Disturbance detection method and device based on wavelet entropy denoising and empirical mode decomposition

Abstract

The invention discloses a disturbance detection method and device based on wavelet entropy denoising and empirical mode decomposition. The method comprises the following steps: acquiring an initial voltage signal; performing multi-scale analysis on the time domain local feature of the initial voltage signal to obtain multi-scale high and low frequency components; according to the high-frequency coefficients in the multi-scale high and low frequency components, analyzing wavelet entropies of different scales to obtain corresponding wavelet thresholds, and performing denoising reconstruction on the multi-scale high and low frequency components in combination with the high-frequency coefficients of different scales to obtain voltage reconstruction signals; decomposing the voltage reconstruction signal by adopting an empirical mode decomposition model to obtain a plurality of intrinsic mode function components; in combination with sensitivity evaluation indexes, sorting the intrinsic mode function components, screening out disturbance characteristic components, determining disturbance starting and ending time of the disturbance characteristic components, and completing disturbance detection, so that rapid and accurate identification of the voltage disturbance event is realized; and the adaptability and reliability of voltage disturbance detection in a complex noise environment and under a multi-type disturbance condition are improved.

Description

Technical Field

[0001] This invention belongs to the technical field of voltage disturbance detection in power systems, specifically relating to a disturbance detection method and apparatus based on wavelet entropy denoising and empirical mode decomposition. Background Technology

[0002] With the large-scale integration of power electronic equipment and various sensitive loads into new power systems, power quality issues are becoming increasingly prominent in power system operation. In particular, disturbances such as voltage sags, voltage swells, and voltage flicker can severely impact industrial production processes, automated control systems, and user terminal equipment. Therefore, the rapid and accurate identification of voltage disturbances is a crucial foundation for power quality monitoring and interference source localization, and has significant engineering application value.

[0003] Existing methods for detecting voltage disturbance timing mainly include those based on amplitude thresholding, spectral analysis, and wavelet transform. Amplitude thresholding methods are simple to implement but sensitive to noise, prone to false triggering or missed detection. Frequency domain analysis methods require preset analysis window parameters and have weak ability to extract instantaneous features from non-stationary disturbance signals. While traditional wavelet denoising methods can perform multi-scale processing of power quality signals, their fixed threshold selection can easily lead to loss of signal details or residual noise, thus affecting the accurate location of the disturbance timing.

[0004] Patent application CN111965409A discloses a voltage transient disturbance detection method based on the effective value of a piecewise differential waveform. The method includes: acquiring a voltage signal; resampling the voltage signal waveform according to the actual frequency of the system from which the voltage signal originates to correct frequency offset; the waveform composed of the resampled points is the corrected frequency waveform; detecting voltage transient disturbances in the corrected frequency waveform using the piecewise differential waveform effective value method; extracting the voltage transient components from the voltage transient disturbances; and characterizing the voltage transient components using four indicators: dominant frequency, polarity, amplitude, and duration. By extracting the transient components, the characteristics of the voltage transient disturbances are characterized. The method is simple, intuitive, versatile, and has high detection accuracy.

[0005] The aforementioned methods can cover most voltage dips, swells, and short-term interruptions, but they still have significant limitations for more complex and subtle disturbances. They lack stratification methods for "multi-frequency, non-stationary, and complex disturbances" and lack time-frequency localization capabilities. Once a disturbance contains wideband, frequency drift, or intermittent pulse groups, frequent main frequency jumps can lead to an inability to provide stable values. If the amplitude and duration fluctuate due to window sliding, there will be issues of repeated alarms / missed alarms. Therefore, there is an urgent need for a voltage disturbance time detection method with strong noise reduction capabilities and suitability for non-stationary signal feature extraction, enabling rapid and accurate identification of voltage disturbance events and providing technical support for power quality monitoring and interference source localization. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention provides a disturbance detection method and apparatus based on wavelet entropy denoising and empirical mode decomposition. The method includes: acquiring an initial voltage signal; performing multi-scale analysis on the local temporal features of the initial voltage signal to obtain multi-scale high- and low-frequency components; analyzing wavelet entropy at different scales based on the high-frequency coefficients in the multi-scale high- and low-frequency components to obtain corresponding wavelet thresholds; denoising and reconstructing the multi-scale high- and low-frequency components based on the wavelet thresholds and the high-frequency coefficients at different scales to obtain a reconstructed voltage signal; decomposing the reconstructed voltage signal using an empirical mode decomposition model to obtain multiple intrinsic mode function components; ranking the intrinsic mode function components based on a sensitivity evaluation index, selecting disturbance feature components, determining the disturbance start and end times of the disturbance feature components, and completing the disturbance detection. By acquiring the initial voltage signal and using wavelet entropy to determine the wavelet threshold, multi-scale noise adaptive suppression is achieved, improving the signal-to-noise ratio of the voltage reconstruction signal. Then, combined with the empirical mode decomposition model, local features of the voltage reconstruction signal are extracted. By constructing a sensitivity evaluation index, the intrinsic mode function component that best reflects the disturbance characteristics is selected. The start and end times of the disturbance are accurately located by using the modulus maxima. This improves the adaptability and reliability of voltage disturbance detection in complex noise environments and various disturbance scenarios.

