Virtual synchronous generator grid-connected current harmonic suppression method and system

Abstract

The invention discloses a virtual synchronous generator grid-connected current harmonic suppression method and system, and relates to the technical field of power systems, and the method comprises the steps: S1, carrying out the digital sampling of intermittent high-frequency disturbance through a wind power plant grid-connected current signal in a power grid through an analog-to-digital converter, and obtaining an original digital current sequence; s2, according to the original digital current sequence, analyzing a wind load frequency component by adopting a fast Fourier transform algorithm, and determining amplitude and phase characteristics of a high-frequency harmonic component; s3, if the amplitude of the high-frequency harmonic component exceeds a preset threshold value, extracting an intermittent disturbance harmonic sequence from the frequency component to obtain a to-be-suppressed harmonic subset; according to the virtual synchronous generator grid-connected current harmonic suppression method and system, the stability of a remote wind power plant power grid is remarkably improved, continuous monitoring and real-time suppression of intermittent disturbance are achieved, and high efficiency and reliability of power grid operation are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for suppressing grid-connected current harmonics of a virtual synchronous generator. Background Art

[0002] As the fundamental infrastructure of modern society, the stable operation of power systems is crucial to economic development and quality of life. With the rapid development of renewable energy generation and smart grids, the stability of grid-connected current in power systems has become a core requirement. Virtual synchronous generators, an emerging technology, can simulate the dynamic characteristics of traditional generators, providing stability and inertia support for the grid. However, the presence of current harmonics can disrupt normal grid operation, increase equipment losses, and even cause system failures. Therefore, research on how to effectively suppress harmonics and ensure the purity of grid-connected current has become a key topic in the power system field.

[0003] Traditional harmonic suppression methods primarily rely on analog control circuits, using hardware filters or controllers with fixed parameters to reduce the impact of harmonics. These methods exhibit significant limitations when faced with complex and changing grid environments. For example, analog control systems struggle to adapt to dynamic changes in grid loads, especially with the nonlinear loads introduced by renewable energy generation, such as wind power and photovoltaics. The frequency and amplitude of harmonics fluctuate frequently, and traditional methods are unable to adjust parameters in real time, resulting in poor suppression effects. Furthermore, analog circuits have complex hardware designs, high maintenance costs, and do not support remote monitoring and dynamic optimization, which is particularly insufficient given the distributed management requirements of modern smart grids.

[0004] In the context of digital transformation, converting harmonic suppression control systems from analog to digital signal processing has become a key research direction. However, this transformation faces significant technical difficulties. The primary issue is to achieve real-time digital sampling and analysis of grid-connected current harmonics. Harmonic signals in the power grid are characterized by high frequency and transient conditions. Traditional sampling methods are prone to distortion in high-frequency environments and have difficulty capturing the dynamic changes of harmonics. For example, in a wind power grid-connected scenario, when a sudden change in wind speed causes power generation to fluctuate, the frequency and amplitude of the harmonics will change rapidly in a short period of time. Existing sampling technologies may not be able to accurately identify harmonic components due to insufficient processing speed.

[0005] The existence of this problem directly affects the subsequent harmonic suppression effect. Accurate harmonic identification is a prerequisite for achieving directional suppression, and directional suppression requires the control system to be able to dynamically adjust the suppression parameters based on real-time analysis results. If there are deviations in the sampling and analysis links, the suppression parameters will not be able to accurately match the harmonic characteristics, resulting in reduced or even ineffective suppression effect. For example, in a distributed photovoltaic power station, if the control system cannot quickly identify the specific high-order harmonics introduced by the inverter, it may cause current waveform distortion, thereby affecting the stability of the power grid. In addition, the implementation of remote monitoring and parameter adjustment functions further exacerbates this challenge. Remote operation requires the system to have efficient data transmission and real-time performance, but the current digital control system still has bottlenecks in data processing speed and network latency. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for suppressing grid-connected current harmonics of virtual synchronous generators, which can achieve high-precision real-time sampling and analysis of harmonic signals in a complex and changeable power grid environment, and perform dynamic directional suppression based on this, while supporting remote monitoring and parameter adjustment.

[0007] To achieve the above-mentioned object, the present invention provides the following technical solutions: a virtual synchronous generator grid-connected current harmonic suppression method, comprising S1, using the wind farm grid-connected current signal in the power grid, using an analog-to-digital converter to digitally sample the intermittent high-frequency disturbance to obtain an original digital current sequence; S2, based on the original digital current sequence, using a fast Fourier transform algorithm to analyze the wind load frequency component and determine the amplitude and phase characteristics of the high-frequency harmonic component; S3, if the amplitude of the high-frequency harmonic component exceeds a preset threshold, extracting the intermittent disturbance harmonic sequence from the frequency component to obtain a harmonic subset to be suppressed; S4, using the harmonic subset to be suppressed, using an adaptive filtering algorithm to dynamically adjust the filtering parameters to generate a compensation current sequence; S5, Obtain phase matching adjustment and amplitude scaling parameters from the compensation current sequence, determine the degree of matching between the compensation current sequence and the original digital current sequence, and if the degree of matching is higher than the threshold, fuse the compensation current sequence into the main control loop through signal superposition operation and noise suppression filtering to obtain an optimized current signal; S6. Based on the optimized current signal, use a digital signal processor to calculate the suppression instruction in real time, and perform loop feedback integration and stability verification checks at the same time to generate a directional suppression pulse sequence. Through the directional suppression pulse sequence, obtain network transmission data packets, perform delay threshold judgment and network bandwidth evaluation, and if the delay is lower than the threshold, use data packet encryption transmission and transmission protocol optimization to the remote server to obtain a remote adjustment parameter set.

[0008] Preferably, S1 includes digitally sampling the wind farm grid-connected current signal through an analog-to-digital converter to generate an original digital current sequence; performing frequency domain analysis on the original digital current sequence using a fast Fourier transform algorithm to obtain the frequency component of the current signal; if there is an abnormal high-frequency component higher than a preset threshold in the frequency component, filtering the original digital current sequence through a bandpass filter to obtain a filtered current sequence; based on the filtered current sequence, extracting the time-frequency characteristics using a short-time Fourier transform algorithm to generate a time-frequency distribution of the current signal; if the duration of the intermittent high-frequency disturbance detected in the time-frequency distribution exceeds a preset threshold, decomposing the filtered current sequence through a wavelet transform algorithm to obtain a high-frequency disturbance component; based on the high-frequency disturbance component, calculating its energy distribution characteristics to determine the disturbance intensity and occurrence location; analyzing the disturbance intensity and occurrence location through a preset classification model to judge the operation status of the power grid.

