Power distribution cabinet electric energy quality monitoring system based on internet of things communication

Abstract

This invention relates to the field of electrical variable measurement technology and discloses a power quality monitoring system for distribution cabinets based on Internet of Things (IoT) communication. The system includes edge acquisition nodes for the distribution cabinet, an IoT communication link, and a cloud-based reconstruction platform. The edge acquisition nodes acquire voltage and current signals, separate the steady-state fundamental signal to generate a sparse basis matrix, and use a pseudo-random matrix to compress and sample the original mixed signal to generate a low-dimensional measurement vector. The low-dimensional measurement vector is combined with the feature parameters of the steady-state fundamental signal to form a multi-mode data packet, which is then transmitted via the IoT communication link. The cloud-based reconstruction platform extracts the feature parameters to reconstruct the sparse basis matrix and combines it with the low-dimensional measurement vector to execute an orthogonal matching pursuit algorithm to reconstruct the high-frequency power quality disturbance waveform. This invention reduces the amount of high-frequency sampled data transmitted in the IoT link, overcomes the narrowband communication bandwidth problem, preserves the complete waveform characteristics of high-frequency disturbances, and avoids power quality characteristic distortion.

Description

Technical Field

[0001] This invention relates to the field of electrical variable measurement technology and discloses a power quality monitoring system for distribution cabinets based on Internet of Things (IoT) communication. Background Technology

[0002] Existing power quality monitoring systems for distribution cabinets based on IoT communication typically deploy edge acquisition terminals within the distribution cabinet to perform high-frequency sampling of the grid's voltage and current signals, capturing disturbances such as voltage drops and transient harmonics. After acquiring the raw mixed signal, the acquisition terminal directly packages the full sampled data and sends it to a cloud server via a narrowband IoT communication link for subsequent analysis. Due to the massive amount of data generated by high-frequency sampling and the limited bandwidth of the IoT communication link, when the full high-frequency data exceeds the link's capacity, existing systems generally employ strategies such as reducing the sampling rate or actively discarding some data packets to maintain uninterrupted communication and complete the routine reporting of power quality data from the distribution cabinet.

[0003] Based on the aforementioned existing technologies, reducing the sampling rate makes it impossible for the acquisition terminal to meet the requirement of complete sampling of high-frequency transient signals, while discarding data packets directly disrupts the continuity of high-frequency disturbances in the time series. Both of these conventional strategies for dealing with bandwidth limitations result in the cloud server receiving data with missing waveform features and distortion when high-frequency transient disturbances occur in the distribution cabinet, making it impossible to reconstruct the true evolution of power quality disturbances based on the incomplete data. Summary of the Invention

[0004] The purpose of this invention is to provide a power quality monitoring system for distribution cabinets based on Internet of Things (IoT) communication, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A power quality monitoring system for distribution cabinets based on Internet of Things (IoT) communication includes edge acquisition nodes of distribution cabinets, IoT communication links, and a cloud-based reconfiguration platform.

[0007] The edge acquisition node of the power distribution cabinet acquires the voltage and current signals of the power distribution cabinet, separates the steady-state fundamental signal from the voltage and current signals, generates a sparse basis matrix using the steady-state fundamental signal as a priori condition, compresses and samples the original mixed signal composed of the voltage and current signals using a pseudo-random matrix to generate a low-dimensional measurement vector, and combines the low-dimensional measurement vector with the feature parameters of the steady-state fundamental signal to form a multi-mode data packet, which is then sent via the Internet of Things communication link.

[0008] The cloud-based reconstruction platform receives the multimodal data packets, extracts the feature parameters of the steady-state fundamental wave signal to reconstruct the dynamically updated sparse basis matrix, and performs an orthogonal matching pursuit algorithm in conjunction with the low-dimensional measurement vector to restore the original high-frequency power quality disturbance waveform of the distribution cabinet.

[0009] Preferably, the specific steps for the distribution cabinet edge acquisition node to separate the steady-state fundamental wave signal and generate a sparse basis matrix include: applying a sliding discrete Fourier transform to the voltage and current signals to extract the amplitude and phase sequence corresponding to the power frequency component as the steady-state fundamental wave signal;

[0010] Based on the amplitude and phase sequence of the steady-state fundamental wave signal, a fundamental wave atom library containing the power frequency component is constructed;

[0011] In the fundamental wave atom library, the steady-state fundamental wave signal is periodically extended and scaled at a preset time step to generate multiple fundamental wave derived atoms;

[0012] The fundamental wave-derived atoms are arranged in time index order to construct the column orthogonal sparse basis matrix, and the amplitude, phase, and time step of the steady-state fundamental wave signal are marked as the characteristic parameters.

[0013] Preferably, the process of the power distribution cabinet edge acquisition node generating a low-dimensional measurement vector using a pseudo-random matrix includes: the power distribution cabinet edge acquisition node has a built-in physical noise generation circuit, which collects environmental thermal noise to generate a real random number sequence, and uses the real random number sequence as a seed to generate the pseudo-random matrix based on the Bernoulli distribution mapping;

[0014] The pseudo-random matrix is ​​multiplied by the original mixed signal, wherein the number of rows of the pseudo-random matrix is ​​set to a preset compression ratio of the number of columns of the original mixed signal. The result of the operation is output as a low-dimensional measurement vector with reduced dimension, which retains the projection coefficient characteristics of the original mixed signal under the sparse basis matrix.

[0015] Preferably, the multimodal data packet combination and encapsulation process includes: the power distribution cabinet edge acquisition node dividing the data frame into a fixed length, wherein the data frame sequentially includes a frame header identifier, a timestamp synchronization area, a load area, and a frame tail check area;

[0016] Write the local clock count of the voltage and current signals received by the edge acquisition node of the power distribution cabinet into the timestamp synchronization area;

[0017] The load area is divided into a first storage block and a second storage block. The real part data and imaginary part data of the low-dimensional measurement vector are interleaved and stored in the first storage block. The characteristic parameters of the steady-state fundamental wave signal are converted into fixed-point numbers and then stored in the second storage block.

[0018] Cyclic redundancy check (CRC) calculations are performed on the load area, and the checksum is written into the frame tail check area.

