Harmonic and inter-harmonic parameter detection method and device based on self-multiplication convolution window algorithm

Abstract

The invention discloses a harmonic and inter-harmonic parameter detection method and device based on a self-multiplication convolution window algorithm, and belongs to the field of signal detection, and the method comprises the steps: obtaining a discrete time domain signal, and carrying out the windowing and mapping of the time domain signal based on the self-multiplication convolution window algorithm, and obtaining a frequency domain signal; calling a quantum particle swarm algorithm to estimate the frequency domain signal to obtain an estimation parameter matched with a preset frequency spectrum; and performing nonlinear error compensation on the estimation parameter by using a preset space-time attention residual network to obtain harmonic and inter-harmonic parameters. According to the method, signal leakage can be suppressed by windowing the signal, the energy of the main lobe of the frequency spectrum is more concentrated, the frequency domain signal is estimated by using the quantum particle swarm algorithm, the evaluated frequency spectrum meeting the actual demand can be quickly screened, and finally, the frequency spectrum is compensated, so that the error can be further reduced, and the detection precision can be further improved.

Description

Technical Field

[0001] This invention relates to the technical field of signal detection, and in particular to a method and apparatus for detecting harmonic and interharmonic parameters based on a self-multiplying convolution window algorithm. Background Technology

[0002] With the continuous development of power systems, the integration of numerous new energy power generation and nonlinear power electronic equipment into the grid has led to a significant increase in the harmonic content and complexity of its components in the power grid. Harmonic problems not only pose a significant threat to the safe and stable operation of the power system, but also trigger a series of issues such as power quality degradation, equipment malfunctions, and failures. Accurately detecting harmonic signals and taking effective measures to suppress them can effectively reduce the impact of harmonics on the power system.

[0003] High-precision harmonic analysis is crucial for ensuring signal extraction from high-precision instruments such as electronic current transformers (ECTs), especially in complex signal source environments where rapid and accurate detection of harmonic and interharmonic components is critical. Currently, the commonly used method for harmonic and interharmonic parameter detection involves performing Fourier transform-based algorithms, followed by wavelet transform and frequency domain filtering, to finally obtain the harmonic and interharmonic parameters. Due to its simple principle, fast calculation speed, and ease of implementation in embedded systems, this method is also recommended by the IEC for harmonic and interharmonic detection.

[0004] However, the commonly used methods mentioned above have the following technical problems: the above techniques face the problem of spectrum leakage, especially when there are multiple signals with similar frequencies, one signal may be masked by another signal, which leads to a reduction in frequency resolution and also reduces the detection accuracy. Summary of the Invention

[0005] This invention provides a method and apparatus for detecting harmonic and interharmonic parameters based on a self-multiplying convolution window algorithm, which can solve the technical problem of spectrum leakage in the prior art, resulting in large detection signal errors.

[0006] A first aspect of this invention provides a method for detecting harmonic and interharmonic parameters based on a self-multiplying convolution window algorithm, the method comprising:

[0007] Discrete time-domain signals are obtained, and windowing and mapping of the time-domain signals are performed based on the self-multiplying convolution window algorithm to obtain frequency-domain signals;

[0008] The frequency domain signal is estimated by calling the quantum particle swarm optimization algorithm to obtain estimated parameters that match the preset spectrum;

[0009] The estimated parameters are compensated for nonlinear errors using a preset spatiotemporal attention residual network to obtain harmonic and interharmonic parameters.

[0010] This invention suppresses signal leakage by windowing the signal, making the main lobe energy of the spectrum more concentrated. Furthermore, by using the quantum particle swarm optimization algorithm to estimate the frequency domain signal, it can quickly screen the evaluation spectrum that meets the actual needs. Finally, by compensating the spectrum, errors can be further reduced and the detection accuracy can be improved.

[0011] In conjunction with the first aspect, in one implementation, the step of acquiring a discrete time-domain signal and windowing and mapping the time-domain signal based on a self-multiplicative convolutional window algorithm to obtain a frequency-domain signal includes:

[0012] The signal under test is sampled based on a preset sampling frequency and number of sampling points to obtain discrete time-domain signals;

[0013] The time-domain signal is windowed using a self-multiply fractal dynamic adaptive window function to obtain a frequency candidate set. The self-multiply fractal dynamic adaptive window function is a function constructed by performing a self-multiply operation on a preset basic window function based on a self-multiply convolution window algorithm to obtain a self-multiply convolution window.

[0014] The time-domain signal is mapped to a frequency-domain signal based on the frequency candidate set.

[0015] In conjunction with the first aspect, in one implementation, the step of windowing the time-domain signal using a self-multiplying fractal dynamic adaptive window function to obtain a frequency candidate set includes:

[0016] The time-domain signal is windowed using a self-multiplying fractal dynamic adaptive window function to obtain a suppressed signal.

[0017] Signal features are extracted from the time-domain signal using the preset Da Vinci oscillator equation to obtain dynamic response feature information;

[0018] Effective frequency signals with a synchronization index less than a preset index threshold are selected from the dynamic response feature information to obtain a frequency candidate set.

[0019] In conjunction with the first aspect, in one implementation, mapping the time-domain signal to a frequency-domain signal based on the frequency point candidate set includes:

[0020] The time-domain signal is mapped and Fourier transformed based on the frequency candidate set to obtain a frequency-domain signal with a complex frequency domain.

[0021] In conjunction with the first aspect, in one implementation, the step of invoking the quantum particle swarm optimization algorithm to estimate the frequency domain signal and obtain estimated parameters matching the measured spectrum includes:

[0022] The quantum particle swarm optimization algorithm is invoked to iteratively filter the spectral values ​​of the frequency candidate set, and several adjacent spectral values ​​are combined into a candidate parameter combination.

[0023] The matching degree between each candidate parameter combination and the preset actual spectrum is calculated using the potential energy function, and a target matching degree that meets the preset matching threshold is selected from multiple matching degrees. The candidate parameter combination corresponding to the target matching degree is used as the estimation parameter.

