Method and system for testing multi-electric-quantity output relay protection secondary circuit

Abstract

The invention relates to the technical field of intelligent power grids, and discloses a multi-electric-quantity output relay protection secondary circuit test method and system, and the method comprises the steps: collecting an environment noise time sequence, obtaining a noise spectrum, and setting a filter parameter set; outputting a test signal including the feature code signal based on the noise spectrum and the filter parameter set; executing a plurality of groups of test sequences by using the test signals to obtain segmented signals; performing real-time noise suppression on the segmented signals to obtain enhanced signals; extracting a feature vector from the enhanced signal; obtaining a fault type and a fault type position coordinate based on the feature vector; according to the invention, through noise spectrum analysis and independent component analysis, high anti-interference capability and real-time noise suppression of a test signal are realized, a composite fault is successfully separated, a fault type is identified, and accurate positioning of the fault is realized through a cross-correlation function and a coordinate optimization algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grids, and more particularly, to a method and system for testing a secondary circuit of a multi-electric output relay protection. Background Art

[0002] Relay protection is an important guarantee for the safe and stable operation of the power system. The reliability of its secondary circuit is directly related to the normal operation of the relay protection device.

[0003] The existing Chinese patent with publication number CN110888082A discloses a method and device for locating voltage faults at the node of a relay protection secondary circuit, including constructing a secondary circuit model. The secondary circuit model includes a positive power supply node, a negative power supply node, a plurality of components located between the positive power supply node and the negative power supply node, and a plurality of nodes at both ends of the components. The voltage of each of the nodes in the secondary circuit model is obtained to form a node voltage matrix. A normal state voltage model is provided, the normal state voltage model includes a plurality of normal voltage matrices, and the fault point is located based on the comparison between the node voltage matrix and the normal state voltage model. The method realizes secondary circuit fault diagnosis and fault location by collecting electrical quantities of the secondary circuit, thereby realizing real-time fault monitoring of the secondary circuit, and improving the operation and maintenance management level of the smart substation.

[0004] However, in actual test environments, electromagnetic interference, background noise, etc. will seriously affect the quality of test signals, resulting in the distortion of the true values of various fault signals generated by complex faults that often exist in relay protection secondary circuits. Existing test methods lack adaptive anti-interference capabilities for complex electromagnetic environments, making it difficult to effectively separate and identify multiple fault signals that are mixed together, and even more difficult to accurately locate the specific location of the fault, resulting in low maintenance efficiency. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for testing a secondary circuit of a multi-electric output relay protection in order to solve the above problems.

[0006] The present invention provides a method for testing a secondary circuit of a multi-electric output relay protection, comprising the following steps:

[0007] Collect the time series of environmental noise, map the time series to the frequency domain, obtain the noise spectrum and set the filter parameter set;

[0008] outputting a test signal including a signature signal based on the noise spectrum and the filter parameter set;

[0009] Execute multiple test sequences using the generated test signals, collect response data of the circuit under test in different excitation modes, and process the response data in segments to obtain segmented signals;

[0010] Perform real-time noise suppression on the segmented signal to obtain an enhanced signal;

[0011] Based on the enhanced signal and the signature signal, the signal components containing various fault components are separated from the aliased enhanced signal, and the feature vector that can characterize the fault characteristics is extracted based on the signal components;

[0012] Based on the feature vector, a hierarchical fault identification strategy is adopted to obtain the fault type and fault type location coordinates.

[0013] Furthermore, the method of obtaining the noise spectrum and setting the filter parameter set includes:

[0014] Step 101: Sampling and analyzing the background noise of the test environment to obtain a noise time series; using Fourier transform to map the noise time series to the frequency domain to obtain a noise spectrum;

[0015] Step 102: Determine the main noise frequency band and its corresponding energy distribution based on the noise spectrum analysis, where the main noise frequency band is the frequency band whose energy exceeds a preset threshold; for each noise frequency band, set the filter parameters based on the center frequency, bandwidth parameter, stopband attenuation, and passband ripple;

[0016] Step 103 : combining the filter parameters of multiple noise frequency bands and the adaptive adjustment factors into a filter parameter set, wherein the adaptive adjustment factor is used to dynamically adjust the filter parameters according to noise changes during the test process.

[0017] Furthermore, the method of outputting a test signal includes:

[0018] Step 201: Apply a dynamic phase compensation algorithm to introduce real-time phase calculation for each channel. The real-time phase is the sum of the initial phase, the real-time frequency integral, and the dynamic compensation amount calculated based on the filter parameter set.

[0019] Step 202: Calculate the dynamic compensation amount by using a weighted sum of multiple frequency components, which is achieved by superposition of multiple frequency components, each frequency component including a compensation coefficient, a phase adjustment function, and a sine waveform;

[0020] Step 203: Set a harmonic superposition formula with adaptive noise adjustment. The harmonic superposition formula calculates the test signal of each channel. The test signal of each channel is the sum of the fundamental wave to the higher harmonics. Each harmonic component is the product of the fundamental amplitude, the noise suppression adjustment factor, and the sine wave plus the signature signal.

[0021] Step 204, using the weighted sum of multi-frequency sinusoidal components as the signature signal; the signature signal is the sum of a low amplitude coefficient multiplied by multiple sinusoidal components, each sinusoidal component has a specific amplitude, characteristic frequency and initial phase, where the specific amplitude refers to a relative strength value pre-set for each signature component.

[0022] Furthermore, the specific method of obtaining the segmented signal is as follows:

[0023] Step 301, converting the test signal into various stimulus forms;

[0024] Step 302: For each excitation mode, collect response data of the circuit under test, the response data including voltage data, current data, phase data and frequency data;

[0025] Step 303: segment the loop response data using a time window function to obtain segmented signals.

[0026] Furthermore, the method for obtaining the enhanced signal is:

[0027] Step 401: Use an adaptive Kalman filter to perform real-time noise suppression on the segmented signal to obtain a Kalman filtered signal;

[0028] Step 402: further denoising the Kalman filtered signal using wavelet packet decomposition, representing the Kalman filtered signal as a linear combination of wavelet basis functions and wavelet coefficients at different scales and translations, and performing threshold processing on the wavelet coefficients to obtain processed wavelet coefficients;

[0029] Step 403 : reconstruct the time domain signal through inverse wavelet transform, and map the processed wavelet coefficients back to the time domain to obtain an enhanced signal.

