A dynamic oscillation detection method and system based on an elliptical band-pass filter set and an unscented Kalman filter

Abstract

The application discloses a kind of based on elliptical band-pass filter set and dynamic oscillation detection method and system of unscented Kalman filtering, belong to power system oscillation detection technical field.The method first collects the oscillation current signal of grid-connected point, then is filtered by trap filter to filter out fundamental component, again extract target oscillation band by elliptical band-pass filter set, finally unknown oscillation frequency and amplitude are estimated on-line using unscented Kalman filtering with frequency random walk model.The narrow transition zone, high selectivity characteristics of elliptical band-pass filter are combined with the nonlinear state tracking ability of unscented Kalman filtering, by introducing frequency random walk model in UKF, frequency state quantity can be adaptively adjusted with system operating condition change, effectively solve the problem of low precision of oscillation parameter detection and poor frequency tracking adaptive ability under harmonic interference in traditional method.

Description

Technical Field

[0001] This invention relates to the field of power system oscillation detection technology, specifically to a dynamic oscillation detection method and system based on an elliptic band-pass filter bank (EBP) and an unscented Kalman filter (UKF). Background Technology

[0002] With the widespread application of modular multilevel converter-based high-voltage direct current (MMC-HVDC) technology in large-scale renewable energy grid connection and long-distance power transmission, the dynamic interaction between MMC and AC grid has become increasingly complex. The resulting broadband oscillation problem has become a significant factor threatening the safe and stable operation of the power system. Oscillations in MMC-connected systems often exhibit characteristics such as frequency uncertainty, time-varying amplitude, and complex harmonic background. Accurate and rapid detection of oscillation frequency and amplitude parameters is a prerequisite for implementing effective oscillation suppression measures.

[0003] Currently, the following technical approaches are mainly used for power system oscillation detection:

[0004] (I) Spectrum analysis method based on Fast Fourier Transform (FFT). This method obtains the spectrum distribution by performing an FFT on the sampled signal, and then identifies the oscillation frequency. Its advantages are simple principle and intuitive calculation, but it has the inherent defect that the frequency resolution is limited by the sampling window length. In the analysis of non-stationary oscillation signals, it is difficult to balance the contradiction between frequency resolution and time domain response speed. Moreover, it is sensitive to harmonic interference and is prone to misidentifying adjacent harmonics as oscillation components.

[0005] (II) Frequency tracking method based on phase-locked loop (PLL). This method indirectly detects the oscillation frequency by tracking the frequency change of the system through a PLL. However, under oscillation conditions, the dynamic response of the PLL itself will couple with the oscillation signal, causing frequency tracking distortion. Furthermore, the filtering performance of the PLL degrades under multiple harmonic backgrounds, making it difficult to guarantee detection accuracy.

[0006] (III) State estimation methods based on Kalman filtering and its improved algorithms. Extended Kalman Filter (EKF) achieves state estimation by linearizing the nonlinear system, but in strongly nonlinear systems, the accumulation of linearization errors easily leads to filter divergence. Unscented Kalman Filter (UKF) approximates the nonlinear transformation through a deterministic sampling strategy, avoiding the linearization error of EKF, and has been increasingly widely used in power system state estimation. However, existing UKF-based oscillation detection methods often treat the oscillation frequency as a constant parameter or propagate only through a fixed angular velocity, failing to consider the slow drift of the oscillation center frequency caused by changes in operating conditions in the actual system, resulting in insufficient adaptive frequency tracking capability. Furthermore, existing methods lack effective signal preprocessing techniques; in grid-connected current signals containing multiple harmonic components, these harmonic components severely interfere with the UKF's estimation accuracy of unknown oscillation parameters.

[0007] (iv) Bandpass filter-based frequency band extraction method. Traditional bandpass filters (such as Butterworth and Chebyshev types) often require high order to achieve selective extraction of the target frequency band, and have a wide transition band. In the spectrum environment with dense harmonics, the signal leakage between adjacent frequency bands is serious, making it difficult to achieve accurate separation of the target oscillation frequency band.