[0024] In conjunction with the first aspect, in one implementation, after the step of iteratively filtering the spectral values ​​of the frequency candidate set using the quantum particle swarm optimization algorithm and forming a candidate parameter combination from several adjacent spectral values, the method further includes:

[0025] Calculate the potential energy value of the particle corresponding to each of the candidate parameter combinations;

[0026] The position of the particle corresponding to each candidate parameter combination is updated based on the potential energy value.

[0027] In conjunction with the first aspect, in one implementation, after the step of using a preset spatiotemporal attention residual network to perform nonlinear error compensation on the estimated parameters to obtain the harmonic and interharmonic parameters, the method further includes:

[0028] Calculate the spectral energy entropy values ​​of the harmonic and interharmonic parameters;

[0029] If the spectral energy entropy value is greater than the preset entropy threshold, the historical time-domain signal of the previous moment is obtained, and the harmonic and interharmonic parameters are determined using the historical time-domain signal.

[0030] A second aspect of this invention provides a harmonic and interharmonic parameter detection device based on a self-multiplying convolution window algorithm, the device comprising:

[0031] A windowing module is used to acquire discrete time-domain signals and window and map the time-domain signals based on a self-multiplying convolution window algorithm to obtain frequency-domain signals.

[0032] The estimation module is used to call the quantum particle swarm optimization algorithm to estimate the frequency domain signal and obtain estimation parameters that match the preset spectrum.

[0033] The compensation detection module is used to perform nonlinear error compensation on the estimated parameters using a preset spatiotemporal attention residual network to obtain harmonic and interharmonic parameters.

[0034] Compared to existing technologies, the harmonic and interharmonic parameter detection method and apparatus based on the self-multiplicative convolutional window algorithm provided in this invention have the following advantages: This invention can acquire discrete time-domain signals and window and map these signals using the self-multiplicative convolutional window algorithm to obtain frequency-domain signals; it then uses the quantum particle swarm optimization algorithm to estimate the frequency-domain signals, obtaining estimated parameters that match a preset spectrum; finally, it uses a preset spatiotemporal attention residual network to perform nonlinear error compensation on the estimated parameters, obtaining the harmonic and interharmonic parameters. This invention, by windowing the signal, can suppress signal leakage and concentrate the main lobe energy of the spectrum. Furthermore, by using the quantum particle swarm optimization algorithm to estimate the frequency-domain signals, it can quickly screen evaluation spectra that meet actual needs. Finally, spectrum compensation can further reduce errors and improve detection accuracy. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating a method for detecting harmonic and interharmonic parameters based on a self-multiplying convolution window algorithm according to an embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram of a harmonic and interharmonic parameter detection device based on a self-multiplying convolution window algorithm provided in an embodiment of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] With the continuous development of power systems, the integration of numerous new energy power generation and nonlinear power electronic equipment into the grid has led to a significant increase in the harmonic content and complexity of its components in the power grid. Harmonic problems not only pose a significant threat to the safe and stable operation of the power system, but also trigger a series of issues such as power quality degradation, equipment malfunctions, and failures. Accurately detecting harmonic signals and taking effective measures to suppress them can effectively reduce the impact of harmonics on the power system.

[0039] High-precision harmonic analysis is crucial for ensuring signal extraction from high-precision instruments such as electronic current transformers (ECTs), especially in complex signal source environments where rapid and accurate detection of harmonic and interharmonic components is critical. Currently, the commonly used method for harmonic and interharmonic parameter detection involves performing Fourier transform-based algorithms, followed by wavelet transform and frequency domain filtering, to finally obtain the harmonic and interharmonic parameters. Due to its simple principle, fast calculation speed, and ease of implementation in embedded systems, this method is also recommended by the IEC for harmonic and interharmonic detection.

[0040] However, the commonly used methods mentioned above have the following technical problems: the above techniques face the problem of spectrum leakage, especially when there are multiple signals with similar frequencies, one signal may be masked by another signal, which leads to a reduction in frequency resolution and also reduces the detection accuracy.

[0041] To address the aforementioned issues, the following specific embodiments will provide a detailed description and explanation of a harmonic and interharmonic parameter detection method and apparatus based on a self-multiplicative convolutional window algorithm, as provided in this application.

[0042] To address the technical problem of spectral leakage in existing technologies, which leads to large detection signal errors, refer to... Figure 1 The diagram shows a flowchart of a harmonic and interharmonic parameter detection method based on a self-multiplying convolution window algorithm provided by an embodiment of the present invention.

[0043] As an example, the harmonic and interharmonic parameter detection method based on the self-multiplication convolution window algorithm may include:

[0044] S11. Obtain the discrete time-domain signal, and window and map the time-domain signal based on the self-multiplication convolution window algorithm to obtain the frequency-domain signal.

[0045] In one embodiment, the signal to be tested can be sampled to obtain a discrete time-domain signal, and then the time-domain signal can be windowed and mapped based on the self-multiplication convolution window algorithm to obtain a frequency-domain signal.

[0046] Windowing is applied to discrete time-domain signals to suppress signal leakage and concentrate the energy of the main lobe of the spectrum. Mapping discrete time-domain signals can convert them into complex spectra for subsequent processing.

[0047] As an example, the step of obtaining a discrete time-domain signal and windowing and mapping the time-domain signal based on a self-multiplying convolution window algorithm to obtain a frequency-domain signal may include the following sub-steps:

[0048] S111. Based on the preset sampling frequency and number of sampling points, the signal under test is sampled to obtain discrete time-domain signals.

[0049] S112. The time-domain signal is windowed using a self-multiply fractal dynamic adaptive window function to obtain a frequency candidate set. The self-multiply fractal dynamic adaptive window function is a function constructed by performing a self-multiply operation on a preset basic window function based on a self-multiply convolution window algorithm to obtain a self-multiply convolution window.

[0050] S113. Map the time-domain signal to a frequency-domain signal according to the frequency candidate set.

[0051] In one embodiment, the signal under test is sampled at an appropriate sampling frequency and number of sampling points to obtain a discrete time-domain signal. The sampling frequency must satisfy the Nyquist sampling theorem to avoid aliasing.

