System for analyzing parallel operation impact data of generators

Abstract

The invention relates to the technical field of electrical variable measurement, in particular to an analysis system for generator parallel operation impact data, which comprises a transient signal acquisition module for synchronously monitoring an ampere value of an impact current and a volt value of an impact voltage during generator parallel operation, obtaining an instantaneous reading of a transient electrical signal, and acquiring a transient signal; and the reading is converted into a digital sequence, and a multi-channel transient electrical signal value is recorded to form an original electrical measurement data set. According to the method, the ampere value of the impulse current and the volt value of the impulse voltage are synchronously monitored in the generator parallel operation period, the instantaneous reading of the transient electric signal is obtained and converted into the digital sequence, a high-precision data foundation is laid for subsequent analysis, and the analysis time period of the impulse event is further set based on the original electrical measurement data set; and carrying out synchronous alignment and effective data segment interception on the current and voltage numerical value sequences according to timestamps, and constructing a structured time domain impact data sequence group.

Description

Technical Field

[0001] The present invention relates to the technical field of measuring electric variables, and in particular to an analysis system for impulse data of parallel operation of generators. Background Art

[0002] The analysis system for generator parallel operation impact data is a dedicated system designed to conduct in-depth analysis and evaluation of electrical variable data such as impact current and impact voltage generated during the specific transient process of generator connection to the power grid (i.e. parallel operation).

[0003] Existing technologies focus solely on whether the surge current amplitude exceeds the limit, potentially overlooking subtle features such as severe current waveform distortion or high-frequency oscillations that could indicate poor controller response or system resonance risk. This can lead to misjudgments of paralleling quality. Furthermore, due to the lack of systematic quantification of the signal's non-Gaussian characteristics, time-varying spectral energy distribution, and pre- and post-surge spectrum differences, existing technologies struggle to effectively identify specific abnormal surge patterns caused by improper control parameters, errors in synchronization condition judgment, or background harmonic interference from the power grid. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an analysis system for impulse data of parallel operation of generators.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A system for analyzing impulse data of parallel operation of generators comprises:

[0006] The transient signal acquisition module synchronously monitors the ampere value of the surge current and the volt value of the surge voltage during the parallel operation of the generators, obtains the instantaneous reading of the transient electrical signal, converts the reading into a digital sequence, and records the multi-channel transient electrical signal values ​​to form the original electrical measurement data set;

[0007] a data sequence construction module that, based on the original electrical measurement data set, sets the start and end points of the impulse event to select an analysis period, performs synchronization and alignment processing on the impulse current value sequence and the impulse voltage value sequence according to the timestamp, and intercepts valid data segments according to the selected analysis period to generate a time-domain impulse current sequence and a time-domain impulse voltage sequence to construct an impulse data sequence group;

[0008] a cumulant feature calculation module, which performs centering processing on each sequence data point based on the time-domain impulse current sequence and the time-domain impulse voltage sequence in the impulse data sequence group, calculates the ratio of the third-order central moment of the sequence to the cube of the standard deviation as a skewness value, calculates the ratio of the fourth-order central moment of the sequence to the fourth power of the standard deviation minus three as a kurtosis value, calculates the third-order cumulant statistics and the fourth-order cumulant statistics respectively, and obtains a high-order cumulant parameter set;

[0009] The energy entropy feature extraction module applies a window function to the time-domain impulse current sequence and the time-domain impulse voltage sequence based on the impulse data sequence group, performs Fourier transform, obtains the spectrum of each time window, calculates the energy distribution of each spectrum in different frequency bands, and then calculates the Shannon entropy value of the energy distribution probability within each time window to obtain the instantaneous energy entropy sequence group.

[0010] Preferably, the system further comprises:

[0011] a spectrum difference quantification module, based on the impulse data sequence group, extracting the current and voltage data segments of steady-state operation before the parallel operation and the impulse current and voltage data segments during the parallel operation, performing Fourier transform on the data segments to obtain their respective spectra, comparing the spectrum energy amplitudes of corresponding frequency points one by one, calculating the difference or ratio, and quantifying the spectrum energy deviation;

[0012] The parallel quality assessment module integrates the current kurtosis and voltage skewness of the high-order cumulant parameter set, the entropy transition amplitude and entropy mean of the instantaneous energy entropy sequence group, and the spectrum energy deviation to form a multidimensional feature vector. The module then compares each component of the vector with a preset reference threshold range to determine the electrical performance status of the parallel impact.

[0013] Preferably, the transient signal acquisition module includes:

[0014] The signal synchronization monitoring submodule collects the ampere value waveform of the surge current and the volt value waveform of the surge voltage on the same time basis during the parallel operation of the generators, captures the drastic changes at the moment of parallel operation, obtains the instantaneous reading of the transient electrical signal, and obtains the synchronous monitoring reading;

[0015] A reading sequence conversion submodule, based on the synchronous monitoring readings, quantizes and encodes the analog readings of the surge current in amperes and the surge voltage in volts according to a predetermined sampling rate and accuracy, and outputs corresponding discrete-time digital sequences to form a digital signal sequence set;

[0016] The raw data compilation submodule associates the impulse current digital sequence and impulse voltage digital sequence of each channel with their respective channel identification and timestamp information based on the digital signal sequence set, stores them in the data storage structure, and records the corresponding digital sequence values ​​of the multi-channel transient electrical signals to form the original electrical measurement data set.

[0017] Preferably, the data sequence construction module includes:

[0018] The time period selection and definition submodule retrieves, based on the original electrical measurement data set, the point in the impulse current or impulse voltage numerical sequence where the amplitude first exceeds a preset starting threshold as a starting point timestamp, and determines the ending point timestamp at a point after the impulse stabilizes or a fixed duration, sets the timestamp information of the start and end points of the impulse event, selects an analysis period, and obtains analysis period parameters;

[0019] a data alignment and truncation submodule, which checks the timestamp sequence of each numerical sequence based on the original electrical measurement data set and the analysis period parameter, aligns the data points of different channels on the time axis through interpolation or resampling, and truncates valid data segments according to the start and end timestamps of the selected analysis period to obtain a truncation data segment group;

[0020] The sequence construction and generation submodule extracts the aligned effective data segment values ​​of the impulse current and the effective data segment values ​​of the impulse voltage based on the truncated data segment group, arranges them in chronological order, and forms independent time-domain impulse current sequence arrays and time-domain impulse voltage sequence arrays. The arrays are then organized into time-domain impulse current sequences and time-domain impulse voltage sequences to construct an impulse data sequence group.

[0021] Preferably, the cumulative feature calculation module includes:

[0022] A data centralization submodule, based on the time-domain impulse current sequence and the time-domain impulse voltage sequence in the impulse data sequence group, first calculates the average value of all data points in each independent sequence, then traverses the sequence, subtracts the calculated average value from the original value of each data point, and then centralizes each data point in each sequence by subtracting the mean of the sequence to generate a centralized data sequence;

[0023] The cumulant parameter calculation submodule, based on the centralized data sequence, takes the cube of the deviation of each data point in each sequence to obtain the third-order central moment, then divides it by the cube of the standard deviation of the sequence to obtain the skewness value, and calculates the ratio of the fourth-order central moment of each centralized sequence to the fourth power of the standard deviation and subtracts three from it as the kurtosis value to obtain the sequence-by-sequence cumulant value;

[0024] The characteristic parameter aggregation submodule extracts the impulse current skewness, impulse current kurtosis, impulse voltage skewness, and impulse voltage kurtosis from the skewness and kurtosis values ​​of the current and voltage sequences as core indicators based on the sequence-by-sequence cumulative values. The skewness and kurtosis values ​​of all sequences are aggregated to form sets of third-order cumulative statistics and fourth-order cumulative statistics, respectively, to obtain a set of high-order cumulative parameters.

