An adaptive power quality monitoring method for power supply cabinet

Abstract

The present application relates to the technical field of measuring electric variable, specifically to a power cabinet adaptive power quality monitoring method, comprising the following steps: starting the low-frequency acquisition unit arranged in the power cabinet monitoring node to collect the low-frequency rough scanning voltage signal of the power cabinet in the current monitoring period at a preset low-frequency sampling rate.The present application transmits the compressed observation vector and uses the orthogonal matching pursuit algorithm to perform sparse base inversion calculation, reconstructs high-precision power quality time-domain waveform data, restores rich high-frequency transient details under low data transmission conditions, carries out distortion feature analysis and identifies disturbance types based on the reconstructed waveform, generates a monitoring analysis report for uploading, realizes real-time dynamic adjustment of the sampling scale and matrix structure according to the signal fluctuation state, achieves the optimal balance between data compression ratio and signal reconstruction quality, solves the contradiction between massive high-frequency data storage and limited transmission bandwidth in complex power grid environment, and improves the sensitivity to transient abnormalities.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology, and in particular to an adaptive power quality monitoring method for power supply cabinets. Background Technology

[0002] Current power quality monitoring primarily relies on a fixed high-frequency sampling mechanism that satisfies the Nyquist sampling theorem to capture transient disturbances. In actual operation, to prevent the omission of high-frequency harmonics or transient pulses, constant high-speed data acquisition must be maintained regardless of whether the grid load is fluctuating or operating smoothly. This indiscriminate, all-time high-frequency sampling generates redundant data during long-term steady-state operation, placing a load burden on front-end analog-to-digital converters and storage units. High-frequency sampling also places high demands on hardware specifications, leading to higher equipment manufacturing costs and operating power consumption. Furthermore, in scenarios involving multi-node distributed monitoring, concurrent transmission of raw high-sampling-rate data can easily cause communication network congestion, resulting in data packet loss or transmission delays, hindering the real-time uploading of critical fault information. Therefore, improvements are needed. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by proposing an adaptive power quality monitoring method for power cabinets.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: an adaptive power quality monitoring method for power cabinets, comprising the following steps:

[0005] The low-frequency acquisition unit set at the power cabinet monitoring node is activated to acquire the low-frequency coarse sweep voltage signal of the power cabinet in the current monitoring period at a preset low-frequency sampling rate, and the time-domain fluctuation characteristics of the low-frequency coarse sweep voltage signal are extracted.

[0006] The adaptive observation dimension for the current monitoring period is calculated based on the time-domain fluctuation characteristics. A dynamic Gaussian random observation matrix is ​​constructed based on the adaptive observation dimension, wherein the number of rows of the dynamic Gaussian random observation matrix is ​​equal to the adaptive observation dimension, and the number of columns of the dynamic Gaussian random observation matrix is ​​equal to the preset high-frequency reconstruction length.

[0007] The dynamic Gaussian random observation matrix is ​​used to configure the compressed sampling circuit of the analog front end to perform compressed sensing analog information conversion on the real-time analog voltage signal of the power cabinet to obtain a compressed observation vector, wherein the dimension of the compressed observation vector is consistent with the adaptive observation dimension.

[0008] The compressed observation vector is transmitted to the digital signal processing unit, and the orthogonal matching pursuit algorithm is used to perform sparse basis inversion calculation on the compressed observation vector to reconstruct high-precision power quality time-domain waveform data.

[0009] The high-precision power quality time-domain waveform data is analyzed for waveform distortion characteristics to identify the power quality disturbance type of the power cabinet, generate a power quality monitoring and analysis report, and upload it to the monitoring terminal.

[0010] Preferably, the steps of activating the low-frequency acquisition unit set at the power cabinet monitoring node, acquiring the low-frequency coarse-scan voltage signal of the power cabinet in the current monitoring period at a preset low-frequency sampling rate, and extracting the time-domain fluctuation characteristics of the low-frequency coarse-scan voltage signal are as follows:

[0011] Initialize the analog-to-digital converter inside the low-frequency acquisition unit, and set the sampling frequency of the analog-to-digital converter to a preset multiple of the power frequency;

[0012] The analog-to-digital converter is used to poll and sample the three-phase voltage ports of the power cabinet to obtain a discrete voltage sampling sequence within a preset time window, and the discrete voltage sampling sequence is marked as the low-frequency coarse sweep voltage signal.

[0013] Calculate the rate of change of voltage amplitude and the total energy value of voltage waveform of the low-frequency coarse sweep voltage signal within the preset time window;

[0014] The voltage amplitude change rate and the total energy value are combined to form the time-domain fluctuation characteristic.

[0015] Preferably, the step of calculating the adaptive observation dimension for the current monitoring period based on the time-domain fluctuation characteristics, and constructing a dynamic Gaussian random observation matrix based on the adaptive observation dimension, specifically includes:

[0016] Read the voltage amplitude change rate and total energy value of the voltage waveform contained in the time-domain fluctuation characteristics, and obtain the pre-stored rated voltage reference value, reference energy value and reference voltage amplitude change rate of the power cabinet;

[0017] The adaptive observation dimension is calculated by processing the voltage amplitude change rate and the total energy value using an adaptive dimension calculation model. The calculation formula in the adaptive dimension calculation model is as follows:

[0018] ;

[0019] in, This represents the adaptive observation dimension. This represents the preset basic observation dimension. This indicates the floor function. This represents the energy weighting coefficient. This represents the ratio of the total energy value to the reference energy value. This represents the total energy value. This represents the reference energy value. Represents the sensitivity coefficient to the rate of change. Represents the logarithmic function with base 10. This represents the absolute value of the rate of change of the voltage amplitude. This represents the rate of change of the reference voltage amplitude;

[0020] Initialize a zero matrix with the number of rows equal to the adaptive observation dimension and the number of columns equal to the preset high-frequency reconstruction length;

[0021] A Gaussian random number sequence is generated using a Gaussian distribution random number generator, and the Gaussian random number sequence is filled into the zero matrix to obtain the dynamic Gaussian random observation matrix.