[0030] Furthermore, the specific method for extracting the feature vector that can characterize the fault characteristics based on the signal components is:

[0031] Step 501, applying a blind signal separation algorithm to separate the aliased enhanced signal to obtain separated signal components;

[0032] Step 502: Calculate the cross-correlation coefficient between each separated signal component and the signature signal. When the cross-correlation coefficient is greater than a set threshold, the component is considered to contain valid fault information, and the separated signal component is set as a valid signal component.

[0033] Step 503: extracting multi-dimensional features including time domain features, frequency domain features, and time-frequency domain features from the effective signal components;

[0034] In step 504, the multi-dimensional features are combined to form a high-dimensional feature vector. The high-dimensional feature vector includes time domain features, frequency domain features, and time-frequency domain features. Principal component analysis is applied to perform feature dimensionality reduction to obtain a reduced-dimensional feature vector. The reduced-dimensional feature vector is output as a feature vector.

[0035] Furthermore, a method for applying a blind signal separation algorithm to separate the aliased enhanced signal to obtain separated signal components includes:

[0036] The enhanced signal is considered as the observation signal matrix; various independent fault signals are considered as independent fault source signal matrices; the coefficient matrix describing how the fault signals are mixed into the observation signal is considered as the aliasing matrix, and a linear model is established between the observation signal matrix and the independent fault source signal matrix. In the linear model, the observation signal matrix is a linear combination of multiple independent fault source signal matrices and the aliasing matrix, where the aliasing matrix determines the contribution of each fault source to the observation signal.

[0037] The fast independent component analysis algorithm is used to solve the separation matrix; the separation matrix is the inverse matrix or pseudo-inverse matrix of the aliasing matrix. The separation matrix is used to convert the mixed observation signals back into independent fault signals to obtain an estimated source signal matrix. Each row in the estimated source signal matrix represents a separated signal component.

[0038] Furthermore, the specific method for obtaining the fault type and the fault type location coordinates includes:

[0039] Step 601: Using a three-level classifier architecture, the fault type is refined to obtain a hierarchical classification result;

[0040] Step 602: Use a Bayesian network to reason about the correlation of the hierarchical classification results, determine the causal relationship and co-occurrence probability between multiple faults, and obtain multiple simultaneously existing fault types;

[0041] Step 603: For each fault type, use the signal propagation time difference to locate the fault. The distance from the fault type to the test point is obtained by multiplying the product of the signal propagation speed in the medium and the time difference between the fault signal and the test point, divided by 2.

[0042] Step 604 : Using the distance information of the multiple test points, the position coordinates of the fault type are obtained by minimizing the fault coordinate optimization function.

[0043] Furthermore, the time difference is determined by the cross-correlation function, which measures the change in the similarity of two signals as the time offset changes. When the offset makes the two signals most matched, the cross-correlation function reaches its maximum value, and this offset is the time difference; the cross-correlation function calculates the correlation between the enhanced signal related to the fault type and the reference signal of the test point, and finds the time delay value that maximizes the correlation as the time difference.

[0044] The present invention provides a multi-electricity output relay protection secondary circuit testing system, which is used to store computer-readable instructions. When the computer-readable instructions are read, the aforementioned multi-electricity output relay protection secondary circuit testing method can be executed.

[0045] The beneficial effects of the present invention are as follows: the present invention realizes high anti-interference capability of the test signal through noise spectrum analysis and adaptive filter parameter setting; adopts a dynamic phase compensation algorithm to ensure the synchronization accuracy of multi-channel signals; improves the quality of signal acquisition through feature code embedding and multi-band segmented acquisition technology; adopts a method combining adaptive Kalman filtering and wavelet packet decomposition to achieve real-time noise suppression; applies independent component analysis and Bayesian network reasoning to successfully separate complex faults and identify fault types; and realizes accurate fault positioning through cross-correlation function and coordinate optimization algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flowchart of a method for testing a secondary circuit of a multi-electric output relay protection system according to the present invention;

[0047] Figure 2 This is an example diagram of a method for extracting a feature vector capable of characterizing fault characteristics based on signal components according to the present invention;

[0048] Figure 3 This is an example diagram of a method for obtaining a fault type and a fault type location coordinate according to the present invention;

[0049] Figure 4 This is a module example diagram of a multi-electric output relay protection secondary circuit test system of the present invention. DETAILED DESCRIPTION

[0050] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0051] A method and system for testing a secondary circuit of a multi-power output relay protection system, including the following embodiments:

[0052] Example 1

[0053] A method for testing secondary circuits of multi-power output relay protection, such as Figure 1 As shown, the following steps are included:

[0054] Step 100: Collect the time series of environmental noise, map the time series to the frequency domain, obtain the noise spectrum and set the filter parameter set.

[0055] Methods for obtaining the noise spectrum and setting the filter parameter set include:

[0056] Step 101: Sample and analyze the background noise of the test environment to obtain a noise time series. In order to convert the time domain noise signal in the noise time series into the frequency domain, Fourier transform is used to map the noise time series into the frequency domain to obtain a noise spectrum.

[0057] Step 102: Determine the main noise frequency bands and their corresponding energy distribution based on the noise spectrum analysis. The specific method is as follows: first, calculate the energy density of the noise spectrum, identify the frequency bands whose energy exceeds a preset threshold (twice the average energy), and regard these frequency bands as the main noise frequency bands; then apply the peak detection algorithm to locate the local maximum points of energy density, and regard these points as the center frequencies of the main noise frequency bands; then calculate the energy distribution characteristics of each noise frequency band, including the band width (the frequency range where the energy drops to half of the peak value) and energy concentration (the ratio of the peak energy to the total energy of the band).

[0058] For the Ni-th noise frequency band, the filter parameter θ of the Ni-th noise frequency band is set based on the center frequency, bandwidth parameter, stopband attenuation, and passband fluctuation Ni :

[0059] θ Ni ={f c,Ni ,BW Ni ,A s,Ni ,δ Ni}

[0060] Among them, θ Ni is the filter parameter of the Ni-th noise frequency band, f c,Ni is the center frequency or cutoff frequency of the Ni-th noise frequency band (determines the center position of the frequency band where the filter acts), BW Ni is the bandwidth parameter of the Ni-th noise frequency band (the frequency band range covered by the control filter, which needs to cover the noise frequency band width), A s,Ni is the stopband attenuation of the Ni-th noise frequency band (determines the suppression strength outside the noise frequency band and must meet the noise suppression requirements), δ Ni is the passband fluctuation of the Ni-th noise frequency band (limiting the allowable distortion of the signal within the filter passband to ensure the quality of the useful signal).