[0025] Preferably, the energy entropy feature extraction module includes:

[0026] A time-spectrum conversion submodule, based on the impulse data sequence group, applies an overlapping segmentation method to each time domain sequence, multiplies each segment by a Hamming window, and then performs a discrete Fourier transform calculation on each segmented data to obtain the spectrum of each time window to form a time-spectrum data frame;

[0027] The window energy distribution calculation submodule, based on the time-spectrum data frame, accumulates and sums the squares of the amplitudes of the frequency points within each time window within a plurality of pre-determined frequency intervals of interest, obtains the energy value of the time window within the frequency interval, calculates the numerical distribution of the energy of the spectrum of each time window within different preset frequency segments, and forms a window-by-window energy distribution set;

[0028] The sequence Shannon entropy acquisition submodule, based on the window-by-window energy distribution set, normalizes the energy of each frequency band in each window to obtain the probability, multiplies the probability by the logarithm and then sums the negative value to obtain the Shannon entropy value, and obtains the instantaneous energy entropy sequence group.

[0029] Preferably, the spectrum difference quantization module includes:

[0030] An event segment extraction submodule, based on the shock data sequence group, selects steady-state current and voltage waveforms of several cycles before the shock occurs, and current and voltage waveforms including the complete transient process after the shock occurs, and uses these as the current and voltage data segments for extracting the steady-state operation before the parallel operation and the shock current data segments and shock voltage data segments during the parallel operation, respectively, to obtain the preceding and following characteristic data segments;

[0031] A spectrum energy calculation submodule, based on the preceding and following characteristic data segments, independently applies discrete Fourier transforms to the current and voltage sequences in the steady-state operation data segment and the impulse period data segment to calculate the amplitude spectrum of each sequence at different frequencies. Fourier transforms are then performed on each extracted data segment to obtain the respective current and voltage spectra, forming a two-state spectrum energy diagram.

[0032] The energy deviation quantification submodule calculates the difference or ratio between the impact state and steady-state energies of each corresponding frequency point based on the two-state spectrum energy diagram, and calculates the total energy change or mean increase in the interharmonic frequency band. This submodule compares the spectrum energy amplitudes of the corresponding frequency points before and after the parallel operation one by one to quantify the spectrum energy deviation.

[0033] Preferably, the parallel quality assessment module includes:

[0034] The multidimensional feature fusion submodule fuses the current kurtosis and voltage skewness values ​​of the high-order cumulant parameter set, extracts the maximum entropy transition amplitude at the impact starting point and the average entropy value of the impact segment from the instantaneous energy entropy sequence group, and the quantized value of the spectrum energy deviation, arranges them in a predetermined order, and combines them into a multidimensional feature vector to establish the impact feature vector;

[0035] A benchmark comparison and discrimination submodule, based on the impact feature vector, compares each component value in the vector with a preset threshold value to see if it exceeds the limit, and generates a comparison and discrimination code;

[0036] The performance status determination submodule determines the electrical performance level of the current parallel impact based on the comparison and discrimination code, and discriminates the electrical performance status of the parallel impact.

[0037] Compared with the prior art, the advantages and positive effects of the present invention are:

[0038] In this invention, by synchronously monitoring the surge current (amperes) and surge voltage (volts) during parallel operation of the generators, instantaneous readings of transient electrical signals are obtained and converted into digital sequences, laying a high-precision data foundation for subsequent analysis. Furthermore, based on the original electrical measurement data set, the analysis period for the surge event is set. The current and voltage numerical sequences are synchronously aligned and valid data segments are truncated based on timestamps to construct a structured set of time-domain surge data sequences, ensuring temporal consistency and data integrity for the analysis. After data point centering, these sequences are calculated as skewness (the ratio of the third-order central moment to the cube of the standard deviation) and as kurtosis (the ratio of the fourth-order central moment to the fourth power of the standard deviation minus three). This yields a set of high-order cumulants, which reveal the non-Gaussian characteristics of the surge signal and the degree of waveform asymmetry and sharpness, addressing the shortcomings of traditional mean-variance analysis. Furthermore, a window function is applied to the time-domain sequence and Fourier transform is performed to obtain the spectrum of each time window. The energy distribution of each spectrum in different frequency bands is calculated, and the Shannon entropy value is calculated for the energy distribution probability of each time window. An instantaneous energy entropy sequence group is formed to dynamically characterize the complexity and uncertainty of energy changes in the frequency domain during the impact process. Combined with a comparative analysis of the spectral energy amplitudes of the current and voltage data segments before and after the parallel operation, the spectrum energy deviation is quantified, especially the energy changes in the interharmonic frequency band. Finally, a multidimensional feature vector is constructed by integrating high-order cumulants, key features of instantaneous energy entropy, and spectral energy deviation. Compared with a baseline threshold or historical excellent data, this allows for a refined and multidimensional quantitative assessment of the electrical performance status of the parallel impact, enhancing the depth of the impact transient process analysis and the accuracy of the assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0040] To further clarify the objectives, technical solutions, and advantages of the present invention, the present invention is further described below in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention.

[0041] See also Figure 1 The present invention provides a technical solution: a system for analyzing impact data of parallel operation of generators, comprising:

[0042] The transient signal acquisition module simultaneously monitors the ampere value of the surge current and the volt value of the surge voltage during parallel operation of the generators, obtains instantaneous readings of the transient electrical signals, converts the readings into a digital sequence, and records the multi-channel transient electrical signal values ​​to form an original electrical measurement data set;

[0043] The data sequence construction module, based on the original electrical measurement data set, sets the start and end points of the impact event to select the analysis period, synchronizes the impact current value series and the impact voltage value series according to the timestamp, and intercepts the valid data segments according to the selected analysis period to generate the time domain impact current series and time domain impact voltage series to construct the impact data series group;

[0044] The cumulant feature calculation module performs centering on each sequence data point based on the time-domain impulse current sequence and the time-domain impulse voltage sequence within the impulse data sequence group. The skewness value is calculated as the ratio of the third-order central moment of the sequence to the cube of the standard deviation. The kurtosis value is calculated as the ratio of the fourth-order central moment of the sequence to the fourth power of the standard deviation minus three. The third-order and fourth-order cumulant statistics are calculated separately to obtain a set of high-order cumulant parameters.

[0045] The energy entropy feature extraction module applies a window function to the time-domain impulse current sequence and the time-domain impulse voltage sequence based on the impulse data sequence group, performs a Fourier transform, and obtains the spectrum of each time window. The energy distribution of each spectrum in different frequency bands is calculated, and the Shannon entropy value of the energy distribution probability within each time window is calculated to obtain the instantaneous energy entropy sequence group.

[0046] The spectrum difference quantification module extracts the current and voltage data segments of steady-state operation before parallel operation and the impulse current and voltage data segments during parallel operation based on the impulse data sequence group. It performs Fourier transform on the data segments to obtain their respective spectra, compares the spectrum energy amplitudes of corresponding frequency points one by one, calculates the difference or ratio, and quantifies the spectrum energy deviation.