[0008] In summary, existing technologies have the following shortcomings: First, they lack effective signal preprocessing methods, making it difficult to accurately extract the target oscillation frequency band signal under multiple harmonic interferences; second, traditional UKFs treat the oscillation frequency as a constant parameter, failing to track the slow drift of the oscillation center frequency with changing operating conditions in actual systems, resulting in insufficient adaptive frequency estimation; third, there is a lack of coordinated design between the bandpass filter and the UKF, with front-end filtering and back-end state estimation being disconnected, limiting overall detection performance. Therefore, a dynamic oscillation detection method capable of achieving high-precision online estimation of oscillation frequency and amplitude under harmonic interference is urgently needed. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a dynamic oscillation detection method and system based on elliptic bandpass filter banks and unscented Kalman filtering, so as to solve the technical problems of low accuracy of oscillation parameter detection and poor frequency tracking and adaptive capability of the existing methods under harmonic interference.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] A dynamic oscillation detection method based on an elliptic bandpass filter bank and an unscented Kalman filter includes the following steps:

[0012] S1: Acquire the oscillating current signal at the grid connection point;

[0013] S2: The oscillating current signal acquired in step S1 is filtered by a notch filter. The fundamental frequency is used as the notch filter center frequency to filter out the fundamental frequency component, and the residual signal after filtering out the fundamental frequency is output.

[0014] S3: Construct an elliptic bandpass filter bank to extract the target oscillation frequency band from the residual signal output in step S2, and output the observation signal after bandpass filtering;

[0015] S4: Construct an unscented Kalman filter state tracking model with a frequency random walk model, and perform online estimation of the unknown oscillation frequency and amplitude in the observed signal output in step S3, and output the estimated values ​​of oscillation frequency and amplitude.

[0016] Furthermore, in step S1, a current transformer is connected in series at the common coupling point of the grid-connected system to collect the oscillating current signal. The current transformer converts the large current in the high-voltage grid into a small current signal, which is then filtered and amplified by the signal conditioning circuit.

[0017] Furthermore, in step S2, the fundamental frequency f1 output by the phase-locked loop is used as the notch center frequency to construct a second-order infinite impulse response notch filter, and the normalized notch center frequency is:

[0018] ;

[0019] The normalized bandwidth is:

[0020] ;

[0021] Where Fs is the sampling frequency. This is the quality factor for notch filtering;

[0022] The transfer function of a second-order IIR notch filter is:

[0023] ;

[0024] in, Let z represent the z-domain transfer function of a second-order IIR notch filter. Denotes the unit sampling delay operator, b i a i i=0,1,2 are the numerator and denominator coefficients of the filter, respectively; f1 is the fundamental frequency of the PLL output; Fs is the sampling frequency; ω0 is the normalized notch filter center frequency; b w For normalized bandwidth; Q notch This is the quality factor for notch filtering.

[0025] Furthermore, in step S3, the elliptic bandpass filter bank is constructed based on the principle of elliptic function approximation, and both its passband and stopband have equiripple characteristics. Given a passband ripple R...p and stopband attenuation R s A steep amplitude-frequency transition characteristic is achieved with the lowest order under constraints;

[0026] Let the fundamental frequency be f1, and the target detection frequency band be selected based on the harmonic distribution. The stopband boundary is set to ,in, , To create transition bands on both sides of the target passband; the transfer function of the elliptic bandpass filter is expressed as:

[0027] ;

[0028] in and These are the numerator and denominator polynomials of the filter, respectively.

[0029] The observed signal after elliptic bandpass filtering is represented as follows:

[0030] ;

[0031] in The observed signal is used to enter the state estimator.

[0032] Furthermore, in step S4, the unscented Kalman filter with frequency random walk model introduces the unknown oscillation frequency as a variable state variable into the state space model, and uses the frequency random walk mechanism to characterize the slow drift of the oscillation center frequency as the system operating conditions change.

[0033] The state vector is defined as:

[0034] ;

[0035] in, , Let represent the sine and cosine components of the nth residual harmonic, respectively. , Represents the sine and cosine components of an unknown oscillation. This indicates an unknown oscillation frequency.

[0036] Furthermore, in step S4, the state of the unknown oscillation frequency... Introducing a random walk model:

[0037] ;

[0038] ;

[0039] In the formula, For frequency state process noise, The variance of the frequency random walk is used to characterize the intensity of the frequency random walk. This indicates that the mean is 0 and the variance is 0. The model uses a Gaussian distribution; the random walk model allows the unknown oscillation frequency between adjacent sampling times to vary randomly within a small range while maintaining the estimate of the previous time, thus enabling UKF to track the slow drift of the oscillation center frequency online.