[0009] Preferably, S2 includes using a fast Fourier transform algorithm to perform frequency domain analysis on the original digital current sequence to obtain the amplitude and phase characteristics of the high-frequency harmonic components; processing the original digital current sequence through a bandpass filter based on the amplitude and phase characteristics of the high-frequency harmonic components to obtain a filtered current sequence; using a short-time Fourier transform algorithm to perform time-frequency analysis on the filtered current sequence to obtain a time-frequency distribution characteristic; if the energy concentration of the frequency component of the high-frequency disturbance is detected in the time-frequency distribution characteristic, decomposing the filtered current sequence through a wavelet transform algorithm to obtain a high-frequency component sequence; calculating the time energy distribution based on the high-frequency component sequence to determine the disturbance time characteristic; if the duration of the disturbance time characteristic exceeds a preset threshold, analyzing the high-frequency component sequence through a preset classification model to judge the grid operation status; generating status assessment data based on the grid operation status to determine the grid connection stability of the wind farm.

[0010] Preferably, S3 includes: if the amplitude of the high-frequency harmonic component exceeds a preset threshold, the original current sequence is decomposed in the frequency domain by a fast Fourier transform algorithm to obtain a frequency component set; based on the frequency component set, the intermittent disturbance sequence is extracted by a spectral analysis method to generate a harmonic sequence to be suppressed; the harmonic sequence to be suppressed is processed by a bandpass filter to obtain a filtered harmonic sequence; if the energy of the filtered harmonic sequence is concentrated in the frequency range of the high-frequency disturbance, the short-time Fourier transform algorithm is used to perform time-frequency analysis to obtain the time-frequency distribution characteristics; based on the time-frequency distribution characteristics, the energy concentration sequence is extracted to generate a high-frequency disturbance sequence; the high-frequency disturbance sequence is analyzed by a preset classification model to judge the operation status of the power grid and obtain status evaluation data; based on the status evaluation data, the high-frequency disturbance sequence is suppressed by an adaptive filtering algorithm to generate a stable current sequence.

[0011] Preferably, the S4 includes dynamically adjusting the filtering parameters by using a minimum mean square adaptive filtering algorithm through a subset of harmonics to be suppressed to generate a compensation current sequence; decomposing the signal by using a discrete Fourier transform according to the compensation current sequence to obtain a set of frequency components; if there are components in the frequency component set within a high-frequency disturbance frequency range, extracting the harmonic sequence in the frequency range through a bandpass filter to obtain a filtered harmonic sequence; analyzing the time-frequency distribution by using a short-time Fourier transform according to the filtered harmonic sequence to obtain a time-frequency feature sequence; judging the state of the power grid by using a preset classification model through the time-frequency feature sequence to obtain state evaluation data; if the state evaluation data indicates that the power grid state is abnormal, performing secondary processing on the filtered harmonic sequence by using a minimum mean square adaptive filtering algorithm to generate a stable current sequence; and reconstructing the time domain signal by using an inverse Fourier transform according to the stable current sequence to obtain an optimized current sequence.

[0012] Preferably, the S5 includes extracting phase adjustment parameters and amplitude scaling parameters from the compensation current sequence, decomposing the signal using fast Fourier transform, obtaining phase and amplitude characteristics, and obtaining phase adjustment parameters and amplitude scaling parameters; calculating the phase difference and amplitude ratio between the compensation current sequence and the original digital current sequence based on the phase adjustment parameters and amplitude scaling parameters, and evaluating the matching degree between the two using the cosine similarity algorithm to obtain a matching value; if the matching value is higher than a preset threshold, the compensation current sequence is fused into the main control loop through a weighted signal superposition operation to obtain a fused current signal; for the fused current signal, a filter is used to remove the noise component to obtain a denoised current signal; based on the denoised current signal, the time domain statistical characteristics of the signal are calculated, and the signal stability is judged using a support vector machine classification model to obtain a stability evaluation result; if the stability evaluation result indicates that the signal is unstable, the denoised current signal is secondary adjusted through a least mean square adaptive filtering algorithm to obtain a stable current signal; based on the stable current signal, the time domain signal is reconstructed using an inverse fast Fourier transform to obtain an optimized current signal.

[0013] Preferably, the S6 includes obtaining time domain features from the optimized current signal, using a digital signal processor to perform real-time calculations to obtain a suppression instruction sequence; using the suppression instruction sequence, using a feedback loop to calculate the deviation from the main control loop to obtain a deviation correction sequence; if the amplitude of the deviation correction sequence exceeds a preset threshold, using a Kalman filter algorithm to optimize the suppression instruction sequence to obtain an optimized pulse sequence; based on the optimized pulse sequence, calculating the statistical characteristics of the time series, using a support vector machine classification model to judge the stability of the sequence, and obtaining a stability judgment result; based on the stability judgment result, extracting unstable sequence fragments, using an adaptive filtering algorithm for secondary adjustment to obtain a stable pulse sequence; based on the stable pulse sequence, generating a directional suppression pulse sequence, using a digital signal processor for real-time output to obtain a final pulse sequence; using the final pulse sequence, calculating the frequency distribution characteristics of the sequence, Fast Fourier transform is used to verify frequency consistency and obtain a verification pulse sequence; network transmission data packets are extracted from the directional pulse sequence, and timestamp analysis is used to calculate the data packet delay to obtain a delay judgment result; if the delay judgment result is lower than a preset threshold, an encryption standard algorithm is used to encrypt the network transmission data packet to obtain an encrypted data packet; based on the encrypted data packet, a transmission control protocol optimization strategy is used to adjust the data packet sending order to generate an optimized data stream; through the optimized data stream, a sliding window protocol is used to analyze the network bandwidth occupancy to obtain bandwidth allocation parameters; based on the bandwidth allocation parameters, a flow control algorithm is used to adjust the data packet transmission rate to generate an adjusted data stream; remote adjustment parameters are extracted from the adjusted data stream, and a data compression algorithm is used to generate a compression parameter set to obtain a remote adjustment parameter set; through the remote adjustment parameter set, a verification algorithm is used to verify the integrity of the parameter set to obtain a verification parameter set.

