A photovoltaic direct-current system fault detection method and device based on line spectrum suppression

Abstract

The application discloses a photovoltaic direct-current system fault detection method and device based on line spectrum suppression, the method extracts signal peak frequency through time-frequency spectrum and analyzes time-frequency stability, and identifies periodic line spectrum signals. Combined with envelope and minimum value point detection on the spectrum, frequency bands are divided, the signal is decomposed by using an empirical wavelet transform, modal components containing line spectrum signals are removed, only non-stationary modal components related to fault characteristics are reserved, and fault characteristic signals are obtained by reconstructing the non-stationary modal components. The statistical characteristics such as root mean square value, variance and kurtosis of the fault characteristic signals are compared with feature thresholds to detect fault arcs. And based on the algorithm, a photovoltaic system direct-current arc detection device is designed, which effectively solves the problem that periodic signals and ripple noises in the photovoltaic direct-current system interfere with fault characteristics in the case that fault characteristics are not obvious under low-voltage operation, and improves the universality of photovoltaic direct-current system fault identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic direct-current system fault diagnosis, and particularly relates to a photovoltaic direct-current system fault detection method and device based on line spectrum suppression. BACKGROUND

[0002] The arc fault of the direct-current system is a non-stationary electrical discharge phenomenon caused by poor contact, line aging or insulation damage, which is characterized by severe fluctuations of current and voltage, and is accompanied by high-frequency nonlinear signal characteristics. The frequency and amplitude of the arc fault randomly change over time. This arc fault is usually superimposed on the normal operating signal, mixed with periodic signals and ripple noise, and is easily hidden and unstable. If the arc fault is not detected in time, it may cause power loss, equipment overheating, component burning, and even serious fire accidents, directly threatening the operation safety and equipment life of the photovoltaic system.

[0003] The arc fault detection mainly relies on the analysis of time domain, frequency domain or time-frequency domain characteristics of the signal. In the time domain analysis, the voltage and current waveform changes on the line when the arc occurs are used for detection. By taking the arc current change rate under a specific time window as a characteristic quantity, the arc can be judged to a certain extent. However, due to the great influence of environmental factors (such as light intensity) on the photovoltaic system, the current changes significantly, and the current changes caused by different arc burning conditions differ greatly, which leads to the fact that the method based on time domain characteristics is easily disturbed by external interference, has a high misjudgment rate, and has insufficient anti-interference ability, making it difficult to achieve accurate arc fault detection. In the frequency domain analysis, the change of harmonic content at high frequency before and after the occurrence of the arc is used as the detection basis, which can avoid the problem that the time domain method is easily disturbed. The frequency domain detection based on Fourier transform has high precision and is not affected by the type of arc, but it lacks time domain information, has complex calculation, and has high requirements for waveform stability. When the bus series arc occurs in the photovoltaic system, the low-frequency part of the current signal will be significantly reduced, and this phenomenon is greatly affected by the arc position. At present, there is an urgent need for a detection method that can accurately retain the non-stationary characteristics of the arc fault, so as to highlight the fault signal characteristics under complex working conditions and improve the accuracy and reliability of fault diagnosis. SUMMARY

[0004] The purpose of the present application is to provide a photovoltaic direct-current system fault detection method and device based on line spectrum suppression, which can accurately extract the non-stationary characteristics of the arc fault by suppressing periodic line spectrum signals and noise interference, realize reliable fault detection of the photovoltaic direct-current system under complex working conditions, and improve the accuracy and robustness of fault diagnosis.