[0040] Furthermore, in step S4, the online estimation of the unscented Kalman filter specifically includes:

[0041] (1) Set the initial values ​​for the extended state equations: ;

[0042] (2) Calculate the Sigma point and make its sample mean and covariance equal to the statistics of the prior state distribution;

[0043] (3) After the state equation propagates through each Sigma point, the predicted particles are obtained, and the predicted state variables are calculated;

[0044] (4) Calculate the predicted observations, observation covariance, and state-observation cross-covariance;

[0045] (5) Solve for the filter gain and update the state estimate;

[0046] (6) Calculate the oscillation amplitude and frequency based on the state estimate.

[0047] A dynamic oscillation detection system based on an elliptic bandpass filter bank and an unscented Kalman filter, for performing the method described above, includes:

[0048] The current signal acquisition unit is used to acquire the oscillating current signal at the grid connection point;

[0049] The notch filter unit is used to filter the oscillating current signal with the fundamental frequency as the notch center frequency, filter out the fundamental frequency component, and output the residual signal after filtering out the fundamental frequency.

[0050] Elliptical bandpass filter unit is used to construct an elliptical bandpass filter bank, extract the target oscillation frequency band from the output residual signal, and output the bandpass filtered observation signal.

[0051] The unscented Kalman filter estimation unit is used to construct an unscented Kalman filter state tracking model with a frequency random walk model, to estimate the unknown oscillation frequency and amplitude in the observed signal online, and to output the oscillation frequency estimate and amplitude estimate.

[0052] Furthermore, the current signal acquisition unit includes a current transformer and a signal conditioning circuit connected in series at the common coupling point of the grid-connected system. The current transformer is used to convert the large current in the high-voltage grid into a small current signal, and the signal conditioning circuit is used to filter and amplify the converted current signal.

[0053] Furthermore, the unscented Kalman filter estimation unit introduces the unknown oscillation frequency as a variable state variable into the state space model. The state vector includes the sinusoidal component of the residual harmonics, the cosine component of the residual harmonics, the sinusoidal component of the unknown oscillation, the cosine component of the unknown oscillation, and the unknown oscillation frequency state. A random walk model is introduced into the frequency state, enabling the filter to track the slow drift of the oscillation center frequency as the system operating conditions change online.

[0054] Compared with the prior art, the present invention has the following advantages:

[0055] 1. Elliptic bandpass filters enable highly selective extraction of the target frequency band and effectively suppress harmonic interference.

[0056] This invention employs an elliptical bandpass filter bank to preprocess the oscillating current signal. Taking advantage of the equal ripple characteristics in both the passband and stopband of the elliptical filter and its steep transition band, it achieves high selectivity extraction of the target oscillation frequency band signal at a lower order, with less signal leakage between adjacent frequency bands. This effectively suppresses the interference of harmonic components on the UKF parameter estimation and significantly improves the signal-to-noise ratio of the oscillation signal.

[0057] 2. The frequency random walk model gives UKF frequency tracking adaptive capabilities.

[0058] This invention introduces the unknown oscillation frequency as a variable state variable into the UKF state-space model and innovatively introduces a frequency random walk mechanism, allowing the frequency state variable to vary randomly within a small range while maintaining the estimated value of the previous time step. This design enables the UKF to track the slow drift of the oscillation center frequency in the actual system as the operating conditions change, overcoming the shortcomings of traditional UKF models that suffer from insufficient adaptive capability due to treating the frequency as a constant parameter.

[0059] 3. Coordinated design of filtering preprocessing and state estimation enables high-precision online detection.

[0060] This invention employs a three-stage collaborative architecture: a notch filter to remove the fundamental frequency, an elliptic bandpass filter to extract the target frequency band, and a UKF (Ultra-Knowledge-Focused Filter) for parameter estimation. This transforms the original broadband observation problem into a narrowband parameter tracking problem step by step, reducing the dimensionality of state estimation and improving estimation accuracy and convergence speed. The signal flow transmission relationship among the three stages is clear—the output of the notch filter serves as the input of the bandpass filter, and the output of the bandpass filter serves as the observation input of the UKF—forming an organically collaborative whole that achieves high-precision online estimation of oscillation frequency and amplitude under harmonic interference.

[0061] 4. Multiple simulations have verified that the detection accuracy and response speed meet engineering requirements.