[0007] In a first aspect, the present invention provides a disturbance detection method based on wavelet entropy denoising and empirical mode decomposition, specifically including the following steps: Acquire the initial voltage signal; Multi-scale analysis of the time-domain local features of the initial voltage signal is performed to obtain multi-scale high- and low-frequency components; Based on the high-frequency coefficients of the multi-scale high and low-frequency components, the wavelet entropy at different scales is analyzed to obtain the corresponding wavelet thresholds. Based on wavelet thresholding and combined with high-frequency coefficients at different scales, the high and low frequency components at multiple scales are denoised and reconstructed to obtain the voltage reconstruction signal. An empirical mode decomposition model is used to decompose the voltage reconstruction signal to obtain multiple intrinsic mode function components; By combining sensitivity evaluation indicators, the intrinsic mode function components are sorted, the perturbation feature components are screened out, the start and end times of the perturbation feature components are determined, and the perturbation detection is completed.

[0008] Furthermore, multi-scale analysis is performed on the time-domain local features of the initial voltage signal to obtain multi-scale high- and low-frequency components, specifically including: The initial voltage signal is decomposed into multiple layers using a low-pass filter to obtain multiple low-frequency coefficients; A high-pass filter is used to perform multi-level decomposition on the initial voltage signal to obtain multi-level high-frequency coefficients; By fusing high-frequency and low-frequency coefficients, multi-scale high- and low-frequency components are obtained.

[0009] Furthermore, the low-frequency coefficient is specifically expressed as:

[0010] Where k is the time index of the initial voltage signal, n is the index after multi-level decomposition, j is the number of wavelet decomposition levels, and c j [k] represents the low-frequency coefficients of the j-th layer, and h() represents the low-pass filter.

[0011] Furthermore, the high-frequency coefficients are specifically represented as follows:

[0012] Where k is the time index of the initial voltage signal, n is the index after multi-level decomposition, j is the number of wavelet decomposition levels, and d j [k] represents the high-frequency coefficients of the j-th layer, and g() represents the high-pass filter.

[0013] Furthermore, the multi-scale high and low frequency components are specifically represented as follows:

[0014] Where x[n] is the initial voltage signal, (c j ,d j ,…,d1) represent multi-scale high and low frequency components, c j For the j-th layer low-frequency coefficients, the signal detail features are mainly concentrated in c. j In the middle, d j For the j-th layer high-frequency coefficient, noise is concentrated in the high-frequency coefficient d. j middle.

[0015] Furthermore, based on the high-frequency coefficients in the multi-scale high and low-frequency components, wavelet entropy at different scales is analyzed to obtain the corresponding wavelet thresholds, specifically including: The wavelet energy of each layer is calculated based on the high-frequency coefficients of each layer in the multi-scale high and low frequency components. The wavelet energy of each layer is normalized to obtain the wavelet energy probability distribution of each layer. Based on the wavelet energy probability distribution, the wavelet entropy of each layer is determined; The wavelet threshold for each layer is determined by combining the wavelet entropy of each layer with the corresponding noise standard deviation.

[0016] Furthermore, wavelet energy is specifically expressed as:

[0017] Among them, E j Let d be the wavelet energy of the j-th layer. j [k] represents the high-frequency coefficient with time index k in the j-th layer, where k is the time index of the initial voltage signal.

[0018] Furthermore, the wavelet energy probability distribution is specifically expressed as:

[0019] Among them, P j Let E be the wavelet energy probability distribution of the j-th layer. j Let J be the wavelet energy of the j-th layer, and J be the total number of layers.

[0020] Furthermore, wavelet entropy is specifically expressed as:

[0021] Among them, WE j Let P be the wavelet entropy of the j-th layer. j Let be the wavelet energy probability distribution of the j-th layer, and log() be the logarithmic function.

[0022] Furthermore, the wavelet threshold is specifically expressed as:

[0023] Among them, T j Let α be the wavelet threshold of the j-th layer. j Let α be the adjustment coefficient for the j-th layer. j ∈[0.5,1.5], σ j Let N be the noise standard deviation of the j-th layer, and N be the signal length of the j-th layer.

[0024] Furthermore, based on the wavelet entropy of each layer and the corresponding noise standard deviation, the wavelet threshold for each layer is determined, specifically as follows: The wavelet entropy of each layer is normalized to obtain the adjustment coefficient of each layer; Determine the high-frequency median based on the high-frequency coefficients of each layer; The noise standard deviation is estimated by using the median absolute deviation method combined with the high-frequency median. The wavelet threshold is obtained by fusing the modulation coefficient and the noise standard deviation, combined with the signal length.

[0025] Furthermore, based on wavelet thresholding and combined with high-frequency coefficients at different scales, denoising and reconstruction of multi-scale high and low frequency components are performed to obtain the voltage reconstruction signal, specifically including: Based on the wavelet threshold of each layer, the high-frequency coefficients are judged and updated, and smooth convergence processing is performed to obtain the updated high-frequency coefficients. Inverse discrete wavelet transform is used to reconstruct the high-frequency coefficients and the low-frequency coefficients in the multi-scale high and low frequency components, resulting in a reconstructed time-domain signal and an output voltage reconstructed signal.

[0026] Furthermore, the high-frequency coefficients are updated, specifically as follows:

[0027] in, For the high-frequency coefficients updated at time index k in the j-th layer, sign() is the sign function, and d j [k] represents the high-frequency coefficients of the j-th layer with time index k, T j is the wavelet threshold of the j-th layer.