[0047] The parallel quality assessment module integrates the current kurtosis and voltage skewness of the high-order cumulant parameter set, the entropy transition amplitude and entropy mean of the instantaneous energy entropy sequence group, and the spectrum energy deviation to form a multidimensional feature vector. It then compares each component of the vector with a preset baseline threshold range to determine the electrical performance status of the parallel impact.

[0048] The transient signal acquisition module includes:

[0049] The signal synchronization monitoring submodule collects the ampere value waveform of the surge current and the volt value waveform of the surge voltage on the same time basis during the parallel operation of the generators, captures the drastic changes at the moment of parallel operation, obtains the instantaneous reading of the transient electrical signal, and obtains the synchronous monitoring reading;

[0050] The reading sequence conversion submodule quantizes and encodes the analog readings of the surge current in amperes and the surge voltage in volts based on the synchronous monitoring readings according to a predetermined sampling rate and accuracy, and outputs the corresponding discrete-time digital sequence to form a digital signal sequence set.

[0051] The raw data compilation submodule associates the impulse current digital sequence and impulse voltage digital sequence of each channel with their respective channel identification and timestamp information based on the digital signal sequence set, stores them in the data storage structure, and records the corresponding digital sequence values ​​of the multi-channel transient electrical signals to form the original electrical measurement data set.

[0052] Specifically, after the generator is commanded to connect to the grid in parallel, a high-speed data acquisition process is immediately initiated. For the surge current (amperes) and surge voltage (volts), a multi-channel analog input device with synchronized triggering is used to ensure that all channels are sampled using a single high-frequency clock reference, such as 1 MHz, to ensure accurate and consistent timestamps. During the acquisition process, a high-precision shunt or current transformer is configured for the current channel, with a response bandwidth covering at least 0 Hz to 100 kHz to accurately capture DC components and high-frequency transients. The voltage channel is connected via a precision voltage divider, similarly ensuring sufficient measurement bandwidth and accuracy. For example, the current measurement range is ±5 times the rated current, and the voltage measurement range is ±2 times the rated voltage. Within a preset period around the expected closing moment of the paralleling switch, for example, from 100 milliseconds before the closing command is issued to 500 milliseconds after closing, the instantaneous analog waveforms of current and voltage are continuously recorded at an extremely high sampling frequency, such as 50,000 samples per second (50 kHz), with particular attention paid to the instant of paralleling. That is, within the microsecond to millisecond timescale from the expected contact of the switch contacts to the beginning of a significant increase in current, all possible non-periodic and violent fluctuations in current and voltage, peak overshoots, and high-frequency oscillations are captured in real time, forming a continuous time series data stream. This immediate, unprocessed sequence of raw measurements constitutes the synchronized monitoring readings required for subsequent analysis.

[0053] Based on the synchronous monitoring readings obtained in the previous step, which include a continuous time series of the impulse current ampere value analog waveform and the impulse voltage volt value analog waveform, these analog signals are then digitally converted. This process is performed separately for the analog readings of each channel. The impulse current ampere value analog reading is processed using a 16-bit resolution analog-to-digital converter (ADC). Its input range is based on the aforementioned monitoring configuration, for example, corresponding to ±2000 amperes. This means that the minimum resolvable current change is approximately

[0054] For the analog reading of the impulse voltage, a 16-bit resolution analog-to-digital converter is also used. Its input range corresponds to ±800 volts, for example, and the minimum resolvable voltage change is approximately Quantization and encoding are performed strictly according to a pre-set sampling rate. This sampling rate is set based on an estimate of the frequency content of transient phenomena, for example, 50,000 samples per second (50kHz) to ensure distortion-free reconstruction of signal components up to 25kHz. Each sample point is assigned a precise timestamp during encoding, derived from the aforementioned synchronous high-frequency clock reference. After quantization and encoding, the original continuous analog waveform is converted into a series of discrete digital values ​​that accurately reflect the current or voltage at the specific sampling moment. Ultimately, a separate discrete-time digital sequence is generated for each monitoring channel (such as three-phase current or three-phase voltage). All these sequences are combined to form a digital signal sequence set.

[0055] Based on the digital signal sequence set generated in the previous process, that is, the discrete-time digital sequences corresponding to each channel, such as phase A current, phase B current, phase C current, phase A voltage, phase B voltage, and phase C voltage, are systematically compiled and organized. Specifically, each digital sequence is clearly associated with its physical source information, namely, the channel identifier, such as "current channel IA" and "voltage channel UA." Furthermore, because all sequences are generated under synchronous acquisition, each data point corresponds to a precise timestamp based on a common time base. This timestamp information is also strictly bound to each digital value in the respective sequence. Subsequently, these impulse current digital sequences and impulse voltage digital sequences with channel identifiers and detailed timestamp information are systematically organized and stored in a structured data storage area. This data storage structure can be a multidimensional array in memory or a specially designed time-series database table to ensure efficient data access and management. During this process, the precise digital sequence value of the transient electrical signal of each channel at each sampling moment is recorded in detail. For example, if there are six channels, each sampled at 50kHz for 1 second, 300,000 numerical points with complete identification and time information will be recorded. Through such compilation and storage, a comprehensive raw electrical measurement data set containing all relevant electrical measurement parameters will be constructed.

[0056] The data series building blocks include:

[0057] The time period selection and definition submodule, based on the original electrical measurement data set, retrieves the point in the impulse current or impulse voltage numerical sequence where the amplitude first exceeds the preset starting threshold as the starting point timestamp, and determines the ending point timestamp at the point after the impulse stabilizes or a fixed duration. The timestamp information of the start and end points of the impulse event is set, the analysis period is selected, and the analysis period parameters are obtained.

[0058] The data alignment and interception submodule checks the timestamp sequence of each numerical sequence based on the original electrical measurement data set and the analysis period parameters, aligns the data points of different channels on the time axis through interpolation or resampling, and intercepts the valid data segments according to the start and end timestamps of the selected analysis period to obtain the intercepted data segment group;

[0059] The sequence construction and generation submodule extracts the aligned effective data segment values ​​of the impulse current and the impulse voltage based on the intercepted data segment group, arranges them in chronological order, and forms independent time-domain impulse current sequence arrays and time-domain impulse voltage sequence arrays. These are then organized into time-domain impulse current sequences and time-domain impulse voltage sequences to construct an impulse data sequence group.