[0052] Next, a base window function can be selected, and a self-multiplication operation can be performed on the base window function to obtain a self-multiplicative convolution window. The self-multiplicative convolution window is combined with a fractal structure to obtain a self-multiplicative fractal dynamic adaptive window function.

[0053] In one operating mode, the Blackman-Harris window can be chosen as the base window function. The Blackman-Harris window has a low sidelobe level and a relatively narrow main lobe width, which allows it to more accurately distinguish signal components of different frequencies in harmonic and interharmonic parameter estimation, reducing the effects of spectral leakage and the picket fence effect. Moreover, the function curve is relatively smooth, and the truncation effect on the signal is relatively mild, which can preserve the detailed information of the signal and avoid introducing additional high-frequency components due to the discontinuity of the window function.

[0054] The specific formula for the basic window function can be:

[0055]

[0056] In the above formula, n represents the discrete time points in the window function; N represents the length of the window function, i.e., the number of discrete points contained in the window function, which determines the time span and frequency resolution of the window function; a0 is a constant term that determines the basic amplitude level of the window function; a1 controls the amplitude of the first harmonic cosine component in the window function, which plays an important role in adjusting the shape of the window function, and its value affects the characteristics of the sidelobes and main lobe; a2 controls the amplitude of the second harmonic cosine component, and works with other coefficients to optimize the frequency response of the window function and reduce spectral leakage; a3 controls the amplitude of the third harmonic cosine component, adjusts the fine shape of the window function, and plays a role in suppressing sidelobes and balancing the width of the main lobe; π is the mathematical constant pi.

[0057] Next, we can work on the basic window function w. b (n) Perform a self-multiplication operation to compress the dynamic range of the window function, resulting in a significant attenuation of sidelobe energy. A square window is generated by squaring point by point, where the square window can be represented by the following formula:

[0058] w sq(n)=[w sq (n)] 2 n = 0, 1, ..., N-1;

[0059] Adding a window can reduce spectral leakage and improve spectral resolution.

[0060] The square window is optimized by convolving it with itself, as shown in the following formula:

[0061] w conv (n)=w sq (n)*w sq (n) |0:N-1 ;

[0062] The convolution result can be obtained by truncating the first N points. In the above formula, * represents the convolution operation. The convolution operation is equivalent to the square of the amplitude response in the frequency domain in the time domain.

[0063] Convolution operations can further suppress sidelobes, while truncation operations prevent excessive broadening of the main lobe, effectively reducing the impact of sidelobes without sacrificing too much main lobe resolution. The dual operations of self-multiplication and convolution make sidelobe attenuation significantly better than traditional windows, effectively suppressing spectral leakage and improving the resolution of dense harmonics.

[0064] By combining a self-multiplying convolutional window with a fractal structure, a self-multiplying fractal dynamic adaptive window function is obtained. The specific formula for the self-multiplying fractal dynamic adaptive window function is as follows:

[0065]

[0066] In the above formula, x(n) represents the value of the original time-domain signal at discrete time point n. Introducing the hyperbolic tangent function tanh, the average amplitude and variance of the original time-domain signal within a local time window are used as the energy of the original signal. The window function w... conv (n) Perform dynamic adjustment of the window function shape to optimize w conv (n) The adaptive capability of the window function to the non-stationary characteristics of the signal. If the output value of tanh() is close to 0, then the original signal has low energy and is a stationary signal. conv (n) The window function is slightly compressed, the main lobe is wider, and high frequency resolution is maintained; when the output value of tanh() is close to 1, the original signal energy is high, with strong noise or transient impact. The window function is strongly compressed, the amplitude is as low as 0, the side lobes are further attenuated, and spectrum leakage is suppressed.

[0067] in, The adaptive compression coefficient is dynamically adjusted based on the local energy variance of the signal. This is the local window average amplitude term; ρ is the sensitivity factor; std(x(n)) represents the standard deviation of the signal x(n). The larger the standard deviation, the greater the deviation of the signal from the average value. Strong noise or transient interference will lead to an increase in the deviation. The smaller the standard deviation, the more stable the signal. ρ is the fractal weight coefficient; Hurst() is the exponent for calculating the local signal. By analyzing the growth rate of the signal variance with the time scale, its self-similarity and long-term correlation are judged, thereby deciding whether to trust the signal segment and retain its details. The larger the Hurst() value, the larger the fractal weight term, which enhances the amplitude of the window function and relatively reduces compression. The lower the Hurst(), the closer the fractal weight term is to 1, the less the amplitude of the window function is enhanced, and the window function remains in its original form or is compressed by the compression term. (nL:n+L) is the local window, which is obtained by taking L local window lengths before and after the current time point n. The fractal weight ρ is an adjustment factor that controls the degree of influence of the fractal structure on the window function and is directly proportional to the Hurst exponent value.

[0068] Windowing the acquired time-domain signal can suppress signal leakage and concentrate the energy of the main lobe of the spectrum. Next, a candidate frequency point set can be generated using a chaotic synchronization mechanism, and the time-domain signal can be mapped to a frequency-domain signal based on this candidate set.

[0069] High-sensitivity, noise-resistant frequency pre-detection can be achieved through chaotic synchronization mechanisms to generate a candidate set of frequency points.

[0070] In an optional embodiment, the step of windowing the time-domain signal using a self-multiplying fractal dynamic adaptive window function to obtain a frequency candidate set may include the following sub-steps:

[0071] S1121. The time-domain signal is windowed using a self-multiplying fractal dynamic adaptive window function to obtain a suppressed signal.

[0072] S1122. Extract signal features from the time-domain signal using the preset Da Vinci oscillator equation to obtain dynamic response feature information.

[0073] S1123. Select effective frequency signals with a synchronization index less than a preset index threshold from the dynamic response feature information to obtain a frequency candidate set.

[0074] The specific formula for adding a window is:

[0075] X win (n)=x(n)·w final (n), n=0, 1,..., N-1;

[0076] In the above formula, x(n) is the time-domain signal; w final (n) is the self-multiplying fractal dynamic adaptive window function; w win (n) represents the suppression signal.