[0022] Preferably, the specific steps for configuring the compressed sampling circuit of the analog front end using the dynamic Gaussian random observation matrix to perform compressed sensing analog information conversion on the real-time analog voltage signal of the power supply cabinet and obtain the compressed observation vector are as follows:

[0023] The dynamic Gaussian random observation matrix is ​​loaded into the random demodulator controller of the analog front end;

[0024] The random demodulator controller uses the timing logic of the dynamic Gaussian random observation matrix to control the switching state of the mixer in the compressed sampling circuit;

[0025] The real-time analog voltage signal of the power cabinet is multiplied by the pseudo-random sequence through the mixer to obtain the mixed analog modulated signal, wherein the pseudo-random sequence is obtained by the dynamic Gaussian random observation matrix.

[0026] The integrator is controlled to perform integration of the analog modulation signal for a preset time period to obtain the integrated analog voltage value;

[0027] At the end of each integration cycle, a low-speed analog-to-digital converter is triggered to quantize the integrated analog voltage value to obtain a digitized observation value. The observation values ​​are then arranged in chronological order to form the compressed observation vector.

[0028] Preferably, the specific steps for transmitting the compressed observation vector to a digital signal processing unit and using an orthogonal matching pursuit algorithm to perform sparse basis inversion calculation on the compressed observation vector to reconstruct high-precision power quality time-domain waveform data are as follows:

[0029] The compressed observation vector is input into the computing memory of the digital signal processing unit, and a preset discrete cosine transform sparse basis matrix is ​​loaded.

[0030] Initialize the residual vector to the compressed observation vector, initialize the index set to an empty set, and initialize the iteration counter;

[0031] The sensing matrix is ​​obtained by multiplying the dynamic Gaussian random observation matrix with the discrete cosine transform sparse basis matrix.

[0032] In each iteration, the inner product correlation coefficient between each column of the sensing matrix and the current residual vector is calculated, and the column index corresponding to the column with the largest absolute value of the inner product correlation coefficient is found.

[0033] Add the column index to the index set, and extract the corresponding sub-matrix from the sensing matrix according to the index set;

[0034] The approximate solution of the submatrix and the compressed observation vector is obtained by using the least squares method to obtain the current sparse coefficient estimate;

[0035] Update the residual vector and determine whether the norm of the residual vector is less than a preset reconstruction stopping threshold. If it is less, stop the iteration; otherwise, continue to the next iteration.

[0036] The high-precision power quality time-domain waveform data is synthesized by using the final sparse coefficient estimate and the discrete cosine transform sparse basis matrix for inverse transformation.

[0037] Preferably, in each iteration, the step of calculating the inner product correlation coefficient between each column of the sensing matrix and the current residual vector, and finding the column index corresponding to the column with the largest absolute value of the inner product correlation coefficient, specifically involves:

[0038] Normalize each column vector of the sensing matrix to obtain a set of normalized sensing column vectors;

[0039] Calculate the dot product between each vector in the normalized sensing column vector set and the residual vector to obtain the correlation coefficient vector;

[0040] Traverse the correlation coefficient vector and locate the position of the maximum value in the correlation coefficient vector through comparison operations;

[0041] The column number of the matrix corresponding to the position of the maximum value is determined as the column index.

[0042] Preferably, the steps of performing waveform distortion feature analysis on the high-precision power quality time-domain waveform data, identifying the power quality disturbance type of the power cabinet, generating a power quality monitoring and analysis report, and uploading it to the monitoring terminal are as follows:

[0043] The high-precision power quality time-domain waveform data is subjected to Fourier transform to obtain power quality spectrum data;

[0044] Calculate the amplitude of the fundamental component and the amplitude of the harmonic component in the power quality spectrum data, and calculate the total harmonic distortion rate based on the amplitude of the fundamental component and the amplitude of the harmonic component.

[0045] Detect voltage sag characteristics, voltage swell characteristics, and voltage interruption characteristics in the high-precision power quality time-domain waveform data;

[0046] If the total harmonic distortion rate exceeds the preset harmonic threshold, the power quality disturbance type is determined to be harmonic pollution.

[0047] If the voltage sag characteristic is detected, the power quality disturbance type is determined to be a voltage sag;

[0048] If the voltage sag characteristic is detected, the power quality disturbance type is determined to be a voltage sag;

[0049] If the voltage interruption feature is detected, the power quality disturbance type is determined to be a voltage interruption;

[0050] The determined power quality disturbance type, total harmonic distortion rate, and timestamp of the disturbance occurrence are encapsulated into a standard format data packet and used as the power quality monitoring and analysis report.

[0051] The power quality monitoring and analysis report is sent to the remote monitoring terminal via an industrial Ethernet interface.

[0052] Preferably, the specific steps for generating a Gaussian random number sequence using a Gaussian distributed random number generator and filling the zero matrix with the Gaussian random number sequence to obtain the dynamic Gaussian random observation matrix are as follows:

[0053] The hardware random number generator is invoked to generate an initial random seed, and the initial random seed is injected into the Gaussian distribution random number generator.