[0014] Preferably, it also includes S7, extracting feedback values ​​from the remote adjustment parameter set, parsing the response of the server and downloading and updating the parameter set, and updating the local filtering parameters by using the proportional integral differential control algorithm in combination with the error retransmission mechanism and the feedback value extraction process to generate a final suppressed output sequence, specifically including extracting feedback values ​​from the remote adjustment parameter set, separating the effective feedback data by using the data parsing algorithm to obtain a feedback data set; extracting update parameters by using the server response parsing based on the feedback data set to generate an updated parameter set; if the updated parameter set meets the preset threshold, adjusting the feedback data set by using the proportional integral differential control algorithm to obtain an adjusted data set; detecting data integrity by combining the error retransmission mechanism based on the adjusted data set to generate a retransmission correction data set; extracting local filtering parameters from the retransmission correction data set, optimizing parameter smoothness by using the sliding average algorithm to obtain smoothing filtering parameters; adjusting the suppressed output sequence by using the smoothing filtering parameters, and forming the final output sequence by using the sequence generation algorithm; verifying the sequence integrity by using the verification algorithm based on the final output sequence to obtain a verified output sequence.

[0015] Preferably, it also includes S8, judging whether the intermittent disturbance stability index of the wind farm power grid meets the requirements based on the final suppression output sequence, and if so, looping back to the sampling process of the original digital current sequence to maintain a continuous monitoring state, specifically including extracting the grid stability characteristics from the suppression output sequence, and using a feature extraction algorithm to separate the intermittent disturbance data to obtain a disturbance feature set; if the disturbance feature set is within the deviation range of the preset threshold, using a threshold comparison algorithm to verify the stability index and generate a verification result set; based on the verification result set, using a cyclic sampling mechanism to re-collect data from the digital current sequence to obtain an updated sampling data set; extracting time series features from the updated sampling data set, and using a sliding window algorithm to analyze the continuity of the sequence to obtain a continuity analysis result; if the continuity analysis result meets the preset continuity condition, using a data integration algorithm to merge the updated sampling data set and the disturbance feature set to generate an integrated monitoring data set; based on the integrated monitoring data set, using a state monitoring algorithm to judge the operation state of the power grid to obtain a state evaluation result; extracting abnormal fluctuation features from the state evaluation result, and using an anomaly detection algorithm to generate an abnormal marking sequence.

[0016] Preferably, a virtual synchronous generator grid-connected current harmonic suppression system is used to implement the steps of the virtual synchronous generator grid-connected current harmonic suppression method. The system includes a sampling module for digitally sampling intermittent high-frequency disturbances using an analog-to-digital converter through the grid-connected current signal of the wind farm in the power grid to obtain an original digital current sequence; a spectrum analysis module for analyzing the wind load frequency component based on the original digital current sequence using a fast Fourier transform algorithm to determine the amplitude and phase characteristics of the high-frequency harmonic component; a disturbance extraction module for extracting the intermittent disturbance harmonic sequence from the frequency component when the amplitude of the high-frequency harmonic component exceeds a preset threshold to obtain a harmonic subset to be suppressed; a filtering control module for dynamically adjusting the filtering parameters using an adaptive filtering algorithm through the harmonic subset to be suppressed to generate a compensation current sequence; a matching calculation module for obtaining phase matching adjustment and amplitude scaling parameters from the compensation current sequence to determine the matching degree between the compensation current sequence and the original digital current sequence; if the matching degree is higher than the threshold, the signal superposition operation is performed The compensation current sequence is integrated into the main control loop through calculation and noise suppression filtering to obtain an optimized current signal; the digital signal processing module is used to calculate the suppression instruction in real time using a digital signal processor based on the optimized current signal, and simultaneously perform loop feedback integration and stability verification checks to generate a directional suppression pulse sequence; the remote transmission module is used to obtain network transmission data packets, perform delay threshold judgment and network bandwidth evaluation, and if the delay is lower than the threshold, the data packet is encrypted and sent to the remote server through transmission protocol optimization to obtain a remote adjustment parameter set; the parameter update module is used to extract feedback values ​​from the remote adjustment parameter set, and update the local filtering parameters through server response analysis and parameter set download and update using a proportional integral differential control algorithm combined with an error retransmission mechanism and feedback value extraction processing to generate a final suppression output sequence; the stability monitoring module is used to determine whether the intermittent disturbance stability index of the remote wind farm power grid meets the requirements based on the final suppression output sequence. If so, the sampling process of the original digital current sequence is looped back to maintain a continuous monitoring state.

[0017] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0018] This virtual synchronous generator grid-connected current harmonic suppression method and system solves the problem that harmonic disturbances caused by changes in wind load frequency in remote wind farms are difficult to suppress in real time and that network transmission delays affect stability. The present invention uses a high-speed analog-to-digital converter to digitally sample the current signal and adopts a fast Fourier transform algorithm to accurately extract the amplitude and phase characteristics of high-frequency harmonic components. When the harmonic amplitude exceeds the threshold, a subset of harmonics to be suppressed is generated, and an adaptive filtering algorithm is used to dynamically adjust the parameters to generate a compensation current sequence. The present invention optimizes the current signal through signal superposition and noise suppression fusion, and combines a digital signal processor to calculate the suppression instructions in real time to ensure loop stability. At the same time, data packets are transmitted through a remote network, the protocol is optimized and encrypted, and the proportional integral differential control algorithm and error retransmission mechanism are combined to update the filter parameters to generate the final suppressed output sequence. The present invention significantly improves the stability of the remote wind farm power grid, realizes continuous monitoring and real-time suppression of intermittent disturbances, and ensures the efficiency and reliability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of the method for suppressing grid-connected current harmonics of a virtual synchronous generator according to the present invention;