[0077] Next, the Da Vinci oscillator equation can be used to detect the suppression signal based on the chaotic state of the oscillator, thereby extracting dynamic response feature information from the suppression signal.

[0078] The windowed suppressed signal is detected using the Da Vinci oscillator equation. The Da Vinci oscillator equation itself is insensitive to noise and interference, and can still process signals relatively accurately even in complex electromagnetic environments or with multiple interference sources. It can also capture the nonlinear characteristics of the signal and has a strong detection capability for weak signals. For the windowed suppressed signal, the complex characteristics of the signal can be described more precisely, and feature information that is difficult to obtain by conventional methods can be extracted. Therefore, useful signal features can be extracted from noisy backgrounds, further improving the signal-to-noise ratio.

[0079] The specific formula can be:

[0080]

[0081] The acceleration of the oscillator is the second derivative term. The oscillator is a nonlinear vibration system described by the Da Vinci equation. Its dynamic behavior can help identify and analyze the harmonic and interharmonic components in the input signal. X is the velocity of the oscillator and is the first derivative term; m X is the oscillator displacement, a state variable; γ is the damping coefficient, used to control the energy dissipation rate; 3 m The nonlinear restoring force term leads to the chaotic dynamics of the oscillator; ε is the coupling strength, which modulates the driving capability of the external signal on the oscillator; X min (n) represents the input discrete-time signal, i.e., the acquired discrete-time signal; Given the natural frequency of the oscillator, distributed according to the golden ratio, when the external signal X... min (n) contains frequencies close to f m When the component is [value], the oscillator transitions from a chaotic state to a periodic synchronous state; f base For reference frequency; The golden ratio is given by the interval f of the golden ratio distribution. m It can exceed the Rayleigh limit and detect dense harmonics with an interval of less than 0.3Δf; n represents the nth time step; Δt is the sampling time interval. f s The sampling rate.

[0082] Quantitative analysis is performed based on dynamic response characteristic information to calculate the synchronization index. Effective frequency signals with synchronization indices lower than a preset threshold are selected to obtain a candidate set of frequencies. By filtering out effective frequency components below the threshold, a simplified candidate set of frequencies can be formed.

[0083] Based on the dynamic response characteristics, a quantitative analysis is performed to calculate the synchronization index, using the following formula:

[0084]

[0085] In the above formula, where λ m is the detection index of the m-th oscillator, characterizing the stability of the trajectory. T is the observation time. The elements of the Jacobian matrix in the state space reflect the state of the system at state point x. m Local linear response characteristics at (t). For the small perturbation of the oscillator state, Let λ be the partial derivative of the oscillator velocity. m <-τ, where τ is a threshold; if τ is greater than 0, the disturbance decays, and the oscillator enters a stable periodic synchronization state, indicating that there is a frequency f in the signal. m Filter out all that satisfy λ m <-τ of f m This forms a concise candidate set of frequency points F. candidate This significantly reduces the amount of subsequent calculations; if λ m If the value is greater than 0, the disturbance diverges, the oscillator remains in a chaotic state, indicating that there is no target frequency component.

[0086] In terms of accuracy improvement, the self-multiplicative convolution window possesses unique spectral refinement capabilities, effectively suppressing spectral leakage and improving frequency resolution compared to traditional window functions. Combined with the self-similarity properties of fractal structures, it can accurately capture the complex characteristics of harmonic and interharmonic signals, improving the detection sensitivity of weak interharmonic components. The chaotic synchronization mechanism endows the system with strong adaptive capabilities, enabling it to quickly lock harmonic parameters even under complex operating conditions such as power grid frequency fluctuations and non-stationary signals, avoiding estimation errors caused by the time-varying nature of signals in traditional methods.

[0087] In an optional embodiment, mapping the time-domain signal to a frequency-domain signal based on the frequency candidate set may include the following sub-steps:

[0088] S1131. Perform a Fourier transform on the time-domain signal mapping based on the frequency candidate set to obtain a frequency-domain signal with a complex frequency domain.

[0089] In one embodiment, a Fourier transform can be performed on the time-domain signal based on a frequency candidate set to transform the time-domain signal X. min (n) is mapped to the frequency domain complex number X(k), providing basic data for harmonic analysis;

[0090] The specific formula for mapping is:

[0091]

[0092] In the above formula, X(k) is the complex representation of the k-th frequency component in the frequency domain, including the amplitude |X(k)| and the phase ∠X(k), and X(k) = |X(k)|·e j∠X(k) ;X min (n) represents the signal value at the nth sampling point in the time domain after windowing; N is the total number of sampling points of the signal, which is the Fourier transform length; j is the imaginary unit, used to construct the complex exponential function to achieve a linear transformation from the time domain to the frequency domain; k is the frequency domain index; n is the time domain index; and k is the frequency candidate set F. candidate The corresponding frequency point is used to specify the frequency point to be calculated. round() is an integer function; f c For candidate frequencies; f s e is the sampling frequency; -j2πkn / N Let f be a complex twitch factor, representing a frequency of f. k Orthogonal basis functions are used to project time-domain signals onto frequency-domain orthogonal bases to separate different frequency components.

[0093] S12. The quantum particle swarm optimization algorithm is invoked to estimate the frequency domain signal to obtain estimation parameters that match the preset spectrum.

[0094] In one embodiment, the quantum particle swarm optimization algorithm can be invoked to estimate the frequency domain signal, thereby selecting several spectral values ​​that match the preset spectrum from the multiple spectral values ​​contained in the frequency domain signal to obtain the estimation parameters.

[0095] In an optional embodiment, the step of invoking the quantum particle swarm optimization algorithm to estimate the frequency domain signal and obtain estimated parameters that match the measured spectrum may include the following sub-steps:

[0096] S121. The quantum particle swarm optimization algorithm is invoked to iteratively filter the spectral values ​​of the frequency candidate set, and several adjacent spectral values ​​are combined into a candidate parameter combination.

[0097] S122. Calculate the matching degree between each candidate parameter combination and the preset actual spectrum using the potential energy function, and select the target matching degree that meets the preset matching threshold from multiple matching degrees, and use the candidate parameter combination corresponding to the target matching degree as the estimation parameter.