[0061] The center frequency, bandwidth parameters, stopband attenuation, and passband fluctuation are determined by minimizing the optimization objective function, which is the product integral of the filter transfer function and the noise spectrum. The optimization goal of the optimization objective function is to find the filter parameter combination that can most effectively suppress a specific noise frequency band. The specific process is: take the center of the noise frequency band f Noise,Ni As the core, within the bandwidth (±Δf), the center frequency, bandwidth parameters, stopband attenuation and passband fluctuation of the filter are optimized to minimize the product integral of the filter transfer function and the noise spectrum.

[0062] For the filter parameter θ of the Ni-th noise frequency band Ni , the specific calculation method of the optimization objective function is:

[0063]

[0064] Among them, H(f,θ Ni ) is the transfer function of the filter at frequency f, Δf is the bandwidth considered, f Noise,Ni Indicates the noise frequency band center of the Ni-th noise frequency band, N NS (f) represents the noise spectrum obtained by Fourier transform. The transfer function is composed of the filter parameters θ of the Ni-th noise frequency band Ni Completely determined; for example, for a bandpass filter, its transfer function H(f,θ Ni ) and the parameters of the filter, which determine the shape and characteristics of the transfer function. Specifically, the center frequency determines the frequency location of the filter's action, the bandwidth parameter controls the width of the passband, the stopband attenuation, reflected by the filter order, determines the degree of filter suppression of stopband signals, and the passband ripple affects the fidelity of the signal within the passband. Together, these parameters constitute the complete characteristics of the bandpass filter, enabling it to effectively suppress noise in a specific frequency band while retaining the useful signal. The specific calculation formula is:

[0065]

[0066] Where n is the filter order, which is determined by the stopband attenuation A of the Ni-th noise band required. s,Ni Determines (the higher the order, the steeper the transition band and the greater the stop band attenuation).

[0067] The center frequency, bandwidth parameters, stopband attenuation, and passband fluctuation parameters are optimized using the gradient descent method combined with the grid search strategy. First, the center frequency f c,Ni Initialized to the center frequency of the noise band, bandwidth parameter BW Ni Initialize to the width of the noise band, and then perform an iterative search in the parameter space. Each iteration calculates the gradient of the optimization objective function with respect to each parameter, and updates the parameter value in the opposite direction of the gradient.s,Ni This is achieved by adjusting the filter order n. The higher the order, the greater the attenuation. The passband fluctuation δ Ni This is achieved by introducing a normalization coefficient control in the transfer function, which is directly related to the gain at the passband edge.

[0068] Step 103: Combine the filter parameters of multiple noise frequency bands and the adaptive adjustment factor into a filter parameter set. The adaptive adjustment factor is used to dynamically adjust the filter parameters according to noise changes during the test process.

[0069] The adaptive adjustment factor is set based on the difference between the real-time noise spectrum and the initial ambient noise spectrum. By calculating the correlation coefficient and energy ratio between the two, the system automatically increases the adjustment factor to speed up the filter parameter update when a significant change in noise characteristics is detected. This allows the filter system to quickly adapt to the new noise environment and ensure the quality and reliability of the test signal. The correlation coefficient refers to the statistical similarity between the real-time noise spectrum and the initial ambient noise spectrum. It is calculated by calculating the ratio of the covariance and standard deviation of the two spectra at each frequency point. The value range is [-1, 1]. The closer the value is to 1, the more similar the two spectra are. The energy ratio is the ratio of the total energy of the real-time noise spectrum to the total energy of the initial ambient noise spectrum. It reflects the overall change in noise intensity. When this ratio deviates significantly from 1, it indicates that the noise environment has changed significantly and the filter parameters need to be updated.

[0070] Step 200: output a test signal including a signature signal based on a noise spectrum and a filter parameter set.

[0071] Methods for outputting anti-interference test signals include:

[0072] Step 201: In order to ensure the synchronization of multi-channel signals and dynamically adapt to the noise environment, a dynamic phase compensation algorithm is applied to introduce real-time phase calculation for each channel. The real-time phase is the sum of the initial phase, the real-time frequency integral and the dynamic compensation amount calculated based on the filter parameter set; for the i-th channel, its real-time phase θ i (t) is:

[0073]

[0074] Among them, θ i,0 is the initial phase (to ensure the initial synchronization of multi-channel signals), is the real-time frequency integral (tracking the change of frequency over time to ensure phase continuity and dynamic synchronization), f(t) is the real-time frequency at time t, represents the integral from time 0 to time t, dt represents the integral variable, Δθ i (t,Θ Filter ) is based on the filter parameter set ΘFilter The calculated dynamic compensation amount (adapting to the noise environment based on the filter parameter set to compensate for the phase shift caused by noise).

[0075] In this embodiment, a channel refers to an independent path for transmitting a specific type of electrical signal. Specifically, the channel can be divided into a voltage channel and a current channel, each channel corresponding to a physical output port, which is used to provide an analog signal to the relay protection device under test. For example, in a three-phase power system, it usually includes three voltage channels (corresponding to A, B, and C three-phase voltages) and three current channels (corresponding to A, B, and C three-phase currents), as well as possible zero-sequence or neutral point channels. Each channel can independently control the amplitude, phase, frequency and other parameters of its output signal, but it needs to maintain a precise synchronization relationship with other channels to simulate the operating state of a real power system.

[0076] Step 202, in order to dynamically adjust the phase compensation according to the noise spectrum characteristics, the dynamic compensation amount is calculated using the weighted sum of multiple frequency components. This is achieved by superimposing multiple frequency components, and each frequency component consists of three parts: a compensation coefficient, a phase adjustment function, and a sine waveform. The compensation coefficient is used to control the weight of each frequency component; the phase adjustment function dynamically adjusts the compensation amount according to the noise intensity at the corresponding frequency; and the sine waveform provides periodic compensation that matches the noise frequency. The center frequencies of 3-5 major noise frequency bands are selected as compensation frequency points. When the noise intensity at these frequency points increases, the corresponding compensation amount will also increase, thereby achieving targeted phase compensation and effectively improving the anti-interference ability of the test signal in a complex electromagnetic environment. The calculation formula is:

[0077]

[0078] Among them, α k is the compensation coefficient (used to weight the compensation strength of different noise frequency bands. The specific calculation method is to multiply the ratio of the energy of the kth noise frequency band to the total energy of all selected noise frequency bands by the maximum compensation coefficient (usually 0.2)), φ k is the phase adjustment function (dynamically calculates the compensation amount according to the noise spectrum characteristics), K is the number of compensation items (covers multi-band noise to improve robustness, the value range is [3,5], covers the main noise band to improve the system's anti-interference ability), f k is the compensation frequency point (directional compensation for the frequency band with significant noise, which is the center frequency of the main noise frequency band), sin(2πf k t) is a sinusoidal waveform that provides periodic compensation that matches the noise frequency.

