Harmonic detection in power systems using an improved compressed sensing model
The improved compressed sensing model addresses the inefficiencies of conventional harmonic detection by using a sparse random measurement matrix and windowing technique for precise harmonic parameter estimation, reducing resource needs and enhancing power quality monitoring.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- HANGZHOU INNOVATION RES INST OF BEIJING UNIV OF AERONAUTICS & ASTRONAUTICS
- Filing Date
- 2024-11-12
- Publication Date
- 2026-05-21
AI Technical Summary
Conventional harmonic detection techniques in power systems require high sampling rates, leading to excessive data storage and communication bandwidth needs, and are inadequate for complex harmonic components in smart grids and high-frequency power switching devices.
An improved compressed sensing model utilizing a sparse random measurement matrix and windowing technique for harmonic signal compressive sampling, combined with fixed-point continuation and double spectrum line interpolation for accurate parameter estimation below the Nyquist rate.
Enables efficient harmonic detection with reduced hardware and storage requirements, while providing precise estimation of harmonic parameters without reconstructing the original signal, thus improving power quality monitoring.
Smart Images

Figure CN2024131477_21052026_PF_FP_ABST
Abstract
Description
Harmonic Detection in Power Systems Using an Improved Compressed Sensing ModelTECHNICALFIELD
[0001] The present invention relates to the field of power system signal processing, and more particularly to a harmonic detection method using an improved compressed sensing model in power systems.BACKGROUND
[0002] Increased power demands in recent years have led to significant changes in grid structures and loads, making power quality problems more common. Impact and nonlinear loads produce serious harmonic pollution in power systems, resulting in the continuous deterioration of power quality. Harmonics (sinusoidal components with frequencies that are integer multiples of the fundamental frequency) are one of the primary causes of power quality issues and can reduce system capacity, lead to relay misoperations, affect power metering, and even cause electrical equipment fires. The detection and analysis of harmonics is thus critical for quality monitoring in power systems. Conventional detection techniques are based on Nyquist sampling theory in which the input signal sampling rate must be at least twice the value of the highest frequency. The sampling process begins with the collection of harmonic data at a high sampling rate through analog-to-digital (A / D) conversion, which produces large amounts of data requiring large storage space and a high communication bandwidth. In addition, recent developments in smart grids and high-frequency power switching electronic devices have made harmonic components more complex, which requires a higher sampling frequency in the Nyquist framework. Such developments continue to increase the need for hardware and storage space resources.SUMMARY
[0003] Compressed sensing (CS) , proposed by Candes and Donoho in 2004, is a signal processing technique that overcomes the limitations of Nyquist sampling theory. In this process, a signal with a sparse representation in a transform domain can be projected into a low-dimensional space by a measurement matrix that is not correlated with the transform basis. The signal can then be recovered with high probability using a nonlinear reconstruction algorithm. The present application relates to a harmonic detection approach using an improved compressed sensing model, which can perform harmonic signal compressive sampling and parameter estimation below the Nyquist rate. A sparse basis is typically selected, due to the sparseness of harmonic signals in the frequency domain, to serve as the discrete Fourier transform (DFT) basis. We combined a windowing technique with a sparse random measurement matrix in the sampling process, which facilitated simultaneous compression sampling and signal windowing. A fixed-point continuation method was then applied in the recovery process to estimate the sparse coefficients, from which the signal parameters were obtained directly. The harmonic parameters were corrected using a double spectrum line interpolation technique, in which the arithmetic expressions of the fundamental and harmonic frequencies, amplitudes, and phases were deduced by polynomial curve fitting. Windowing and double spectrum line interpolation techniques were used to suppress the impact of spectrum leakage and fence effect.BRIEF DESCRIPTION OF THE DRAWINGS
[0004] FIG. 1 is the implementation of the improved CS sampling process.
[0005] FIG. 2 is a schematic diagram of signal compression sampling.
[0006] FIG. 3 is a flowchart of the harmonic detection method using the improved CS model.DETAILED DESCRIPTION
[0007] An N×1 sparse signal in an orthogonal basis Ψ can be expressed as
[0008] Where α= [α1, α2, …αN] is a sparse coefficient vector obtained from the orthogonal basis Ψ.
