the fault.

8. The method according to claim 6, characterized in that The method also includes determining the specific location where the fault occurs based on the result of the strong classifier and in combination with the topological structure of the power system.

9. The method according to claim 1, characterized in that: Fault type identification is performed according to the fault characteristic data to obtain a comprehensive analysis result of the power grid fault, including: Standardizing each feature quantity in the obstacle feature quantity data; Determining the weight of each of the feature quantities; Calculating a fault score of a power grid fault according to the characteristic quantities and their weights; According to the result of the fault scoring, determining the fault type corresponding to the currently observed feature quantity combination; The fault score satisfies: L = ω1·I peak +ω2·T fault +ω3·R rate +ω4·R drop +ω n ·N m Wherein, L represents the fault score, ω1, ω2, ω3, ω4, ω n Represents the weight of each feature; I peak is the peak value of the fault current; T fault is the fault duration; R rate is the rate of change of resistance; R drop is the voltage drop; N m Indicates other feature quantities.

10. The method according to claim 1, characterized in that According to the pre-processed data, a plurality of fault characteristic quantity data are obtained, including: Obtaining the fault characteristic quantity data by performing frequency domain analysis on the preprocessed data; The fault characteristic quantity data include: the amplitude of the steady-state quantity, the amplitude of the transient quantity, the amplitude and change rate of the transition resistance value, the phase-to-phase coupling characteristic quantity, the energy spectrum distribution and the transient waveform.

Images