[0020] Figure 2 This is a connection diagram of the virtual synchronous generator grid-connected current harmonic suppression system of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] like Figure 1As shown, the present invention provides a technical solution: a virtual synchronous generator grid-connected current harmonic suppression method, including S1, using the wind farm grid-connected current signal in the power grid, using an analog-to-digital converter to digitally sample the intermittent high-frequency disturbance to obtain an original digital current sequence; S2, according to the original digital current sequence, using the fast Fourier transform algorithm to analyze the wind load frequency component, and determine the amplitude and phase characteristics of the high-frequency harmonic component; S3, if the amplitude of the high-frequency harmonic component exceeds a preset threshold, extracting the intermittent disturbance harmonic sequence from the frequency component to obtain a harmonic subset to be suppressed; S4, using the adaptive filtering algorithm to dynamically adjust the filtering parameters through the harmonic subset to be suppressed, and generating a compensation current sequence; S5, obtaining phase matching adjustment and amplitude scaling parameters from the compensation current sequence, judging the matching degree of the compensation current sequence with the original digital current sequence, and if the matching degree is higher than the threshold, fusing the compensation current sequence through signal superposition operation and noise suppression filtering. Sequence to the main control loop to obtain an optimized current signal; S6. According to the optimized current signal, a digital signal processor is used to calculate the suppression instruction in real time, and loop feedback integration and stability verification are performed at the same time to generate a directional suppression pulse sequence. Through the directional suppression pulse sequence, the network transmission data packet is obtained, and the delay threshold judgment and network bandwidth evaluation are performed. If the delay is lower than the threshold, the data packet encryption transmission and transmission protocol optimization are used to transmit to the remote server to obtain a remote adjustment parameter set; S7. The feedback value is extracted from the remote adjustment parameter set, and the proportional integral differential control algorithm is used to update the local filter parameters through server response analysis and parameter set download and update, and the local filter parameters are updated using the proportional integral differential control algorithm combined with the error retransmission mechanism and feedback value extraction processing to generate the final suppression output sequence; S8. According to the final suppression output sequence, it is judged whether the intermittent disturbance stability index of the wind farm power grid meets the requirements. If it meets the requirements, the sampling process of the original digital current sequence is looped back to maintain a continuous monitoring state.

[0023] This embodiment implements dynamic suppression of harmonics by performing multi-stage detection, analysis and compensation control on intermittent high-frequency harmonic disturbances in the wind farm grid-connected current. In step S1, the actual current signal is digitally sampled in real time using an analog-to-digital converter to obtain an original current sequence with complete timing characteristics. In step S2, the spectral characteristics of the current sequence are analyzed using a fast Fourier transform algorithm to effectively identify high-frequency harmonic components and accurately extract their amplitude and phase information. In step S3, the intermittent disturbance frequency band to be processed is screened based on threshold judgment logic to extract a subset of harmonics to be suppressed. In step S4, an adaptive filtering algorithm is applied to this subset to adjust the filter parameters to improve the filter's response to discontinuous high-frequency components, thereby generating a compensation current to offset the disturbance. In step S5, a dynamic matching evaluation is performed based on the phase difference and amplitude relationship between the original sequence and the compensation sequence to ensure accurate superposition of the compensation signal, and the current main control loop signal quality is optimized through fusion operations. In steps S6 and S7, a digital signal processor is further introduced to perform suppression instruction generation and feedback stability verification. At the same time, a refined adjustment parameter set is obtained from a remote server through a network transmission mechanism and locally updated. Finally, in step S8, the current disturbance stability index is determined based on the updated suppression sequence to determine whether to continue the closed-loop cycle.

[0024] High-frequency disturbances refer to short-term, intermittent, and high-frequency abnormal current fluctuations in wind farm grid-connected current caused by factors including sudden changes in wind speed, inverter switching, and nonlinear loads. These disturbances are typically non-periodic and sudden, with frequencies ranging from 500 Hz to 30 kHz and durations ranging from a few milliseconds to tens of milliseconds. High-frequency harmonic components are harmonic signal components in high-frequency disturbances that appear as high integer multiples (e.g., 25th order or higher) of the fundamental wave (e.g., 50 Hz). These components exhibit stable periodicity and characteristic characteristics and serve as an important basis for determining grid-connected current anomalies. In this patent, these components generally refer to current components with frequencies greater than 2 kHz and amplitudes exceeding the fundamental wave amplitude by 1%-3%.

[0025] This implementation combines fast Fourier transform (FFT) and adaptive filtering technology to achieve accurate identification and real-time compensation of high-frequency harmonics, significantly improving the power quality and operational stability of virtual synchronous generators during wind power grid integration. Highly dynamic and real-time, it can quickly respond to intermittent disturbances and effectively suppress transient current harmonics caused by wind speed fluctuations. A remote parameter tuning mechanism enhances adaptive and remote maintenance capabilities, while achieving highly reliable transmission through packet delay and bandwidth control. Proportional-integral-derivative control combined with an error retransmission mechanism improves local filter response accuracy and control stability, ensuring long-term operational reliability and consistency.

[0026] S1 includes digitally sampling the wind farm grid-connected current signal through an analog-to-digital converter to generate an original digital current sequence; using the fast Fourier transform algorithm to perform frequency domain analysis on the original digital current sequence to obtain the frequency component of the current signal; if there is an abnormal high-frequency component higher than a preset threshold in the frequency component, the original digital current sequence is filtered through a bandpass filter to obtain a filtered current sequence; based on the filtered current sequence, the short-time Fourier transform algorithm is used to extract the time-frequency characteristics to generate the time-frequency distribution of the current signal; if the duration of the intermittent high-frequency disturbance detected in the time-frequency distribution exceeds the preset threshold, the filtered current sequence is decomposed through the wavelet transform algorithm to obtain the high-frequency disturbance component; based on the high-frequency disturbance component, its energy distribution characteristics are calculated to determine the disturbance intensity and occurrence location; the disturbance intensity and occurrence location are analyzed through a preset classification model to judge the operation status of the power grid.

[0027] In one possible implementation, the wind farm grid-connected current signal is first sampled using an analog-to-digital converter. The sampling frequency is set to 50,000 sample points per second, the sampling accuracy is set to 16 bits, and the sampling process lasts for 2 seconds. A total of 100,000 equally spaced current data points are obtained to form an original digital current sequence.