[0054] The Box-Muller transformation algorithm is used to convert uniformly distributed random numbers into standard normally distributed random numbers.

[0055] The standard normal distribution random numbers are scaled to make the variance meet the numerical requirement of 1 divided by the adaptive observation dimension;

[0056] Fill each element position of the zero matrix sequentially until all positions are filled;

[0057] The filled matrix is ​​orthogonalized and verified. If the verification fails, it is regenerated until the dynamic Gaussian random observation matrix that satisfies the restricted isometry property is obtained.

[0058] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0059] In this invention, a coarse-scan voltage signal of the current period is acquired using a preset low-frequency sampling rate, and time-domain fluctuation features are extracted. The observation dimension is adaptively calculated based on the fluctuation feature values, and a dynamic Gaussian random observation matrix with rows strictly matching this dimension is constructed. This ensures real-time coupling between sampling resource allocation and the actual complexity of the signal, avoids the use of redundant resources by steady-state signals, and guarantees the accuracy of transient disturbance capture. The dynamic matrix configuration of the compressed sampling circuit directly performs compressed sensing analog information conversion on the real-time analog voltage signal, obtaining a compressed observation vector consistent with the adaptive dimension. This overcomes the limitations of the Nyquist sampling theorem at the hardware physical layer, reducing the bandwidth of high-bandwidth signals. To address the demands of analog-to-digital conversion rates and the pressure of backend data throughput, the system compresses observation vectors for transmission and utilizes an orthogonal matching pursuit algorithm to perform sparse basis inversion calculations. This reconstructs high-precision power quality time-domain waveform data, restoring rich high-frequency transient details under low data transmission conditions. Based on the reconstructed waveform, distortion feature analysis is performed, and disturbance types are identified. Monitoring and analysis reports are generated and uploaded, enabling real-time dynamic adjustment of sampling scale and matrix structure according to signal fluctuations. This achieves an optimal balance between data compression ratio and signal reconstruction quality, resolving the contradiction between massive high-frequency data storage and limited transmission bandwidth in complex power grid environments, and improving sensitivity to transient anomalies. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0062] Please see Figure 1 This invention provides a technical solution: an adaptive power quality monitoring method for power cabinets, comprising the following steps:

[0063] The low-frequency acquisition unit set at the power cabinet monitoring node is activated to acquire the low-frequency coarse sweep voltage signal of the power cabinet in the current monitoring period at a preset low-frequency sampling rate, and extract the time-domain fluctuation characteristics of the low-frequency coarse sweep voltage signal.

[0064] The adaptive observation dimension for the current monitoring period is calculated based on the time-domain fluctuation characteristics. A dynamic Gaussian random observation matrix is ​​constructed based on the adaptive observation dimension, where the number of rows of the dynamic Gaussian random observation matrix is ​​equal to the adaptive observation dimension and the number of columns of the dynamic Gaussian random observation matrix is ​​equal to the preset high-frequency reconstruction length.

[0065] By configuring the compressed sampling circuit of the analog front end using a dynamic Gaussian random observation matrix, the real-time analog voltage signal of the power cabinet is converted into compressed sensing analog information to obtain a compressed observation vector, wherein the dimension of the compressed observation vector is consistent with the dimension of the adaptive observation.

[0066] The compressed observation vector is transmitted to the digital signal processing unit, and the orthogonal matching pursuit algorithm is used to perform sparse basis inversion calculation on the compressed observation vector to reconstruct high-precision power quality time-domain waveform data.

[0067] Waveform distortion feature analysis is performed on high-precision power quality time-domain waveform data to identify the power quality disturbance type of the power cabinet, generate a power quality monitoring and analysis report, and upload it to the monitoring terminal.

[0068] In this embodiment, the steps of activating the low-frequency acquisition unit set at the power cabinet monitoring node, acquiring the low-frequency coarse sweep voltage signal of the power cabinet in the current monitoring period at a preset low-frequency sampling rate, and extracting the time-domain fluctuation characteristics of the low-frequency coarse sweep voltage signal are as follows: initializing the analog-to-digital converter inside the low-frequency acquisition unit, setting the sampling frequency of the analog-to-digital converter to a preset multiple of the power frequency, polling and sampling the three-phase voltage ports of the power cabinet through the analog-to-digital converter, obtaining the discrete voltage sampling sequence within a preset time window length, marking the discrete voltage sampling sequence as the low-frequency coarse sweep voltage signal, calculating the voltage amplitude change rate and the total energy value of the voltage waveform within the preset time window length of the low-frequency coarse sweep voltage signal, and combining the voltage amplitude change rate and the total energy value into time-domain fluctuation characteristics.