[0009] The signal x K-sparse in the Ψ domain if there are K non-zero or larger coefficients in α. The compressed sampling process can then be represented as:
[0010] y=Φx (2)
[0011] where y is an M×1 vector of signal measurements, M<N, and Φ is a measurement matrix.
[0012] By combining Eqs. (1) and (2) , y can be expressed as:
[0013] y=Φx=Aα (3)
[0014] Where A=ΦΨ is a sensing matrix.
[0015] A sparse signal representation x in the basis Ψ can be acquired from:
[0016] minα||α||0 s.t. y=Aα (4)
[0017] The formulation above defines an ill-posed problem with an infinite solution space, a result of the rectangular shape of matrix A. However, this system can be approximately solved with a greedy pursuit algorithm, such as matching pursuit or orthogonal matching pursuit. In addition, it has been shown that Eq. (4) can be solved by approximating the l0-norm term as an l1-norm:
[0018] minα||α||1 s.t. y=Aα (5)
[0019] This optimization problem can then be solved using convex relaxation methods.
[0020] The application of CS theory assumes a signal to be sparse. If the signal x exhibits a sparse representation in the basis Ψ, it can be accurately recovered from a measurement vector y using the CS reconstruction algorithm. The selection of an appropriate sparse basis is therefore a significant component of CS theory.
[0021] Harmonic signals in power systems can be represented mathematically as:
[0022] where A, f, and φh denote the amplitude, frequency, and phase, respectively. DFT basis is chosen as the sparse basis, due to the sparseness of harmonic signals in the frequency domain.
[0023] In the sampling process, an N×1 signal x can be projected into a low-dimensional space using a measurement matrix, to obtain an M×1 vector y. A sparse random measurement matrix was used as μM elements were randomly set to 1 in each column vector, with all other elements set to 0. The term μ<<1 denotes the sparse ratio of the measurement matrix. In this present, μM is set to 4, indicating there are only 4 non-zero values in each column vector of the sparse random measurement matrix. Since the sparse random measurement matrix consists of binary terms (0 or 1) , it can be represented by simple switching hardware circuits or a switch transistor.
[0024] In practical power grids, the frequency of harmonic is prone to minor oscillations around the power frequency, which poses difficulties in achieving synchronized sampling and integer period truncation when using FFT for harmonic analysis. To suppress the effects of spectrum leakage and fence effect, an improved CS model was developed by introducing a windowing technique in the sampling process. The included Hanning window can be defined as:
[0025] where wH is a 1×N vector. Hanning windows offer the best universality and are especially suitable to signals with multiple frequency components. A window function matrix W can then be constructed as:
[0026]
[0027] The improved CS model is given by:
[0028] where denotes the Hadamard product (i.e., element-by-element multiplication) . The term αin the improved CS model is equivalent to the sparse coefficients of the windowed signal xw=xwH, with respect to the basis Ψ. In other words, the model is capable of implementing compression sampling and signal windowing simultaneously. The random modulation preintegration (RMPI) architecture is applied to implement the signal compression sampling, as shown in Fig. 1.
[0029] In the recovery phase, the frequencies, amplitudes, and phases of the fundamental and harmonic components were calculated directly from the sparse coefficients obtained by the CS reconstruction algorithm. This was achieved without reconstructing the original harmonic signal, which can reduce the required runtime. Convex relaxation methods are thus capable of achieving higher reconstruction accuracy than greedy methods. A fixed-point continuation (FPC) technique was applied to solve the convex optimization problem defined in Eq. (10)
[0030] where M is a positive definite matrix, is the associated M-norm, and μ>0 is a regularization parameter. Unlike other convex optimization techniques, the FPC algorithm only requires vector operations and matrix-vector multiplication, without any linear system solutions or matrix factorization steps.