[0079] φ k (N NS (f kThe phase adjustment function uses a nonlinear mapping function to convert the noise spectrum intensity into a phase adjustment amount, ensuring that a small compensation is provided when the noise is weak, and a larger compensation is provided when the noise is strong but does not exceed the safety threshold. The nonlinear mapping function is: take the smaller of 0.5 and the multiplication value, which is 0.1 times the logarithm of the following value: 1 plus 10 times the frequency f k The noise spectrum intensity at is divided by the logarithm of the reference noise level, where the reference noise level is usually taken as the average value of the ambient noise.

[0080] Step 203: In order to generate a test signal that contains multiple harmonics and can avoid the noise frequency band, a harmonic superposition formula with noise adaptive adjustment is set. The harmonic superposition formula calculates the test signal of the i-th channel by summing the harmonics from the 1st to the 24th order. Each harmonic component consists of the product of the basic amplitude, the noise suppression adjustment factor and the sine wave plus the characteristic code signal. Among them, the noise suppression adjustment factor is dynamically adjusted according to the noise spectrum intensity at the harmonic frequency, so that the harmonic energy automatically avoids the frequency band where the noise is concentrated; the sine wave is generated according to the real-time phase; the characteristic code signal is used for subsequent signal recognition, and the specific calculation method is as follows:

[0081]

[0082] Among them, V i (t) is the test signal of channel i, in volts (V) or amperes (A), depending on the channel type; Indicates the sum of harmonics from the 1st to the 24th order, n is the harmonic order index, A n is the fundamental amplitude of the nth harmonic, in the same unit as V i (t) are the same; β is the noise suppression coefficient (determined based on the system's sensitivity to noise and the importance of the test signal, with a value range of 0.5 to 2.0. In conventional test scenarios, β takes a moderate value (e.g., 0.8-1.2) to provide a balanced noise suppression effect; in scenarios where signal integrity is paramount, β takes a smaller value (e.g., 0.5-0.7) to moderately suppress only the strong noise frequency band. When the noise environment is detected to be deteriorating, the β value is automatically increased to enhance the suppression effect). It is a dimensionless coefficient; N NS (nh) represents the noise spectrum at frequency nh, (1-β·N NS (nh)) is the noise suppression adjustment factor, which is used to dynamically adjust the actual amplitude of each harmonic so that the harmonic energy avoids the frequency band where the noise is concentrated. It is a dimensionless quantity; h is the fundamental frequency, sin(n×θ i (t)) is a sine wave with a value range of [-1,1] and dimensionless; S Mark (t) is the characteristic code signal, the unit is V i (t) are the same, ensuring that they can be directly added to the harmonic components.

[0083] Basic amplitude A n and the noise suppression adjustment factor (1-β·N NS (nh)) and the sine wave sin(n×θ i (t)) is multiplied, and the result is the same as A n Same, that is, volts or amperes; characteristic code signal S Mark (t) also have the same unit, so they can be directly added to form the final test signal V of the i-th channel i (t). Test signal V of multiple channels i (t) The complete test signal V(P) composed of

[0084] Step 204: In order to embed a low-amplitude signature code in the signal for subsequent identification of valid signals, a weighted sum of multi-frequency sinusoidal components is used as the signature code signal. The signature code signal is composed of a low-amplitude coefficient (ranging from 0.05 to 0.1) multiplied by the sum of multiple sinusoidal components. Each sinusoidal component has a specific amplitude, characteristic frequency, and initial phase. The specific amplitude refers to a relative strength value preset for each signature code component. These values are optimized and designed based on signal recognition requirements and anti-interference requirements, and are usually allocated according to a specific pattern (such as a decreasing sequence or a golden ratio) to ensure the uniqueness and identifiability of the signature code. The characteristic frequency is selected from the frequency band with the lowest noise identified in step 100 to ensure that the signature code is not interfered with by noise and remains identifiable. The specific calculation method is as follows:

[0085]

[0086] Among them, γ is the low amplitude coefficient (ranging from 0.05 to 0.1), and P is the number of signature code components (determined based on the system's requirements for signature code recognition accuracy and the balance between computational complexity. In a standard test environment, P is usually set to 3-5, which provides sufficient signature code complexity to ensure reliable recognition while maintaining computational efficiency. In a high-interference environment, P can be increased to 6-8 to improve anti-interference capabilities by increasing the redundancy of the signature code. In scenarios requiring fast processing, P can be reduced to 2-3 to reduce the computational burden. In addition, the P value is also related to the required uniqueness of the signature code. When the system needs to distinguish more different types of test signals, a larger P value should be used to provide more coding space. The characteristic frequency f of each signature code component m Evenly distributed in the frequency band with minimum noise to maximize the spectrum utilization efficiency and recognition reliability of the signature code), B m is the amplitude of the mth component, f m is the characteristic frequency of the mth component (select the minimum noise frequency band identified in step 100), is the initial phase of the mth component.

[0087] Step 300: Execute multiple test sequences using the generated test signals, collect response data of the circuit under test in different excitation modes, and process the response data in segments to obtain segmented signals. The response data includes voltage, current, phase, and frequency data.

[0088] The specific method of obtaining the segmented signal is as follows:

[0089] Step 301: In order to comprehensively test the circuit under test using different excitation modes, the basic test signal is converted into various excitation forms, specifically including the following modes:

[0090] Steady-state mode: Directly outputs the original test signal without any transformation processing, used to test the response characteristics of the circuit under test in a stable state;

[0091] Step mode: The test signal is suddenly applied at the step moment to simulate the sudden change in the power system. The step moment refers to the time point when the signal suddenly changes from zero to the set value.

[0092] Sweep frequency mode: The frequency of the test signal changes continuously within a certain range, gradually changing from the starting frequency to the ending frequency. The entire change process is completed within one sweep frequency cycle. It is used to test the response characteristics of the circuit under test at different frequencies.