[0062] This invention demonstrates the effectiveness of its method through multiple verification experiments: in grid-connected current signals containing 5th, 7th, and 11th harmonics (total THD = 5%), this method can quickly and accurately detect the oscillation frequency and amplitude; under conditions where oscillations are induced by adding 400μs and 600μs delays to the MMC control loop, the UKF can quickly track changes in the oscillation frequency and amplitude. Experimental results show that this method has high detection accuracy and fast response speed under harmonic interference, and can meet the engineering requirements for online oscillation detection in MMC grid-connected systems. Attached Figure Description

[0063] Figure 1 This is a flowchart of the dynamic oscillation detection method based on elliptic bandpass filter bank and unscented Kalman filter of the present invention;

[0064] Figure 2 A comparison of the amplitude-frequency characteristics of Butterworth bandpass filter banks and elliptic bandpass filter banks with the same filter order.

[0065] Figure 3 This is a diagram showing the effect of the method of the present invention in estimating the frequency and amplitude of the oscillation signal;

[0066] Figure 4 The FFT analysis spectra of the original signal, the notch-filtered signal, and the bandpass-filtered signal are shown.

[0067] Figure 5 This is a diagram showing the effect of the method of the present invention in tracking the oscillation frequency and amplitude under a delay of 400μs;

[0068] Figure 6 This is a diagram showing the effect of the method of the present invention in tracking the oscillation frequency and amplitude under a delay of 600μs. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0070] like Figure 1 As shown, the dynamic oscillation detection method based on elliptic bandpass filter banks and unscented Kalman filtering provided by this invention includes four core steps: current signal acquisition (S1), notch filter filtering (S2), elliptic bandpass filtering (S3), and unscented Kalman filter estimation (S4). There is a clear signal flow transmission relationship between each step: the oscillating current signal acquired in S1 is output to S2 for notch filtering; the residual signal after filtering out the fundamental frequency in S2 is output to S3 for bandpass filtering; the observed signal after extracting the target frequency band in S3 is output to S4 for UKF state estimation; and S4 outputs the final oscillation frequency and amplitude estimation results.

[0071] S1: Acquire the oscillation current signal at the MMC grid connection point

[0072] A current transformer is connected in series at the point of common coupling (PCC) of the MMC system to acquire the oscillating current signal. The current transformer is used to convert the large current (typically hundreds to thousands of amperes) in the high-voltage power grid into a small current signal (usually 0~5A or 4~20mA standard signal). Then, the signal is appropriately filtered and amplified by the signal conditioning circuit to filter out high-frequency noise interference and adjust the signal to a voltage range suitable for the input of the analog-to-digital converter (ADC) (such as 0~10V or ±5V) to facilitate subsequent digital processing and analysis of harmonic and oscillating components.

[0073] S2: The oscillating current signal acquired in step S1 is filtered by a notch filter. The fundamental frequency is used as the notch filter center frequency to filter out the fundamental frequency component, and the residual signal after filtering out the fundamental frequency is output.

[0074] Step S2 specifically includes: filtering the oscillating current signal using a notch filter; constructing a second-order impulse response (IIR) notch filter using the fundamental frequency f1 output by the phase-locked loop (PLL) as the notch center frequency; and filtering out the fundamental frequency component in the input signal. Let the sampling frequency be Fs, then the normalized notch center frequency is:

[0075] (1)

[0076] The normalized bandwidth is:

[0077] (2)

[0078] in For notch filter quality factor, The larger the value, the narrower the notch bandwidth, and the smaller the impact on other frequency components near the fundamental frequency. The transfer function of a second-order IIR notch filter is:

[0079] (3).

[0080] S3: Construct an elliptic bandpass filter bank to extract the target oscillation frequency band from the residual signal output in step S2, and output the observed signal after bandpass filtering.

[0081] Step S3 specifically includes: designing an elliptic bandpass filter bank (EBP) to extract the target oscillation frequency band of the current signal.

[0082] Elliptic bandpass filters are a class of infinite impulse response filters constructed based on the principle of elliptic function approximation, exhibiting equal ripple characteristics in both their passband and stopband. To illustrate the selectivity differences between different types of bandpass filters in target frequency band extraction, the amplitude-frequency characteristics of Butterworth and elliptic bandpass filter banks are compared under the same filter order and continuous subband conditions. The results are as follows: Figure 2 As shown. By Figure 2 It can be seen that the amplitude-frequency response of the Butterworth bandpass filter bank is relatively smooth, but the attenuation is relatively slow at the edge of the passband, and the transition band is relatively wide. When there are adjacent harmonics or other interfering frequency components near the target oscillation frequency band, the wide transition band is prone to causing frequency band leakage, which affects the separation effect between the target oscillation component and the adjacent frequency components.