[0098] In one embodiment, high-precision harmonic and interharmonic parameter estimation is performed using a quantum particle swarm optimization algorithm based on the frequency domain signal.

[0099] The quantum particle swarm optimization algorithm can solve problems such as the mutual interference between adjacent peaks when the harmonic or interharmonic interval is less than 2 frequency resolutions, which leads to a sharp increase in frequency offset estimation error; the failure of interpolation formula due to noise-induced spectral fluctuations; and the nonlinear coupling between amplitude, phase and frequency, which requires joint optimization.

[0100] Specifically, the quantum particle swarm optimization algorithm replaces signals with particles and simultaneously estimates the amplitude, frequency, and phase parameters of harmonics and interharmonics, avoiding the accumulation of step-by-step estimation errors. It provides a global, noise-resistant matching metric through potential energy function evaluation, supports sub-resolution high-precision estimation, and uses a quantum tunneling mechanism to update particle positions, breaking through the local convergence bottleneck of traditional optimization and ensuring that the globally optimal parameter solution is found.

[0101] This mechanism stems from the nonlocality and probabilistic distribution characteristics of quantum particles, enabling them to explore efficiently on a global scale and significantly improve convergence stability and accuracy. Because of this, the quantum particle swarm optimization algorithm can overcome the resolution limitations and premature convergence bottlenecks of traditional methods. Even in complex scenarios with strong noise interference, dense harmonics, or interharmonics, it can accurately separate and estimate weak components, greatly improving the accuracy and robustness of power quality analysis.

[0102] For each candidate frequency f in the frequency candidate set c In the frequency domain, the m-th frequency point k corresponds to m Extract its three nearest frequency points k m-1 ,k m ,k m+1 The spectral value X(k) is used to divide the three frequency points k. m-1 ,k m ,k m+1 The spectral value X(k) is used as a candidate parameter combination. Next, a particle swarm is defined, assuming there are s particles in the swarm, each particle representing a candidate parameter combination of harmonics and interharmonics, denoted as q. k =(δ k A k ,φ k ), where δ k Frequency offset is relative to f. c The small deviation, initially within the range of δ k ∈[-0.5,0.5], A k The amplitude comes from X(k) m The initial range is [0.8|X(k)]. m ),1.2|X(k m )],φ k As phase, it is based on Xk( m Centered on the phase angle of ), The phase is randomly initialized within a certain range; if it exceeds (-π, π), it is adjusted by ±2π. The δ of each particle... k A k ,φ k Initialize uniformly and randomly within the above range.

[0103] Next, the potential energy function can be used to measure the degree of matching between the candidate parameters and the actual spectrum. Using the potential energy function, the physical properties of the signal and the parameter space are directly correlated through mathematical modeling at three neighboring frequency points k. m-1 ,k m ,k m+1 The sum of squared complex differences between the theoretical and measured spectra generated from the candidate parameters is used as the potential energy to measure the potential energy of the parameter combination (δ). k A k ,φ k The fitting error to the actual spectrum is measured by the parameter combination (δ). k A k ,φ k The fitting error of the actual spectrum is given by the formula:

[0104]

[0105] In the above formula, X(k) m For the frequency point k extracted from the Fourier transform m-1 ,k m ,k m+1 The complex spectrum value of G(δ); k W(k) is the amplitude attenuation compensation factor, used to correct the amplitude attenuation caused by windowing; m -δ k The discrete spectral response model is a self-multiplicative convolutional window, accurately capturing the main lobe morphology and side lobe leakage characteristics; complex exponential terms... Phase error is incorporated into the optimization objective to solve the coupling error of amplitude and phase separation estimation in traditional methods; j is the imaginary unit; when the harmonic interval is less than the frequency resolution Δf, the potential energy function separates the overlapping main lobe through multi-frequency joint optimization m=-1,0,1, and the random spectral fluctuations caused by noise are suppressed by multi-frequency average in the potential energy function.

[0106] The potential energy function continuously evaluates the matching degree across the entire frequency domain, and can be finely adjusted by adjusting the frequency offset δ. k Find the non-integer frequency that minimizes the potential energy, and the potential energy value U(q). k The lower the potential energy, the higher the matching degree, and the closer the parameters are to the true value. Therefore, particles with low potential energy are considered excellent individuals, and their position information is used to guide other particles. Through group cooperation, they gradually approach the global minimum potential energy point, i.e., the optimal parameter combination. To this end, target matching degrees that meet a preset matching threshold can be selected from multiple matching degrees. The candidate parameter combination corresponding to the target matching degree is taken as the optimal parameter combination, and the estimated parameters can be obtained. Furthermore, the potential energy function is fitted based on the entire frequency band, rather than relying on a single peak, so local deviations caused by random noise are weakened by overall averaging.

[0107] In extremely challenging scenarios, when the harmonic spacing is less than the frequency resolution Δf, two or more harmonics become too close. The potential energy function is optimized through multi-frequency joint optimization, simultaneously fitting the overall spectral shape of three neighboring frequencies. Even if the main lobes of multiple harmonics overlap, the distribution of each harmonic component in the frequency domain is determined by the main lobe shape W(k) of the window function. m -δ k The superposition pattern of these variables is unique, and the potential energy function will attempt to use multiple (δ) values. k A k ,φ k The parameters are combined to fit the overlapping spectral envelope. Only when the parameters of all components are close to the true values ​​can the total error be minimized. Therefore, it is possible to separate the harmonic components with very small intervals from the severely overlapping main lobes.

[0108] Standard particle swarm optimization is prone to getting trapped in local optima in complex multi-peak potential energy surfaces. In scenarios where harmonics and interharmonics coexist, it may converge to an erroneous peak. Therefore, a quantum tunneling mechanism is needed to update particle positions. In one embodiment, after the step of iteratively filtering the spectral values ​​of the frequency candidate set by calling the quantum particle swarm algorithm and forming a candidate parameter combination from several adjacent spectral values, the method may further include the following sub-steps:

[0109] S21. Calculate the potential energy value of the particle corresponding to each candidate parameter combination.