[0060] Specifically, based on the previously constructed original electrical measurement data set, which contains the impulse current and impulse voltage digital sequences with time stamps for all channels, the analysis period is precisely defined. First, the impulse current numerical sequence, such as the A-phase current sequence, or the impulse voltage numerical sequence, such as the A-phase voltage sequence, is retrieved in the data set to determine the starting point of the impulse event. The criterion for determining the starting point is that the amplitude in the sequence continuously exceeds a "preset starting threshold" for the first time. This threshold is set based on historical data and experience. For example, for a generator with a rated current of 1000 amperes, the steady-state operating current before paralleling may be 50 amperes. The "preset starting threshold" can be set to 5 times the steady-state current, that is, 250 amperes. When it is detected that the current value jumps rapidly from less than 250 amperes to 250 amperes or above, and maintains this state for at least 3 consecutive sampling points (corresponding to 60 microseconds at a 50kHz sampling rate), the timestamp of the first sampling point exceeding the threshold is The starting point timestamp of the impact event is recorded, and then the ending point timestamp of the impact event is determined. The ending point can be determined in one of two ways: the first is to monitor the impact current or voltage and gradually decay after reaching the peak and tend to a new stable state. When the fluctuation rate of the signal, such as the standard deviation within the 10 millisecond (500 sampling points) window, is less than 1% of the peak impact current or 5% of the new steady-state current for 20 milliseconds (1000 sampling points), the time point is regarded as the impact stable point, and its timestamp is the ending point timestamp; the second way is to use a "fixed time length", which is judged to be sufficient to cover the entire parallel impact transient process based on experience, for example, it is set to 300 milliseconds from the starting point timestamp. After selecting one of the methods to determine the end, the starting point and ending point timestamp information of the impact event are set. These two timestamps together define the analysis period required for subsequent analysis, and finally the analysis period parameters are obtained, such as the starting time t start and end time t end .

[0061] Using the original electrical measurement dataset obtained in the previous step and the determined analysis period parameters, namely the start timestamp and end timestamp of the impact event, perform data alignment and truncation operations. First, check the timestamp sequences of each impact current numerical sequence and impact voltage numerical sequence in the original electrical measurement dataset. Although designed for synchronous acquisition, minor clock drifts or startup delays of different acquisition boards may result in incomplete alignment of the timestamps of each channel at the microsecond level. To ensure the accuracy of subsequent calculations, these timestamps need to be precisely calibrated. By selecting the timestamp sequence of a reference channel (such as the phase A voltage channel) as the benchmark, perform time-axis alignment processing on the numerical sequences of all other channels. This processing uses the linear interpolation method, that is, for the case where there is no direct sampling point on a non-reference channel at a certain reference time point t ref , then calculate according to the two nearest sampling points (t1, value1) and (t2, value2) before and after it. The interpolation formula is ref where t1 < t ref < t2. In this way, generate data values at the common time point for all channels, ensuring that all data points are precisely aligned on the time axis. After completion of the alignment, according to the previously determined analysis period parameters, that is, the start timestamp and end timestamp, precisely truncate the data segments that fall within this period from the aligned numerical sequences of each channel. Any data earlier than the start timestamp or later than the end timestamp will be discarded. Thus, obtain a truncated data segment group that contains the valid data of all relevant channels within the selected analysis period.

[0062] Based on the truncated data segment group obtained from the previous step processing, this data group contains the valid data segments of the impact current and impact voltage that are aligned on the common time axis and truncated within the predetermined analysis period, perform the final construction of the time-domain sequence. The specific operation is to extract the data of each channel one by one from the truncated data segment group. For example, extract the valid data segment values of the phase A impact current, the valid data segment values of the phase B impact current, the valid data segment values of the phase C impact current, as well as the valid data segment values of the phase A impact voltage, the valid data segment values of the phase B impact voltage, and the valid data segment values of the phase C impact voltage. For each extracted valid data segment value, arrange them strictly in the chronological order of their original sampling to ensure that the time series characteristics between data points are retained. In this way, form an independent, time-ordered numerical array for each current channel and voltage channel. For example, the numerical sequence of the phase A current may be represented as [I A (t1), I A (t2),..., I A (t N )], where t1 is the start point of the analysis period or the first sampling point after it, t​N It is the end point of the analysis period or the last sampling point before it. These independent one-dimensional time series arrays are sorted and their corresponding physical quantities (impact current or impulse voltage) and phases (such as phase A, phase B, and phase C) are clearly identified. They are eventually organized into structured time-domain impulse current sequences and time-domain impulse voltage sequences. All these sequences together constitute the impulse data sequence group for subsequent feature extraction and analysis.

[0063] The cumulative feature calculation module includes:

[0064] The data centralization submodule, based on the time-domain impulse current sequence and time-domain impulse voltage sequence within the impulse data sequence group, first calculates the average value of all data points in each independent sequence, then traverses the sequence, subtracts the calculated average value from the original value of each data point, and then subtracts the mean value of the sequence from each data point in each sequence to perform centralization processing to generate a centralized data sequence;

[0065] The cumulant parameter calculation submodule, based on the centralized data sequence, takes the cube of the deviation of each data point in each sequence to obtain the third-order central moment, then divides it by the cube of the standard deviation of the sequence to obtain the skewness value. It also calculates the ratio of the fourth-order central moment of each centralized sequence to the fourth power of the standard deviation and subtracts three as the kurtosis value to obtain the sequence-by-sequence cumulant value.

[0066] The characteristic parameter aggregation submodule extracts the impulse current skewness, impulse current kurtosis, impulse voltage skewness, and impulse voltage kurtosis as core indicators from the skewness and kurtosis values ​​of the current and voltage sequences based on the sequence-by-sequence cumulative values. The skewness and kurtosis values ​​of all sequences are aggregated to form sets of third-order and fourth-order cumulative statistics, respectively, to obtain a set of high-order cumulative parameters.

[0067] Specifically, based on the previously constructed impulse data series group, which includes multiple independent digital time series such as the A-phase time-domain impulse current series, the B-phase time-domain impulse current series, the C-phase time-domain impulse current series, and the A-phase time-domain impulse voltage series, the B-phase time-domain impulse voltage series, and the C-phase time-domain impulse voltage series, data centering is performed on each independent series. The specific operation is as follows: first, for a selected independent series, such as the A-phase time-domain impulse current series, all N data points (x1, x2, ..., x N ), calculate the arithmetic mean of these data points, that is, the mean After the calculation is completed, the A phase time domain impulse current sequence is traversed again. For each original data point x in the sequence i , subtract the previously calculated mean of the sequence from its original value Get a new value This new value x′i is the centered data point. This mean subtraction operation is performed on all data points in the phase A time-domain impulse current sequence to obtain a new sequence with a mean of zero, namely the phase A centralized impulse current sequence. The same processing flow is independently applied to the phase B time-domain impulse current sequence, the phase C time-domain impulse current sequence, and the time-domain impulse voltage sequences of all phases to ensure that each original sequence is converted into its corresponding centralized version. All these processed sequences together generate the centralized data sequence.

[0068] Based on the centralized data sequence generated in the previous process, that is, the time series with zero mean of each current channel and voltage channel, the cumulative parameter is calculated independently for each centralized sequence. Taking the centralized impulse current sequence of phase A as an example, the sequence contains N data points (x′1, x′2,…, x′ N ), first calculate the standard deviation σ of the sequence, and then calculate the third-order central moment m3 of the sequence, that is, the average value of the cube of the deviation of each data point in the sequence (because it has been centered, that is, the data point itself), specifically Then, the calculated third-order central moment m3 is divided by the cube of the standard deviation σ (i.e. σ 3 ), the result is the skewness value of the A phase centralized impulse current sequence, which reflects the asymmetry of the sequence data distribution. Then, the fourth-order central moment m4 of the sequence is calculated, that is, the average value of the fourth power of the deviation of each data point in the sequence, specifically: Divide the calculated fourth-order central moment m4 by the fourth power of the standard deviation σ (i.e. σ 4 ), and then subtract the constant 3 from the ratio result. The resulting value is the kurtosis value (also known as excess kurtosis) of the phase A centralized impulse current sequence. The kurtosis value describes the degree of sharpness or tail thickness of the sequence data distribution. For each sequence in the centralized data sequence (for example, the phase B centralized current sequence, the phase C centralized current sequence, and the phase centralized voltage sequence), the above complete process of calculating the third-order central moment, standard deviation, skewness value, and the fourth-order central moment and kurtosis value is repeated. Thus, a set of skewness and kurtosis values ​​are obtained for each sequence, which together constitute the sequence-by-sequence cumulative value.