[0019] Preferably, the step of the cloud reconstruction platform reconstructing the dynamically updated sparse basis matrix includes: the cloud reconstruction platform parsing the multimodal data packet and reading the amplitude, phase and time step from the characteristic parameters of the steady-state fundamental wave signal;

[0020] In the general fundamental wave atom library pre-stored locally on the cloud reconstruction platform, the atom matching and filtering are performed based on the read amplitude and phase to retrieve the corresponding initial fundamental wave atom;

[0021] According to the read time step, the initial fundamental atoms are subjected to the same periodic extension and scaling calculations as the edge acquisition node of the power distribution cabinet. The derived atoms generated by the calculation are rearranged to construct the dynamically updated sparse basis matrix that is consistent with the internal structure of the edge acquisition node of the power distribution cabinet.

[0022] Preferably, the process of the cloud-based reconstruction platform executing the orthogonal matching tracking algorithm includes: initializing the residual vector as the low-dimensional measurement vector and initializing the support set as an empty set;

[0023] Calculate the inner product between the residual vector and each column vector in the dynamically updated sparse basis matrix, and select the column vector index with the largest absolute value of the inner product to add to the support set;

[0024] Based on the column vectors contained in the support set, extract the submatrix of the dynamically updated sparse basis matrix, and calculate the sparse coefficient vector under the current residual using the least squares method.

[0025] The current estimated signal is calculated using the sparse coefficient vector and the submatrix. The difference between the low-dimensional measurement vector and the current estimated signal is updated to the residual vector. The iteration stops when the L2 norm of the residual vector is lower than a preset threshold, and the sparse coefficient vector is output.

[0026] Preferably, when the edge acquisition node of the power distribution cabinet performs the sliding discrete Fourier transform, it calculates the frequency deviation between the power frequency components extracted in two adjacent sliding windows in real time.

[0027] When the frequency deviation value exceeds the preset frequency deviation range, it is determined that the steady-state fundamental signal has experienced frequency drift.

[0028] The edge acquisition node of the power distribution cabinet is triggered to stop using the fundamental wave atom library containing power frequency components and switch to calling the locally stored cosine envelope atom library as the alternative orthogonal basis dictionary;

[0029] The fundamental wave-derived atoms are replaced with standard atoms from the cosine envelope atom library, the column orthogonality is recalculated, the replaced sparse basis matrix is ​​generated, and the frequency drift flag and the index number of the cosine envelope atom library are appended to the feature parameters.

[0030] Preferably, the process of updating the pseudo-random matrix after the edge acquisition node of the power distribution cabinet completes the output of a low-dimensional measurement vector includes: reading the last random value output by the physical noise generation circuit at the end of the current matrix multiplication operation;

[0031] Using the last random value as the shift step, perform a cyclic left shift operation on the real random number sequence to obtain a new seed sequence after shifting.

[0032] The Bernoulli distribution-based mapping generation operation is re-executed using the new seed sequence, overwriting the pseudo-random matrix in the current memory;

[0033] When the voltage and current signals are input in the next time period, the updated pseudo-random matrix is ​​directly called to participate in the matrix multiplication operation, and the shift step size is appended to the tail of the low-dimensional measurement vector for transmission.

[0034] Preferably, before storing the low-dimensional measurement vector into the first storage block, the edge acquisition node of the power distribution cabinet performs adaptive load segment division: scanning each element in the low-dimensional measurement vector and counting the number and position index of non-zero elements whose absolute value is greater than a preset zero threshold;

[0035] The current sparsity ratio is calculated based on the ratio of the number of non-zero elements to the total number of elements in the low-dimensional measurement vector.

[0036] When the sparsity ratio is lower than a preset lower threshold, the first storage block is divided into multiple non-contiguous sub-blocks. Only the non-zero elements and their corresponding position indices are written into the sub-blocks. The positions of the remaining elements whose absolute values ​​are less than or equal to the preset zero threshold are left empty, and the checksum of the frame tail check area is updated.

[0037] Preferably, when the cloud-based reconstruction platform selects the column vector index with the largest absolute inner product value to add to the support set, it performs a backtracking elimination calculation: extracting multiple candidate column vector indices with the highest-ranking absolute inner product value in the current iteration;

[0038] Calculate the correlation matrix among the candidate column vectors corresponding to the multiple candidate column vector indices, and extract the correlation coefficients off-diagonal in the correlation matrix;

[0039] When there are any two candidate column vectors whose correlation coefficient is greater than the preset correlation threshold, compare the absolute value of the inner product of the two candidate column vectors, remove the index of the candidate column vector with the smaller absolute value of the inner product, add the index of the retained candidate column vector to the support set, and then perform the least squares calculation.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] 1. This invention generates a sparse basis matrix by separating the steady-state fundamental wave signal from the voltage and current signals as a priori condition, and uses a pseudo-random matrix to compress and sample the original mixed signal to generate a low-dimensional measurement vector. This transforms high-dimensional data into a multi-mode data packet combining the low-dimensional measurement vector and feature parameters for transmission. The cloud reconstructs the sparse basis matrix based on the feature parameters and uses an orthogonal matching pursuit algorithm to reconstruct the waveform using the low-dimensional measurement vector. This processing logic reduces the amount of high-frequency sampled data transmitted in the IoT link, overcoming the transmission problem of high-frequency power quality monitoring data in narrowband IoT communication bandwidth. Simultaneously, the sparse basis matrix constructed based on the steady-state fundamental wave prior condition retains the complete waveform characteristics of high-frequency disturbances such as voltage drops and transient harmonics, avoiding power quality characteristic distortion caused by data compression or discarding.

[0042] 2. This invention extracts the power frequency component using sliding discrete Fourier transform and generates a sparse basis matrix through periodic extension and scaling, thus clarifying the physical foundation of compressed sensing. A real random number sequence is generated using a physical noise generation circuit and mapped to a pseudo-random matrix. This matrix is ​​updated after each operation by shifting the step size, increasing the randomness of compressed sampling. When encapsulating multimodal data packets, the load segment division is adaptively adjusted according to the sparsity ratio, writing only non-zero elements and position indices to the storage segment, reducing redundant load on data frames. During the execution of the orthogonal matching pursuit algorithm in the cloud, the correlation matrix is ​​calculated for candidate column vectors. Column vectors with excessive correlation coefficients are eliminated through backtracking, avoiding the introduction of highly correlated redundant atoms in the support set. When the frequency deviation of the power frequency component in adjacent windows exceeds the interval, the cosine envelope atom library is switched to replace the fundamental frequency derived atoms to regenerate the sparse basis matrix, overcoming the sparse decomposition mismatch problem caused by fundamental frequency drift and ensuring the convergence stability of the iterative solution process. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating the overall workflow of the present invention;