[0083] In contrast, elliptic bandpass filter banks exhibit steeper passband edges and higher band selectivity at the same filter order, enabling greater amplitude attenuation within a narrower transition band. Although elliptic filters exhibit equiripple characteristics in both the passband and stopband, they achieve higher stopband attenuation at lower orders, making them more suitable for extracting target oscillation frequency bands against a background of dense harmonics. By employing elliptic bandpass filter banks, harmonic and noise components outside the target frequency band can be effectively attenuated, improving the signal-to-noise ratio of the observed signal entering the unscented Kalman filter, thus facilitating subsequent online estimation of unknown oscillation frequencies and amplitudes.

[0084] Let the sampled discrete signal of the grid-connected current be:

[0085] (4)

[0086] in, For the fundamental component, For harmonic components, The oscillation component to be identified, This represents the noise component.

[0087] Let the fundamental frequency be f1, then the target detection frequency band can be selected based on the harmonic distribution as follows:

[0088] (5)

[0089] The stopband boundary is set as follows:

[0090] (6)

[0091] in, , This creates transition bands on both sides of the target passband, thus forming transition bands. The transfer function of an elliptic bandpass filter can be expressed as:

[0092] (7)

[0093] in and These are the numerator and denominator polynomials of the filter, respectively. An elliptic filter with a given passband ripple R... p and stopband attenuation R s Under the constraints of [the specific parameter], a steep amplitude-frequency transition characteristic can be achieved with a lower order, thus making it suitable for rapidly separating target frequency band signals in a background of dense harmonics. The signal after bandpass processing can be expressed as:

[0094] (8)

[0095] in, The observed signal is used to enter the state estimator.

[0096] The role of bandpass preprocessing is to transform the original broadband observation problem into a narrowband parameter tracking problem, thereby reducing the state estimation dimension and improving the signal-to-noise ratio of the unknown oscillation components. Since residual harmonic components such as the 7th and 11th harmonics may still exist within the target frequency band, bandpass filtering alone is insufficient to accurately separate the unknown oscillations. Therefore, a joint state-space model incorporating both residual harmonics and the unknown oscillations is further constructed. For example... Figure 2 As shown, the transition band of a conventional bandpass filter bank is relatively wide, and it is prone to band leakage near adjacent harmonics. In contrast, an elliptical bandpass filter bank has a steeper amplitude-frequency transition characteristic under the same passband ripple and stopband attenuation requirements, and can more effectively suppress harmonic components outside the target frequency band, thus providing a high signal-to-noise ratio observation signal for subsequent UKF state estimation.

[0097] S4: Construct an unscented Kalman filter state tracking model with a frequency random walk model, and perform online estimation of the unknown oscillation frequency and amplitude in the observed signal output in step S3, and output the estimated values ​​of oscillation frequency and amplitude.

[0098] Step S4 specifically includes: designing an unscented Kalman filter (UKF) state tracking method with a frequency random walk model to estimate the unknown oscillation frequency and amplitude within the target frequency band online. Unlike traditional UKF methods that treat the oscillation frequency as a constant parameter or propagate only through a fixed angular velocity, this step introduces the unknown oscillation frequency as a variable state variable into the state-space model separately, and uses the frequency random walk mechanism to characterize the slow drift of the oscillation center frequency in the actual system as the operating conditions change, thereby improving the adaptive capability of frequency estimation and the accuracy of online tracking.

[0099] Based on the orthogonal decomposition concept, each sinusoidal component is written as a combination of in-phase and orthogonal components. The state vector is then defined as:

[0100] (9)

[0101] in, , Let represent the sine and cosine components of the nth residual harmonic, respectively. , Represents the sine and cosine components of an unknown oscillation. This represents the state of an unknown oscillation frequency. Let the unknown oscillation be:

[0102] (10)

[0103] (11)

[0104] in, Indicates the amplitude of the unknown oscillation. Indicates the phase of an unknown oscillation.

[0105] The steps for estimating the frequency and amplitude of unknown oscillations within the target frequency band using UKF with a random walk model are as follows:

[0106] (1) Initial values ​​of the extended state equations:

[0107] (12)

[0108] in, For error covariance; .