[0093] Pulse mode: Converts the test signal into a series of periodic pulse sequences. The time interval between each pulse is the pulse period, generating a specific number of pulses in total. This mode is used to test the response characteristics of the circuit under test to transient signals.

[0094] Step 302: For each excitation mode, collect response data of the circuit under test, the response data including voltage data, current data, phase data and frequency data:

[0095] Voltage data: voltage of three phases A, B, C and neutral point;

[0096] Current data: A, B, C three-phase current;

[0097] Phase data: the phase of each voltage and current;

[0098] Frequency data: A, B, and C three-phase frequency changes.

[0099] Step 303: In order to reduce spectrum leakage and improve the analysis accuracy of different frequency bands, a multi-band segmented acquisition method is adopted, and the time window function w is used. j (t) The loop response data X(t) is segmented to obtain segmented signals. The specific expression formula is:

[0100] X Se,j(t) = X(t) × w j (t)

[0101] Among them, the time window function w j (t) Includes three types of windows: rectangular window, Hanning window, and Blackman window; different segments cover low-frequency bands, mid-frequency bands, and high-frequency bands. Each frequency band uses a different sampling rate and window length, and is processed using different window functions to reduce spectrum leakage and interference.

[0102] Different segments cover different frequency bands. An example of the specific division is as follows:

[0103] Low frequency band: f∈[10,100]Hz, using rectangular window and sampling rate f s =2kHz, window length T w =200ms;

[0104] Mid-frequency band: f∈[100,500]Hz, using Hanning window, sampling rate f s =5kHz, window length T w =100ms;

[0105] High frequency band: f∈[500,1200]Hz, using Blackman window, sampling rate f s =10kHz, window length T w =50ms.

[0106] Step 400: Perform real-time noise suppression on the segmented signal to obtain an enhanced signal.

[0107] The specific method for obtaining the enhanced signal is:

[0108] Step 401: Use an adaptive Kalman filter to perform real-time noise suppression on the segmented signal to obtain a Kalman filtered signal.

[0109] Step 402: Although Kalman filtering can effectively suppress Gaussian noise, it has limited effect on non-Gaussian noise and transient interference. Wavelet packet decomposition is used to further denoise the Kalman filtered signal, and the Kalman filtered signal is represented as a linear combination of wavelet basis functions and wavelet coefficients of different scales and translations. The specific expression is:

[0110]

[0111] Among them, ∑ j,p represents the sum of all scales j and translations p, c j,p is the wavelet coefficient, ψ j,p(t) is the wavelet basis function; the wavelet basis function is a special function with compact support (locally finite), oscillation and attenuation. Through scaling and translation operations, it can form a complete function family, which can effectively represent the local time-frequency characteristics of the signal; and the wavelet coefficient is the projection value of the signal on the corresponding wavelet basis function, reflecting the energy distribution of the signal at a specific time scale and position. Larger coefficients usually correspond to important features in the signal, while smaller coefficients may represent noise components.

[0112] In order to remove the noise components in the wavelet domain, hard threshold processing is applied to the wavelet coefficients to obtain the processed wavelet coefficients. The coefficients greater than the threshold are retained and the coefficients less than the threshold are suppressed. This can effectively remove the noise components while retaining the important features of the signal. The specific calculation method is:

[0113]

[0114] in, represents the processed wavelet coefficients, c j,p represents the original wavelet coefficients, λ j represents the threshold value of the j-th scale.

[0115] In order to determine the optimal threshold according to the signal length and noise level, a general threshold calculation formula based on the noise standard deviation is adopted. The threshold of the jth scale is the product of the noise standard deviation and the optimal threshold factor. The optimal threshold factor is the natural logarithm of 2 times the signal length N under the square root. The specific calculation formula is:

[0116]

[0117] Among them, σ j is the standard deviation of the noise at the jth scale, which is calculated by the energy distribution of the noise spectrum in the corresponding frequency band (the specific calculation method is to take the square root of the energy value of the noise spectrum in the frequency band corresponding to the jth scale), N is the signal length (the number of sampling points in the signal), It is the optimal threshold factor derived based on statistical theory. lnN is the natural logarithm of the signal length N. The threshold increases appropriately with the increase of signal length.

[0118] For segmented signals in different frequency bands, different wavelet basis functions are used:

[0119] Low frequency band (10-100Hz): using Daubechies-4 wavelet;

[0120] Mid-frequency band (100-500Hz): using Daubechies-8 wavelet;

[0121] High frequency band (500-1200Hz): Use Symlet-6 wavelet.

[0122] Step 403: Finally, reconstruct the time domain signal through inverse wavelet transform, map the processed wavelet coefficients back to the time domain, and obtain the enhanced signal; the enhanced signal is the sum of the products of all wavelet coefficients after threshold processing and the corresponding wavelet basis functions, and the specific calculation method is:

[0123]

[0124] Among them, X En (t) is the enhanced signal.

[0125] Step 500: Based on the enhanced signal and the signature signal, separate the signal components containing various fault components from the aliased enhanced signal, and extract the signature vector that can characterize the fault characteristics based on the signal components.

[0126] Methods for extracting feature vectors that can characterize fault characteristics based on signal components include Figure 2 As shown, specifically including:

[0127] Step 501: In the relay protection secondary circuit, multiple faults often exist simultaneously and overlap with each other, forming a composite fault signal. En (t), apply the blind signal separation algorithm to separate the aliased enhanced signal and obtain the separated signal components.

[0128] Methods for obtaining separated signal components include:

[0129] Will enhance signal X En (t) is regarded as the observation signal matrix X; various independent fault signals are regarded as the independent fault source signal matrix S; the coefficient matrix describing how the fault signals are mixed into the observation signals is regarded as the aliasing matrix A.

[0130] To represent the linear mixing relationship of multiple fault signals, a linear model is established between the observation signal matrix and the independent fault source signal matrix. The observed signal in the linear model is a linear combination of multiple independent fault signals, where the aliasing matrix determines the contribution of each fault source to the observed signal. The specific expression formula is as follows:

[0131] X=AS

[0132] In order to recover the original independent fault signal from the mixed signal, the separation matrix is calculated; the separation matrix is the inverse matrix or pseudo-inverse matrix of the aliasing matrix. The separation matrix is used to convert the mixed observation signals back to independent fault signals to obtain the estimated source signal matrix. Each row in the estimated source signal matrix represents a separated signal component. By applying the separation matrix to the observation signal matrix, the conversion from mixed observation signals to independent fault components is achieved. The specific expression formula is as follows:

[0133]

[0134] in, is the estimated source signal matrix, that is, the independent fault signal matrix obtained after separation, W represents the separation matrix, and X represents the observation signal matrix.