[0110] S22. Update the position of the particle corresponding to each candidate parameter combination according to the potential energy value.

[0111] The quantum tunneling mechanism endows the particle swarm optimization algorithm with the ability to dynamically escape local optima. In each iteration, the potential energy value U(q) of each particle can be calculated. k Based on the current temperature coefficient and the tunneling probability, a decision is made on whether to accept the new location. The tunneling probability can be expressed as follows:

[0112]

[0113] In the above formula, σ T It is a dynamic temperature coefficient; The search range can be adaptively adjusted, allowing for large-scale exploration at high temperatures and focusing on local optimization at low temperatures; K represents the number of states.

[0114] If P tunnel (q k If the potential energy of the new position is lower than ξ (ξ ~ μ(0,1)), or if there is a certain probability of accepting a worse solution (to avoid getting trapped in a local optimum), then the particle position is updated.

[0115] Where ξ is a random threshold, and the updated position is... δbest This is the frequency offset reference value; A best This is the amplitude reference value; φ best For phase reference value; δ k This is the initial value of the frequency offset; A k φ is the initial value of the amplitude. k η is the initial phase value; υ is the learning rate; η controls the convergence speed; υ adjusts the tunneling intensity; α(0,1) is the standard normal distribution; jυα is the imaginary term, which introduces the quantum tunneling effect and breaks through the real number space limitation.

[0116] A maximum number of iterations can be set. If the change in the global optimal solution after 5 consecutive iterations satisfies... If the iteration fails, the process terminates prematurely. After the iteration is complete, the particle with the lowest potential energy is selected as the final solution. Frequency calculation can be performed as follows: f true =(k m +δ best )·Δf; Amplitude is taken as A best Phase taken as φ best , φ best Δf represents the frequency point; Δf represents the frequency resolution.

[0117] S13. Use a preset spatiotemporal attention residual network to perform nonlinear error compensation on the estimated parameters to obtain harmonic and interharmonic parameters.

[0118] After obtaining the estimated parameters, a spatiotemporal attention residual network can be used to compensate for nonlinear errors in the estimated parameters.

[0119] Assume the input is a frequency domain neighborhood vector X k =[X(k-1),X(k),X(k+1)] R X k To estimate the candidate frequencies f of the parameters c The corresponding complex spectral values ​​of three adjacent frequency points, including amplitude, phase, and leakage information; X(k) is the complex spectrum output of the Fourier transform after windowing the self-multiplicative convolution window; X(k-1) is the candidate frequency f. c The adjacent preceding frequency point, X(k+1) candidate frequency f c The next adjacent frequency point; R is the transpose matrix. The spatiotemporal attention residual network includes dilated convolutional layers, multi-head attention mechanism, and residual output structure layer.

[0120] The preliminarily estimated harmonic parameter sequence is processed by dilated convolutional layers to expand the receptive field without increasing computational cost, capture long-range temporal dependencies, and extract multi-scale time-frequency features. The specific formula is as follows:

[0121]

[0122] In the above formula, σ is the weight matrix of the dilated convolution kernel; * represents the convolution operation with a dilation rate of d = 3, i.e., sampling at intervals in the frequency domain to expand the receptive field to 7 frequency points; σ is the activation function. This is the bias vector.

[0123] The multi-head attention mechanism automatically suppresses spectral fluctuations caused by noise; the specific formula is as follows:

[0124] Q = W Q H (1) K = W K H (1) V = W V H (1) ,

[0125] In the above formula, Q is the query matrix, which learns the correlation between frequency domain features; K is the key matrix, which encodes feature location information; V is the value matrix, which preserves feature details; K R Let K be the transpose of the key matrix. d is a scaling factor to prevent the gradient from vanishing due to an excessively large dot product; k W represents the dimension of the key matrix K; Q The weight matrix is ​​a learnable matrix used to weight the input feature matrix H. (1) Mapped to the query matrix space; W K The weight matrix is ​​a learnable matrix used to weight the input feature matrix H. (1) Convert to a key matrix; W V The weight matrix is ​​a learnable matrix used to weight the input feature matrix H. (1) Transform it into a value matrix.

[0126] The residual output structure layer directly transmits input features to deeper layers through cross-layer skip connections, allowing gradients to be losslessly backpropagated via identity mapping. This enhances feature representation capabilities and reduces amplitude and phase estimation errors. The specific formula is as follows:

[0127] ΔA,Δφ=W o (H (1) +H (2) );

[0128] In the above formula, H (1) The local frequency domain features extracted by dilated convolution; H (2) Global relevance features extracted for attention mechanisms; W o ΔA represents the weights of the fully connected layer, mapping high-dimensional features to correction values; ΔA is the magnitude correction factor; and Δφ is the phase offset.

[0129] In one embodiment, the corrected harmonic and interharmonic parameters are A. true =A best (1+ΔA), φtrue =φ best +Δφ; where A true , φ true This is for the corrected high-precision amplitude and phase parameters.

[0130] Attention-based residual networks, by introducing an attention mechanism, can automatically focus on key harmonic features, reducing redundant computations. While maintaining high accuracy, this reduces computational complexity, significantly improving algorithm response speed and meeting the needs of real-time monitoring. This method also exhibits good anti-interference performance, stably outputting reliable parameter estimation results through the synergistic effect of chaotic synchronization mechanisms and fractal structures.

[0131] In harmonic and interharmonic analysis, noise and transient disturbances can cause abrupt changes or instability in parameter estimates. To determine whether corrected harmonic and interharmonic parameters are usable, in one embodiment, the method may include the following steps:

[0132] S14. Calculate the spectral energy entropy values ​​of the harmonic and interharmonic parameters.

[0133] S15. If the spectral energy entropy value is greater than the preset entropy threshold, the historical time domain signal of the previous moment is obtained, and the harmonic and interharmonic parameters are determined using the historical time domain signal.

[0134] Specifically, the stability of the current spectrum can be evaluated by verifying the dynamic error entropy. If the change in the spectrum energy entropy value is greater than the preset entropy threshold, it will revert to the previous state to avoid error accumulation.