[0109] (2) Select several weighted Sigma points to form a particle set, and make its sample mean and covariance equal to the statistics of the prior state distribution. The Sigma points are calculated as follows:

[0110] (13)

[0111] In the formula: , representing a matrix The i-th row or column vector of the square root. n is the dimension of the state space, and Sigma is the point scaling parameter. Sigma point distribution range adjustment parameters Secondary adjustment parameters .

[0112] (3) Construct a state prediction model with random walk. For the extended state vector To ensure that the filter maintains the continuity of frequency estimation while tracking the slow shift in the actual oscillation frequency, the frequency state... Introducing a random walk model:

[0113] (14)

[0114] (15)

[0115] In the formula, For frequency state process noise, The variance of the frequency random walk is used to characterize the intensity of the frequency random walk. This indicates that the mean is 0 and the variance is 0. The equation shows that the unknown oscillation frequency generally maintains the estimated value of the previous moment between adjacent sampling times, but allows for random variation within a small range, thus enabling UKF to track the slow drift of the oscillation center frequency during system operation online.

[0116] The predicted particles are obtained from each Sigma point after propagation through the equation of state:

[0117] (16)

[0118] Predicted state variables :

[0119] (17)

[0120] In the formula , , .

[0121] (4) The observation equation predicts the following:

[0122] (18)

[0123] (19)

[0124] in, Representation matrix The i-th row or column vector of the square root. Predicted observations Observation covariance State-observation cross-covariance Calculated by the following formula:

[0125] (20)

[0126] (twenty one)

[0127] (twenty two)

[0128] in, Let be the noise covariance.

[0129] (5) Solve for the filter gain in sequence Update the state estimate accordingly. :

[0130] (twenty three)

[0131] (twenty four)

[0132] Among them, P y,y To predict the observed covariance matrix, P x,y Let be the cross-covariance matrix between state variables and observations.

[0133] To measure the noise covariance matrix, the predicted value is updated using the estimate at time k, and the system prediction for the entire period is obtained by rolling calculation based on the time series.

[0134] (6) The amplitude and frequency estimation results are obtained according to equation (25):

[0135] (25).

[0136] in, and These are the orthogonal state components of the unknown oscillation component. Let f be the frequency of the unknown oscillation state, and A be the amplitude of the unknown oscillation.

[0137] To verify the effectiveness of the method of the present invention, the following three sets of verification experiments were conducted:

[0138] Experiment 1: Detection of Oscillation Signals with Harmonic Interference

[0139] Set a verification signal to verify the effectiveness of the detection method: sampling frequency Sampling period Fundamental frequency , Fundamental amplitude The fundamental initial phase is 0.30 rad. The expression for the grid-connected current signal containing noisy signals is set as follows:

[0140] (31)

[0141] In the formula, the oscillation frequency is set to 520Hz, the oscillation amplitude is 0.05pu, the total THD of the 5th, 7th, and 11th harmonics is 5%, and the amplitude of the 5th harmonic is... 7th harmonic amplitude 11th harmonic amplitude The corresponding initial phases are -0.80 rad, 1.10 rad, and -0.35 rad, respectively. For noise with a standard deviation of 0.01, the fundamental notch filter quality factor is... The elliptic bandpass filter has a passband of 350~550Hz, a stopband of below 320Hz and above 580Hz, a passband ripple of 0.5dB, and a stopband attenuation of 60dB.

[0142] like Figure 4 As shown, the fundamental component dominates the original signal, while also containing the 7th and 11th harmonics and a 520 Hz oscillation component. After notch filtering, the fundamental component is significantly suppressed. Further processing with an elliptic bandpass filter preserves the signal within the target oscillation band while effectively reducing out-of-band interference. This result demonstrates that cascaded preprocessing with a notch filter and an elliptic bandpass filter can improve the signal-to-noise ratio of the target oscillation component, which is beneficial for UKF to stably estimate unknown oscillation frequencies and amplitudes.

[0143] The method of this invention effectively estimates the frequency and amplitude of oscillating signals as follows: Figure 3 As shown, it can be seen that even when adjacent frequency harmonics are present, the method of the present invention can still detect the frequency and amplitude of the oscillation relatively quickly and accurately.