[0005] To achieve the above purpose, the present application adopts the following technical scheme: a photovoltaic direct-current system fault detection method based on line spectrum suppression, characterized in that it comprises the following steps:

[0006] S1: initialize the arc fault detection device, use the acquisition module to collect the voltage signal of the photovoltaic DC system in real time from the voltage sampling point, and the collection frequency is f0;

[0007] S2: perform fast Fourier transform (FFT) and short-time Fourier transform (STFT) on the signal to generate the single-sided spectrum and time-frequency spectrum Y1 of the signal, find the local peak frequency in each time frame, and form a time-frequency spectrum matrix Y4 containing only the peak frequency;

[0008] S3: count the number of occurrences of each frequency point in different time frames as the frequency stability, select the points of stable frequency as the starting frequency points, and perform dynamic programming to generate a path using the frequency offset between time frames and the offset of the path point from the starting frequency as the total deviation constraint. This path is the path of the main frequency changing with time frames;

[0009] S4: perform line spectrum recognition using the path offset length and node energy entropy as threshold indicators, perform Fourier transform (FFT) on the signal, extract the minimum value points of the envelope on the frequency spectrum, and perform interval merging according to the extracted line spectrum frequency information to obtain the frequency band division boundary;

[0010] S5: perform empirical wavelet transform (EWT) on the signal according to the divided frequency band to obtain a plurality of modal components (IMFs), and reconstruct the modal components without line spectrum frequency to obtain a fault characteristic signal;

[0011] S6: extract the threshold values of the root mean square value, variance, kurtosis, peak factor, pulse factor and waveform factor of the reconstructed signal to detect arc faults, and determine whether an arc is generated.

[0012] Further, the step S1 is specifically: initializing the arc fault detection device, using an ADC acquisition module to acquire the voltage signal in real time from the voltage sampling point of the photovoltaic DC system, and the sampling frequency f0 is at least twice the maximum frequency of the effective fault arc characteristic frequency band. In the case where the sampling hardware allows, a higher sampling frequency can more accurately capture the high frequency characteristics and non-stationarity of the fault arc signal, and f0 is set to 100 kHz-1 MHz, and the sampling time is T.

[0013] Further, the step S2 is specifically: in the optional embodiment S2 of the present application, the collected signal x(t) is subjected to fast Fourier transform (FFT) to generate a single-sided spectrum X(f).

[0014] The collected signal is subjected to short-time Fourier transform (STFT), the signal is segmented and weighted using a Hanning window, the window length is set to N win , the length of the window overlap part is set to N overlap , and the number of fast Fourier transform points is N FFT .

[0015] After applying the window function to each segment of signal, fast Fourier transform (FFT) is performed to obtain the spectrum of the time frame. The spectra of all time frames are stacked to obtain the time-frequency spectrum matrix Y1.

[0016] In each time frame of Y1, local peak frequency in frequency dimension is detected point by point, and the local maximum frequency point is determined by the following condition:

[0017]

[0018] In the formula, f is the current frequency point, is the spectral resolution, t is the time frame index, and N t is the total number of time frames.

[0019] The local peak frequency point is retained, and the rest is set to zero to obtain the peak frequency matrix Y4.

[0020] Further, the step S3 is specifically: for the peak frequency matrix Y4 generated in S2, all time frames are traversed, and the number of non-zero values of each frequency point in all time frames is counted as the frequency stability N(f) of the frequency point.

[0021] In order to reduce the amount of calculation, preliminary screening is performed to filter out non-stationary frequency points, and the preliminary screening process is as follows:

[0022]

[0023] In the formula, Td is the time proportion threshold, and S'(f) is the initial frequency point set.

[0024] Taking the frequency point of S'(f) as the starting frequency point, a line spectrum stationary path is searched in the peak frequency matrix Y4 by using dynamic programming, and the total deviation constraint Cost(f t ) of the dynamic programming is as follows:

[0025] Cost(f t ) = a · |f t -f t-1 | + β · |f t -f start |

[0026] In the formula, f start is the starting frequency point of the initial frequency point set, |f t -f t-1 | is the frequency offset cost between time frames, |f t -f start | is the global offset cost of the current path frequency point and the starting frequency point, and a and β are cost weight parameters.

[0027] The cumulative minimum cost value C(f, t) of each frequency point is calculated frame by frame, and the recursive formula is as follows:

[0028] C(f, t) = min(C(f t-1 ,t-1) + Cost(f t )

[0029] Wherein, C(f, t) is the cumulative minimum cost value of the current time frame t, C(f t-1 ,t-1) is the cumulative minimum cost value of the previous time frame t-1, and Cost(f t ) is the total deviation constraint of the current time frame.