[0069] Based on the sequence-by-sequence cumulative value calculated in the previous step, which includes the corresponding skewness value and kurtosis value of each centralized current sequence (such as the centralized impulse current sequence of phase A, phase B, and phase C) and each centralized voltage sequence (such as the centralized impulse voltage sequence of phase A, phase B, and phase C), the core feature parameters are extracted and aggregated. Specifically, the skewness value of the phase A impulse current sequence is extracted from the cumulative value, recorded as "phase A impulse current skewness", and its kurtosis value is extracted, recorded as "phase A impulse current kurtosis". Similarly, the "phase B impulse current skewness" and "phase B impulse current kurtosis" are extracted from the phase B impulse current sequence, and the "phase C impulse current skewness" and "phase C impulse current kurtosis" are extracted from the phase C impulse current sequence. The same extraction operation is performed on the voltage sequence to obtain, for example, "phase A impulse voltage skewness", "phase A impulse voltage kurtosis", "phase B impulse voltage skewness", "phase B impulse voltage kurtosis", "C-phase impulse voltage skewness" and "C-phase impulse voltage kurtosis" are collected together to form a set of all the extracted skewness values, namely, "A-phase impulse current skewness", "B-phase impulse current skewness", "C-phase impulse current skewness", "A-phase impulse voltage skewness", "B-phase impulse voltage skewness", and "C-phase impulse voltage skewness". This set is used as a third-order cumulant statistical value set. At the same time, all the extracted kurtosis values, namely, "A-phase impulse current kurtosis", "B-phase impulse current kurtosis", "C-phase impulse current kurtosis", "A-phase impulse voltage kurtosis", "B-phase impulse voltage kurtosis", and "C-phase impulse voltage kurtosis" are collected together to form another set. This set is used as a fourth-order cumulant statistical value set. These two sets together constitute the required high-order cumulant parameter set.

[0070] The energy entropy feature extraction module includes:

[0071] The time-spectrum conversion submodule uses an overlapping segmentation method on each time domain sequence based on the impulse data sequence group, multiplies each segment by a Hamming window, and then performs a discrete Fourier transform on each segmented data to obtain the spectrum of each time window to form a time-spectrum data frame;

[0072] The window energy distribution calculation submodule, based on the time-spectrum data frame, accumulates and sums the squared amplitude values ​​of each frequency point within a plurality of pre-determined frequency intervals of interest for the amplitude-spectrum data within each time window to obtain the energy value of the time window within that frequency interval. It also calculates the numerical distribution of the energy of each time window spectrum within different preset frequency segments to form a window-by-window energy distribution set.

[0073] The sequence Shannon entropy acquisition submodule, based on the window-by-window energy distribution set, normalizes the energy of each frequency band in each window to obtain the probability, multiplies the probability by the logarithm and then sums and takes the negative value to obtain the Shannon entropy value, thereby obtaining the instantaneous energy entropy sequence group.

[0074] Specifically, based on the impulse data sequence group, in particular, each time-domain impulse current sequence and time-domain impulse voltage sequence contained therein, such as the time-domain impulse current sequence of phase A, time-frequency spectrum conversion is independently performed on each such time-domain sequence. First, the selected time-domain sequence is processed using an overlapping segmentation method, and the length of the data segment is set. For example, 1024 sampling points are selected as a data segment. If the sampling rate is 50 kHz, the time length represented by each data segment is 1024 / 50,000 seconds ≈ 20.48 milliseconds. The overlap ratio between data segments is also set, for example, 50%. This means that the second data segment will begin at the midpoint of the first data segment, overlapping by 512 samples. Each segment thus divided is weighted using a window function. A Hamming window is selected, and each of the 1024 samples within the segment is multiplied by the coefficient at the corresponding position of the Hamming window. The purpose of windowing is to reduce spectral leakage caused by the subsequent Fourier transform. A discrete Fourier transform (DFT) is performed on each data segment after the Hamming window weighting process. The fast Fourier transform (FFT) algorithm is typically used to improve computational efficiency. This yields the complex spectrum of the data segment, which contains the amplitude and phase information of each frequency component. Since the time domain sequence is divided into multiple (possibly overlapping) time windows (i.e., data segments), the above processing is performed on each time window, resulting in a series of spectra corresponding to different time periods. These chronologically arranged spectrum sequences together constitute the time-spectral data frame of the original time domain sequence.

[0075] Based on the time-spectrum data frame generated in the previous process, which contains, for example, the spectrum information of the A-phase impulse current sequence in each time window, the energy distribution of the amplitude-spectrum data in each time window is calculated. First, several "pre-divided frequency intervals of interest" are defined. The setting of these intervals is based on experience and understanding of the characteristic frequencies that may appear when the generators are paralleled. For example, for a 50 Hz power system, the frequency intervals of interest can be divided into: Interval 1 (fundamental frequency band): 45 Hz to 55 Hz; Interval 2 (low-order harmonic band): 90 Hz to 160 Hz, covering the 2nd and 3rd harmonics; Interval 3 (medium and high frequency disturbance band): 500 Hz to 2000 Hz; Interval 4 (high-frequency disturbance band): 2000 Hz to 10000 Hz. The boundaries and number of specific intervals can be adjusted according to actual analysis requirements. For example, a total of K = 4 frequency intervals of interest are divided. For a specific time window in the time-spectral data frame, its amplitude spectrum data is extracted. Then, in each preset frequency interval of interest, for example, in interval 1 (45 Hz to 55 Hz), the spectrum amplitudes |X(f)| of all frequency points falling within this frequency range are squared to obtain the approximate power spectrum density of each frequency point. These squared values ​​are then accumulated and summed within the frequency interval. The formula is: where f low和 f high are the lower and upper frequencies of the current frequency interval of interest respectively. The calculation result is the total energy value of the time window in this specific frequency interval. This energy accumulation operation is performed on all K preset frequency intervals of interest in the time window to obtain the numerical distribution of the spectrum energy in these different frequency segments in the time window. This process is repeated for each time window in the time-spectrum data frame, and finally a window-by-window energy distribution set is formed that records the energy distribution of each time window in each frequency segment of interest.

[0076] Based on the window-by-window energy distribution set formed in the previous step, the energy value of each time window in the K preset frequency intervals of interest is recorded in the set. For example, for the jth time window, its energy distribution is (E j1 ,E j2 ,…,E jK ), where E jk is the energy of the jth time window in the kth frequency interval. Next, the energy distribution data of each time window is processed to calculate its Shannon entropy value. First, for the jth time window, the total energy in all K frequency intervals of interest is calculated. Then, the energy E of this time window in each frequency interval is jk Divide by the total energy E j,total , perform normalization processing to obtain the energy distribution probability p in each frequency band jk =E jk / E j,total ,make sure If the energy E of a certain frequency band jk If it is zero, then its probability pjk is also zero. When calculating the logarithm, it is agreed that 0log0=0. Then, these probability values ​​are used to calculate the Shannon entropy value H of the time window. j , the calculation formula is Here, the base-2 logarithm is used so that the unit of entropy is bits. This formula calculates the uncertainty or complexity of the energy distribution within the time window. The above normalization and Shannon entropy calculation steps are performed for each time window in the window-by-window energy distribution concentration, resulting in a sequence of Shannon entropy values ​​that changes over time (arranged in time window order). This sequence is the instantaneous energy entropy sequence group.