[0031] If in FPC, M must be explicitly constructed as:
[0032] where σ1 and σ2 denote the standard deviation of signal and measurement noise, respectively. The regularization parameter μ can be calculated as:
[0033] where is the 1-ν critical value for a chi-square distribution with M degrees of freedom, and In addition, A and y should be substituted with the following quantities:
[0034] A forward-backward splitting technique and a continuation strategy can then be applied to solve the convex optimization problem represented by Eq. (10) , to estimate the sparse coefficient vector α. The frequencies, amplitudes, and phases of the fundamental and harmonic components can further be calculated from α.
[0035] The International Electrotechnical Commission (IEC protocol 61000-4-7) requires input signals to be expressed with a frequency resolution Δf of 5 Hz and the length of observation intervals Tw to exceed 10 periods. In other words, in a 50 Hz power system, Tw=1 / Δf=200 ms. Fig. 2 (a) shows a sample signal containing four harmonic components, with a fundamental frequency of 49.7 Hz. The signal measurements acquired from the measurement matrix Φ are shown in Fig. 2 (b) , in which M is set to 200 and N is selected to be 1024. Fig. 2 (c) shows the sparse coefficients estimated by the FPC algorithm. The amplitude spectrum for the input signal is shown in Fig. 2 (e) .
[0036] It is evident that each peak corresponds to a harmonic component in the signal, with the maximum peak representing the fundamental. Extracting harmonic parameters directly from the peaks may lead to inaccurate results. As such, we adopted a double spectrum line interpolation algorithm to correct the estimation of frequency, amplitude, and phase for fundamental and harmonic components. Spectral lines on the left and right sides of the peak are denoted as k1 and k2, respectively, representing the maximum and second-maximum spectral lines near the peak. Arithmetic expressions of the fundamental and harmonic frequencies, amplitudes, and phases were deduced by a polynomial curve fit:
[0037] where y1 is the amplitude of the k1-th spectral line, y2 is the amplitude of the k2-th spectral line, and δ=1.5 (y2-y1) / (y2+y1) . The value of i in the phase interpolation formula can be either 1 or 2, indicating the estimation of phase involves a single peak spectrum correction algorithm. The workflow for harmonic detection using the improved CS model is shown in Fig. 3 and involves compression sampling, the recovery of sparse coefficients, and interpolation correction.
Claims
1.An improved CS model for harmonic detection in power systems, characterized by comprising:A sampling module for performing compressed sampling on the harmonic signals in the power system, combining the window function technique to achieve simultaneous compressed sampling and signal windowing;A recovery module for estimating the sparse coefficients through the FPC technique and directly calculating the frequencies, amplitudes, and phases of the fundamental and harmonic components from the sparse coefficients;A correction module for correcting the estimation of the frequencies, amplitudes, and phases of the fundamental and harmonic components using a double spectrum line interpolation algorithm.2.According to the improved compressed sensing model described in claim 1, characterized in that a sparse random measurement matrix is used in the sampling module to project the signal into a low-dimensional space to obtain the measurement vector.3.According to the improved compressed sensing model described in claim 1, characterized in that the recovery module solves the convex optimization problem through the convex relaxation method to estimate the sparse coefficients.4.According to the improved compressed sensing model described in claim 1, characterized in that the correction module deduces the arithmetic expressions of the fundamental and harmonic frequencies, amplitudes, andphases through polynomial curve fitting.5.A method for harmonic detection in power systems, characterized by using the improved compressed sensing model described in any one ofclaims 1-4, including the following steps:Compressed sampling: performing compressed sampling on the harmonic signals in the power system while achieving signal windowing;Recovery of sparse coefficients: estimating the sparse coefficients through the FPC technique;Interpolation correction: using the double spectrum line interpolation algorithm to correct the parameters of the fundamental and harmonic components.6.According to the method for harmonic detection in power systems described in claim 5, characterized in that the Hanning window is used as the window function in the compressed sampling process.7.According to the method for harmonic detection in power systems described in claim 5, characterized in that in the process of recovering the sparse coefficients, the FPC algorithm only requires vector operations and matrix-vector multiplication, without any linear system solution or matrix factorization steps.8.According to the method for harmonic detection in power systems described in claim 5, characterized in that in the interpolation correction process, the double spectrum line interpolation algorithm is used to correct the inaccurate results that may be caused by directly extracting the harmonic parameters from the peaks.