[0028] This sequence is input into the fast Fourier transform (FFT) processing module, which first applies a window function to the 100,000 sampling points. A rectangular window with a fixed window width of 2048 points is used to segment the data, with each segment overlapping 512 points, for spectral decomposition. During the transformation process, each signal segment is converted into a frequency component, and its corresponding amplitude is extracted. An amplitude threshold is then set to identify abnormal high-frequency components. This threshold is calculated based on the average value of the fundamental amplitude measured on-site and is specifically set to 3% of this average value. For example, if the average amplitude of the fundamental is 20 amps, the amplitude threshold is set to 0.6 amps. If a frequency component has an amplitude greater than this value, it is marked as an abnormal high-frequency component. Next, a bandpass filter is constructed using the identified abnormal frequency value as the center frequency. The filter bandwidth is set to 1000 Hz, with a passband range of 500 Hz on each side of the center frequency. The original digital current sequence is filtered. The new current sequence obtained after filtering retains the high-frequency information of the specified frequency band.

[0029] The sequence then enters the short-time Fourier transform (SFT) processing module, where it is processed using a Hamming window with a window width of 4096 points and a sliding step of 512 points. A fast Fourier transform is performed on each sliding window to generate a frequency distribution within each time segment, ultimately synthesizing a time-frequency distribution diagram for the entire time interval. The energy trajectory of each frequency in the image is detected. If an energy peak appears continuously at the same frequency for more than 1500 sampling points, meaning it lasts for 30 milliseconds, the disturbance corresponding to that frequency is determined to be an intermittent high-frequency disturbance. The wavelet transform module is then called to perform a four-layer discrete wavelet decomposition of the filtered current sequence using a fixed Daubechies fourth-order function. This decomposition yields four sets of high-frequency coefficient sequences, and the top-level high-frequency sequence is selected as the high-frequency disturbance component for the subsequent energy analysis step.

[0030] In the energy analysis phase, the high-frequency disturbance component is divided into time windows of 100 points each, and the square of each group of values ​​is accumulated to obtain the energy value, thereby constructing a complete energy distribution sequence. The starting point of the time window corresponding to the maximum energy value is marked as the location where the disturbance occurs, and the maximum value is the disturbance intensity. The disturbance intensity value is judged by comparing it with the preset level table. The level table pre-sets the energy value thresholds for mild disturbance, moderate disturbance and severe disturbance, which are 10, 50 and 100 ampere-square seconds respectively. If the disturbance energy is within this range, the corresponding label is marked according to its level. Finally, the disturbance intensity and occurrence location are used as input parameters and imported into the preset classification model. The model is based on the polynomial condition judgment process and compares the disturbance intensity level and the disturbance location item by item to see if they are in the sensitive range. If the conditions of moderate or above intensity and critical time period are met at the same time, it is determined that the power grid is currently at risk of disturbance, otherwise it is determined to be in normal operation.

[0031] According to the high-frequency disturbance component after wavelet transformation, the calculation of the disturbance energy distribution is completed, and two key parameters, the disturbance intensity and the disturbance occurrence location, are extracted. The disturbance intensity is obtained by accumulating the square of the high-frequency current value in each time window to obtain the energy value, and the unit is ampere square second. The entire disturbance sequence is divided into several continuous windows, each window contains 100 sampling points, and the total energy of each window is calculated. Finally, the corresponding value of the window with the largest energy is selected as the disturbance intensity of the disturbance. The disturbance occurrence location is the time point corresponding to the starting position of the maximum energy window in the entire sampling sequence, in milliseconds, with an accuracy of each sampling cycle, that is, 0.02 milliseconds. Subsequently, these two parameters are passed as input variables into the preset classification model module. This model is a judgment model built based on a fixed decision logic structure, and the model contains several clear classification rules. Specifically, it includes: the first step, judging whether the disturbance intensity is greater than 10 ampere-square seconds. If it is less than or equal to 10 ampere-square seconds, it is directly judged as a mild disturbance and the grid operation status is normal; if it is greater than 10 ampere-square seconds, it proceeds to the next step of judgment; the second step, judging whether the disturbance intensity is greater than 100 ampere-square seconds. If so, it is judged as a severe disturbance; if it is between 10 and 100 ampere-square seconds, it is judged as a moderate disturbance; the third step, for moderate and severe disturbances, further judge whether the location of the disturbance is within the set critical operation time period. The critical operation time period is pre-configured, usually for key control windows such as load switching, high wind speed warning, and grid synchronization, such as between the 500th and 600th milliseconds of the sampling time. If the disturbance occurs within this time period, it is judged that the grid operation status is abnormal and the suppression mechanism needs to be activated; if it is not within this time period, it is judged as a tolerable disturbance and the grid operation status is to be observed.

[0032] The amplitude threshold for abnormal high-frequency components is set at 3% of the fundamental amplitude. This value is determined based on the actual operating characteristics of wind farm grid-connected current signals and spectral analysis results. Under normal operating conditions, the high-frequency portion of the current signal is affected by noise, measurement errors, and control disturbances, and its amplitude generally does not exceed 1% to 2% of the fundamental amplitude. To avoid misidentifying normal noise as an abnormal signal, a threshold must be set that is above the noise ceiling but sensitive enough to detect true high-frequency harmonic disturbances. Statistical analysis of field operating data and extensive experimental comparisons have shown that when the high-frequency component amplitude exceeds 3% of the fundamental amplitude, it is highly correlated with actual grid disturbances such as sudden wind speed changes and load switching. Therefore, the 3% ratio effectively balances the risks of false positives and false negatives, improving identification accuracy and engineering practicality, and is a proven and stable threshold setting.