[0028] Furthermore, inverse discrete wavelet transform is used to reconstruct the updated high-frequency coefficients and the low-frequency coefficients in the multi-scale high and low-frequency components, resulting in a reconstructed time-domain signal and an output voltage reconstructed signal, specifically including: Upsample the high-frequency coefficients and the low-frequency coefficients in the multi-scale high- and low-frequency components respectively to obtain the first high-frequency coefficients and the first low-frequency coefficients; The first low-frequency coefficients are synthesized and convolved using a low-pass synthesis filter to obtain a low-frequency coefficient sequence; A high-pass synthesizing filter is used to synthesize and convolve the first high-frequency coefficients to obtain a high-frequency coefficient sequence; The low-frequency coefficient sequence and the high-frequency coefficient sequence are added point by point to obtain the current reconstructed time-domain signal; Repeat the above process. If the current reconstructed time-domain signal has the same sampling frequency as the initial voltage signal, output the reconstructed voltage signal.

[0029] Furthermore, by combining sensitivity evaluation indicators, the intrinsic mode function components are sorted, perturbation feature components are selected, and the start and end times of the perturbation feature components are determined to complete the perturbation detection. Specifically, this includes: Based on the sensitivity evaluation index, the components of each intrinsic mode function are analyzed to obtain the sensitivity values ​​of each intrinsic mode function component. The perturbation feature component corresponding to the maximum sensitivity value is obtained by sorting the sensitivity values ​​of each intrinsic mode function component in descending order. The extreme values ​​in the perturbation characteristic components are obtained, and the magnitudes of each extreme value are compared. The start and end times of the perturbation are obtained by combining the time points of each extreme value.

[0030] Furthermore, the sensitivity evaluation index is obtained through the following steps: Calculate the total energy corresponding to each intrinsic mode function component; Based on the first sensitivity coefficient, the energy ratio of each intrinsic mode function is obtained through the total energy of each intrinsic mode function component, and the first sensitivity term is determined. Based on the second sensitivity coefficient, the correlation between the voltage reconstruction signal and the intrinsic mode function components is analyzed, the correlation coefficient is given, and the second sensitivity term is determined. By integrating the first and second sensitivity items, sensitivity evaluation indicators are determined.

[0031] Furthermore, the sensitivity evaluation index is specifically expressed as follows:

[0032] Among them, s i Let Q be the sensitivity evaluation index for the i-th intrinsic mode function component, where λ1 is the first sensitivity coefficient, λ2 is the second sensitivity coefficient, and Q is the third sensitivity coefficient. i Let ρ be the total energy of the i-th intrinsic mode function component. i Let x'[n] be the correlation coefficient between the i-th intrinsic mode function component and the voltage reconstructed signal, and let x'[n] be the voltage reconstructed signal. i [k] represents the energy value of the i-th intrinsic mode function component with time index k, and K is the total time length. Let be the average energy of the i-th intrinsic mode function component. This represents the average frequency of the voltage reconstruction signal.

[0033] Furthermore, the first sensitivity term is the product of the first sensitivity coefficient and the energy proportion of the intrinsic mode function; the second sensitivity term is the product of the second sensitivity coefficient and the absolute value of the correlation coefficient.

[0034] Secondly, the present invention also provides a disturbance detection device based on wavelet entropy denoising and empirical mode decomposition, employing a disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described above, comprising: The signal acquisition module is used to acquire the initial voltage signal; The signal analysis module is used to perform multi-scale analysis on the local time-domain features of the initial voltage signal to obtain multi-scale high and low frequency components; The threshold calculation module is used to analyze the wavelet entropy at different scales based on the high-frequency coefficients in the multi-scale high and low frequency components, and obtain the corresponding wavelet thresholds. The signal reconstruction module is used to denoise and reconstruct multi-scale high and low frequency components based on wavelet thresholds and combined with high-frequency coefficients at different scales to obtain the voltage reconstruction signal. The mode decomposition module is used to decompose the voltage reconstruction signal using an empirical mode decomposition model to obtain multiple intrinsic mode function components; The disturbance analysis module is used to sort the intrinsic mode function components by combining sensitivity evaluation indicators, screen out disturbance feature components, determine the start and end times of disturbance feature components, and complete disturbance detection.

[0035] The disturbance detection method and apparatus based on wavelet entropy denoising and empirical mode decomposition provided by this invention have at least the following beneficial effects: (1) By collecting the initial voltage signal and using wavelet entropy to determine the wavelet threshold, multi-scale noise adaptive suppression is achieved, which improves the signal-to-noise ratio of the voltage reconstruction signal. Then, the local feature extraction of the voltage reconstruction signal is performed by combining the empirical mode decomposition model. By constructing a sensitivity evaluation index, the inherent mode function component that best reflects the disturbance characteristics is selected. The start and end times of the disturbance are accurately located by using the modulus maxima. This improves the adaptability and reliability of voltage disturbance detection in complex noise environments and various types of disturbances.

[0036] (2) The energy distribution of high-frequency components at each scale is calculated by wavelet entropy, and the noise is denoised by adaptive wavelet threshold. This effectively suppresses noise interference and preserves the transient characteristics of disturbance, improves the quality of voltage reconstruction signal and preserves signal details.

[0037] (3) By introducing an empirical mode decomposition model to adaptively decompose the denoised voltage reconstruction signal, the limitations of the traditional fixed window function method are avoided, and the local variation features of non-stationary voltage disturbances can be extracted more accurately.