[0135] Specifically, by calculating the spectral energy entropy value, the stability of the current spectrum's energy distribution can be assessed. If the spectral energy entropy value changes too much, there may be sudden interference or estimation errors. In this case, reverting to the previous state can prevent the accumulation of errors. The formula for the spectral energy entropy value is:

[0136]

[0137] In the above formula, H e p represents the spectral energy entropy, which quantifies the disorder of the energy distribution in the frequency domain. k The normalized energy percentage at the k-th frequency point. |X(k)| 2 is the energy at frequency k; N is the number of Fourier transform points, which determines the frequency resolution.

[0138] The stability judgment formula is:

[0139]

[0140] In the above formula, This represents the spectral energy entropy value at the current time (t). θ represents the spectral energy entropy value at the previous time (t-1); θ is the threshold value, which controls the allowable range of entropy change.

[0141] If the entropy change is less than the threshold, the parameter estimation is considered reliable and the current result is accepted. If the entropy change is greater than the threshold, it is determined to be a noise impact or algorithm failure, triggering parameter rollback and reverting the parameters to the state at time t-1. The harmonic and interharmonic parameters are then determined using the time domain signal at the previous time.

[0142] Through the above operations, the dynamic changes of transient harmonics and interharmonics can be accurately tracked, providing reliable data support for harmonic control in power systems. This helps reduce electrical equipment losses, improve the safety and stability of power grid operation, and has broad application prospects and significant economic benefits.

[0143] In this embodiment, the present invention provides a method for detecting harmonic and interharmonic parameters based on a self-multiplicative convolutional window algorithm. Its advantages are as follows: the present invention can acquire discrete time-domain signals, and window and map these signals using a self-multiplicative convolutional window algorithm to obtain frequency-domain signals; it then uses a quantum particle swarm optimization algorithm to estimate the frequency-domain signals, obtaining estimated parameters that match a preset spectrum; finally, it uses a preset spatiotemporal attention residual network to perform nonlinear error compensation on the estimated parameters, obtaining the harmonic and interharmonic parameters. By windowing the signal, the present invention can suppress signal leakage, making the main lobe energy of the spectrum more concentrated. Furthermore, by using the quantum particle swarm optimization algorithm to estimate the frequency-domain signals, it can quickly screen evaluation spectra that meet actual needs. Finally, spectrum compensation can further reduce errors and improve detection accuracy.

[0144] This invention also provides a harmonic and interharmonic parameter detection device based on a self-multiplying convolution window algorithm, see [link to relevant documentation]. Figure 2 The diagram shows a schematic of the structure of a harmonic and interharmonic parameter detection device based on a self-multiplying convolution window algorithm according to an embodiment of the present invention.

[0145] As an example, the harmonic and interharmonic parameter detection device based on the self-multiplication convolution window algorithm may include:

[0146] The windowing module 201 is used to acquire discrete time-domain signals and window and map the time-domain signals based on the self-multiplication convolution window algorithm to obtain frequency-domain signals;

[0147] The estimation module 202 is used to call the quantum particle swarm optimization algorithm to perform estimation processing on the frequency domain signal to obtain estimation parameters that match the preset spectrum;

[0148] The compensation detection module 203 is used to perform nonlinear error compensation on the estimated parameters using a preset spatiotemporal attention residual network to obtain harmonic and interharmonic parameters.

[0149] Optionally, the step of acquiring a discrete time-domain signal and windowing and mapping the time-domain signal based on a self-multiplying convolution window algorithm to obtain a frequency-domain signal includes:

[0150] The signal under test is sampled based on a preset sampling frequency and number of sampling points to obtain discrete time-domain signals;

[0151] The time-domain signal is windowed using a self-multiply fractal dynamic adaptive window function to obtain a frequency candidate set. The self-multiply fractal dynamic adaptive window function is a function constructed by performing a self-multiply operation on a preset basic window function based on a self-multiply convolution window algorithm to obtain a self-multiply convolution window.

[0152] The time-domain signal is mapped to a frequency-domain signal based on the frequency candidate set.

[0153] Optionally, the step of windowing the time-domain signal using a self-multiplying fractal dynamic adaptive window function to obtain a frequency candidate set includes:

[0154] The time-domain signal is windowed using a self-multiplying fractal dynamic adaptive window function to obtain a suppressed signal.

[0155] Signal features are extracted from the time-domain signal using the preset Da Vinci oscillator equation to obtain dynamic response feature information;

[0156] Effective frequency signals with a synchronization index less than a preset index threshold are selected from the dynamic response feature information to obtain a frequency candidate set.

[0157] Optionally, mapping the time-domain signal to a frequency-domain signal based on the frequency candidate set includes:

[0158] The time-domain signal is mapped and Fourier transformed based on the frequency candidate set to obtain a frequency-domain signal with a complex frequency domain.

[0159] Optionally, the step of using the quantum particle swarm optimization algorithm to estimate the frequency domain signal to obtain estimated parameters that match the measured spectrum includes:

[0160] The quantum particle swarm optimization algorithm is invoked to iteratively filter the spectral values ​​of the frequency candidate set, and several adjacent spectral values ​​are combined into a candidate parameter combination.

[0161] The matching degree between each candidate parameter combination and the preset actual spectrum is calculated using the potential energy function, and a target matching degree that meets the preset matching threshold is selected from multiple matching degrees. The candidate parameter combination corresponding to the target matching degree is used as the estimation parameter.

[0162] Optionally, the device further includes:

[0163] The potential energy module is used to calculate the potential energy value of the particle corresponding to each candidate parameter combination after the step of calling the quantum particle swarm algorithm to iteratively filter the spectral values ​​of the frequency candidate set and forming a candidate parameter combination with several adjacent spectral values.

[0164] An update module is used to update the position of the particle corresponding to each candidate parameter combination based on the potential energy value.

[0165] Optionally, the device further includes:

[0166] The entropy module is used to calculate the spectral energy entropy value of the harmonic and interharmonic parameters after the step of using a preset spatiotemporal attention residual network to perform nonlinear error compensation on the estimated parameters to obtain the harmonic and interharmonic parameters.