[0144] Experiment 2: Oscillation Detection under a 400μs Delay Condition

[0145] At 2.0s, a 400µs delay is added to the MMC control loop, causing the system to oscillate. FFT analysis yields an oscillation frequency of 1855.4Hz. At 2.2s, a damping device is activated, causing the oscillation to decay. The method of this invention effectively estimates the frequency and amplitude of the oscillation signal as follows: Figure 5 As shown, after the oscillation starts, the UKF can still track the frequency and amplitude of the oscillation relatively quickly and accurately.

[0146] Experiment 3: Oscillation Detection under a 600μs Delay Condition

[0147] At 2.0s, a 600µs delay is added to the MMC control loop, causing the system to oscillate. FFT analysis yields an oscillation frequency of 732.4Hz. At 2.2s, a damping device is activated, causing the oscillation to decay. The method of this invention effectively estimates the frequency and amplitude of the oscillation signal as follows: Figure 6 As shown, after the oscillation starts, the UKF can still track the frequency and amplitude of the oscillation relatively quickly and accurately.

[0148] The above three sets of experiments verified the detection accuracy and response speed of the method of the present invention under different oscillation frequencies and different harmonic backgrounds, proving that the method of the present invention can meet the engineering requirements of broadband oscillation online detection in MMC grid-connected systems.

[0149] This invention also provides a dynamic oscillation detection system based on an elliptic bandpass filter bank and an unscented Kalman filter, for performing the method described above, including:

[0150] The current signal acquisition unit is used to acquire the oscillating current signal at the grid connection point;

[0151] The notch filter unit is used to filter the oscillating current signal with the fundamental frequency as the notch center frequency, filter out the fundamental frequency component, and output the residual signal after filtering out the fundamental frequency.

[0152] Elliptical bandpass filter unit is used to construct an elliptical bandpass filter bank, extract the target oscillation frequency band from the output residual signal, and output the bandpass filtered observation signal.

[0153] The unscented Kalman filter estimation unit is used to construct an unscented Kalman filter state tracking model with a frequency random walk model, to estimate the unknown oscillation frequency and amplitude in the observed signal online, and to output the oscillation frequency estimate and amplitude estimate.

[0154] It should be noted that the above embodiments use oscillation detection in an MMC grid-connected system as an example to illustrate the present invention, but the application of the present invention is not limited to MMC grid-connected systems. For flexible DC transmission converters, new energy grid-connected inverters, energy storage converters, static var compensators, SVG / STATCOM, and other grid-connected systems containing power electronic conversion devices, as long as there is a broadband oscillation component to be detected in the current or voltage signal at the grid connection point, the notch filtering, elliptic bandpass filter group frequency band extraction, and unscented Kalman filtering online estimation methods described in this invention can be used to achieve dynamic detection of oscillation frequency and amplitude. Therefore, equivalent substitutions or adaptive adjustments made to the application object, sampling signal type, or grid-connected equipment type based on the technical concept of this invention should all fall within the protection scope of this invention.

[0155] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic oscillation detection method based on an elliptic bandpass filter bank and an unscented Kalman filter, characterized in that, Includes the following steps: S1: Acquire the oscillating current signal at the grid connection point; S2: The oscillating current signal acquired in step S1 is filtered by a notch filter. The fundamental frequency is used as the notch filter center frequency to filter out the fundamental frequency component, and the residual signal after filtering out the fundamental frequency is output. S3: Construct an elliptic bandpass filter bank to extract the target oscillation frequency band from the residual signal output in step S2, and output the observation signal after bandpass filtering; S4: Construct an unscented Kalman filter state tracking model with a frequency random walk model, and perform online estimation of the unknown oscillation frequency and amplitude in the observed signal output in step S3, and output the estimated values ​​of oscillation frequency and amplitude.

2. The method according to claim 1, characterized in that, In step S1, a current transformer is connected in series at the common coupling point of the grid-connected system to collect the oscillating current signal. The current transformer converts the large current in the high-voltage grid into a small current signal, which is then filtered and amplified by the signal conditioning circuit.