[0069] Specifically, first, a reset command is sent to the analog-to-digital converter (ADC) to clear residual data in its internal registers. Then, the frequency multiplier parameter, typically set to 128 or 256, is read from the pre-written configuration file. Based on the rated power frequency of the grid connected to the power cabinet (e.g., 50Hz), the sampling clock frequency of the ADC is calculated through multiplication and written into the clock control register. The ADC is then started to cyclically poll the voltage ports of phases A, B, and C. At each sampling point, the voltage channels of each phase are sequentially connected, maintaining the channel connection time to meet the setup time requirement of the sample-and-hold circuit (e.g., 10 microseconds). Continuous voltage values ​​within a set time window (e.g., 0.2 seconds) are acquired. These values ​​are stored in a buffer in chronological order to form a discrete voltage sampling sequence. The process is then iterated through... All values ​​are calculated by comparing the difference between two adjacent sampling points and dividing by the sampling time interval to obtain the instantaneous rate of change at each moment. The value with the largest absolute value is selected as the voltage amplitude rate of change. Alternatively, the average rate of change can be obtained by calculating the difference between the maximum and minimum voltage values ​​in the sequence and dividing by the time window length. At the same time, each voltage value in the sequence is squared, and all squared values ​​are summed to obtain the total energy of the voltage signal within the time window. This summed value is the total energy value of the voltage waveform. The calculated voltage amplitude rate of change value and the total energy value of the voltage waveform are packaged in a predetermined format, such as constructing a structure containing two floating-point data points, as a time-domain fluctuation feature representing the current power grid fluctuation state, which is used for subsequent dynamic adjustment of compressed sensing parameters.

[0070] In this embodiment, the steps of calculating the adaptive observation dimension for the current monitoring period based on the time-domain fluctuation characteristics and constructing a dynamic Gaussian random observation matrix based on the adaptive observation dimension are as follows: Read the voltage amplitude change rate and the total energy value of the voltage waveform contained in the time-domain fluctuation characteristics, and obtain the pre-stored rated voltage reference value, reference energy value, and reference voltage amplitude change rate of the power supply cabinet. Process the voltage amplitude change rate and total energy value using the adaptive dimension calculation model to calculate the adaptive observation dimension. The calculation formula in the adaptive dimension calculation model is: ,in, Indicates the adaptive observation dimension. This represents the preset basic observation dimension. This indicates the floor function. This represents the energy weighting coefficient. This represents the ratio of the total energy value to the reference energy value. Indicates the total energy value. Indicates the reference energy value. Represents the sensitivity coefficient to the rate of change. Represents the logarithmic function with base 10. Represents the absolute value of the rate of change of voltage amplitude. The reference voltage amplitude change rate is represented by a zero matrix with the number of rows being the adaptive observation dimension and the number of columns being the preset high-frequency reconstruction length. A Gaussian random number sequence is generated using a Gaussian distributed random number generator and filled into the zero matrix to obtain the dynamic Gaussian random observation matrix.

[0071] Specifically, the newly generated time-domain fluctuation characteristic data packet is retrieved from memory, and the voltage amplitude change rate and the total energy value of the voltage waveform are parsed out. Simultaneously, the non-volatile storage space is accessed to read the rated voltage reference value, the reference energy value calculated under standard operating conditions, and the allowable reference voltage amplitude change rate, obtained from at least 24 hours of no-load and full-load operation tests conducted during the initial operation of the power cabinet. These real-time measured values ​​and historical reference values ​​are then substituted into the adaptive dimension calculation model for calculation, yielding the adaptive observation dimension. The calculation formula in the adaptive dimension calculation model is as follows: ,in, This represents the adaptive observation dimension, which determines the number of rows in the final generated observation matrix, i.e., the size of the compressed data. The result must be a positive integer. This represents the preset basic observation dimension, which is determined based on the minimum number of measurements required to guarantee the lowest reconstruction probability in compressed sensing theory. It is usually set to 3 to 4 times the signal sparsity K, for example, in this embodiment, it is set to 64 through sparsity analysis of historical data. This indicates a rounding down operation, which rounds the result within the parentheses to the nearest integer in the direction of negative infinity. For example, if the result is 80.9, the value will be 80. This is used to ensure that the dimension is an integer. This represents the energy weighting coefficient, which reflects the degree of influence of signal energy fluctuations on the observation dimension. Its value is usually between 0 and 1. It is calibrated through a large number of experiments. For example, it is set to 0.5 to represent the contribution weight of the energy term to the dimension adjustment. It represents the ratio of the total energy value to the reference energy value, and is used to measure the degree of deviation of the current signal energy from the standard operating condition; This represents the total energy value, which is the real-time energy value calculated in the previous steps by summing the squares of the sampled sequences. The reference energy value is the average energy statistical value recorded during the equipment installation and commissioning phase when the grid voltage is under an ideal sinusoidal state. This represents the rate of change sensitivity coefficient, used to adjust the sensitivity of the observation dimension to voltage fluctuations. It is usually taken as a small value, such as 0.3, to avoid excessive dimensional oscillations. This represents a base-10 logarithmic function used to compress the dynamic range of the ratio of rates of change, making dimension adjustment smoother. This represents the absolute value of the rate of change of voltage amplitude. That is, the rate of change of voltage monitored in real time is taken as positive to ensure that the value substituted into the logarithmic function is non-negative. This represents the rate of change of the reference voltage amplitude, a reference benchmark set according to the maximum voltage fluctuation rate allowed by power grid quality standards, for example, 10 volts per second; based on the calculated... Allocate memory space for the value, and create a row with the number of rows. A two-dimensional array with a preset high-frequency reconstruction length (e.g., 512, corresponding to the number of signal points required for high-frequency reconstruction) is generated, and all elements in the array are initialized to zero. A random number generation algorithm is then called to fill the array to obtain a dynamic Gaussian random observation matrix.