[0030] The frequency point with the minimum cost value of the last time frame is obtained The main frequency path is obtained

[0031] Further, the step S4 is specifically: the line spectrum signal has frequency stability and periodic stability in the process of changing with time, and the path offset length L path = C(f, t) and the node energy entropy H are selected for line spectrum identification, and the node energy entropy calculation formula is as follows:

[0032]

[0033] Wherein, E i is the energy value of the node of the path P(t, f).

[0034] The stable line spectrum frequency point in the signal is determined by the path offset length threshold L0 and the node energy entropy threshold H0, and the distribution interval of the line spectrum frequency is determined by taking the line spectrum frequency point as a reference.

[0035] The single side spectrum X(f) obtained in the step S2 is analyzed by a sliding window to extract the upper envelope X upper (f), and the initial frequency band division boundary is obtained by detecting the envelope minimum value point.

[0036] The frequency band interval containing the line spectrum frequency point and adjacent to each other is merged as a frequency band interval, and the final frequency band division boundary B = {[f j ] | j = 1, 2,...} is obtained.

[0037] Further, the step S5 is specifically: according to the frequency band division boundary B, the EWT decomposition is carried out, for each frequency band interval [f i ,f i+1 ], a band-pass filter W i (f) is constructed, and the modal decomposition is carried out on the collected signal x(t):

[0038]

[0039] wherein, and denote the Fourier transform and inverse transform of the signal respectively, each IMF i (t) corresponds to the feature of the signal in the i-th frequency band.

[0040] The reconstructed signal x r (t) in the frequency band interval not containing the line spectrum frequency is obtained as the fault feature signal.

[0041]

[0042] Further, the step S6 is specifically: extracting statistical features of the reconstructed signal x r (t) for analysis, and judging whether the arc fault occurs by setting a threshold of the feature quantity, and the selection of the statistical features should be based on the following aspects:

[0043] Energy variation: reflecting the enhancement of the energy level and fluctuation of the fault signal;

[0044] Waveform characteristics: capturing the characteristics of the instantaneous peak or complex waveform in the fault signal;

[0045] Transient characteristics: reflecting the short-time drastic change caused by the arc fault.

[0046] The application has the following beneficial effects:

[0047] (1) The application can accurately identify and eliminate the line spectrum frequency interference stably existing in the photovoltaic direct-current system by frequency stability and energy stability analysis combined with the dynamic programming algorithm, avoid the line spectrum frequency from covering the fault feature signal, and more clearly display the fault feature.

[0048] (2) The application can effectively separate the normal signal and the fault signal under the complex working conditions such as the low-voltage operation of the photovoltaic system and the starting of the inverter, retain the characteristics such as the starting of the inverter and the arc fault, accurately judge the occurrence of the arc fault, reduce the misjudgment and the missed judgment, and has strong robustness and anti-interference ability. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the application, and therefore should not be regarded as a limitation to the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0050] Figure 1 A photovoltaic direct-current system fault detection device based on line spectrum suppression is provided for the embodiments of the application.

[0051] Figure 2 A flowchart for suppressing the spectral frequency of a signal line is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0053] Figure 1 A method and device for photovoltaic DC system fault detection based on line spectrum suppression are provided, which include a signal acquisition module S101, a signal processing unit S102, a fault judgment module S103, and a display alarm device S104.

[0054] The signal acquisition module S101 is distributed and arranged at the grid-connected end and the inverter side of the photovoltaic DC system, and real-time voltage or current signals are acquired;

[0055] The signal processing unit S102 receives the acquired signals, and performs line spectrum interference suppression, frequency band decomposition, and signal reconstruction;

[0056] The fault judgment module S103 extracts statistical characteristic values (such as root mean square value, variance, peak factor, etc.) and combines a preset threshold to judge whether there is a fault arc;

[0057] The display alarm device S104 is used for real-time display of system operation state and fault information, and triggers an alarm when a fault occurs.