[0077] The spectrum difference quantization module includes:

[0078] The event segment extraction submodule, based on the impulse data sequence group, selects the steady-state current and voltage waveforms of several cycles before the impulse occurs, as well as the current and voltage waveforms containing the complete transient process after the impulse occurs. These are used to extract the steady-state current and voltage data segments before parallel operation and the impulse current and voltage data segments during parallel operation, respectively, to obtain the preceding and following feature data segments.

[0079] The spectrum energy calculation submodule applies discrete Fourier transforms to the current and voltage sequences in the steady-state and surge data segments based on the preceding and following characteristic data segments. This calculates the amplitude spectra of each sequence at different frequencies. Fourier transforms are then performed on each extracted data segment to obtain its respective current and voltage spectra, forming a two-state spectrum energy map.

[0080] The energy deviation quantification submodule calculates the difference or ratio between the impact-state and steady-state energies at each corresponding frequency point based on the two-state spectrum energy diagram. It also calculates the total energy change or mean increase in the interharmonic frequency band. It then compares the spectrum energy amplitudes of the corresponding frequency points before and after the parallel operation to quantify the degree of spectrum energy deviation.

[0081] Specifically, based on the impulse data sequence group obtained in the previous step, which contains synchronized time series data for all channels (e.g., current and voltage of phases A, B, and C) during the parallel operation of the generators, as well as the determined start and end timestamps of the impulse event, the steady-state current and voltage waveforms before the impulse are first selected. Specifically, using the start timestamp of the impulse event as a reference, a period of time is traced back, for example, the time segment from 100 milliseconds to 20 milliseconds before the start timestamp is selected. This segment corresponds to four complete power frequency cycles in a 50 Hz system. For example, at this time, the generators are in stable operation but not yet paralleled. Current and voltage waveform data for all phases are extracted from this segment to form the "steady-state current and voltage data segment before parallel operation." Subsequently, current and voltage waveforms covering the complete transient process after the impulse are selected. This segment directly uses the range defined by the previously determined start and end timestamps of the impulse event. Current and voltage waveform data for all phases are extracted from this range to form the "inrush current data segment and surge voltage data segment during parallel operation." Through such selection and extraction, we finally clearly obtained the "steady-state current and voltage data segment before parallel operation" and the "inrush current data segment and surge voltage data segment during parallel operation" for subsequent spectrum comparison analysis. These two sets of data are collectively referred to as the before and after characteristic data segments.

[0082] Based on the preceding and following feature data segments extracted in the previous process, namely the "steady-state current and voltage data segment before parallel operation" and the "inrush current and voltage data segment during parallel operation," a discrete Fourier transform (DFT) is independently applied to each current sequence (e.g., phase A current, phase B current, and phase C current) and voltage sequence (e.g., phase A voltage, phase B voltage, and phase C voltage) in these two data segments. A fast Fourier transform (FFT) algorithm is typically used for efficient calculations. Taking phase A current as an example, an FFT is first performed on the phase A current subsequence in the "steady-state current and voltage data segment before parallel operation" to obtain the amplitude spectrum of the steady-state phase A current. This amplitude spectrum shows the energy amplitude of the current signal at different frequency points. Similarly, an FFT is performed on the phase A current subsequence in the "inrush current and voltage data segment during parallel operation" to obtain the amplitude spectrum of the phase A current during the inrush period. When performing the FFT, the lengths of the data segments in the two states (steady state and inrush state) are ensured to be consistent. Alternatively, zero padding can be used to ensure that the transformed spectra have the same frequency resolution and frequency points. For example, if the steady-state data segment is 1024 points long and the impulse data segment is 2048 points long, the steady-state data segment can be padded to 2048 points before performing the FFT. This independent FFT process is repeated for the current and voltage sequences of all phases, and the amplitude spectrum at different frequencies is calculated for each sequence in the steady and impulse states. Finally, the spectrum of all steady-state sequences is combined to form a steady-state spectrum diagram, and the spectrum of all impulse-state sequences is combined to form a impulse-state spectrum diagram. These two sets of spectrum diagrams together constitute the dual-state spectrum energy diagram.

[0083] Based on the two-state spectrum energy diagram formed in the previous step, which contains detailed spectrum amplitude information of each current and voltage sequence in steady state and impulse state before and after the parallel operation, the spectrum energy deviation is quantitatively calculated. First, for each corresponding signal (for example, the current of phase A), the difference between its impulse state spectrum energy and steady state spectrum energy at each corresponding frequency point is compared. This difference can be expressed by calculating the energy ratio. For example, at the frequency point f i At the energy ratio R(f i )=E impact (f i ) / E steady (f i ), where E(f i) is the square of the amplitude spectrum at the corresponding frequency point, representing the energy at that frequency point. At the same time, special attention is paid to the energy change of the "interharmonic frequency band". This frequency band is pre-defined according to the system characteristics. For example, for a 50 Hz system, it can be defined as a subsynchronous harmonic frequency band covering 5 Hz to 45 Hz, as well as the frequency bands between each integer harmonic, such as 55 Hz to 95 Hz, 105 Hz to 145 Hz, etc., up to half of the Nyquist frequency. The total energy E of the impulse state in these combined interharmonic frequency bands is calculated. impact,IH and the total energy E in steady state steady,IH , calculate the percentage change of its total energy, for example If E steady,IH Close to zero, you can directly use E impact,IH As the variation, the energy amplitude change of each corresponding harmonic frequency point (such as power frequency, 2nd, 3rd, 5th, 7th harmonic) and the total energy change of the interharmonic frequency band are integrated to form a comprehensive "spectrum energy deviation" index. For example, the deviation S dev It can be calculated as: Among them A impact,k and A steady,k are the amplitudes of the kth harmonic in the impact state and steady state, A steady,k,nom is the expected normal amplitude of the kth harmonic in steady state (for example, the fundamental is the rated value, and other harmonics are a few percent of the rated value), H represents the set of harmonic orders of interest {1, 2, 3, 5, 7}, and E IH ,ref is a reference interharmonic total energy level, such as a benchmark value of 1 joule (such as energy unit) set according to historical data, and the weight w h and w ih The harmonics and interharmonics are set according to their importance to the parallel quality, for example, h =0.6 and w ih =0.4. These weights are determined through expert experience or sensitivity analysis of historical fault data, and their sum is 1. In this way, the spectrum energy amplitude and interharmonic energy of the corresponding frequency points before and after the parallel operation are compared one by one, and the spectrum energy deviation is finally quantified.