[0072] In this embodiment, the steps of using a dynamic Gaussian random observation matrix to configure the compressed sampling circuit of the analog front end to perform compressed sensing analog information conversion on the real-time analog voltage signal of the power cabinet and obtain the compressed observation vector are as follows: The dynamic Gaussian random observation matrix is ​​loaded into the random demodulator controller of the analog front end. The random demodulator controller controls the switching state of the mixer in the compressed sampling circuit according to the timing logic of the dynamic Gaussian random observation matrix. The mixer performs analog multiplication operation on the real-time analog voltage signal of the power cabinet and the pseudo-random sequence to obtain the mixed analog modulation signal, wherein the pseudo-random sequence is converted by the dynamic Gaussian random observation matrix. The integrator is controlled to perform integration operation on the analog modulation signal for a preset time period to obtain the integrated analog voltage value. At the end of each integration period, the low-speed analog-to-digital converter is triggered to quantize the integrated analog voltage value to obtain the digitized observation value. The observation values ​​are arranged in chronological order to form the compressed observation vector.

[0073] Specifically, the generated dynamic Gaussian random observation matrix is ​​written row by row into the control register of the random demodulator via SPI or I2C bus, or stored in the FPGA's block RAM as a lookup table for the pseudo-random sequence. The timing control logic is then activated. Based on the value of each element in the observation matrix (usually binarized to +1 and -1, or corresponding to 1 and 0 for on / off switching), a control level is output in each clock cycle, connected to the switching control terminal of the analog mixer circuit. When the control level is high, the mixer is driven through; when it is low, the mixer is inverted or deactivated. This achieves the multiplication of the continuous analog voltage signal input from the power supply cabinet with the pseudo-random sequence corresponding to the observation matrix in the analog domain. This process shifts and expands the voltage signal's spectrum across the entire bandwidth, thus enabling the mixing... The analog signal is input to the analog integrator circuit. The parameters of the integrating capacitor and integrating resistor are set to match the preset integration time period, which is equal to the time period corresponding to the high-frequency reconstruction length. After reset, the integrator starts to accumulate charge on the signal, continuously integrating the analog modulated signal and accumulating the signal fluctuations over a period of time into a single voltage value. When the integration timer reaches the end of the preset period, a trigger signal is immediately sent to the sample-and-hold terminal of the low-speed analog-to-digital converter to hold the current analog voltage output by the integrator and start the quantization conversion process to convert the analog voltage into a digital quantity. Then, a reset signal is immediately sent to clear the charge on the integrator capacitor, preparing for the next integration cycle. The converted digital quantity is sequentially stored in a first-in-first-out queue, and the process continues until completion. After multiple integration and quantization operations, the data in the queue are arranged in chronological order to form a compressed observation vector.

[0074] In this embodiment, the steps of transmitting the compressed observation vector to the digital signal processing unit and using the orthogonal matching pursuit algorithm to perform sparse basis inversion calculation on the compressed observation vector to reconstruct high-precision power quality time-domain waveform data are as follows: The compressed observation vector is input into the computing memory of the digital signal processing unit, and a preset discrete cosine transform sparse basis matrix is ​​loaded. The residual vector is initialized as the compressed observation vector, the index set is initialized as an empty set, the iteration counter is initialized, and the product of the dynamic Gaussian random observation matrix and the discrete cosine transform sparse basis matrix is ​​calculated to obtain the sensing matrix. In each iteration, each step of the sensing matrix is ​​calculated. The inner product correlation coefficient between the column and the current residual vector is used to find the column index corresponding to the column with the largest absolute value of the inner product correlation coefficient. The column index is added to the index set. The corresponding sub-matrix is ​​extracted from the sensing matrix according to the index set. The least squares method is used to solve the approximate solution between the sub-matrix and the compressed observation vector to obtain the current sparse coefficient estimate. The residual vector is updated. It is determined whether the norm of the residual vector is less than the preset reconstruction stopping threshold. If it is less, the iteration stops. Otherwise, the next iteration continues. The final sparse coefficient estimate and the discrete cosine transform sparse basis matrix are used to perform inverse transformation to synthesize high-precision power quality time-domain waveform data.

[0075] Specifically, the compressed observation vectors in the FIFO are transferred in batches to the computational memory of the DSP or GPU via the DMA controller, and the pre-computed Discrete Cosine Transform (DCT) sparse basis matrix is ​​loaded from the read-only memory. The dimension of this matrix is ,in To determine the high-frequency reconstruction length, initialize a length of... residual vector Its initial value is directly assigned to the input compressed observation vector. Create an empty list to store column indexes. And the iteration count counter Set to 1, perform matrix multiplication, and convert the dynamic Gaussian random observation matrix... With sparse basis matrix Multiply to obtain the sensing matrix Enter the iterative loop, and calculate the sensing matrix in each iteration. Each column vector and the current residual vector The inner product, which represents the correlation between the basis vectors and the residuals, is selected by choosing the index of the column with the largest absolute value of the inner product. Add the index to the index list In the middle, based on the updated index list, from the sensor matrix Select the corresponding column vectors to form a submatrix Constructing the least squares problem The problem can be solved using matrix inversion or QR decomposition algorithms to obtain the current estimates of sparse coefficients. Use this estimate to update the residuals. Calculate the updated residual vector Norm, i.e., its energy magnitude, is compared with a preset reconstruction stopping threshold (e.g., This threshold is pre-set based on the required reconstruction accuracy and noise level. If the residual norm is less than the threshold or the number of iterations reaches the upper limit (e.g., ...), the comparison is performed. If the iteration terminates after a certain number of iterations (e.g., 1000), the final sparse coefficient vector is output. Finally, the inverse transform operation is performed to calculate... By projecting the sparse coefficients back into the time domain, high-precision power quality time-domain waveform data is obtained.