[0038] (4) By constructing a sensitivity evaluation index, the correlation between the energy of the inherent mode function component and the initial voltage signal is comprehensively considered, so as to realize the adaptive selection of the characteristic mode, effectively avoid the influence of mode aliasing on the disturbance identification accuracy, and improve the accuracy and stability of disturbance start and end time identification. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the disturbance detection method based on wavelet entropy denoising and empirical mode decomposition provided in an embodiment of the present invention. Figure 2 A flowchart of multi-scale analysis provided for embodiments of the present invention; Figure 3 A flowchart for determining the wavelet threshold provided in an embodiment of the present invention; Figure 4 A flowchart for obtaining the voltage reconstruction signal provided in an embodiment of the present invention; Figure 5 A flowchart for disturbance detection provided in an embodiment of the present invention; Figure 6 A flowchart for constructing sensitivity evaluation metrics provided in an embodiment of the present invention; Figure 7 This is a disturbance start and end time detection diagram provided in an embodiment of the present invention during a voltage swell. Figure 8 This is a disturbance start and end time detection diagram provided in an embodiment of the present invention during a voltage sag. Figure 9 This is a disturbance start and end time detection diagram provided in an embodiment of the present invention during voltage flicker; Figure 10 A flowchart of a disturbance detection method based on wavelet entropy denoising and empirical mode decomposition provided in an embodiment of the present invention; Figure 11 This is a structural block diagram of a disturbance detection device based on wavelet entropy denoising and empirical mode decomposition provided in an embodiment of the present invention.

[0040] Among them, 201 is the signal acquisition module; 202 is the signal analysis module; 203 is the threshold calculation module; 204 is the signal reconstruction module; 205 is the mode decomposition module; and 206 is the disturbance analysis module. Detailed Implementation

[0041] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0042] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0043] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0044] To address the problems of strong noise interference, unclear disturbance characteristics, and difficulty in accurately identifying the start and end times of disturbances in distribution network voltage disturbance signals, this invention proposes a disturbance detection method based on wavelet entropy denoising and empirical mode decomposition. The method includes: acquiring an initial voltage signal; performing multi-scale analysis on the temporal local features of the initial voltage signal to obtain multi-scale high- and low-frequency components; analyzing wavelet entropy at different scales based on the high-frequency coefficients in the multi-scale high- and low-frequency components to obtain corresponding wavelet thresholds; based on the wavelet thresholds and combined with the high-frequency coefficients at different scales, denoising and reconstructing the multi-scale high- and low-frequency components to obtain a reconstructed voltage signal; using an empirical mode decomposition model to decompose the reconstructed voltage signal to obtain multiple intrinsic mode function (IMF) components; and ranking the IMF components using a sensitivity evaluation index to select disturbance feature components, determining the disturbance start and end times of the disturbance feature components, and completing the disturbance detection. This invention utilizes wavelet entropy to achieve multi-scale noise adaptive suppression, improving the signal-to-noise ratio of disturbance signals. It also combines empirical mode decomposition to extract local features of non-stationary signals, constructs a sensitivity evaluation index to select the mode components that best reflect the disturbance characteristics, and uses the modal maxima to accurately locate the start and end times of disturbances. This improves the adaptability and reliability of voltage disturbance detection in complex noise environments and various disturbance scenarios, providing effective technical support for online monitoring of power quality in distribution networks, disturbance event tracing, and operation control.

[0045] like Figure 1 As shown in the figure, this embodiment of the invention provides a disturbance detection method based on wavelet entropy denoising and empirical mode decomposition, the specific steps of which are as follows: S101: Acquire the initial voltage signal.

[0046] The initial voltage signal is a discrete power quality signal. By using a voltage transformer and a data acquisition card to "cut" a continuous voltage curve into a series of digital samples at regular intervals, the signal becomes a finite-length, equally spaced, quantized discrete sequence, resulting in the discrete power quality signal. The discrete power quality signal is the array of digital samples into which the continuous voltage / current is cut into equally spaced digital samples using a transformer and a data acquisition card, preserving all the characteristics of power quality events (sags, harmonics, oscillations).

[0047] S102: Perform multi-scale analysis on the time-domain local features of the initial voltage signal to obtain multi-scale high and low frequency components.

[0048] Furthermore, multi-scale analysis is performed on the time-domain local features of the initial voltage signal to obtain multi-scale high- and low-frequency components, referring to... Figure 2 Specifically, it includes: The initial voltage signal is decomposed into multiple layers using a low-pass filter to obtain multiple low-frequency coefficients; A high-pass filter is used to perform multi-level decomposition on the initial voltage signal to obtain multi-level high-frequency coefficients; By fusing high-frequency and low-frequency coefficients, multi-scale high- and low-frequency components are obtained.

[0049]

[0050] Where k is the time index of the initial voltage signal, n is the index after multi-level decomposition, j is the number of wavelet decomposition levels, and c j [k] represents the low-frequency coefficients of the j-th layer, and h() represents the low-pass filter.

[0051] Furthermore, the high-frequency coefficients are specifically represented as follows:

[0052] Where k is the time index of the initial voltage signal, n is the index after multi-level decomposition, j is the number of wavelet decomposition levels, and d j [k] represents the high-frequency coefficients of the j-th layer, and g() represents the high-pass filter.

[0053] In one specific implementation, using discrete wavelet transform to perform multi-scale analysis on the time-domain local features of the acquired initial voltage signal can effectively filter out noise and retain the original detailed features of the signal, thereby obtaining multi-scale high and low frequency components.