[0084] The parallel quality assessment modules include:

[0085] The multidimensional feature fusion submodule fuses the current kurtosis and voltage skewness values ​​of the high-order cumulant parameter set, extracts the maximum entropy transition amplitude at the impact starting point and the average entropy value of the impact segment from the instantaneous energy entropy sequence group, and the quantized value of the spectrum energy deviation, arranges them in a predetermined order, and combines them into a multidimensional feature vector to establish the impact feature vector;

[0086] The benchmark comparison and discrimination submodule compares the value of each component in the vector with the preset threshold based on the impact feature vector to see if it exceeds the limit, and generates a comparison and discrimination code;

[0087] The performance status determination submodule determines the electrical performance level of the current parallel impact based on the comparison and discrimination code, and discriminates the electrical performance status of the parallel impact.

[0088] Specifically, a fusion operation of multi-dimensional features is performed. First, specific current kurtosis values ​​and voltage skewness values ​​are selected from the previously calculated and aggregated high-order cumulative quantity parameter set. For example, the average value of the impulse current kurtosis of phase A, the impulse current kurtosis of phase B, and the impulse current kurtosis of phase C is selected as the "average current kurtosis" indicator, and the maximum absolute value of the impulse voltage skewness of phase A, the impulse voltage skewness of phase B, and the impulse voltage skewness of phase C is selected as the "maximum voltage skewness" indicator. Secondly, two key features are extracted from the previously obtained instantaneous energy entropy sequence group: one is the impulse starting point (i.e., the previously determined t start The maximum entropy transition amplitude near the moment t start One power frequency cycle before and after (for example, t start ±20ms), the difference between the maximum and minimum values ​​of the instantaneous energy entropy sequence; the second is the entire impact section (from t start to t end ) is recorded as the "average entropy value of the impact section". Then, the specific value of the spectrum energy deviation obtained by quantification in the previous step is obtained, and these characteristic values ​​from different analysis dimensions: namely, "average current peak tones", "maximum voltage skewness", "maximum entropy transition amplitude at the impact starting point", "average entropy value of the impact section", and "spectral energy deviation" are arranged in a predetermined fixed order. For example, they are arranged in this order into a five-dimensional vector. This ordered numerical combination containing multiple key electrical characteristics is combined into a multidimensional feature vector, which is the impact feature vector used in subsequent evaluation.

[0089] Based on the impact characteristic vector established in the previous process, the vector contains multiple components such as "average current peak value", "maximum voltage skewness", "maximum entropy transition amplitude at the impact starting point", "average entropy value of the impact section" and "spectral energy deviation". Each component value in the vector is compared with its corresponding "preset threshold value" to determine whether it exceeds the limit. These "preset threshold values" are set based on a large amount of historical generator parallel operation data (including normal and various fault conditions), relevant industry standards and expert experience. Each threshold value defines the acceptable range of the corresponding characteristic component. For example, the preset threshold range of "average current peak value" may be set to 1.5 to 4.5. If it exceeds this range, it is considered abnormal. The specific setting process is: collect "average current peak value" data of at least 100 successful parallel operations, and calculate its mean μ k and standard deviation σ k , the upper threshold is set to μ k +2σ k , the lower limit is set to μ k -2σ k , if the historical data shows that the normal kurtosis mean of a specific generator is 3.0 and the standard deviation is 0.5, then the threshold range is 3.0±2×0.5, that is, [2.0, 4.0]; for the "maximum voltage skewness", the preset threshold range may be -0.8 to +0.8; the preset upper threshold limit of the "maximum entropy transition amplitude at the impact starting point" is set to 0.5 bits. This value is based on the observed characteristic that the entropy value changes in normal parallel operation are usually relatively slow, and the 95th percentile of the normal transition amplitude is selected as the threshold; the preset upper threshold limit of the "average entropy value in the impact section" is set to 1.2 bits. A value higher than this may indicate chaotic energy distribution; "spectral energy deviation" S dev The preset threshold upper limit is set to 0.75 (unitless, the normalized value calculated according to the above formula). When a component value in the feature vector exceeds the upper or lower limit of its corresponding preset threshold range, it is marked as exceeding the limit. According to the comparison result, a discrimination bit is generated for each component. For example, if the component value is within the threshold range, the discrimination bit is 0, if it exceeds the upper limit, it is 1, if it is lower than the lower limit (if applicable), it is -1, or uniformly set to 1 to indicate exceeding the limit in any direction. All these discrimination bits are arranged in the order of the original feature vector to jointly generate a comparison discrimination code, such as a five-bit code such as [0, 1, 0, 0, 1].

[0090] Based on the comparison and discrimination code generated in the previous step, the code reflects the compliance of each component in the impact feature vector with respect to its preset threshold. For example, the code [0, 1, 0, 0, 1] indicates that the second and fifth feature components are out of limit, while the remaining components are within the normal range. Next, the electrical performance level of the current parallel impact is determined based on this comparison and discrimination code in combination with predefined evaluation rules. These evaluation rules map different code combinations to different performance levels. For example, four performance levels can be defined: "excellent", "qualified", "concern", and "abnormal". The rule examples are as follows: If all bits in the comparison and discrimination code are 0, the electrical performance level is determined to be "excellent"; if only one non-critical feature (such as "impact segment average entropy value") in the code corresponds to 1 (out of limit), the level is "qualified"; if the bits corresponding to two or more features in the code are 1, or if one of the features considered to be critical (such as "average current peak value" or "spectral energy deviation") corresponds to 1, the level is "concern"; if the bits corresponding to more than three features in the code are 1, Or if any key feature is seriously out of limit (for example, exceeding 150% of the threshold), the electrical performance level is determined to be "abnormal". By applying this series of clear logical judgment rules to the comparison and discrimination codes, the electrical performance status of the parallel impact is ultimately judged and the corresponding performance level is output.

Claims

1. A system for analyzing impact data of parallel operation of generators, characterized in that: The system comprises: The transient signal acquisition module synchronously monitors the ampere value of the surge current and the volt value of the surge voltage during the parallel operation of the generators, obtains the instantaneous reading of the transient electrical signal, converts the reading into a digital sequence, and records the multi-channel transient electrical signal values ​​to form the original electrical measurement data set; A data sequence construction module, based on the original electrical measurement data set, sets the starting point and ending point of the impact event to select an analysis period, performs synchronization alignment processing on the impact current value sequence and the impact voltage value sequence according to the timestamp, and intercepts the valid data segment according to the selected analysis period to generate a time domain impact current sequence and a time domain impact voltage sequence, and construct an impact data sequence group; a cumulant feature calculation module, which performs centering processing on each sequence data point based on the time-domain impulse current sequence and the time-domain impulse voltage sequence in the impulse data sequence group, calculates the ratio of the third-order central moment of the sequence to the cube of the standard deviation as the skewness value, calculates the ratio of the fourth-order central moment of the sequence to the fourth power of the standard deviation minus three as the kurtosis value, calculates the third-order cumulant statistics and the fourth-order cumulant statistics respectively, and obtains a high-order cumulant parameter set; The energy entropy feature extraction module applies a window function to the time domain impulse current sequence and the time domain impulse voltage sequence based on the impulse data sequence group, performs Fourier transform to obtain the spectrum of each time window, calculates the energy distribution of each spectrum in different frequency bands, and then calculates the Shannon entropy value of the energy distribution probability in each time window to obtain the instantaneous energy entropy sequence group.