[0044] Figure 2 This is a flowchart of the steady-state fundamental signal separation and sparse basis matrix generation process of the present invention;

[0045] Figure 3 This is a flowchart of the pseudo-random matrix generation and compressed sampling process of the present invention;

[0046] Figure 4 This is a flowchart of the multimodal data packet encapsulation process of the present invention;

[0047] Figure 5 This is a flowchart of the cloud-based dynamic sparse basis matrix reconstruction process of the present invention;

[0048] Figure 6 This is a flowchart of the orthogonal matching tracking algorithm with backtracking elimination in this invention. Detailed Implementation

[0049] Please refer to the attached document. Figure 1 This embodiment provides a power quality monitoring system for distribution cabinets based on Internet of Things (IoT) communication. Edge acquisition nodes are deployed at the electrical circuit sampling end of the distribution cabinet, establishing electrical connections with the voltage and current sampling units within the cabinet to acquire real-time voltage and current signals during cabinet operation. These voltage and current signals are discrete-time series, containing steady-state power frequency components and transient disturbance components during grid operation, forming a raw mixed signal. The edge acquisition nodes preprocess the acquired voltage and current signals, separating the steady-state fundamental signal. This steady-state fundamental signal corresponds to the disturbance-free component at the grid's rated power frequency, reflecting the fundamental steady-state characteristics of the grid operation.

[0050] The edge acquisition node of the distribution cabinet uses the separated steady-state fundamental signal as a priori condition to generate a sparse basis matrix. The sparse basis matrix is ​​a column-orthogonal square matrix, and the orthogonal basis dictionary formed by its column vectors can sparsely represent the original mixed signal, so that the original mixed signal can be mapped to a sparse vector containing a small number of non-zero coefficients under the sparse basis matrix. The non-zero coefficients correspond to the transient disturbance components in the original mixed signal, and the zero coefficients correspond to the steady-state fundamental component, thereby realizing the sparse decomposition of the original mixed signal.

[0051] Furthermore, the edge acquisition node of the distribution cabinet uses a pseudo-random matrix to compress and sample the original mixed signal composed of the voltage and current signals, generating a low-dimensional measurement vector. The pseudo-random matrix is ​​a measurement matrix with a dimension lower than the length of the original mixed signal, and it satisfies the incoherence constraint with the sparse basis matrix, ensuring that the core features of the original mixed signal are not lost during the compression sampling process. The compression sampling process is implemented through matrix multiplication, mapping the high-dimensional original mixed signal to a low-dimensional measurement vector, reducing the dimensionality of the data to be transmitted.

[0052] The edge acquisition node of the power distribution cabinet combines the low-dimensional measurement vector with the characteristic parameters of the steady-state fundamental signal to form a multimodal data packet. The characteristic parameters are a set of parameters that uniquely determine the sparse basis matrix structure, including the core features of the steady-state fundamental signal and the generation rules parameters of the sparse basis matrix. The encapsulation process of the multimodal data packet follows a preset frame structure specification to ensure synchronization and integrity during data transmission. The encapsulated multimodal data packet is sent to the cloud reconstruction platform via an IoT communication link. This IoT communication link is a wide-area communication network that supports bidirectional data transmission, enabling data interaction between the edge and the cloud.

[0053] The cloud-based reconstruction platform continuously monitors downlink data in the IoT communication link and receives multimodal data packets sent by the edge acquisition nodes of the power distribution cabinet. The platform parses and verifies the received multimodal data packets. Upon successful verification, it extracts the characteristic parameters of the steady-state fundamental wave signal within the data packet and reconstructs a dynamically updated sparse basis matrix based on these parameters. The reconstructed sparse basis matrix has a completely identical structure and column vector content to the sparse basis matrix generated internally by the edge acquisition nodes of the power distribution cabinet, ensuring complete synchronization between the sparse basis dictionaries on the cloud and the edge.

[0054] Furthermore, the cloud-based reconstruction platform combines the reconstructed sparse basis matrix with the analytically obtained low-dimensional measurement vector, and executes an orthogonal matching pursuit algorithm to solve for the sparse coefficient vector of the original mixed signal under the sparse basis matrix. Based on the sparse coefficient vector and the sparse basis matrix, the original high-frequency power quality disturbance waveform of the distribution cabinet is reconstructed. The reconstructed waveform has consistent time series characteristics and disturbance component characteristics with the original mixed signal acquired by the edge acquisition nodes of the distribution cabinet, and fully preserves the waveform details of high-frequency disturbances such as voltage drops and transient harmonics.

[0055] Please refer to the attached document. Figure 4In the encapsulation process of multimodal data packets, the edge acquisition node of the power distribution cabinet divides data frames of fixed length. Each data frame sequentially includes a frame header identifier, a timestamp synchronization area, a load area, and a frame tail check area. The local clock count of the voltage and current signals received by the edge acquisition node is written into the timestamp synchronization area. This local clock count is the output value of a counter driven by the local crystal oscillator of the edge acquisition node, used for time synchronization between the cloud and the edge. The load area is divided into a first storage block and a second storage block. The real and imaginary parts of the low-dimensional measurement vector are interleaved and stored in the first storage block. This interleaved storage reduces the impact of sudden errors on continuous data segments during data transmission. The characteristic parameters of the steady-state fundamental signal are converted into fixed-point numbers and stored in the second storage block. The fixed-point number conversion process follows a preset quantization bit width rule to ensure the transmission accuracy of the characteristic parameters. Cyclic redundancy check (CRC) calculation is performed on all data in the load area, and the calculated checksum is written into the frame tail check area for integrity verification of the data packets by the cloud reconstruction platform.

[0056] Table 1. Frame structure definition table for multimodal data packets:

[0057] Frame structure segment Length (bytes) Data types Storage content description Frame header identifier 2 Unsigned integer A fixed synchronization code is used for frame boundary identification of data packets. The receiving end uses the synchronization code to complete frame synchronization locking. Timestamp synchronization area 6 Unsigned integer The local clock count at the moment when the voltage and current signals are acquired by the edge acquisition node of the power distribution cabinet is measured in local clock cycles. Load Area - First Storage Block variable fixed point number The real and imaginary parts of a low-dimensional measurement vector are interleaved and stored sequentially, with the data length adjusted according to the dimension of the measurement vector. Load Area - Second Storage Block fixed fixed point number The characteristic parameters of the steady-state fundamental wave signal, such as amplitude, phase, and time step, include the generation rule parameters of the sparse basis matrix. Frame end check area 4 Unsigned integer A 32-bit cyclic redundancy check (CRC) code is used for all data in the payload area to verify data integrity at the receiving end.