[0054] The initial voltage signal x[n] is decomposed into multiple layers using a low-pass filter h(n) and a high-pass filter g(n). The discrete wavelet transform employs a doubling downsampling, reducing the number of data points by half after each decomposition. This achieves efficient signal analysis while reducing computational load. By continuously decomposing the low-frequency components of the signal, a multi-layered wavelet decomposition structure (i.e., multi-scale high and low frequency components) can be obtained, specifically expressed as:

[0055] Where x[n] is the initial voltage signal, (c j ,d j ,…,d1) represent multi-scale high and low frequency components, c j For the j-th layer low-frequency coefficients, the signal detail features are mainly concentrated in c. j In the middle, d j For the j-th layer high-frequency coefficient, noise is concentrated in the high-frequency coefficient d. j middle.

[0056] Denoising of the original signal can be achieved by threshold filtering the high-frequency coefficients. In the example provided in this invention, the wavelet decomposition uses the Daubechies wavelet basis, preferably the db4 or db6 wavelet basis, and the number of decomposition layers is 3 to 6 to enhance the noise suppression capability and the ability to preserve perturbation details.

[0057] S103: Based on the high-frequency coefficients of the multi-scale high and low frequency components, analyze the wavelet entropy at different scales to obtain the corresponding wavelet threshold.

[0058] Furthermore, based on the high-frequency coefficients in the multi-scale high and low-frequency components, wavelet entropy at different scales is analyzed to obtain the corresponding wavelet thresholds, referring to... Figure 3 Specifically, it includes: The wavelet energy of each layer is calculated based on the high-frequency coefficients of each layer in the multi-scale high and low frequency components. The wavelet energy of each layer is normalized to obtain the wavelet energy probability distribution of each layer. Based on the wavelet energy probability distribution, the wavelet entropy of each layer is determined; The wavelet threshold for each layer is determined by combining the wavelet entropy of each layer with the corresponding noise standard deviation.

[0059] In one specific implementation, the wavelet energy of each layer is first calculated based on the high-frequency coefficients of that layer. Then, the wavelet energy of each layer is normalized to obtain the wavelet energy probability distribution for each layer, which represents the proportion of that layer in the overall signal energy. Next, the wavelet entropy of each layer can be calculated based on the wavelet energy probability distribution. Finally, the wavelet entropy of each layer is normalized, and the wavelet threshold corresponding to the high-frequency coefficients of each layer is calculated based on the magnitude of the wavelet entropy. A larger wavelet entropy value indicates a higher noise content at that scale; in this case, the adjustment coefficient is less than 1, and a smaller wavelet threshold is selected. Conversely, a smaller wavelet entropy value indicates a higher wavelet threshold, thereby achieving noise filtering and preservation of the original signal's detailed features.

[0060] Furthermore, wavelet energy is specifically expressed as:

[0061] Among them, E j Let d be the wavelet energy of the j-th layer. j [k] represents the high-frequency coefficient with time index k in the j-th layer, where k is the time index of the initial voltage signal.

[0062] Furthermore, the wavelet energy probability distribution is specifically expressed as:

[0063] Among them, P j Let E be the wavelet energy probability distribution of the j-th layer. j Let J be the wavelet energy of the j-th layer, and J be the total number of layers.

[0064] Furthermore, wavelet entropy is specifically expressed as:

[0065] Among them, WEj Let P be the wavelet entropy of the j-th layer. j Let be the wavelet energy probability distribution of the j-th layer, and log() be the logarithmic function.

[0066] Furthermore, based on the wavelet entropy of each layer and the corresponding noise standard deviation, the wavelet threshold for each layer is determined, specifically as follows: The wavelet entropy of each layer is normalized to obtain the adjustment coefficient of each layer; Determine the high-frequency median based on the high-frequency coefficients of each layer; The noise standard deviation is estimated by using the median absolute deviation method combined with the high-frequency median. The wavelet threshold is obtained by fusing the modulation coefficient and the noise standard deviation, combined with the signal length.

[0067] In one specific implementation, the median absolute deviation (MAD) method is used to determine the noise standard deviation corresponding to the wavelet entropy. First, the high-frequency coefficients d of each layer are calculated. j The median(d) of [k] j [k]) (i.e., the high-frequency median). Then calculate the high-frequency coefficients d for each layer. j The absolute deviation of [k] from the high-frequency median |d j [k]-median(d j [k])|. Then calculate the median of these absolute deviations, i.e., MAD(d j [k])=median(|d j [k]-median(d j [k])|). Because the median of absolute deviation is almost immune to outliers, it is often used as a benchmark for robust standard deviation. Under the Gaussian assumption, σ j ≈MAD(d j [k]) / 0.6745.

[0068] Furthermore, the wavelet threshold is specifically expressed as:

[0069] Among them, T j Let α be the wavelet threshold of the j-th layer. j Let α be the adjustment coefficient for the j-th layer. j ∈[0.5,1.5], σ j Let N be the noise standard deviation of the j-th layer, and N be the signal length of the j-th layer.

[0070] S104: Based on wavelet thresholding and combined with high-frequency coefficients at different scales, the high and low frequency components at multiple scales are denoised and reconstructed to obtain the voltage reconstruction signal.

[0071] Furthermore, based on wavelet thresholding and combined with high-frequency coefficients at different scales, denoising and reconstruction of multi-scale high and low frequency components are performed to obtain the voltage reconstruction signal, referring to... Figure 4 Specifically, it includes: Based on the wavelet threshold of each layer, the high-frequency coefficients are judged and updated, and smooth convergence processing is performed to obtain the updated high-frequency coefficients. Inverse discrete wavelet transform is used to reconstruct the high-frequency coefficients and the low-frequency coefficients in the multi-scale high and low frequency components, resulting in a reconstructed time-domain signal and an output voltage reconstructed signal.