[0076] In this embodiment, the specific steps for calculating the correlation coefficient of the inner product between each column of the sensing matrix and the current residual vector, and finding the column index corresponding to the column with the largest absolute value of the inner product correlation coefficient during each iteration are as follows: normalize each column vector of the sensing matrix to obtain a set of normalized sensing column vectors; calculate the dot product between each vector in the set of normalized sensing column vectors and the residual vector to obtain the correlation coefficient vector; traverse the correlation coefficient vectors; locate the position of the maximum value in the correlation coefficient vectors through comparison operations; and determine the column index of the matrix corresponding to the position of the maximum value as the column index.

[0077] Specifically, traversing the sensor matrix All For each column vector, ... Perform a normalization operation, that is, calculate the Euclidean norm of the vector. Then, each element in the vector is divided by the norm to obtain the unit direction vector. This step is to eliminate the influence of different column vector magnitudes on the correlation calculation. The processed normalized column vector set is temporarily stored in the cache, and parallel computing instructions are started to process the current residual vector. Perform a dot product operation with each normalized column vector, that is, multiply corresponding elements and then sum them to obtain a vector containing... The correlation coefficient vector of each element , of which The element is the first The dot product of a normalized column vector and a residual vector is obtained by multiplying the corresponding elements of the normalized column vector and the residual vector and summing the results. A maximum value variable is initialized to 0 and an index variable is initialized to 0. The correlation coefficient vector is scanned sequentially, and the absolute value of the current element is compared with the maximum value variable. If the absolute value of the current element is greater than the maximum value variable, the maximum value variable is updated to the absolute value of the current element, and the index variable is updated to the current position number. This process continues until all elements have been scanned. The final locked index variable is the position of the atom that best matches the current residual in the dictionary. The column number of the matrix corresponding to this position is determined as the column index.

[0078] In this embodiment, the steps of performing waveform distortion feature analysis on high-precision power quality time-domain waveform data, identifying the power quality disturbance type of the power cabinet, generating a power quality monitoring and analysis report, and uploading it to the monitoring terminal are as follows: Perform Fourier transform on the high-precision power quality time-domain waveform data to obtain power quality spectrum data; calculate the amplitude of the fundamental component and harmonic components in the power quality spectrum data; calculate the total harmonic distortion rate based on the amplitude of the fundamental component and harmonic components; detect voltage sag characteristics, voltage swell characteristics, and voltage interruption characteristics in the high-precision power quality time-domain waveform data; if the total harmonic distortion rate is... If the harmonic distortion rate exceeds the preset harmonic threshold, the power quality disturbance type is determined to be harmonic pollution. If a voltage sag is detected, the power quality disturbance type is determined to be a voltage sag. If a voltage swell is detected, the power quality disturbance type is determined to be a voltage swell. If a voltage interruption is detected, the power quality disturbance type is determined to be a voltage interruption. The determined power quality disturbance type, total harmonic distortion rate, and timestamp of the disturbance occurrence are encapsulated into a standard format data packet as a power quality monitoring and analysis report. The power quality monitoring and analysis report is sent to the remote monitoring terminal through the industrial Ethernet interface.

[0079] Specifically, the Fast Fourier Transform (FFT) algorithm library is invoked to perform frequency domain transformation on the reconstructed time-domain waveform data. The number of transformation points is set to 512 or 1024 to obtain sufficient frequency resolution. The magnitude of the transformation result is calculated to obtain the amplitude spectrum. The amplitude at the 50Hz power frequency is extracted from the amplitude spectrum as the fundamental component amplitude. The amplitudes at integer multiples of frequencies such as 100Hz and 150Hz are then used as the amplitudes of each harmonic component. Based on the total harmonic distortion (THD) calculation formula, the square root of the sum of the squares of all harmonic amplitudes is divided by the fundamental amplitude to obtain the THD value. Simultaneously, a point-by-point scan is performed on the time-domain waveform to calculate the root mean square (RMS) value. This RMS value is compared with a preset nominal voltage value. For example, if the nominal value is 220V, the voltage sag threshold is set to 90% of the nominal value, i.e., 198V, and the voltage swell threshold is set to 110% of the nominal value. That is, 242V. The voltage interruption judgment threshold is set to 10% of the nominal value, i.e., 22V. If the RMS value of multiple consecutive cycles is lower than 198V and higher than 22V, it is marked as a voltage sag. If it is higher than 242V, it is marked as a voltage swell. If it is lower than 22V, it is marked as a voltage interruption. Then, the logic judgment process is entered. If the calculated total harmonic distortion rate exceeds the limit specified by the national standard, such as 5%, it is judged that there is harmonic pollution. If the above sag, swell or interruption characteristics are identified, they are respectively judged as the corresponding disturbance type. The determined disturbance type code, the calculated distortion rate value accurate to two decimal places, and the start timestamp of the fault occurrence are encapsulated according to IEC61850 or a custom JSON format, and a check code and frame header and trailer are added. The encapsulated data packet is sent to the IP address of the remote monitoring terminal through the TCP / IP protocol stack to drive the physical network port chip.

[0080] In this embodiment, the steps of generating a Gaussian random number sequence using a Gaussian distributed random number generator and filling the zero matrix with the Gaussian random number sequence to obtain a dynamic Gaussian random observation matrix are as follows: The hardware random number generator is called to generate an initial random seed; the initial random seed is injected into the Gaussian distributed random number generator; the uniformly distributed random numbers are converted to standard normally distributed random numbers using the Box-Muller transformation algorithm; the standard normally distributed random numbers are scaled to ensure that the variance meets the requirement of 1 divided by the adaptive observation dimension; each element position of the zero matrix is ​​sequentially filled until all positions are filled; the filled matrix is ​​orthogonalized and verified; if the verification fails, it is regenerated until a dynamic Gaussian random observation matrix satisfying the restricted isometry property is obtained.