[0058] The table above defines the complete frame structure of a multimodal data packet. The length and storage content of each data segment follow fixed rules. The length of the first storage block is dynamically adjusted according to the dimension of the low-dimensional measurement vector, while the lengths of the remaining segments remain fixed, ensuring that the receiving end can complete the parsing of each data segment based on a fixed offset.

[0059] In this embodiment, when the cloud-based reconstruction platform parses the received multimodal data packets, it first completes frame synchronization based on the frame header identifier, extracts the local clock count from the timestamp synchronization area, and completes time synchronization alignment with the edge end. Then, it reads all the data in the payload area and performs cyclic redundancy check on the payload data based on the checksum in the frame tail check area. Data packets that fail the check are discarded, while those that pass the check are further parsed into the contents of the first and second storage blocks. When parsing the first storage block, the real and imaginary parts of the low-dimensional measurement vector are separated according to the cross-storage rule, and the complete low-dimensional measurement vector is reconstructed. When parsing the second storage block, the fixed-point numbers are restored to floating-point values ​​of the feature parameters according to the preset quantization bit width rule, obtaining all the parameters used to reconstruct the sparse basis matrix.

[0060] In this embodiment, when the cloud-based reconstruction platform executes the orthogonal matching pursuit algorithm to reconstruct the original waveform, it first initializes the residual vector to the low-dimensional measurement vector obtained analytically and initializes the support set to an empty set. Then, it proceeds in an iterative process, calculating the inner product of the current residual vector with each column vector in the reconstructed sparse basis matrix, and selecting the column vector index with the largest absolute value of the inner product to add to the support set. Based on the column vector indices contained in the support set, the corresponding column vectors in the sparse basis matrix are extracted to form a submatrix, and the sparse coefficient vector under the current residual is calculated using the least squares method.

[0061] The calculated sparse coefficient vector and submatrix are used to calculate the current estimated signal, and the difference between the low-dimensional measurement vector and the current estimated signal is updated as the residual vector. After each iteration, it is determined whether the L2 norm of the residual vector is lower than a preset threshold. When the L2 norm of the residual vector is lower than the preset threshold, the iteration stops, and the final sparse coefficient vector is output. Based on the output sparse coefficient vector and the reconstructed sparse basis matrix, matrix multiplication is performed to restore the original mixed signal. The transient components in the original mixed signal are extracted to obtain the original high-frequency power quality disturbance waveform of the distribution cabinet.

[0062] This embodiment fully realizes the entire process of edge compression sampling, IoT link transmission, and cloud waveform reconstruction of power quality signals from distribution cabinets. By constructing a sparse basis matrix using the steady-state fundamental signal as a priori condition, and combining compression sampling technology, the high-dimensional original signal is transformed into a low-dimensional measurement vector and a small number of feature parameters for transmission. This reduces the amount of high-frequency sampled data transmitted in the IoT link, and overcomes the problem of transmitting high-frequency power quality monitoring data in the narrowband communication bandwidth of the IoT. At the same time, the sparse basis matrix constructed based on the steady-state fundamental priori condition retains the complete waveform characteristics of high-frequency disturbances such as voltage drops and transient harmonics, avoiding power quality characteristic distortion caused by data compression or discarding.

[0063] Please refer to the attached document. Figure 2 In a preferred embodiment, the technical implementation process of separating the steady-state fundamental signal and generating a sparse basis matrix at the edge acquisition node of the distribution cabinet is further refined. The edge acquisition node of the distribution cabinet applies a sliding discrete Fourier transform to the acquired voltage and current signals. The operation process of the sliding discrete Fourier transform is defined by the following formula:

[0064]

[0065] in, Let be the discrete Fourier transform coefficients of the k-th harmonic of the m-th sliding window. Let m be the discrete voltage and current signal sequence within the m-th sliding window. The length of the sliding window. The sliding step size, For the time index within the window, For harmonic order, It is the imaginary unit.

[0066] The length of the sliding window is an integer multiple of the power grid frequency period, ensuring that the sliding discrete Fourier transform can accurately extract the characteristics of the power frequency component. The edge acquisition node of the distribution cabinet extracts the discrete Fourier transform coefficients of the power frequency component corresponding to k=1, and calculates the amplitude and phase sequence corresponding to the power frequency component based on the transform coefficients. These amplitude and phase sequences constitute the steady-state fundamental signal. The discrete-time expression of the steady-state fundamental signal is defined by the following formula:

[0067]

[0068] in, For steady-state fundamental frequency signal, This represents the amplitude of the power frequency component. The rated power frequency, The sampling period is This represents the initial phase of the power frequency component. For time indexing.

[0069] Furthermore, the edge acquisition node of the distribution cabinet constructs a fundamental wave atom library containing the power frequency component, based on the amplitude and phase sequence of the steady-state fundamental wave signal. The basic atoms in the fundamental wave atom library are cosine signals with the same frequency, amplitude, and phase as the steady-state fundamental wave signal. The discrete-time expression of the fundamental wave atom is defined by the following formula:

[0070]

[0071] in, For fundamental wave atoms, This is the time delay parameter, corresponding to the time index of the periodic extension. This is the scaling factor; the remaining parameters are consistent with the definition of the steady-state fundamental signal.

[0072] In the fundamental wave atom library, the edge acquisition node of the distribution cabinet performs periodic extension and scaling on the steady-state fundamental wave signal at a preset time step, generating multiple fundamental wave derived atoms. The periodic extension process adjusts the time delay parameter. This is achieved by assigning a time delay parameter to each time step, generating fundamental atoms with different time offsets. The scaling process involves adjusting the scaling factor. This allows for the generation of fundamental-wave derived atoms at different frequency scales, covering the harmonic frequency range of the power frequency component.

[0073] The edge acquisition nodes of the distribution cabinet arrange all the generated fundamental-wave derived atoms in time index order to construct a column-orthogonal sparse basis matrix. Column orthogonalization is achieved using the Gram-Schmidt orthogonalization method, defined by the following formula:

[0074]

[0075]

[0076] in, Let i be the i-th column vector of the sparse basis matrix. This is an intermediate vector in the orthogonalization process. For inner product operations, This refers to the 2-norm operation.