[0072] In one specific implementation, after obtaining the wavelet thresholds for each layer, a soft thresholding method is used to smooth and shrink the high-frequency coefficients. High-frequency coefficients smaller than the wavelet threshold are directly set to zero, while high-frequency coefficients larger than the wavelet threshold are shrunk towards zero, resulting in updated high-frequency coefficients. This is specifically expressed as follows:

[0073] in, For the high-frequency coefficients updated at time index k in the j-th layer, sign() is the sign function, and d j [k] represents the high-frequency coefficients of the j-th layer with time index k, T j is the wavelet threshold of the j-th layer.

[0074] After thresholding the high-frequency coefficients of each layer, the inverse discrete wavelet transform signal is used for reconstruction to obtain the denoised voltage reconstruction signal x′[n].

[0075] Furthermore, inverse discrete wavelet transform is used to reconstruct the updated high-frequency coefficients and the low-frequency coefficients in the multi-scale high and low-frequency components, resulting in a reconstructed time-domain signal and an output voltage reconstructed signal, specifically including: Upsample the high-frequency coefficients and the low-frequency coefficients in the multi-scale high- and low-frequency components respectively to obtain the first high-frequency coefficients and the first low-frequency coefficients; The first low-frequency coefficients are synthesized and convolved using a low-pass synthesis filter to obtain a low-frequency coefficient sequence; A high-pass synthesizing filter is used to synthesize and convolve the first high-frequency coefficients to obtain a high-frequency coefficient sequence; The low-frequency coefficient sequence and the high-frequency coefficient sequence are added point by point to obtain the current reconstructed time-domain signal; Repeat the above process. If the current reconstructed time-domain signal has the same sampling frequency as the initial voltage signal, output the reconstructed voltage signal.

[0076] It's important to understand that the inverse discrete wavelet transform (IDWT) resynthesizes processed wavelet coefficients (thresholding, filtering, quantization, etc.) into a time-domain signal, completing a "decomposition-processing-reconstruction" closed loop to achieve denoising, compression, and feature extraction. In one specific implementation, firstly, the updated high-frequency and low-frequency coefficients are upsampled. For example, zeros are inserted into both the low-frequency and updated high-frequency coefficients to double their length, resulting in first high-frequency and first low-frequency coefficients. Then, a low-pass synthesis filter is used to convolve the first low-frequency coefficients to obtain a low-frequency coefficient sequence, and a high-pass synthesis filter is used to convolve the first high-frequency coefficients to obtain a high-frequency coefficient sequence. Finally, the two convolution results (i.e., the low-frequency and high-frequency coefficient sequences) are summed point-by-point to obtain the next-level (or final) time-domain sequence. This process is repeated until the top layer, outputting a voltage reconstruction signal x′[n] with the same sampling rate as the original.

[0077] S105: The voltage reconstruction signal is decomposed using an empirical mode decomposition model to obtain multiple intrinsic mode function components.

[0078] In one specific implementation, the denoised voltage reconstructed signal is decomposed using an empirical mode decomposition model to obtain multiple intrinsic mode function (IMF) components. The voltage reconstructed signal x′[n] is specifically represented as follows:

[0079] Among them, IMF i [k] represents the i-th intrinsic mode function component, r n (k) represents the residual component. The IMF satisfies two conditions: ① the difference between the number of local maxima and local minima in the data sequence does not exceed 1; ② the local means of the upper and lower envelopes of the IMF are close to 0.

[0080] S106: Combining the sensitivity evaluation index, sort the intrinsic mode function components, screen out the perturbation feature components, determine the start and end times of the perturbation of the perturbation feature components, and complete the perturbation detection.

[0081] Furthermore, by combining sensitivity evaluation indicators, the intrinsic mode function components are sorted, disturbance characteristic components are selected, and the start and end times of the disturbance characteristic components are determined to complete the disturbance detection. Figure 5 Specifically, it includes: Based on the sensitivity evaluation index, the components of each intrinsic mode function are analyzed to obtain the sensitivity values ​​of each intrinsic mode function component. The perturbation feature component corresponding to the maximum sensitivity value is obtained by sorting the sensitivity values ​​of each intrinsic mode function component in descending order. The extreme values ​​in the perturbation characteristic components are obtained, and the magnitudes of each extreme value are compared. The start and end times of the perturbation are obtained by combining the time points of each extreme value.

[0082] Furthermore, the sensitivity evaluation indicators refer to... Figure 6 This is obtained through the following steps: Calculate the total energy corresponding to each intrinsic mode function component; Based on the first sensitivity coefficient, the energy proportion of each intrinsic mode function is obtained through the total energy of each intrinsic mode function component, and the first sensitivity term is determined, wherein the first sensitivity term is the product of the first sensitivity coefficient and the energy proportion of the intrinsic mode function; Based on the second sensitivity coefficient, the correlation between the voltage reconstructed signal and the intrinsic mode function components is analyzed, the correlation coefficient is given, and the second sensitivity term is determined, wherein the second sensitivity term is the product of the absolute value of the second sensitivity coefficient and the correlation coefficient; By integrating the first and second sensitivity items, sensitivity evaluation indicators are determined.