[0081] Specifically, the register of the True Random Number Generator (TRNG) integrated within the microcontroller or FPGA is read. This generator typically uses thermal noise or oscillator jitter as an entropy source to obtain a 32-bit unsigned integer as the initial random seed. This seed is written into the status register of the pseudo-random number generator to initialize the sequence. Two independent random numbers uniformly distributed in the interval (0, 1) are generated using the linear congruential method or the Mason tween rotation algorithm. and The calculation is performed using the Box-Muller transformation formula, i.e. and ,in It is the natural logarithm. Pi (value 3.1415926). and These are cosine and sine trigonometric functions, respectively, which transform the uniform distribution into random numbers from a standard normal distribution. The currently calculated adaptive observation dimension is then read. Calculate the scaling factor Multiply the generated standard normal random numbers by the scaling factor to make their variance satisfy According to the requirements, the scaled random numbers are written into the previously initialized zero matrix one by one in row priority or column priority order. After each row or column is filled, an orthogonality check is performed to calculate the inner product between the row vectors of the matrix and determine whether its off-diagonal elements are close to zero, or to calculate the limiting equidistant constant (RIP) of the matrix. If the preset error range is not met, the current matrix is ​​discarded and a new sequence is generated by calling the seed again until the matrix meets the incoherence requirements required by compressed sensing. Finally, the matrix is ​​output as the dynamic Gaussian random observation matrix.

[0082] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An adaptive power quality monitoring method for power cabinets, characterized in that, Includes the following steps: The low-frequency acquisition unit set at the power cabinet monitoring node is activated to acquire the low-frequency coarse sweep voltage signal of the power cabinet in the current monitoring period at a preset low-frequency sampling rate, and the time-domain fluctuation characteristics of the low-frequency coarse sweep voltage signal are extracted. The adaptive observation dimension for the current monitoring period is calculated based on the time-domain fluctuation characteristics. A dynamic Gaussian random observation matrix is ​​constructed based on the adaptive observation dimension, wherein the number of rows of the dynamic Gaussian random observation matrix is ​​equal to the adaptive observation dimension, and the number of columns of the dynamic Gaussian random observation matrix is ​​equal to the preset high-frequency reconstruction length. The dynamic Gaussian random observation matrix is ​​used to configure the compressed sampling circuit of the analog front end to perform compressed sensing analog information conversion on the real-time analog voltage signal of the power cabinet to obtain a compressed observation vector, wherein the dimension of the compressed observation vector is consistent with the adaptive observation dimension. The compressed observation vector is transmitted to the digital signal processing unit, and the orthogonal matching pursuit algorithm is used to perform sparse basis inversion calculation on the compressed observation vector to reconstruct high-precision power quality time-domain waveform data. The high-precision power quality time-domain waveform data is subjected to waveform distortion feature analysis to identify the power quality disturbance type of the power cabinet, generate a power quality monitoring and analysis report and upload it to the monitoring terminal; The steps of activating the low-frequency acquisition unit set at the power cabinet monitoring node, acquiring the low-frequency coarse-scan voltage signal of the power cabinet in the current monitoring period at a preset low-frequency sampling rate, and extracting the time-domain fluctuation characteristics of the low-frequency coarse-scan voltage signal are as follows: Initialize the analog-to-digital converter inside the low-frequency acquisition unit, and set the sampling frequency of the analog-to-digital converter to a preset multiple of the power frequency; The analog-to-digital converter is used to poll and sample the three-phase voltage ports of the power cabinet to obtain a discrete voltage sampling sequence within a preset time window, and the discrete voltage sampling sequence is marked as the low-frequency coarse sweep voltage signal. Calculate the rate of change of voltage amplitude and the total energy value of voltage waveform of the low-frequency coarse sweep voltage signal within the preset time window; The voltage amplitude change rate and the total energy value are combined to form the time-domain fluctuation characteristic.

2. The adaptive power quality monitoring method for power cabinets according to claim 1, characterized in that, The steps for calculating the adaptive observation dimension for the current monitoring period based on the time-domain fluctuation characteristics, and constructing a dynamic Gaussian random observation matrix based on the adaptive observation dimension, are as follows: Read the voltage amplitude change rate and total energy value of the voltage waveform contained in the time-domain fluctuation characteristics, and obtain the pre-stored rated voltage reference value, reference energy value and reference voltage amplitude change rate of the power cabinet; The adaptive observation dimension is calculated by processing the voltage amplitude change rate and the total energy value using an adaptive dimension calculation model. The calculation formula in the adaptive dimension calculation model is as follows: ； in, This represents the adaptive observation dimension. This represents the preset basic observation dimension. This indicates the floor function. This represents the energy weighting coefficient. This represents the ratio of the total energy value to the reference energy value. This represents the total energy value. This represents the reference energy value. Represents the sensitivity coefficient to the rate of change. Represents the logarithmic function with base 10. This represents the absolute value of the rate of change of the voltage amplitude. This represents the rate of change of the reference voltage amplitude; Initialize a zero matrix with the number of rows equal to the adaptive observation dimension and the number of columns equal to the preset high-frequency reconstruction length; A Gaussian random number sequence is generated using a Gaussian distribution random number generator, and the Gaussian random number sequence is filled into the zero matrix to obtain the dynamic Gaussian random observation matrix.