3. The method according to claim 1, characterized in that, In step S2, the fundamental frequency f1 output by the phase-locked loop is used as the notch center frequency to construct a second-order infinite impulse response notch filter. The normalized notch center frequency is: ； The normalized bandwidth is: ； Where Fs is the sampling frequency. This is the quality factor for notch filtering; The transfer function of a second-order IIR notch filter is: ； in, Let z represent the z-domain transfer function of a second-order IIR notch filter. Denotes the unit sampling delay operator, b i a i i=0,1,2 are the numerator and denominator coefficients of the filter, respectively; f1 is the fundamental frequency of the PLL output; Fs is the sampling frequency; ω0 is the normalized notch filter center frequency; b w For normalized bandwidth; Q notch This is the quality factor for notch filtering.

4. The method according to claim 1, characterized in that, In step S3, the elliptic bandpass filter bank is constructed based on the principle of elliptic function approximation. Both its passband and stopband have equiripple characteristics. Given a passband ripple R... p and stopband attenuation R s A steep amplitude-frequency transition characteristic is achieved with the lowest order under constraints; Let the fundamental frequency be f1, and the target detection frequency band be selected based on the harmonic distribution. The stopband boundary is set to ,in, , To create transition bands on both sides of the target passband; the transfer function of the elliptic bandpass filter is expressed as: ； in and These are the numerator and denominator polynomials of the filter, respectively. The observed signal after elliptic bandpass filtering is represented as follows: ； in The observed signal is used to enter the state estimator.

5. The method according to claim 1, characterized in that, In step S4, the unscented Kalman filter with frequency random walk model introduces the unknown oscillation frequency as a variable state variable into the state space model, and uses the frequency random walk mechanism to characterize the slow drift of the oscillation center frequency as the system operating conditions change. The state vector is defined as: ； in, , Let represent the sine and cosine components of the nth residual harmonic, respectively. , Represents the sine and cosine components of an unknown oscillation. This indicates an unknown oscillation frequency.

6. The method according to claim 5, characterized in that, In step S4, the state of the unknown oscillation frequency... Introducing a random walk model: ； ； In the formula, For frequency state process noise, The variance of the frequency random walk is used to characterize the intensity of the frequency random walk. This indicates that the mean is 0 and the variance is 0. The model uses a Gaussian distribution; the random walk model allows the unknown oscillation frequency between adjacent sampling times to vary randomly within a small range while maintaining the estimate of the previous time, thus enabling UKF to track the slow drift of the oscillation center frequency online.

7. The method according to claim 5, characterized in that, In step S4, the online estimation of the unscented Kalman filter specifically includes: (1) Set the initial values ​​for the extended state equations: ; (2) Calculate the Sigma point and make its sample mean and covariance equal to the statistics of the prior state distribution; (3) After the state equation propagates through each Sigma point, the predicted particles are obtained, and the predicted state variables are calculated; (4) Calculate the predicted observations, observation covariance, and state-observation cross-covariance; (5) Solve for the filter gain and update the state estimate; (6) Calculate the oscillation amplitude and frequency based on the state estimate.

8. A dynamic oscillation detection system based on an elliptic bandpass filter bank and an unscented Kalman filter, used to execute the method according to any one of claims 1-7, characterized in that, include: The current signal acquisition unit is used to acquire the oscillating current signal at the grid connection point; The notch filter unit is used to filter the oscillating current signal with the fundamental frequency as the notch center frequency, filter out the fundamental frequency component, and output the residual signal after filtering out the fundamental frequency. Elliptical bandpass filter unit is used to construct an elliptical bandpass filter bank, extract the target oscillation frequency band from the output residual signal, and output the bandpass filtered observation signal. The unscented Kalman filter estimation unit is used to construct an unscented Kalman filter state tracking model with a frequency random walk model, to estimate the unknown oscillation frequency and amplitude in the observed signal online, and to output the oscillation frequency estimate and amplitude estimate.

9. The system according to claim 8, characterized in that, The current signal acquisition unit includes a current transformer and a signal conditioning circuit connected in series at the common coupling point of the grid-connected system. The current transformer is used to convert the large current in the high-voltage grid into a small current signal, and the signal conditioning circuit is used to filter and amplify the converted current signal.

10. The system according to claim 8, characterized in that, The unscented Kalman filter estimation unit introduces the unknown oscillation frequency as a variable state variable into the state space model. The state vector includes the sinusoidal component of the residual harmonic, the cosine component of the residual harmonic, the sinusoidal component of the unknown oscillation, the cosine component of the unknown oscillation, and the state of the unknown oscillation frequency. By introducing a random walk model to the frequency state, the filter can track the slow drift of the oscillation center frequency as the system operating conditions change online.

Images