[0083] Furthermore, the sensitivity evaluation index is specifically expressed as follows:

[0084] Among them, s i Let Q be the sensitivity evaluation index for the i-th intrinsic mode function component, where λ1 is the first sensitivity coefficient, λ2 is the second sensitivity coefficient, and Q is the third sensitivity coefficient. i Let ρ be the total energy of the i-th intrinsic mode function component. i Let x' be the correlation coefficient between the i-th intrinsic mode function component and the voltage reconstructed signal, and let x'[n] be the voltage reconstructed signal. i [k] represents the energy value of the i-th intrinsic mode function component with time index k, and K is the total time length. Let be the average energy of the i-th intrinsic mode function component. This represents the average frequency of the voltage reconstruction signal.

[0085] In this example, λ1=0.6 and λ2=0.4. In other examples, the values ​​of the first and second sensitivity coefficients can be adjusted according to the actual situation, and there is no limitation on this.

[0086] In one specific implementation, the energy percentage of each intrinsic mode function component obtained from the decomposition is calculated separately. and its correlation coefficient ρ with the voltage reconstruction signal i Finally, the sensitivity evaluation index s was obtained. i .

[0087] Based on the sensitivity values ​​of each intrinsic mode function (IMF) component, all IMF components are sorted, and the IMF component with the highest sensitivity value is selected as the feature component for disturbance detection (i.e., the disturbance feature component). Finally, all extreme values ​​of the disturbance feature component are calculated, and the maximum and minimum values ​​of the magnitude are selected, i.e., the magnitude maxima and magnitude minima. The time points corresponding to the magnitude maxima and magnitude minima are taken as the disturbance start and end times of the voltage disturbance. Figure 7 , Figure 8 and Figure 9 , Figure 7 This is a graph showing the start and end times of disturbances during a voltage spurt. Figure 8 This is a graph showing the start and end times of disturbances during a voltage dip. Figure 9 This is a graph showing the start and end times of disturbances during voltage flicker.

[0088] Reference Figure 10 This invention proposes a disturbance detection method based on wavelet entropy denoising and empirical mode decomposition (IMF), applicable to the identification of the start and end times of voltage sag, voltage swell, and voltage flicker disturbance events. First, the acquired initial voltage signal is subjected to discrete wavelet decomposition to obtain multi-scale high- and low-frequency components. Second, the wavelet energy at each scale is calculated based on the high-frequency coefficients, and the wavelet entropy at each scale is further calculated. The corresponding wavelet threshold is adaptively determined based on the magnitude of the wavelet entropy, and the high-frequency coefficients are smoothed using soft thresholding. A high signal-to-noise ratio (SNR) voltage reconstruction signal is then obtained through inverse discrete wavelet transform. Next, the voltage reconstruction signal is input into an empirical mode decomposition model to obtain multiple intrinsic mode function (IMF) components. The energy proportion of each IMF component and its correlation coefficient with the initial voltage signal are calculated, and a sensitivity evaluation index is constructed. The IMF components are ranked, and the IMF component with the highest sensitivity is selected as the disturbance feature component. Finally, by detecting the position of the modulus maxima in the disturbance feature component, the corresponding time point is used as the start and end times of the voltage disturbance event (i.e., the disturbance start and end times), thereby achieving accurate identification of the voltage disturbance time.

[0089] Reference Figure 11 This invention provides a disturbance detection device based on wavelet entropy denoising and empirical mode decomposition, comprising: Signal acquisition module 201 is used to acquire the initial voltage signal; The signal analysis module 202 is used to perform multi-scale analysis on the time-domain local features of the initial voltage signal to obtain multi-scale high and low frequency components; The threshold calculation module 203 is used to analyze the wavelet entropy at different scales based on the high-frequency coefficients in the multi-scale high and low frequency components, and obtain the corresponding wavelet threshold. The signal reconstruction module 204 is used to perform noise reduction and reconstruction of multi-scale high and low frequency components based on wavelet threshold and combined with high frequency coefficients of different scales to obtain the voltage reconstruction signal. The mode decomposition module 205 is used to decompose the voltage reconstruction signal using an empirical mode decomposition model to obtain multiple intrinsic mode function components; The disturbance analysis module 206 is used to sort the intrinsic mode function components by combining sensitivity evaluation indicators, screen out disturbance feature components, determine the disturbance start and end time of the disturbance feature components, and complete the disturbance detection.

[0090] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0091] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and variations of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A disturbance detection method based on wavelet entropy denoising and empirical mode decomposition, characterized in that, include: Acquire the initial voltage signal; Multi-scale analysis of the time-domain local features of the initial voltage signal is performed to obtain multi-scale high- and low-frequency components; Based on the high-frequency coefficients of the multi-scale high and low frequency components, the wavelet entropy at different scales is analyzed to obtain the corresponding wavelet thresholds. Based on wavelet thresholding and combined with high-frequency coefficients at different scales, the high and low frequency components at multiple scales are denoised and reconstructed to obtain the voltage reconstruction signal. An empirical mode decomposition model is used to decompose the voltage reconstruction signal to obtain multiple intrinsic mode function components; By combining sensitivity evaluation indicators, the intrinsic mode function components are sorted, the perturbation feature components are screened out, the start and end times of the perturbation feature components are determined, and the perturbation detection is completed.

2. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 1, characterized in that, Multi-scale analysis of the time-domain local features of the initial voltage signal yields multi-scale high- and low-frequency components, specifically including: The initial voltage signal is decomposed into multiple layers using a low-pass filter to obtain multiple low-frequency coefficients. A high-pass filter is used to perform multi-level decomposition on the initial voltage signal to obtain multi-level high-frequency coefficients; By fusing high-frequency and low-frequency coefficients, multi-scale high- and low-frequency components are obtained.

3. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 1, characterized in that, Based on the high-frequency coefficients in the multi-scale high and low-frequency components, wavelet entropy at different scales is analyzed to obtain the corresponding wavelet thresholds, specifically including: The wavelet energy of each layer is calculated based on the high-frequency coefficients of each layer in the multi-scale high and low frequency components. The wavelet energy of each layer is normalized to obtain the wavelet energy probability distribution of each layer. Based on the wavelet energy probability distribution, the wavelet entropy of each layer is determined; The wavelet threshold for each layer is determined by combining the wavelet entropy of each layer with the corresponding noise standard deviation.

4. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 3, characterized in that, Based on the wavelet entropy of each layer and the corresponding noise standard deviation, the wavelet threshold of each layer is determined, specifically as follows: The wavelet entropy of each layer is normalized to obtain the adjustment coefficient of each layer; Determine the high-frequency median based on the high-frequency coefficients of each layer; The noise standard deviation is estimated by using the median absolute deviation method combined with the high-frequency median. The wavelet threshold is obtained by fusing the modulation coefficient and the noise standard deviation, combined with the signal length.

5. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 1, characterized in that, Based on wavelet thresholding and combined with high-frequency coefficients at different scales, denoising and reconstruction of multi-scale high and low frequency components are performed to obtain the voltage reconstruction signal, specifically including: Based on the wavelet threshold of each layer, the high-frequency coefficients are judged and updated, and smooth convergence processing is performed to obtain the updated high-frequency coefficients. Inverse discrete wavelet transform is used to reconstruct the high-frequency coefficients and the low-frequency coefficients in the multi-scale high and low frequency components, resulting in a reconstructed time-domain signal and an output voltage reconstructed signal.

6. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 5, characterized in that, Inverse discrete wavelet transform is used to reconstruct the high-frequency coefficients and low-frequency coefficients in the multi-scale high and low-frequency components, resulting in a reconstructed time-domain signal and an output voltage reconstructed signal, specifically including: Upsample the updated high-frequency coefficients and the low-frequency coefficients in the multi-scale high- and low-frequency components respectively to obtain the first high-frequency coefficients and the first low-frequency coefficients; The first low-frequency coefficients are synthesized and convolved using a low-pass synthesis filter to obtain a low-frequency coefficient sequence; A high-pass synthesizing filter is used to synthesize and convolve the first high-frequency coefficients to obtain a high-frequency coefficient sequence; The low-frequency coefficient sequence and the high-frequency coefficient sequence are added point by point to obtain the current reconstructed time-domain signal; Repeat the above process. If the current reconstructed time-domain signal has the same sampling frequency as the initial voltage signal, output the reconstructed voltage signal.

7. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 1, characterized in that, By combining sensitivity evaluation metrics, the intrinsic mode function components are sorted, perturbation feature components are selected, and the start and end times of the perturbation feature components are determined to complete the perturbation detection. Specifically, this includes: Based on the sensitivity evaluation index, the components of each intrinsic mode function are analyzed to obtain the sensitivity values ​​of each intrinsic mode function component. The perturbation feature component corresponding to the maximum sensitivity value is obtained by sorting the sensitivity values ​​of each intrinsic mode function component in descending order. The extreme values ​​in the perturbation characteristic components are obtained, and the magnitudes of each extreme value are compared. The start and end times of the perturbation are obtained by combining the time points of each extreme value.

8. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 7, characterized in that, Sensitivity evaluation metrics are obtained through the following steps: Calculate the total energy corresponding to each intrinsic mode function component; Based on the first sensitivity coefficient, the energy ratio of each intrinsic mode function is obtained through the total energy of each intrinsic mode function component, and the first sensitivity term is determined. Based on the second sensitivity coefficient, the correlation between the voltage reconstruction signal and the intrinsic mode function components is analyzed, the correlation coefficient is given, and the second sensitivity term is determined. By integrating the first and second sensitivity items, sensitivity evaluation indicators are determined.

9. The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in claim 8, characterized in that, The first sensitivity term is the product of the first sensitivity coefficient and the energy proportion of the intrinsic mode function; the second sensitivity term is the product of the second sensitivity coefficient and the absolute value of the correlation coefficient.

10. A disturbance detection device based on wavelet entropy denoising and empirical mode decomposition, characterized in that, The disturbance detection method based on wavelet entropy denoising and empirical mode decomposition as described in any one of claims 1-9 includes: The signal acquisition module is used to acquire the initial voltage signal; The signal analysis module is used to perform multi-scale analysis on the local time-domain features of the initial voltage signal to obtain multi-scale high and low frequency components; The threshold calculation module is used to analyze the wavelet entropy at different scales based on the high-frequency coefficients in the multi-scale high and low frequency components, and obtain the corresponding wavelet thresholds. The signal reconstruction module is used to denoise and reconstruct multi-scale high and low frequency components based on wavelet thresholds and combined with high-frequency coefficients at different scales to obtain the voltage reconstruction signal. The mode decomposition module is used to decompose the voltage reconstruction signal using an empirical mode decomposition model to obtain multiple intrinsic mode function components; The disturbance analysis module is used to sort the intrinsic mode function components by combining sensitivity evaluation indicators, screen out disturbance feature components, determine the start and end times of disturbance feature components, and complete disturbance detection.

Images