2. The system for analyzing the impact data of parallel operation of generators according to claim 1, characterized in that: The system further comprises: a spectrum difference quantification module, based on the impulse data sequence group, extracting the current and voltage data segments of steady-state operation before the parallel operation and the impulse current data segments and impulse voltage data segments during the parallel operation, performing Fourier transform on the data segments to obtain their respective spectra, comparing the spectrum energy amplitudes of corresponding frequency points one by one, calculating the difference or ratio, and quantifying the spectrum energy deviation; The parallel quality assessment module integrates the current peak value and voltage skewness of the high-order cumulant parameter set, the entropy transition amplitude and entropy mean of the instantaneous energy entropy sequence group, and the spectrum energy deviation to form a multidimensional feature vector, compares each component of the vector with a preset reference threshold range, and determines the electrical performance status of the parallel impact.

3. The system for analyzing the impact data of parallel operation of generators according to claim 1, characterized in that: The transient signal acquisition module includes: The signal synchronization monitoring submodule collects the ampere value waveform of the impulse current and the volt value waveform of the impulse voltage on the same time basis during the parallel operation of the generators, captures the drastic changes at the moment of parallel operation, obtains the instantaneous reading of the transient electrical signal, and obtains the synchronous monitoring reading; A reading sequence conversion submodule, based on the synchronous monitoring readings, quantizes and encodes the analog readings of the ampere value of the impulse current and the analog readings of the volt value of the impulse voltage according to a predetermined sampling rate and accuracy, and outputs a corresponding discrete time digital sequence to form a digital signal sequence set; The raw data compilation submodule associates the impulse current digital sequence and impulse voltage digital sequence of each channel with their respective channel identification and timestamp information based on the digital signal sequence set, stores them in the data storage structure, and records the corresponding digital sequence values ​​of the multi-channel transient electrical signals to form an original electrical measurement data set.

4. The system for analyzing the impact data of parallel operation of generators according to claim 1, characterized in that: The data sequence building module includes: The time period selection definition submodule retrieves the point in the impulse current or impulse voltage numerical sequence where the amplitude first exceeds the preset starting threshold value based on the original electrical measurement data set as the starting point timestamp, and determines the ending point timestamp at the point after the impulse stabilizes or a fixed duration, sets the timestamp information of the starting point and the ending point of the impulse event, selects the analysis period, and obtains the analysis period parameters; A data alignment and interception submodule, based on the original electrical measurement data set and the analysis period parameters, checks the timestamp sequence of each numerical sequence, aligns the data points of different channels on the time axis through interpolation or resampling, and intercepts valid data segments according to the start and end timestamps of the selected analysis period to obtain a intercepted data segment group; The sequence construction generation submodule extracts the aligned impulse current valid data segment values ​​and impulse voltage valid data segment values ​​based on the intercepted data segment group, arranges them in chronological order, forms independent time domain impulse current sequence arrays and time domain impulse voltage sequence arrays, organizes them into time domain impulse current sequences and time domain impulse voltage sequences, and constructs an impulse data sequence group.

5. The system for analyzing the impact data of parallel operation of generators according to claim 1, characterized in that: The cumulative feature calculation module includes: The data centralization submodule calculates the average value of all data points in each independent sequence based on the time-domain impulse current sequence and the time-domain impulse voltage sequence in the impulse data sequence group, and then traverses the sequence, subtracts the calculated average value from the original value of each data point, and subtracts the mean value of the sequence from each data point in each sequence to perform centralization processing to generate a centralized data sequence; The cumulant parameter calculation submodule, based on the centralized data sequence, takes the cube of the deviation of each data point in each sequence to obtain the third-order central moment, then divides it by the cube of the standard deviation of the sequence to obtain the skewness value, and calculates the ratio of the fourth-order central moment of each centralized sequence to the fourth power of the standard deviation and subtracts three as the kurtosis value to obtain the sequence-by-sequence cumulant value; The characteristic parameter aggregation submodule extracts the impulse current skewness, impulse current kurtosis, impulse voltage skewness and impulse voltage kurtosis as core indicators from the skewness and kurtosis values ​​of the current sequence and the voltage sequence based on the sequence-by-sequence cumulative value, aggregates the skewness values ​​and kurtosis values ​​of all sequences, forms a third-order cumulative value statistical value set and a fourth-order cumulative value statistical value set, and obtains a high-order cumulative value parameter set.

6. The system for analyzing the impact data of parallel operation of generators according to claim 1, characterized in that: The energy entropy feature extraction module includes: The time-spectrum conversion submodule, based on the impulse data sequence group, adopts an overlapping segmentation method for each time domain sequence, multiplies each segment by a Hamming window, and then performs discrete Fourier transform calculation on each segmented data to obtain the spectrum of each time window to form a time-spectrum data frame; The window energy distribution calculation submodule, based on the time-spectrum data frame, accumulates and sums the squares of the amplitudes of the frequency points in each time window within a plurality of pre-divided frequency intervals of interest for the amplitude-spectrum data within each time window, obtains the energy value of the time window in the frequency interval, calculates the numerical distribution of the energy of the spectrum of each time window in different preset frequency segments, and forms a window-by-window energy distribution set; The sequence Shannon entropy acquisition submodule, based on the window-by-window energy distribution set, normalizes the energy of each frequency band in each window to obtain the probability, multiplies the probability by the logarithm and then sums and takes the negative value to obtain the Shannon entropy value, thereby obtaining the instantaneous energy entropy sequence group.

7. The system for analyzing the impact data of parallel operation of generators according to claim 1, characterized in that: The spectrum difference quantization module includes: An event segment extraction submodule, based on the shock data sequence group, selects the steady-state current and voltage waveforms of several cycles before the shock occurs, and the current and voltage waveforms containing the complete transient process after the shock occurs, and uses them as the current and voltage data segments of the steady-state operation before the parallel operation and the shock current data segments and shock voltage data segments during the parallel operation to obtain the front and back feature data segments; The spectrum energy calculation submodule applies discrete Fourier transform to the current and voltage sequences in the steady-state operation data segment and the impulse period data segment based on the preceding and following characteristic data segments, calculates the amplitude spectrum of each sequence at different frequencies, and performs Fourier transform processing on each extracted data segment to obtain the respective current and voltage spectra, thereby forming a two-state spectrum energy diagram; The energy deviation quantification submodule calculates the difference or ratio between the impact state and steady-state energies of each corresponding frequency point based on the two-state spectrum energy diagram, and counts the total energy change or mean increase in the interharmonic frequency band. In this way, the spectrum energy amplitudes of the corresponding frequency points before and after the parallel operation are compared one by one to quantify the spectrum energy deviation.

8. The system for analyzing generator parallel operation impact data according to claim 1, characterized in that: The parallel quality assessment module includes: The multidimensional feature fusion submodule fuses the current peak value and voltage skewness values ​​of the high-order cumulant parameter set, extracts the maximum entropy transition amplitude of the impact starting point and the average entropy value of the impact segment from the instantaneous energy entropy sequence group, as well as the quantized value of the spectrum energy deviation, arranges them in a predetermined order, and combines them into a multidimensional feature vector to establish the impact feature vector; A benchmark comparison and discrimination submodule, based on the impact feature vector, compares each component value in the vector with a preset threshold value to see if it exceeds the limit, and generates a comparison and discrimination code; The performance status determination submodule determines the electrical performance level of the current parallel impact based on the comparison and discrimination code, and discriminates the electrical performance status of the parallel impact.

Images