3. The adaptive power quality monitoring method for power cabinets according to claim 1, characterized in that, The specific steps for configuring the compressed sampling circuit of the analog front end using the dynamic Gaussian random observation matrix to perform compressed sensing analog information conversion on the real-time analog voltage signal of the power cabinet and obtain the compressed observation vector are as follows: The dynamic Gaussian random observation matrix is ​​loaded into the random demodulator controller of the analog front end; The random demodulator controller uses the timing logic of the dynamic Gaussian random observation matrix to control the switching state of the mixer in the compressed sampling circuit; The real-time analog voltage signal of the power cabinet is multiplied by the pseudo-random sequence through the mixer to obtain the mixed analog modulated signal, wherein the pseudo-random sequence is obtained by the dynamic Gaussian random observation matrix. The integrator is controlled to perform integration of the analog modulation signal for a preset time period to obtain the integrated analog voltage value; At the end of each integration cycle, a low-speed analog-to-digital converter is triggered to quantize the integrated analog voltage value to obtain a digitized observation value. The observation values ​​are then arranged in chronological order to form the compressed observation vector.

4. The adaptive power quality monitoring method for power cabinets according to claim 1, characterized in that, The specific steps for transmitting the compressed observation vector to the digital signal processing unit and using the orthogonal matching pursuit algorithm to perform sparse basis inversion calculation on the compressed observation vector to reconstruct high-precision power quality time-domain waveform data are as follows: The compressed observation vector is input into the computing memory of the digital signal processing unit, and a preset discrete cosine transform sparse basis matrix is ​​loaded. Initialize the residual vector to the compressed observation vector, initialize the index set to an empty set, and initialize the iteration counter; The sensing matrix is ​​obtained by multiplying the dynamic Gaussian random observation matrix with the discrete cosine transform sparse basis matrix. In each iteration, the inner product correlation coefficient between each column of the sensing matrix and the current residual vector is calculated, and the column index corresponding to the column with the largest absolute value of the inner product correlation coefficient is found. Add the column index to the index set, and extract the corresponding sub-matrix from the sensing matrix according to the index set; The approximate solution of the submatrix and the compressed observation vector is obtained by using the least squares method to obtain the current sparse coefficient estimate; Update the residual vector and determine whether the norm of the residual vector is less than a preset reconstruction stopping threshold. If it is less, stop the iteration; otherwise, continue to the next iteration. The high-precision power quality time-domain waveform data is synthesized by using the final sparse coefficient estimate and the discrete cosine transform sparse basis matrix for inverse transformation.

5. The adaptive power quality monitoring method for power cabinets according to claim 4, characterized in that, In each iteration, the specific steps for calculating the inner product correlation coefficient between each column of the sensing matrix and the current residual vector, and finding the column index corresponding to the column with the largest absolute value of the inner product correlation coefficient, are as follows: Normalize each column vector of the sensing matrix to obtain a set of normalized sensing column vectors; Calculate the dot product between each vector in the normalized sensing column vector set and the residual vector to obtain the correlation coefficient vector; Traverse the correlation coefficient vector and locate the position of the maximum value in the correlation coefficient vector through comparison operations; The column number of the matrix corresponding to the position of the maximum value is determined as the column index.

6. The adaptive power quality monitoring method for power cabinets according to claim 1, characterized in that, The specific steps for performing waveform distortion feature analysis on the high-precision power quality time-domain waveform data, identifying the power quality disturbance type of the power cabinet, generating a power quality monitoring and analysis report, and uploading it to the monitoring terminal are as follows: The high-precision power quality time-domain waveform data is subjected to Fourier transform to obtain power quality spectrum data; Calculate the amplitude of the fundamental component and the amplitude of the harmonic component in the power quality spectrum data, and calculate the total harmonic distortion rate based on the amplitude of the fundamental component and the amplitude of the harmonic component. Detect voltage sag characteristics, voltage swell characteristics, and voltage interruption characteristics in the high-precision power quality time-domain waveform data; If the total harmonic distortion rate exceeds the preset harmonic threshold, the power quality disturbance type is determined to be harmonic pollution. If the voltage sag characteristic is detected, the power quality disturbance type is determined to be a voltage sag; If the voltage sag characteristic is detected, the power quality disturbance type is determined to be a voltage sag; If the voltage interruption feature is detected, the power quality disturbance type is determined to be a voltage interruption; The determined power quality disturbance type, total harmonic distortion rate, and timestamp of the disturbance occurrence are encapsulated into a standard format data packet and used as the power quality monitoring and analysis report. The power quality monitoring and analysis report is sent to the remote monitoring terminal via an industrial Ethernet interface.

7. The adaptive power quality monitoring method for power cabinets according to claim 2, characterized in that, The specific steps for generating a Gaussian random number sequence using a Gaussian distribution random number generator and filling the zero matrix with the Gaussian random number sequence to obtain the dynamic Gaussian random observation matrix are as follows: The hardware random number generator is invoked to generate an initial random seed, and the initial random seed is injected into the Gaussian distribution random number generator. The Box-Muller transformation algorithm is used to convert uniformly distributed random numbers into standard normally distributed random numbers. The standard normal distribution random numbers are scaled to make the variance meet the numerical requirement of 1 divided by the adaptive observation dimension; Fill each element position of the zero matrix sequentially until all positions are filled; The filled matrix is ​​orthogonalized and verified. If the verification fails, it is regenerated until the dynamic Gaussian random observation matrix that satisfies the restricted isometry property is obtained.

Images