[0135] Methods for calculating the separation matrix include:

[0136] The Fast Independent Component Analysis (FastICA) algorithm is used to solve the separation matrix W, and the non-Gaussianity metric is maximized through an iterative optimization process. The non-Gaussianity metric measures the degree to which a mixed signal deviates from a normal (Gaussian) distribution. The maximization process in the iterative optimization process involves repeated calculations and continuous parameter adjustments to maximize the non-Gaussianity metric.

[0137] To quantify the non-Gaussian nature of the signal and serve as a measure of independence, a contrast function based on a nonlinear function is defined. This function calculates the square of the expected value difference between the input signal vector (i.e., the observed mixed signal) and a standard Gaussian random variable (i.e., a normally distributed random variable with mean 0 and variance 1). Nonlinear functions, such as tanh or exp, are used to enhance the non-Gaussian nature of the signal. The contrast function serves as the optimization objective, with a larger value indicating better separation. The expected value difference reflects the degree to which the signal distribution differs from the Gaussian distribution.

[0138] Step 502: After signal separation is completed, it is necessary to identify which separated components contain valid fault information. By calculating the cross-correlation coefficient between each separated signal component and the characteristic code signal, when the cross-correlation coefficient is greater than a set threshold (the value range is 0.6-0.8), it is considered that the component contains valid fault information, and the separated signal component is set as a valid signal component.

[0139] The cross-correlation coefficient is calculated by taking the sum of the products of the separated signal components and the signature minus their respective means, and dividing it by the product of their standard deviations. The higher the cross-correlation coefficient, the more likely that the component contains valid signal information. The specific calculation method for the cross-correlation coefficient is:

[0140]

[0141] Among them, ρ i is the mutual correlation coefficient, s i (t) is the ith signal component after separation, is the signal component s i The mean of (t), is the signature signal S Mark The mean of (t), Σ t represents the sum of all time points t. The mutual correlation coefficient ρ iThe higher the value, the more likely the component is to contain valid signal information.

[0142] Step 503: After determining the effective signal components, multi-dimensional features are extracted from the effective signal components. Feature extraction includes the following three types of features:

[0143] Time domain characteristics: obtained by directly performing statistical analysis on the time domain signal, including statistical quantities such as mean, variance, peak, kurtosis, and skewness;

[0144] Frequency domain features: obtained by Fourier transforming the signal and then analyzing it in the frequency domain, including power spectrum density, harmonic content, frequency band energy distribution, etc.

[0145] Time-frequency domain features: These features are obtained by simultaneously analyzing the characteristics of the signal in the time domain and frequency domain, including wavelet energy entropy and empirical mode decomposition (EMD) coefficients. These features can reflect the frequency characteristics of the signal that change over time.

[0146] In step 504, to comprehensively characterize the fault characteristics, the different types of features extracted above are combined to form a high-dimensional feature vector. The high-dimensional feature vector contains time domain features, frequency domain features, and time-frequency domain features. However, high-dimensional feature vectors may contain redundant information, which affects the efficiency and accuracy of subsequent fault diagnosis. Therefore, principal component analysis (PCA) is applied to reduce the feature dimensionality. The reduced feature vector is obtained by multiplying the original feature vector with the transpose of the principal component transformation matrix. The reduced feature vector is then output as the feature vector.

[0147] Step 600: Based on the feature vector, a hierarchical fault identification strategy is adopted to obtain the fault type and the coordinates of the fault type location.

[0148] The method for obtaining the fault type and the fault type location coordinates is as follows: Figure 3 As shown, specifically including:

[0149] Step 601 : adopt a three-level classifier architecture to gradually refine the fault type and obtain a hierarchical classification result.

[0150] In the first level classification, the system first processes the feature vector after dimensionality reduction to achieve a rough classification of the fault type. The first level classifier receives the dimensionality reduction feature vector F Dr As input, and according to the preset classifier parameters, the main type of fault C1 is output, such as basic categories such as ground fault, short circuit fault or open circuit fault.

[0151] In the second-level classification, the system further subdivides the fault into subtypes based on the coarse classification. The second-level classifier not only considers the original feature vector information but also uses the first-level classification result C1 as an important reference. By comprehensively analyzing these two parts of information, it outputs a more refined fault subtype classification result C2.

[0152] In the third level of classification, the system performs a final and accurate classification of the fault. The third-level classifier comprehensively considers the feature vector and the classification results C1 and C2 of the first two levels. Through more complex classification algorithms and parameter configurations, it determines the specific fault mode and severity C3, providing detailed guidance for subsequent fault handling.

[0153] This hierarchical classification architecture can gradually improve the accuracy of classification. Through this progressive classification method, complex fault conditions can be identified more accurately and a complete hierarchical classification result (C1, C2, C3) can be obtained.

[0154] For example, for a fault signal in a relay protection secondary circuit:

[0155] The first-level classifier identifies it as a “short-circuit fault” (C1);

[0156] The second-level classifier further subdivides it into “phase-to-phase short circuit” (C2);

[0157] The third-level classifier ultimately determines it as "metallic short circuit between phases A and B, severity is high" (C3).

[0158] Step 602: After obtaining the hierarchical classification results (C1, C2, C3), the Bayesian network is used to infer the correlation of the hierarchical classification results, determine the causal relationship and co-occurrence probability between multiple faults, and obtain multiple simultaneously existing fault types.

[0159] The core of Bayesian network reasoning is to associate observed evidence with the probability of failure to calculate the posterior probability, that is, the probability that a certain failure exists when a specific piece of evidence E is observed. This probability is equal to the probability of observing evidence E when the failure exists (likelihood probability) multiplied by the failure F i The prior probability of is divided by the total probability of observing evidence E under all possible faults (as a normalization factor). The observed evidence E consists of two parts: the feature vector and the hierarchical classification results (C1, C2, C3).

[0160] The formula for Bayes' theorem is:

[0161]

[0162] Among them, P(F i |E) means that under the condition of observing evidence E, fault F i Probability of existence (posterior probability), P(E|F i ) indicates fault F i The probability (likelihood) of observing evidence E when it exists, P(F i ) indicates fault Fi The prior probability is the basic probability of a failure occurring in the absence of any observational evidence, and the denominator is a normalization factor to ensure that the sum of all failure probabilities is 1.