[0058] Further, the signal acquisition module S101 monitors the voltage signal in the photovoltaic DC system in real time, acquires the signals of key positions (such as the input end of the inverter and the grid-connected line) of the system by using a voltage sensor, and digitizes the signals by a high-speed analog-to-digital converter (ADC) at a sampling frequency of f0=100 kHz, so as to ensure the capture of high-frequency fault characteristics.

[0059] Further, the signal processing process of the signal processing unit S102 is as shown in Figure 2

[0060] The acquired voltage signal is subjected to short-time Fourier transform (STFT), the window length N win is 512, the length N overlap of the window overlap part is 256, the fast Fourier transform (FFT) point number N FFT is 1024, and the frequency resolution is as follows:

[0061] ​

[0062] The spike extraction is to compare the amplitude of adjacent frequency points frame by frame on the time-frequency spectrum matrix Y1, extract the local spike frequency, and obtain the spike frequency matrix Y4.

[0063] According to the frequency stability N(f), the initial frequency point S'(f) is preliminarily screened, and the offset process of the initial frequency point S'(f) on the time axis is dynamically planned to obtain the main frequency path P(t,f). The total deviation constraint Cost(f t ) of the dynamic planning is as follows:

[0064] Cost(f t ) = a · |f t -f t-1 | + β · |f t -f start |

[0065] Wherein, |f t -f t-1 | is the frequency offset cost between time frames, |f t -f start | is the global offset cost of the current path frequency point and the initial frequency point, and a and β are cost weight parameters, both of which are set to 0.5.

[0066] The path offset length L path of the main frequency path P(t,f) is calculated, and the line spectrum frequency is identified according to the path offset length threshold L0 and the energy entropy threshold H0.

[0067] Wherein, the path offset length reflects the frequency stability of the line spectrum frequency, and the node energy entropy reflects the periodicity and energy stability of the frequency component.

[0068] The collected voltage signal is subjected to fast Fourier transform (FFT) to obtain a single-sided spectrum X(f).

[0069] According to the minimum value point of the envelope on X(f), the frequency band division boundary is extracted, and the adjacent frequency band intervals containing the line spectrum frequency points are merged as a frequency band interval.

[0070] The signal is decomposed into a plurality of modal components (IMFs) by using empirical wavelet transform (EWT), and the non-line spectrum modal components are superimposed and reconstructed to obtain a feature signal x r (t) without line spectrum interference.

[0071] Further, the fault judgment module S103 performs statistical feature extraction and analysis on the reconstructed signal to realize the judgment of the fault arc.

[0072] The statistical characteristics of the extracted signal, such as root mean square value, variance, kurtosis, peak factor, pulse factor and waveform factor, can reflect the energy level, waveform sharpness and instantaneous pulse characteristics of the signal.

[0073] According to the fault characteristic threshold, the possible fault arc in the system is judged. If it is judged that the circuit produces a fault arc, the display alarm device S104 is triggered to alarm, and if it is judged that the circuit is normal, the detection is continued.

[0074] Further, the display alarm device S104 displays the running state and fault information of the photovoltaic DC system in real time, and triggers an alarm when a fault arc is detected.

[0075] The display alarm device S104 receives the fault judgment result and related signal characteristic data in real time through the communication interface with the fault judgment module S103, and displays the current system running state on the display screen of the device, including the system state, related characteristic value and fault information.