[0167] The history module is used to acquire the historical time-domain signal of the previous moment if the spectral energy entropy value is greater than a preset entropy threshold, and to determine the harmonic and interharmonic parameters using the historical time-domain signal.

[0168] Those skilled in the art will understand that, for ease of description and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0169] Furthermore, this application also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the harmonic and interharmonic parameter detection method based on the self-multiplicative convolution window algorithm as described in the above embodiments.

[0170] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer-executable program, which is used to cause a computer to execute the harmonic and interharmonic parameter detection method based on the self-multiplication convolution window algorithm as described in the above embodiments.

[0171] In the description of the embodiments of the present invention, it should be noted that the terms "above," "below," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. When an element such as a layer, region, or substrate is referred to as being "above" or "on top of" another element, it may be directly on the other element, or there may be an intermediate element. Conversely, when an element is referred to as being "directly on" or "above" another element, there is no intermediate element. It should also be understood that when an element is referred to as being "below" or "under" another element, it may be directly below or under the other element, or there may be an intermediate element. Conversely, when an element is referred to as being "directly below" or "under" another element, there is no intermediate element. Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0172] Those skilled in the art will understand that embodiments of this application may also include computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0173] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), devices, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0174] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0175] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0176] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting harmonic and interharmonic parameters based on a self-multiplying convolution window algorithm, characterized in that, The method includes: Discrete time-domain signals are obtained, and windowing and mapping of the time-domain signals are performed based on the self-multiplying convolution window algorithm to obtain frequency-domain signals; The frequency domain signal is estimated by calling the quantum particle swarm optimization algorithm to obtain estimated parameters that match the preset spectrum; The estimated parameters are compensated for nonlinear errors using a preset spatiotemporal attention residual network to obtain harmonic and interharmonic parameters.

2. The method for detecting harmonic and interharmonic parameters based on the self-multiplying convolution window algorithm according to claim 1, characterized in that, The process of acquiring discrete time-domain signals and windowing and mapping the time-domain signals based on a self-multiplicative convolutional window algorithm to obtain frequency-domain signals includes: The signal under test is sampled based on a preset sampling frequency and number of sampling points to obtain discrete time-domain signals; The time-domain signal is windowed using a self-multiply fractal dynamic adaptive window function to obtain a frequency candidate set. The self-multiply fractal dynamic adaptive window function is a function constructed by performing a self-multiply operation on a preset basic window function based on a self-multiply convolution window algorithm to obtain a self-multiply convolution window. The time-domain signal is mapped to a frequency-domain signal based on the frequency candidate set.

3. The method for detecting harmonic and interharmonic parameters based on the self-multiplying convolution window algorithm according to claim 2, characterized in that, The step of applying a self-multiplying fractal dynamic adaptive window function to the time-domain signal to obtain a frequency candidate set includes: The time-domain signal is windowed using a self-multiplying fractal dynamic adaptive window function to obtain a suppressed signal. Signal features are extracted from the time-domain signal using the preset Da Vinci oscillator equation to obtain dynamic response feature information; Effective frequency signals with a synchronization index less than a preset index threshold are selected from the dynamic response feature information to obtain a frequency candidate set.

4. The method for detecting harmonic and interharmonic parameters based on the self-multiplicative convolution window algorithm according to claim 2, characterized in that, The step of mapping the time-domain signal to a frequency-domain signal based on the frequency candidate set includes: The time-domain signal is mapped and Fourier transformed based on the frequency candidate set to obtain a frequency-domain signal with a complex frequency domain.

5. The method for detecting harmonic and interharmonic parameters based on the self-multiplying convolution window algorithm according to claim 1, characterized in that, The process of using the quantum particle swarm optimization algorithm to estimate the frequency domain signal and obtain estimated parameters that match the measured spectrum includes: The quantum particle swarm optimization algorithm is invoked to iteratively filter the spectral values ​​of the frequency candidate set, and several adjacent spectral values ​​are combined into a candidate parameter combination. The matching degree between each candidate parameter combination and the preset actual spectrum is calculated using the potential energy function, and a target matching degree that meets the preset matching threshold is selected from multiple matching degrees. The candidate parameter combination corresponding to the target matching degree is used as the estimation parameter.

6. The method for detecting harmonic and interharmonic parameters based on the self-multiplying convolution window algorithm according to claim 5, characterized in that, After the step of iteratively filtering the spectral values ​​of the frequency candidate set by invoking the quantum particle swarm optimization algorithm and forming a candidate parameter combination from several adjacent spectral values, the method further includes: Calculate the potential energy value of the particle corresponding to each of the candidate parameter combinations; The position of the particle corresponding to each candidate parameter combination is updated based on the potential energy value.

7. The method for detecting harmonic and interharmonic parameters based on the self-multiplicative convolution window algorithm according to any one of claims 1-6, characterized in that, After the step of using a preset spatiotemporal attention residual network to perform nonlinear error compensation on the estimated parameters to obtain the harmonic and interharmonic parameters, the method further includes: Calculate the spectral energy entropy values ​​of the harmonic and interharmonic parameters; If the spectral energy entropy value is greater than the preset entropy threshold, the historical time-domain signal of the previous moment is obtained, and the harmonic and interharmonic parameters are determined using the historical time-domain signal.

8. A harmonic and interharmonic parameter detection device based on a self-multiplying convolution window algorithm, characterized in that, The device includes: A windowing module is used to acquire discrete time-domain signals and window and map the time-domain signals based on a self-multiplying convolution window algorithm to obtain frequency-domain signals. The estimation module is used to call the quantum particle swarm optimization algorithm to estimate the frequency domain signal and obtain estimation parameters that match the preset spectrum. The compensation detection module is used to perform nonlinear error compensation on the estimated parameters using a preset spatiotemporal attention residual network to obtain harmonic and interharmonic parameters.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the harmonic and interharmonic parameter detection method based on the self-multiplicative convolution window algorithm as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the harmonic and interharmonic parameter detection method based on the self-multiplicative convolution window algorithm as described in any one of claims 1-7.

Images