[0033] In the bandpass filter processing link, the bandwidth is fixedly set to 1000 Hz, and its setting is based on the coverage analysis of the frequency fluctuation range of high-frequency disturbances. During the operation of wind power, the high-frequency components caused by factors such as inverter switching frequency fluctuations and wind speed disturbances are usually concentrated, and the frequency variation range is generally within plus or minus 500 Hz of the target frequency. Setting a bandwidth of 500 Hz above and below the center frequency can completely cover the main disturbance frequency bands, effectively extract the target frequency signal, and at the same time suppress the interference signals in other frequency bands to the greatest extent. If the bandwidth is set too small, the real disturbance information will be filtered out; if it is set too large, invalid spectrum components will be introduced, reducing the accuracy of subsequent analysis. Therefore, the bandwidth value of 1000 Hz is aimed at covering the main disturbance, and the engineering setting with the optimal signal-to-noise ratio is verified through experiments.

[0034] In the process of determining intermittent high-frequency disturbances, a fixed threshold of 30 milliseconds is set for the duration of disturbances in the time-frequency distribution. The setting of this threshold is determined based on the time response characteristics of the wind power control logic and the actual persistence analysis of the disturbance behavior. The rapid adjustment cycle of the control is usually between 20 and 50 milliseconds. 30 milliseconds is in the typical time period for responding to sudden disturbances, which can effectively distinguish short-term noise fluctuations from destructive disturbance events. Through comparative analysis of a large number of time-frequency graph samples, it was found that disturbances less than 30 milliseconds are mostly non-continuous signals and have no control interference significance; while disturbances exceeding 30 milliseconds are often accompanied by power fluctuations and frequency offsets. Therefore, this time threshold can stably identify disturbance behaviors that have a substantial impact on operations, and has good engineering stability and applicability.

[0035] During the wavelet energy disturbance analysis phase, fixed disturbance intensity classification thresholds of 10 ampere-second squared, 50 ampere-second squared, and 100 ampere-second squared are used to classify the high-frequency disturbance intensity. This setting is based on the disturbance energy value, which is reflected as the product of the disturbance duration and amplitude, thereby accurately measuring the degree of its impact. After modeling and analyzing a large number of historical disturbance events, it was found that when the disturbance energy is less than 10 ampere-second squared, it has almost no substantial impact on the control of the virtual synchronous generator, so it is defined as a mild disturbance; when the disturbance energy exceeds 50 ampere-second squared, the frequency and voltage will fluctuate significantly, which is defined as a moderate disturbance; and disturbances exceeding 100 ampere-second squared are generally accompanied by instantaneous instability of the control link, requiring emergency control measures, so it is defined as a severe disturbance.

[0036] S2 includes using the fast Fourier transform algorithm to perform frequency domain analysis on the original digital current sequence to obtain the amplitude and phase characteristics of the high-frequency harmonic components; based on the amplitude and phase characteristics of the high-frequency harmonic components, the original digital current sequence is processed by a bandpass filter to obtain a filtered current sequence; using the short-time Fourier transform algorithm to perform time-frequency analysis on the filtered current sequence to obtain time-frequency distribution characteristics; if energy concentration of a specific frequency component is detected in the time-frequency distribution characteristics, the filtered current sequence is decomposed by the wavelet transform algorithm to obtain a high-frequency component sequence; based on the high-frequency component sequence, the time energy distribution is calculated to determine the disturbance time characteristics; if the duration of the disturbance time characteristics exceeds a preset threshold, the high-frequency component sequence is analyzed by a preset classification model to judge the grid operation status; based on the grid operation status, status assessment data is generated to determine the grid connection stability of the wind farm.

[0037] In this implementation, the wind farm grid-connected current signal is first digitally sampled using an analog-to-digital converter (ADC). The sampling frequency is fixed at 50,000 Hz, 50,000 points are collected per second, the sampling accuracy is 16 bits, and each sampling period is 2 seconds. This yields 100,000 current sampling points, which form the original digital current sequence. The sampled data is sequentially fed into the fast Fourier transform (FFT) module for frequency domain analysis. During the conversion process, a segmented window function is used, with a single window width set to 2048 points and an overlap of 512 points between adjacent windows. A spectrum is calculated for each window, extracting the amplitude and phase values ​​for all frequency points. The amplitude is calculated by taking the square root of the sum of the real and imaginary parts and expressing the energy intensity of the frequency component. The phase is calculated using the inverse tangent function on the imaginary and real parts and expressed in degrees, representing the time offset of the frequency component relative to the fundamental. Amplitude thresholds are applied to all frequency components in the spectrum. If the amplitude of a frequency exceeds 0.6 amperes, it is considered an abnormal high-frequency harmonic component. The 0.6A threshold is determined by multiplying the fundamental current amplitude of 20A by 3%. 3% is the minimum recognizable disturbance value verified in engineering based on the actual wind power grid-connected current background noise distribution and disturbance differences, avoiding false alarms and improving recognition accuracy.

[0038] After identifying the abnormal frequency, its frequency value is extracted as the center frequency of the bandpass filter. The bandwidth of the bandpass filter is fixed at 1000 Hz, that is, the passband range is 500 Hz above and below the center frequency, and a second-order Butterworth filter is constructed. This bandwidth is determined based on the frequency drift range of the identified disturbance. In disturbances caused by wind speed fluctuations and equipment switching, the high-frequency components will not deviate more than 500 Hz. Therefore, setting the total bandwidth to 1000 Hz ensures that the filtering does not lose key components. The filter takes the original digital current sequence as input, performs a differential filtering operation, and outputs a filtered current sequence. Only the signal in the specified high-frequency range is retained, and all other frequency components are filtered out to form a high-frequency disturbance retention sequence for subsequent time-frequency analysis and processing.

[0039] The filtered current sequence is fed into the short-time Fourier transform module for time-frequency feature extraction. The transformation parameters are a window width of 4096 points and a step length of 512 points, meaning that every 4096 sampling points constitute a window, and 512 points slide between each window. Fourier spectrum operations are performed separately, and finally a complete time-frequency distribution diagram is formed by splicing them in chronological order. The energy trajectory corresponding to each frequency in the time-frequency distribution diagram is detected to calculate whether the energy change trajectory forms a concentrated and continuous high-energy area at a certain frequency point. If the energy of a certain frequency continues to rise and maintain a high amplitude in multiple adjacent time windows, it is identified as a frequency energy concentration feature, and the frequency is determined to be a suspected disturbance frequency.