[0077] After orthogonalization, the edge acquisition node of the power distribution cabinet marks the amplitude, phase and time step of the steady-state fundamental wave signal as the characteristic parameters. The characteristic parameters can uniquely determine the generation rules of the fundamental wave atom and the structure of the sparse basis matrix.

[0078] Table 2. Comparison of core parameters between the fundamental wave atomic library and the cosine envelope atomic library:

[0079] Parameters Fundamental wave atomic library Cosine envelope atomic library Core mathematical expressions A cosine signal based on the power frequency component, containing a fixed frequency term. The cosine envelope signal modulated by a Gaussian window contains an adjustable frequency term. Applicable Scenarios Steady-state operation scenario with stable power frequency and no frequency drift. The power grid frequency shifts, resulting in transient scenarios of frequency drift. Atomic generation benchmark The amplitude, phase, and frequency of the steady-state fundamental signal extracted in real time. Locally pre-stored standard cosine envelope template, supporting frequency offset adaptation Orthogonalization Gram-Schmidt column orthogonalization based on time index order Based on the intrinsic orthogonal basis of the discrete cosine transform, no real-time orthogonalization is required. Feature parameter dimension 3D (amplitude, phase, time step) 4-dimensional (envelope width, center frequency, time step, index number) Frequency drift adaptation capability Only supports a very small range of frequency offsets near the rated power frequency. Supports a wide range of frequency offsets, adapting to frequency drift scenarios.

[0080] The table above compares the core parameter differences between the fundamental frequency atomic library and the cosine envelope atomic library, clarifying the applicable scenarios and generation rules of the two atomic libraries, and providing a clear parameter basis for switching atomic libraries. Furthermore, when performing a sliding discrete Fourier transform, the edge acquisition node of the distribution cabinet calculates the frequency deviation between the extracted power frequency components within two adjacent sliding windows in real time. The calculation process of the frequency deviation is defined by the following formula:

[0081]

[0082] in, This is the frequency deviation value. Let m be the phase of the power frequency component in the sliding window. The phase of the power frequency component in the (m-1)th sliding window is given, and the remaining parameters are consistent with the definition of the sliding discrete Fourier transform.

[0083] When the frequency deviation exceeds the preset frequency deviation range, the edge acquisition node of the distribution cabinet determines that the steady-state fundamental signal has experienced frequency drift. At this time, the edge acquisition node stops using the fundamental atom library containing the power frequency component and switches to a locally pre-stored cosine envelope atom library as a candidate orthogonal basis dictionary. The standard atoms in the cosine envelope atom library are cosine envelope signals modulated by Gaussian windows, which have a wide range of frequency adaptation capabilities and can maintain the accuracy of sparse decomposition when the power frequency drifts.

[0084] The edge acquisition node of the power distribution cabinet replaces the fundamental wave derived atoms with standard atoms from the cosine envelope atom library, recalculates the column orthogonality relationship based on the inherent orthogonality of the standard atoms, and generates the replaced sparse basis matrix. Simultaneously, the edge acquisition node appends the frequency drift flag and the index number of the cosine envelope atom library to the feature parameters, enabling the cloud-based reconstruction platform to synchronously reconstruct the sparse basis matrix based on the appended feature parameters.

[0085] This embodiment refines the extraction process of the steady-state fundamental signal and the generation process of the sparse basis matrix. The real-time and accurate extraction of the power frequency component is achieved through sliding discrete Fourier transform. The fundamental atom library constructed with the steady-state fundamental signal as a priori condition can achieve the optimal sparse representation of the original mixed signal. At the same time, a switching mechanism for the cosine envelope atom library is set up for the scenario of power grid frequency drift, which overcomes the sparse decomposition mismatch problem caused by fundamental frequency drift and ensures the accuracy of sparse representation and the precision of subsequent waveform reconstruction.

[0086] Please refer to the attached document. Figure 3 In a preferred embodiment, the process of generating low-dimensional measurement vectors using a pseudo-random matrix at the distribution cabinet edge acquisition node, and the adaptive encapsulation process of multimodal data packets, are further refined. The distribution cabinet edge acquisition node incorporates a physical noise generation circuit. This circuit generates random electrical signals based on the thermal noise effect of semiconductor devices. After analog-to-digital conversion of the random electrical signals, a true random number sequence corresponding to the environmental thermal noise is acquired. This true random number sequence is unpredictable and can provide a highly random seed for the generation of the pseudo-random matrix.

[0087] The edge acquisition node of the distribution cabinet uses the real random number sequence as a seed to generate the pseudo-random matrix based on the Bernoulli distribution mapping. The element generation process of the pseudo-random matrix is ​​defined as follows:

[0088] The probability is 0.5. The probability is 0.5;

[0089] in, For pseudo-random matrices The element in the m-th row and n-th column, with matrix dimension . , , The length of the original mixed signal, is the length of the low-dimensional measurement vector.

[0090] The number of rows in the pseudo-random matrix is ​​set to a preset compression ratio value equal to the number of columns in the original mixed signal. The compression ratio is determined based on the bandwidth capacity of the IoT communication link and the sparsity of the original mixed signal. The pseudo-random matrix and the sparse basis matrix satisfy the incoherence constraint, ensuring that the sparse characteristics of the original mixed signal are completely preserved in the low-dimensional measurement vector during the compression sampling process. The edge acquisition node of the distribution cabinet performs matrix multiplication on the pseudo-random matrix and the original mixed signal. The operation process is defined by the following formula:

[0091]

[0092] in, A low-dimensional measurement vector with dimension . , The matrix is ​​a pseudo-random measurement matrix with dimension 1. , The original mixed signal vector has a dimension of .

[0093] The result of the matrix multiplication operation is the reduced-dimensional measurement vector, which retains the projection coefficient characteristics of the original mixed signal under the sparse basis matrix. The original mixed signal can be restored by the reconstructed sparse basis matrix and the reconstruction algorithm.

[0094] Furthermore, after completing the output of a low-dimensional measurement vector, the edge acquisition node of the distribution cabinet performs a pseudo-random matrix update process. The edge acquisition node reads the last random value output by the physical noise generation circuit at the end of the current matrix multiplication operation, uses this last random value as the shift step size, and performs a cyclic left shift operation on the real random number sequence to obtain a new seed sequence after shifting. The cyclic shift operation is defined by the following formula:

[0095]

[0096] in, The new seed sequence after shifting. It is the original real random number sequence. This is a circular left shift operation. The shift step size is the last random value output by the physical noise generation circuit at the end of the operation.