[0163] Prior probability P(F i ) is obtained through historical fault data statistics, reflecting the frequency of occurrence of different faults in historical records. The Bayesian network stores the dependencies between nodes through conditional probability tables, enabling the system to handle the uncertainty and causal relationships in compound faults, and ultimately identify multiple simultaneous fault types.

[0164] Step 603: Fault signal time difference analysis. i , determine its specific location in the secondary circuit. Use the signal propagation time difference for positioning, and divide the product of the signal propagation speed in the medium and the time difference of the fault signal reaching the test point by 2 to get the fault type F i The distance to the test point l.

[0165] The distance calculation formula from the fault to the test point is:

[0166]

[0167] Among them, d il Indicates fault type F i The distance to the test point l, v represents the propagation speed of the signal in the medium, which is determined by the characteristics of the loop material, Δt il It represents the time difference of the fault signal reaching the test point, divided by 2 because of the characteristics of the round-trip propagation of the signal.

[0168] Time difference Δt il This is determined by the cross-correlation function, which measures how the similarity of two signals changes over time. When the offset makes the two signals most closely match, the cross-correlation function reaches its maximum value, and this offset is the time difference. Specifically, the cross-correlation function calculates the correlation between the enhanced signal associated with the fault type and the reference signal at the test point, finding the time delay value that maximizes the correlation as the time difference. The calculation method is:

[0169]

[0170] in, Fault type F i The cross-correlation function, Is related to fault type F i The associated enhanced signal, is the reference signal of test point l (the reference signal is obtained by collecting the signal of the normal working state at test point j at the initialization time, and is band-pass filtered and normalized to improve the signal quality. The reference signal is used as a benchmark for comparison with the fault signal), argmax τ represents the time delay τ that makes the cross-correlation function reach its maximum value, Represents the integral from negative infinity to positive infinity.

[0171] In step 604, the distance information from multiple test points is used to determine the location coordinates of the fault type by minimizing the fault coordinate optimization function. The optimization goal of the fault coordinate optimization function is to find a spatial coordinate point that minimizes the sum of the squared errors between the theoretical distance from that point to each test point and the actual distance calculated using time difference analysis. This is a minimization problem, which involves calculating the sum of the squares of the differences between the estimated distance (calculated from the fault to the test point) and the actual distance (i.e., the Euclidean distance from the coordinate point to the test point) and finding the coordinate point that minimizes this sum of squares. The squaring operation ensures that positive and negative errors are treated equally and amplifies the impact of large errors.

[0172] The formula of the fault coordinate optimization function is:

[0173]

[0174] Among them, (x i ,y i ,z i ) is the fault type F i The spatial coordinates of il Indicates fault type F i The distance to the test point l, argmin (x,y,z) Represents the coordinates (x, y, z) that minimize the following expression. The expression calculates the sum of the squares of the differences between the estimated distance and the actual distance. The square root term calculates the Euclidean distance from the coordinate point to the test point. The square operation ensures that positive and negative errors are treated equally and amplifies the impact of large errors. represents the sum from l=1 to l=R, (x l ,y l ,z l ) is the coordinate of the lth test point, and R is the number of test points. This minimization problem is solved using the Levenberg-Marquardt algorithm, iteratively updating the coordinate estimates.

[0175] In the secondary circuit of relay protection, the coordinate system definition of spatial coordinates adopts a three-dimensional rectangular coordinate system, where:

[0176] X-axis: along the width of the distribution cabinet or control cabinet, in meters (m);

[0177] Y-axis: along the depth direction of the distribution cabinet or control cabinet, in meters (m);

[0178] Z axis: along the height direction of the power distribution cabinet or control cabinet, the unit is meter (m);

[0179] The coordinate origin (0,0,0) is set at the front of the lower left corner of the distribution cabinet or control cabinet. Test point coordinates (x j ,y j ,z j ) are fixed positions that are measured and recorded in advance in the system. These test points are usually distributed at key nodes of the relay protection secondary circuit, such as terminal blocks, relays, transmitters, etc.

[0180] In order to improve positioning accuracy, the layout of test points should follow the following principles:

[0181] The number of test points R should be no less than 4 to ensure a unique solution for three-dimensional spatial positioning;

[0182] The test points should be distributed on different planes as much as possible to avoid collinear or coplanar arrangements;

[0183] The distance between test points should be appropriate, neither too dense nor too scattered, usually 0.5m to 2m.

[0184] Through this coordinate positioning method, the system can accurately determine the spatial location of the fault, with a positioning accuracy of typically ±10cm, providing maintenance personnel with intuitive fault location information and greatly shortening troubleshooting time.

[0185] Example 2

[0186] See Figure 4 As shown, a multi-power output relay protection secondary circuit test system is provided, which is used to store computer-readable instructions. When the computer-readable instructions are read, the aforementioned multi-power output relay protection secondary circuit test method can be executed. The system includes:

[0187] The environmental perception module 101 collects the environmental noise time series, maps the time series to the frequency domain, obtains the noise spectrum and sets the filter parameter set;

[0188] The collaborative output module 102 outputs a test signal including a signature signal based on the noise spectrum and the filter parameter set;

[0189] The test acquisition module 103 executes multiple test sequences using the generated test signals, collects response data of the circuit under test under different excitation modes, and processes the response data in segments to obtain segmented signals; wherein the response data includes voltage, current, phase, and frequency data;

[0190] The suppression and enhancement module 104 performs real-time noise suppression on the segmented signal to obtain an enhanced signal;

[0191] The fault separation module 105 separates the signal components containing various fault components from the aliased enhanced signal based on the enhanced signal and the characteristic code signal, and extracts the characteristic vector that can characterize the fault characteristics based on the signal components;

[0192] The diagnosis and positioning module 106 adopts a hierarchical fault identification strategy based on the feature vector to obtain the fault type and the location coordinates of the fault type.

[0193] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A method for testing a secondary circuit of a multi-electric output relay protection, characterized in that: The following steps are involved: Collect the time series of environmental noise, map the time series to the frequency domain, obtain the noise spectrum and set the filter parameter set; outputting a test signal including a signature signal based on the noise spectrum and the filter parameter set; Execute multiple test sequences using the generated test signals, collect response data of the circuit under test in different excitation modes, and process the response data in segments to obtain segmented signals; Perform real-time noise suppression on the segmented signal to obtain an enhanced signal; Based on the enhanced signal and the signature signal, the signal components containing various fault components are separated from the aliased enhanced signal, and the feature vector that can characterize the fault characteristics is extracted based on the signal components; Based on the feature vector, a hierarchical fault identification strategy is adopted to obtain the fault type and fault type location coordinates.

2. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 1, characterized in that: Methods for obtaining the noise spectrum and setting the filter parameter set include: Step 101: Sampling and analyzing the background noise of the test environment to obtain a noise time series; using Fourier transform to map the noise time series to the frequency domain to obtain a noise spectrum; Step 102: Determine the main noise frequency band and its corresponding energy distribution based on the noise spectrum analysis, where the main noise frequency band is the frequency band whose energy exceeds a preset threshold; for each noise frequency band, set the filter parameters based on the center frequency, bandwidth parameter, stopband attenuation, and passband ripple; Step 103 : combining the filter parameters of multiple noise frequency bands and the adaptive adjustment factors into a filter parameter set, wherein the adaptive adjustment factor is used to dynamically adjust the filter parameters according to noise changes during the test process.

3. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 1, characterized in that: Methods for outputting test signals include: Step 201: Apply a dynamic phase compensation algorithm to introduce real-time phase calculation for each channel. The real-time phase is the sum of the initial phase, the real-time frequency integral, and the dynamic compensation amount calculated based on the filter parameter set. Step 202: Calculate the dynamic compensation amount by using a weighted sum of multiple frequency components, which is achieved by superposition of multiple frequency components, each frequency component including a compensation coefficient, a phase adjustment function, and a sine waveform; Step 203: Set a harmonic superposition formula with adaptive noise adjustment. The harmonic superposition formula calculates the test signal of each channel. The test signal of each channel is the sum of the fundamental wave to the higher harmonics. Each harmonic component is the product of the fundamental amplitude, the noise suppression adjustment factor, and the sine wave plus the signature signal. Step 204, using the weighted sum of multi-frequency sinusoidal components as the signature signal; the signature signal is the sum of a low amplitude coefficient multiplied by multiple sinusoidal components, each sinusoidal component has a specific amplitude, characteristic frequency and initial phase, where the specific amplitude refers to a relative strength value pre-set for each signature component.

4. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 1, characterized in that: The specific method of obtaining the segmented signal is as follows: Step 301, converting the test signal into various stimulus forms; Step 302: For each excitation mode, collect response data of the circuit under test, the response data including voltage data, current data, phase data and frequency data; Step 303: segment the loop response data using a time window function to obtain segmented signals.

5. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 1, characterized in that: The method to obtain the enhanced signal is: Step 401: Use an adaptive Kalman filter to perform real-time noise suppression on the segmented signal to obtain a Kalman filtered signal; Step 402: further denoising the Kalman filtered signal using wavelet packet decomposition, representing the Kalman filtered signal as a linear combination of wavelet basis functions and wavelet coefficients at different scales and translations, and performing threshold processing on the wavelet coefficients to obtain processed wavelet coefficients; Step 403 : reconstruct the time domain signal through inverse wavelet transform, and map the processed wavelet coefficients back to the time domain to obtain an enhanced signal.

6. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 1, characterized in that: The specific method of extracting the feature vector that can characterize the fault characteristics based on the signal components is: Step 501, applying a blind signal separation algorithm to separate the aliased enhanced signal to obtain separated signal components; Step 502: Calculate the cross-correlation coefficient between each separated signal component and the signature signal. When the cross-correlation coefficient is greater than a set threshold, the component is considered to contain valid fault information, and the separated signal component is set as a valid signal component. Step 503: extracting multi-dimensional features including time domain features, frequency domain features, and time-frequency domain features from the effective signal components; In step 504, the multi-dimensional features are combined to form a high-dimensional feature vector. The high-dimensional feature vector includes time domain features, frequency domain features, and time-frequency domain features. Principal component analysis is applied to perform feature dimensionality reduction to obtain a reduced-dimensional feature vector. The reduced-dimensional feature vector is output as a feature vector.

7. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 6, characterized in that: Methods for applying a blind signal separation algorithm to separate the aliased enhanced signal to obtain separated signal components include: The enhanced signal is considered as the observation signal matrix; various independent fault signals are considered as independent fault source signal matrices; the coefficient matrix describing how the fault signals are mixed into the observation signal is considered as the aliasing matrix, and a linear model is established between the observation signal matrix and the independent fault source signal matrix. In the linear model, the observation signal matrix is a linear combination of multiple independent fault source signal matrices and the aliasing matrix, where the aliasing matrix determines the contribution of each fault source to the observation signal. The fast independent component analysis algorithm is used to solve the separation matrix; the separation matrix is the inverse matrix or pseudo-inverse matrix of the aliasing matrix. The separation matrix is used to convert the mixed observation signals back into independent fault signals to obtain an estimated source signal matrix. Each row in the estimated source signal matrix represents a separated signal component.

8. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 1, characterized in that: The specific methods for obtaining the fault type and the coordinates of the fault type location include: Step 601: Using a three-level classifier architecture, the fault type is refined to obtain a hierarchical classification result; Step 602: Use a Bayesian network to reason about the correlation of the hierarchical classification results, determine the causal relationship and co-occurrence probability between multiple faults, and obtain multiple simultaneously existing fault types; Step 603: For each fault type, use the signal propagation time difference to locate the fault. The distance from the fault type to the test point is obtained by multiplying the product of the signal propagation speed in the medium and the time difference between the fault signal and the test point, divided by 2. Step 604 : Using the distance information of the multiple test points, the position coordinates of the fault type are obtained by minimizing the fault coordinate optimization function.

9. A method for testing a secondary circuit of a multi-electric output relay protection according to claim 8, characterized in that: The time difference is determined by the cross-correlation function, which measures the change in the similarity of two signals as the time offset changes. When the offset makes the two signals most matched, the cross-correlation function reaches its maximum value, and the offset is the time difference; The cross-correlation function calculates the correlation between the enhanced signal related to the fault type and the reference signal of the test point, and finds the time delay value that maximizes the correlation as the time difference.

10. A multi-power output relay protection secondary circuit test system, characterized in that: It is used to store computer-readable instructions, and when the computer-readable instructions are read, it can execute a multi-electric output relay protection secondary circuit testing method as described in any one of claims 1-9.

Images