[0040] The wavelet transform module uses the filtering results near this frequency as input and performs a four-layer wavelet decomposition using Daubechies wavelet basis functions. The highest-level high-frequency components are retained to form a high-frequency component sequence. The high-frequency component sequence is divided into time windows of 100 points each, with the total number of windows being the total number of points in the filtered sequence divided by 100. Within each window, the energy value is calculated by squared summing all data points in amperes per second. The start time and energy value corresponding to each window are recorded to form a complete temporal energy distribution diagram, and disturbance signatures are identified based on a set threshold. The specific judgment rule is: if the energy value in a continuous time window is continuously greater than 10 amperes per second, and the total duration of the continuous window exceeds 30 milliseconds, it is considered a disturbance temporal signature. The energy threshold of 10 amperes per second is determined based on statistical data collected during normal operation and is three times the upper limit of background noise, providing sufficient discrimination capability. The control response threshold of 30 milliseconds is set to ensure that the control module has sufficient time to intervene and process.

[0041] Once a disturbance section that meets the requirements of energy concentration and duration is identified, the high-frequency component sequence is input into a preset classification model for grid status identification. The classification model is a fixed rule structure and does not involve a learning mechanism. It is executed in the following logical order: first, it is determined whether the disturbance energy is greater than 10 amperes per second squared. If not, the grid status is output as normal; if so, the second step is to determine whether the disturbance occurs within a set critical time segment, such as a wind turbine switching window or the moment of grid connection, which is between 1000 milliseconds and 1200 milliseconds. If the disturbance meets the requirements of an intensity higher than the threshold and occurs within a critical period, the grid status is determined to be at operational risk. Subsequently, information such as the disturbance intensity value, disturbance start time, disturbance duration, and grid status mark are combined into structured status assessment data and stored in a local database or uploaded to remote monitoring for dispatch response or equipment linkage operation.

[0042] A fixed amplitude threshold of 0.6 amps is used to identify high-frequency harmonic components. This threshold is derived from 3% of the fundamental current amplitude. Actual measurement data from grid-connected wind farms indicates that the normal amplitude of the fundamental current is approximately 20 amps. Taking into account background noise, electromagnetic interference, and high-frequency jitter caused by equipment switching, its amplitude typically does not exceed 1% to 2% of the fundamental amplitude. If the threshold is set below 3%, invalid noise will be frequently identified as disturbances, resulting in a high false alarm rate and poor stability. If the threshold is set too high, for example, above 5%, it can easily miss real high-frequency disturbances with lower amplitudes but clear frequency structures. Therefore, 0.6 amps, as the specific value for the 3% amplitude ratio, effectively eliminates noise while maintaining sensitivity to key disturbances. This represents the optimal engineering solution, determined through analysis of multiple sets of measured data.

[0043] In the disturbance time feature identification, a duration threshold of 30 milliseconds is set. The setting of this threshold is based on the response time of the virtual synchronous generator and its control to external disturbances. After receiving the disturbance signal, the adjustment response cycle of most wind power control is between 20 milliseconds and 50 milliseconds, and 30 milliseconds is in the typical middle value range. If the disturbance duration is shorter than 30 milliseconds, it may be a transient fluctuation and no control intervention is required; if it exceeds 30 milliseconds, it is very likely to affect stable operation. Therefore, setting 30 milliseconds as the lower limit threshold for disturbance time identification can effectively avoid misidentification of transient fluctuations, while ensuring that disturbance detection and status determination are completed in time before control intervention.

[0044] In the disturbance intensity analysis, 10 amperes per second squared is used as the energy judgment threshold. This value is derived from the energy analysis and calculation of high-frequency disturbance components in actual wind power grid-connected scenarios. The unit is amperes per second squared, representing the square integral value of the current within each disturbance window. During the experimental phase, energy accumulation analysis of thousands of disturbances under different working conditions was conducted. It was found that when the energy value is less than 10 amperes per second squared, its impact on the grid current curve is not significant, and it is a tolerable disturbance; when the energy is greater than 10 amperes per second squared, it is often accompanied by current waveform distortion and frequency offset, which requires attention. Therefore, 10 amperes per second squared is used as the minimum value for disturbance energy identification, which can effectively filter out non-substantial interference while retaining real destructive disturbance data.

[0045] S3 includes: if the amplitude of the high-frequency harmonic component exceeds the preset threshold, the original current sequence is decomposed in the frequency domain by the fast Fourier transform algorithm to obtain a set of frequency components; based on the frequency component set, the intermittent disturbance sequence is extracted by the spectral analysis method to generate a harmonic sequence to be suppressed; the harmonic sequence to be suppressed is processed by a bandpass filter to obtain a filtered harmonic sequence; if the energy of the filtered harmonic sequence is concentrated in the frequency range of the high-frequency disturbance, the short-time Fourier transform algorithm is used to perform time-frequency analysis to obtain the time-frequency distribution characteristics; based on the time-frequency distribution characteristics, the energy concentration sequence is extracted to generate a high-frequency disturbance sequence; the high-frequency disturbance sequence is analyzed by a preset classification model to judge the operation status of the power grid and obtain status evaluation data; based on the status evaluation data, the high-frequency disturbance sequence is suppressed by an adaptive filtering algorithm to generate a stable current sequence.

[0046] During operation, the frequency characteristics of the raw digital current sequence are monitored in real time. If the amplitude of any high-frequency harmonic component exceeds 0.6 amperes, the frequency domain decomposition process is triggered. The 0.6 amperes threshold is determined by multiplying the fundamental current amplitude of 20 amperes by 0.03. The 3% scaling factor is derived from the actual operational distinction between background noise and effective disturbances to ensure that the identified high-frequency signals are meaningful interference signals. After triggering, the raw current sequence is divided into multiple 2048-point analysis windows, with 512 points overlapping each window. A fast Fourier transform is then performed to convert the time-domain data into a set of frequency domain components. This includes frequency, amplitude, and phase values ​​for each frequency, covering a frequency range of 0 to 25,000 Hz. The amplitude, measured in amperes, represents the current intensity at that frequency, and the phase, measured in degrees, represents the relative offset of that frequency in time.