[0097] The distribution cabinet edge acquisition node re-executes the Bernoulli distribution-based mapping generation operation using the new seed sequence to generate a new pseudo-random matrix, overwriting the original pseudo-random matrix in memory. When the voltage and current signals are input in the next time period, the distribution cabinet edge acquisition node directly calls the updated pseudo-random matrix to participate in matrix multiplication operations. Simultaneously, the shift step size is appended to the tail of the low-dimensional measurement vector for transmission, enabling the cloud reconstruction platform to synchronously update the pseudo-random matrix based on the shift step size, ensuring complete consistency between the measurement matrices at the cloud and edge.

[0098] Furthermore, before storing the low-dimensional measurement vector into the first storage block, the edge acquisition node of the distribution cabinet performs adaptive load segmentation. The edge acquisition node scans each element in the low-dimensional measurement vector, counting the number and position index of non-zero elements whose absolute value is greater than a preset zero threshold. The preset zero threshold is determined based on the system's quantization noise level and is used to filter out minute values ​​corresponding to quantization noise in the low-dimensional measurement vector.

[0099] The edge acquisition node of the distribution cabinet calculates the current sparsity ratio based on the ratio of the number of non-zero elements to the total number of elements in the low-dimensional measurement vector. The calculation process of the sparsity ratio is defined by the following formula:

[0100]

[0101] in, The sparsity ratio, This represents the number of non-zero elements in a low-dimensional measurement vector whose absolute value is greater than a preset zero threshold. This represents the total number of elements in the low-dimensional measurement vector.

[0102] When the sparsity ratio is lower than a preset lower threshold, the distribution cabinet edge acquisition node divides the first storage block into multiple non-contiguous sub-blocks, writes only the non-zero elements and their corresponding position indices into the sub-blocks, and sets the positions of elements whose absolute values ​​are less than or equal to the preset zero threshold to empty, without storing redundant data corresponding to zero values. After completing the data writing, the distribution cabinet edge acquisition node recalculates the cyclic redundancy check (CRC) for the load area and updates the checksum of the frame tail check area. The calculation process of the CRC is defined by the following formula:

[0103]

[0104] in, The cyclic redundancy check code corresponding to the data in the load area. The binary polynomial corresponding to the data in the load area. For the preset generator polynomial, For polynomial shift terms, This is a modulo 2 division operation.

[0105] Table 3. Load block partitioning parameters under different sparsity ratios:

[0106] Sparsity ratio range First storage block partitioning method Number of sub-blocks Stored content Single-frame load redundancy ρ≥ preset lower threshold Continuous complete blocks 1 Low-dimensional measurement vector full data No redundancy 1 / 2 of the preset lower threshold ≤ ρ < preset lower threshold Dual-block partitioning 2 Non-zero element data blocks, position index data blocks low redundancy ρ < 1 / 2 of the preset lower threshold value Discrete partitioning of multiple sub-blocks variable Non-zero elements are paired with their corresponding position indices. Extremely low redundancy

[0107] The table above defines the rules for partitioning the first storage block under different sparsity ratios. The storage method is dynamically adjusted according to the changes in the sparsity ratio, so as to minimize the amount of redundant data in the load area while ensuring data integrity.

[0108] This embodiment refines the generation and dynamic update mechanism of the pseudo-random matrix. Using a real random number sequence generated from physical thermal noise as a seed, the randomness and anti-interference capability of the pseudo-random matrix are improved. The dynamic update mechanism avoids the problem of accumulated reconstruction errors caused by a fixed measurement matrix. Simultaneously, through adaptive load segmentation, the storage method is dynamically adjusted according to the sparsity of the low-dimensional measurement vector, eliminating redundant data corresponding to zero values, further reducing the load length of multimodal data packets and adapting to the bandwidth constraints of narrowband IoT communication links.

[0109] Please refer to the attached document. Figure 5 In a preferred embodiment, the process of reconstructing the dynamically updated sparse basis matrix on the cloud-based reconstruction platform and the optimized execution process of the orthogonal matching pursuit algorithm are further refined. After the cloud-based reconstruction platform completes the parsing and verification of the multimodal data packets, it reads the characteristic parameters of the steady-state fundamental wave signal within the multimodal data packets and extracts the amplitude, phase, and time step from the characteristic parameters. When the characteristic parameters contain a frequency drift flag and an index number of the cosine envelope atom library, the corresponding flag and index number parameters are extracted simultaneously.

[0110] The cloud-based reconfiguration platform has a locally pre-stored universal fundamental wave atom library and cosine envelope atom library that are completely identical to those of the edge acquisition nodes of the power distribution cabinet. The generation rules and atom parameters of the atom library are fully synchronized with the edge nodes. The cloud-based reconfiguration platform uses the locally pre-stored universal fundamental wave atom library to perform atom matching and filtering based on the read amplitude and phase parameters, retrieving the initial fundamental wave atom that perfectly matches the amplitude and phase parameters. When the characteristic parameters contain a frequency drift flag, the cloud-based reconfiguration platform retrieves the standard cosine envelope atom with the corresponding index number as the initial atom based on the index number of the cosine envelope atom library.

[0111] Furthermore, the cloud-based reconstruction platform performs the same periodic extension and scaling calculations on the initial fundamental atoms according to the read time step, identical to those performed on the edge acquisition node of the distribution cabinet. The time delay parameters of the periodic extension and the scaling factor of the scaling are completely consistent with the calculation rules at the edge. The derived atoms generated by the calculation have the exact same parameters and discrete time series as the fundamental derived atoms generated at the edge. The cloud-based reconstruction platform rearranges the generated derived atoms according to the same time index order as at the edge, performs column orthogonalization processing identical to that at the edge, and constructs the dynamically updated sparse basis matrix that is consistent with the internal structure of the edge acquisition node of the distribution cabinet. The reconstructed sparse basis matrix has the exact same dimension, column vector content, and orthogonality as the sparse basis matrix at the edge, ensuring complete synchronization between the cloud-based sparse decomposition and the edge-based compressed sampling basis dictionary.

[0112] Please refer to the attached document. Figure 6 During the execution of the orthogonal matching tracking algorithm on the cloud-based reconstruction platform, the initialization operation is first completed. The residual vector is initialized to the low-dimensional measurement vector, and the support set is initialized to an empty set. The initialization process is defined by the following formula:

[0113]

[0114]

[0115] in, The initial residual vector, For the initial support set, It is an empty set.