[0076] Finally, it should be noted that: the above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A line spectrum based photovoltaic DC system fault detection method, characterized in that, The method comprises the following steps: S1: initializing the arc fault detection device, and collecting fault arc signal data through a signal acquisition module; S2: performing fast Fourier transform and short-time Fourier transform on the fault arc signal data respectively, generating a single-side spectrum and a time-frequency spectrum Y1 of the signal respectively, finding a local peak frequency in each time frame, and forming a peak frequency matrix Y4 containing only the peak frequency; S3: counting the number of occurrences of each frequency point in different time frames as a frequency stability, screening a starting frequency point as a stable frequency point, and performing dynamic programming to generate a path by taking the frequency offset between time frames and the offset of a path point from the starting frequency as a total offset constraint, the path being a path of changes in the main frequency over time frames; Step S3 is specifically: for the spike frequency matrix Y4 generated in S2, all time frames are traversed, the number of non-zero values of each frequency point in all time frames is counted as the frequency stability N(f) of the frequency point, and the preliminary screening process is as follows: Wherein, Td is the time proportion threshold, N t is the total number of time frames, and S'(f) is the initial frequency point set. The frequency point of S'(f) is taken as the starting frequency point, and a line spectrum stationary path is searched in the peak frequency matrix Y4 by using dynamic programming, and the total deviation constraint Cost(f t ) of the dynamic programming is as follows: Cost(f t ) = a · |f t -f t-1 | + β · |f t -f start |, where f start is the start frequency point of the initial frequency point set, f t is the current path frequency point, |f t -f t-1 | is the frequency offset cost between time frames, |f t -f start | is the global offset cost of the current path frequency point and the start frequency point, and a and β are cost weight parameters. The cumulative minimum cost value C(f, t) of each frequency point is calculated frame by frame, and the recursive formula is as follows: C(f, t) = min(C(f t-1 ,t-1) + Cost(f t )) Wherein, C(f, t) is the cumulative minimum cost value of the current time frame t, C(f, t-1) is the cumulative minimum cost value of the previous time frame t-1, and Cost(ft) is the total deviation constraint of the current time frame. The frequency point with the minimum cost value of the last time frame is obtained t-1 The dominant frequency path is obtained.​ S4: performing line spectrum recognition by taking the path offset length and the node energy entropy as threshold indicators, determining the stable line spectrum frequency point in the signal, determining the distribution interval of the line spectrum frequency based on the line spectrum frequency point, performing sliding window analysis on the single-side spectrum X(f) obtained in the step S2, and extracting the minimum value point of the envelope on the spectrum, and performing interval merging based on the extracted line spectrum frequency information to obtain the frequency band division boundary; S5: performing empirical wavelet transform on the signal according to the divided frequency band to obtain a plurality of modal components, and reconstructing the modal components without the line spectrum frequency to obtain a fault characteristic signal; S6: Extracting the reconstructed signal x r The statistical features of (t) are analyzed, and whether an arc fault occurs is judged by setting a threshold of the feature quantity.

2. The method of claim 1, wherein, In the local peak frequency detection, the time-frequency spectrum of the signal is analyzed, and in each time frame, the local peak frequency with an amplitude higher than that of the adjacent frequency points is identified by comparing the amplitude changes in the frequency dimension.

3. The method of claim 1, wherein, The implementation steps of the empirical wavelet transform (EWT) are as follows: The minimum value of the upper envelope of the single-side spectrum is taken as an initial boundary, adjacent frequency band intervals containing line spectrum frequency points are merged to obtain a frequency band division boundary, For each frequency band interval, a bandpass filter W is constructed i (f) modal decomposition of the acquired signal: wherein and denote the Fourier transform and inverse transform of a signal, respectively, IMF i (t) is the decomposed modal component.

4. The method of claim 1, wherein, The statistical characteristics include energy changes, waveform characteristics and transient characteristics, and are used to reflect the energy level and fluctuations, waveform sharpness and transient pulse characteristics of the fault signal.

5. A line-spectrum-rejection-based photovoltaic DC system fault detection apparatus for implementing the line-spectrum-rejection-based photovoltaic DC system fault detection method according to any one of claims 1-4, characterized in that, The device comprises: a signal acquisition module for acquiring a voltage signal from a photovoltaic direct-current system; a signal processing unit for performing line spectrum interference suppression, frequency band decomposition and signal reconstruction; a fault judgment module for extracting statistical characteristic values of the reconstructed signal and judging whether an arc fault exists; a display and alarm device for displaying the system operation state and fault information in real time and triggering an alarm when a fault is detected.

Images