[0116] After initialization, the iterative computation process begins. The inner product of the residual vector and each column vector in the dynamically updated sparse basis matrix is ​​calculated. The column vector index with the largest absolute value of the inner product is then added to the support set. The inner product calculation and support set update process is defined by the following formula:

[0117]

[0118]

[0119] in, Let be the absolute value of the inner product of the residual and the i-th column vector. Let be the residual vector of the (t-1)th iteration. Let be the support set after the t-th iteration update. For the number of iterations, It is a pseudo-random matrix. Let be the i-th column vector of the sparse basis matrix.

[0120] Furthermore, the cloud-based reconstruction platform extracts the corresponding column vectors from the dynamically updated sparse basis matrix based on the column vector indices contained in the support set, forming a submatrix. The sparse coefficient vector under the current residual is calculated using the least squares method, the solution process of which is defined by the following formula:

[0121]

[0122] in, Let be the sparse coefficient vector obtained in the t-th iteration. Based on the support set Extracted sparse basis matrix submatrices, superscript For matrix transpose operation, superscript To perform the matrix inversion operation, Let be the sparse coefficient vector to be solved.

[0123] The cloud-based reconstruction platform uses the sparse coefficient vector and the submatrix to calculate the current estimated signal, and updates the residual vector with the difference between the low-dimensional measurement vector and the current estimated signal. The residual update process is defined by the following formula:

[0124]

[0125] in, Let be the residual vector after the t-th iteration update.

[0126] After each iteration, the cloud-based reconstruction platform determines whether the L2 norm of the residual vector is lower than a preset threshold. When the L2 norm of the residual vector is lower than the preset threshold, the iteration stops, and the final sparse coefficient vector is output. Based on the output sparse coefficient vector and the reconstructed sparse basis matrix, matrix multiplication is performed to restore the original mixed signal. The transient disturbance component in the original mixed signal is extracted to obtain the original high-frequency power quality disturbance waveform of the distribution cabinet.

[0127] Furthermore, when selecting the column vector index with the largest absolute inner product to add to the support set, the cloud-based reconstruction platform performs a backtracking elimination calculation. The cloud-based reconstruction platform extracts multiple candidate column vector indices with the highest absolute inner product values ​​in the current iteration, with the preset number of digits determined based on the dimension of the sparse basis matrix and the sparsity of the original mixed signal. Based on the candidate column vector indices, the corresponding candidate column vectors in the sparse basis matrix are extracted to form candidate submatrices. The correlation matrix between the multiple candidate column vectors corresponding to the multiple candidate column vector indices is calculated, and the calculation process of the correlation matrix is ​​defined by the following formula:

[0128]

[0129] in, The correlation matrix, The submatrix formed by the candidate column vectors, with superscripts This is the matrix transpose operation.

[0130] The cloud-based reconstruction platform extracts the correlation coefficients from the off-diagonal lines of the correlation matrix. These correlation coefficients reflect the correlation between two candidate column vectors; the closer the correlation coefficient is to 1, the higher the similarity between the two column vectors. Introducing a support set can lead to redundant atoms. When there are any two candidate column vectors with a correlation coefficient greater than a preset correlation threshold, the cloud-based reconstruction platform compares the absolute value of the inner product of these two candidate column vectors, removes the index of the candidate column vector with the smaller absolute value, adds the remaining candidate column vector index to the support set, and then performs the least squares calculation.

[0131] Table 4. Parameter Changes During the Iterative Process of the Orthogonal Matching Pursuit Algorithm:

[0132] Number of iterations Maximum inner product Number of elements in the support set residual norm Does it trigger backtracking removal? Remove column vector indexes 1 0.92 1 0.38 no none 2 0.87 2 0.21 no none 3 0.79 3 0.15 yes 127 4 0.65 3 0.08 no none 5 0.42 4 0.03 no none Iteration stopped - 4 0.03 - -

[0133] The table above records the parameter changes during the iterative process of the Orthogonal Matching Pursuit algorithm. The third iteration triggered a backtracking elimination operation, removing candidate column vector index 127, which was highly correlated with existing column vectors in the support set. This avoided the introduction of redundant atoms, accelerated the decrease in the residual L2 norm, and reduced the number of iterations. This embodiment refines the sparse basis matrix reconstruction process of the cloud-based reconstruction platform. By using atom generation rules and orthogonalization processing completely consistent with the edge platform, it ensures complete synchronization of the sparse basis matrices between the cloud and edge platforms, eliminating reconstruction errors caused by mismatched base dictionaries. Simultaneously, the Orthogonal Matching Pursuit algorithm is optimized through backtracking elimination calculations. Highly correlated redundant candidate column vectors are removed during the support set update process, avoiding the introduction of highly correlated redundant atoms into the support set. This improves the accuracy of sparse coefficient solving and the convergence stability of the iterative process, ensuring the accuracy of high-frequency power quality disturbance waveform reconstruction.

Claims

1. A power quality monitoring system for distribution cabinets based on Internet of Things (IoT) communication, characterized in that, This includes edge data acquisition nodes for power distribution cabinets, IoT communication links, and cloud-based reconfiguration platforms. The edge acquisition node of the power distribution cabinet acquires the voltage and current signals of the power distribution cabinet, separates the steady-state fundamental signal from the voltage and current signals, generates a sparse basis matrix using the steady-state fundamental signal as a priori condition, compresses and samples the original mixed signal composed of the voltage and current signals using a pseudo-random matrix to generate a low-dimensional measurement vector, and combines the low-dimensional measurement vector with the feature parameters of the steady-state fundamental signal to form a multi-mode data packet, which is then sent via the Internet of Things communication link. The cloud-based reconstruction platform receives the multimodal data packets, extracts the feature parameters of the steady-state fundamental wave signal to reconstruct the dynamically updated sparse basis matrix, and performs an orthogonal matching pursuit algorithm in conjunction with the low-dimensional measurement vector to restore the original high-frequency power quality disturbance waveform of the distribution cabinet.

2. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 1, characterized in that, The specific steps for the distribution cabinet edge acquisition node to separate the steady-state fundamental wave signal and generate a sparse basis matrix include: applying a sliding discrete Fourier transform to the voltage and current signals, and extracting the amplitude and phase sequence corresponding to the power frequency component as the steady-state fundamental wave signal; Based on the amplitude and phase sequence of the steady-state fundamental wave signal, a fundamental wave atom library containing the power frequency component is constructed; In the fundamental wave atom library, the steady-state fundamental wave signal is periodically extended and scaled at a preset time step to generate multiple fundamental wave derived atoms; The fundamental wave-derived atoms are arranged in time index order to construct the column orthogonal sparse basis matrix, and the amplitude, phase, and time step of the steady-state fundamental wave signal are marked as the characteristic parameters.

3. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 1, characterized in that, The process of generating a low-dimensional measurement vector using a pseudo-random matrix by the edge acquisition node of the power distribution cabinet includes: the edge acquisition node of the power distribution cabinet has a built-in physical noise generation circuit, which collects environmental thermal noise to generate a real random number sequence, and uses the real random number sequence as a seed to generate the pseudo-random matrix based on the Bernoulli distribution mapping. The pseudo-random matrix is ​​multiplied by the original mixed signal, wherein the number of rows of the pseudo-random matrix is ​​set to a preset compression ratio of the number of columns of the original mixed signal. The result of the operation is output as a low-dimensional measurement vector with reduced dimension, which retains the projection coefficient characteristics of the original mixed signal under the sparse basis matrix.

4. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 1, characterized in that, The multimodal data packet combination and encapsulation process includes: the power distribution cabinet edge acquisition node divides the data frame into a fixed length, and the data frame sequentially includes a frame header identifier, a timestamp synchronization area, a load area and a frame tail verification area; Write the local clock count of the voltage and current signals received by the edge acquisition node of the power distribution cabinet into the timestamp synchronization area; The load area is divided into a first storage block and a second storage block. The real part data and imaginary part data of the low-dimensional measurement vector are interleaved and stored in the first storage block. The characteristic parameters of the steady-state fundamental wave signal are converted into fixed-point numbers and then stored in the second storage block. Cyclic redundancy check (CRC) calculations are performed on the load area, and the checksum is written into the frame tail check area.

5. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 1, characterized in that, The steps of the cloud-based reconstruction platform to reconstruct the dynamically updated sparse basis matrix include: the cloud-based reconstruction platform parses the multimodal data packet and reads the amplitude, phase, and time step from the characteristic parameters of the steady-state fundamental wave signal; In the general fundamental wave atom library pre-stored locally on the cloud reconstruction platform, the atom matching and filtering are performed based on the read amplitude and phase to retrieve the corresponding initial fundamental wave atom; According to the read time step, the initial fundamental atoms are subjected to the same periodic extension and scaling calculations as the edge acquisition node of the power distribution cabinet. The derived atoms generated by the calculation are rearranged to construct the dynamically updated sparse basis matrix that is consistent with the internal structure of the edge acquisition node of the power distribution cabinet.

6. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 1, characterized in that, The process of the cloud-based reconstruction platform executing the orthogonal matching tracking algorithm includes: initializing the residual vector as the low-dimensional measurement vector and initializing the support set as an empty set; Calculate the inner product between the residual vector and each column vector in the dynamically updated sparse basis matrix, and select the column vector index with the largest absolute value of the inner product to add to the support set; Based on the column vectors contained in the support set, extract the submatrix of the dynamically updated sparse basis matrix, and calculate the sparse coefficient vector under the current residual using the least squares method. The current estimated signal is calculated using the sparse coefficient vector and the submatrix. The difference between the low-dimensional measurement vector and the current estimated signal is updated to the residual vector. The iteration stops when the L2 norm of the residual vector is lower than a preset threshold, and the sparse coefficient vector is output.

7. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 2, characterized in that, When performing a sliding discrete Fourier transform, the edge acquisition node of the power distribution cabinet calculates the frequency deviation between the power frequency components extracted in two adjacent sliding windows in real time. When the frequency deviation value exceeds the preset frequency deviation range, it is determined that the steady-state fundamental signal has experienced frequency drift. The edge acquisition node of the power distribution cabinet is triggered to stop using the fundamental wave atom library containing power frequency components and switch to calling the locally stored cosine envelope atom library as the alternative orthogonal basis dictionary; The fundamental wave-derived atoms are replaced with standard atoms from the cosine envelope atom library, the column orthogonality is recalculated, the replaced sparse basis matrix is ​​generated, and the frequency drift flag and the index number of the cosine envelope atom library are appended to the feature parameters.

8. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 3, characterized in that, After the edge acquisition node of the power distribution cabinet completes the output of a low-dimensional measurement vector, the process of updating the pseudo-random matrix includes: reading the last random value output by the physical noise generation circuit at the end of the current matrix multiplication operation; Using the last random value as the shift step, perform a cyclic left shift operation on the real random number sequence to obtain a new seed sequence after shifting. The Bernoulli distribution-based mapping generation operation is re-executed using the new seed sequence, overwriting the pseudo-random matrix in the current memory; When the voltage and current signals are input in the next time period, the updated pseudo-random matrix is ​​directly called to participate in the matrix multiplication operation, and the shift step size is appended to the tail of the low-dimensional measurement vector for transmission.

9. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 4, characterized in that, Before storing the low-dimensional measurement vector into the first storage block, the edge acquisition node of the power distribution cabinet performs adaptive load segmentation: scanning each element in the low-dimensional measurement vector and counting the number and position index of non-zero elements whose absolute value is greater than the preset zero threshold. The current sparsity ratio is calculated based on the ratio of the number of non-zero elements to the total number of elements in the low-dimensional measurement vector. When the sparsity ratio is lower than a preset lower threshold, the first storage block is divided into multiple non-contiguous sub-blocks. Only the non-zero elements and their corresponding position indices are written into the sub-blocks. The positions of the remaining elements whose absolute values ​​are less than or equal to the preset zero threshold are left empty, and the checksum of the frame tail check area is updated.

10. The power quality monitoring system for distribution cabinets based on Internet of Things communication according to claim 6, characterized in that, When the cloud-based reconstruction platform selects the column vector index with the largest absolute inner product value to add to the support set, it performs a backtracking elimination calculation: extracting multiple candidate column vector indices with the highest-ranking absolute inner product value in the current iteration; Calculate the correlation matrix among the candidate column vectors corresponding to the multiple candidate column vector indices, and extract the correlation coefficients off-diagonal in the correlation matrix; When there are any two candidate column vectors whose correlation coefficient is greater than the preset correlation threshold, compare the absolute value of the inner product of the two candidate column vectors, remove the index of the candidate column vector with the smaller absolute value of the inner product, add the index of the retained candidate column vector to the support set, and then perform the least squares calculation.

Images