[0047] The part of the frequency component set with a frequency higher than 2000 Hz is screened as a high-frequency candidate area, and the amplitudes of all frequency points therein are subjected to local peak extraction and discontinuity detection, and the power spectrum density difference calculation method with a sliding window length of 5 frequency points is used for analysis. In each sliding window, if the amplitude of the current frequency point is more than 20% higher than the average value of the previous and next frequency points, and the frequency point is discontinuous with the adjacent interference frequency, it is regarded as an intermittent disturbance frequency point. All frequency points that meet the conditions are combined into an intermittent disturbance sequence, that is, the harmonic sequence to be suppressed. This sequence retains all frequency components that meet the characteristics of strong amplitude, discontinuity and high frequency, providing a clear processing target for subsequent filtering and time-frequency analysis.

[0048] The harmonic sequence to be suppressed is input into the bandpass filter module, and the center frequency is set to the energy peak frequency in the sequence, the bandwidth is 500 Hz above and below, and the total bandwidth is 1000 Hz. The bandwidth is selected based on experimental statistics of the frequency drift range of high-frequency disturbances. It is found that the main disturbances are concentrated within 500 Hz of the center frequency. Therefore, setting this fixed range ensures that the disturbance components are not missed. The filter type is a second-order Butterworth filter. By performing point-to-point filtering on the original current sequence, non-target frequency components are eliminated to generate a filtered harmonic sequence. A local integral energy calculation is performed on the sequence. The energy of each frequency segment is ratioed to the total energy of the entire sequence. If the energy proportion of the passband frequency segment exceeds 90%, it is determined that the high-frequency disturbance energy is concentrated, and the conditions for the next step of analysis are met.

[0049] The filtered harmonic sequence is subjected to short-time Fourier transform processing, with a window width of 4096 points and a sliding step of 512 points. The window function uses a fixed Hamming window, and the spectrum of each sliding window is transformed and spliced ​​to construct a time-frequency spectrum. Subsequently, the energy distribution of the part of the spectrum with a frequency higher than 2000 Hz is calculated window by window. The frequency energy in each time window is the square of the current value and then accumulated, with the unit being ampere-square-second, to form a time-energy distribution sequence. The energy threshold is set to 10 ampere-square-seconds, that is, if the energy value of a frequency point in more than three consecutive time windows is greater than the threshold, it is judged that the disturbance energy is concentrated; this threshold comes from the actual grid-connected control tolerance of the wind farm. If it is lower than this value, there will be no control response demand. Through the time index of the energy concentration area, the continuous disturbance fragments are extracted to form a high-frequency disturbance sequence, which is used as the input for subsequent classification judgment.

[0050] The high-frequency disturbance sequence is input into the classification model, which is a logical judgment process based on a fixed rule structure. First, the maximum energy value in the disturbance sequence is determined to be greater than 10 amperes per second squared. If not, the grid operation status is considered normal. If so, the second step is to determine whether the disturbance lasts for more than 30 milliseconds, that is, whether it spans 1500 consecutive sampling points. This time threshold sets the critical response time, ensuring that identification is completed before the disturbance affects the control logic. The third step determines whether the disturbance occurs within a critical window, such as within 100 milliseconds after grid connection. If two or more of the three conditions are met, the grid operation status is output as abnormal. A status report is generated based on the evaluation data and input into the adaptive filtering module. This module uses the least mean square algorithm to extract parameters from the high-frequency disturbance sequence, calculate the target frequency, amplitude, and duration, and then automatically adjust the filter gain parameters, response speed, and time constant. Compensatory filtering is performed on the disturbance waveform to generate a stable current sequence free of high-frequency disturbances. This stable current sequence is used in subsequent control operations of the main controller to ensure power quality stability and safe operation during the wind turbine grid connection process.

[0051] S4 includes dynamically adjusting the filtering parameters using the minimum mean square adaptive filtering algorithm through the subset of harmonics to be suppressed to generate a compensation current sequence; decomposing the signal using discrete Fourier transform according to the compensation current sequence to obtain a set of frequency components; if there are components in the frequency range of high-frequency disturbances in the frequency component set, extracting the harmonic sequence in the frequency range through a bandpass filter to obtain a filtered harmonic sequence; analyzing the time-frequency distribution based on the filtered harmonic sequence using short-time Fourier transform to obtain a time-frequency feature sequence; judging the power grid state using a preset classification model based on the time-frequency feature sequence to obtain state evaluation data; if the state evaluation data indicates that the power grid state is abnormal, performing secondary processing on the filtered harmonic sequence using the minimum mean square adaptive filtering algorithm to generate a stable current sequence; reconstructing the time domain signal through inverse Fourier transform based on the stable current sequence to obtain an optimized current sequence.

[0052] In one possible implementation, the least mean square adaptive filtering algorithm is first called to process the subset of harmonics to be suppressed identified in the previous process, and the filter initialization step is initiated. Specifically, the subset of harmonics to be suppressed is used as the reference input signal, while the original current sequence is introduced into the least mean square filter as the main input signal. The filter weight coefficient is initialized to zero, and the step size parameter is set to 0.01. This step size value is determined experimentally and is the minimum value that achieves a reasonable convergence speed while ensuring the stability of the algorithm. After receiving two signals, the process is iterated step by step at each sampling point. At each sampling point, the filter output value under the current weight is first calculated and the difference between it and the corresponding point of the reference input signal is calculated to obtain an instantaneous error signal. Then, according to the weight update formula of the least mean square algorithm, the error signal is multiplied by the current reference input value and then multiplied by the step size coefficient to obtain a new incremental value. This incremental value is added to the original weight coefficient and used as the weight coefficient input at the next moment. This step is continuously executed until all sampling points are processed, that is, the complete compensation current sequence is output.

[0053] After the compensation current sequence is generated, it enters the frequency domain analysis stage, and the discrete Fourier transform is used to decompose the compensation current sequence. The transformation operation parameters are a window length of 2048 points, a sampling frequency set to 50,000 Hz, and a transformation frequency resolution of 50,000 divided by 2048, which is approximately 24.41 Hz. The entire compensation sequence is divided into multiple non-overlapping windows, and Fourier transform calculations are performed in each window to extract the amplitude components of each frequency point to form a frequency component set. Amplitude analysis is performed on all frequency points in the set, and special attention is paid to checking whether there are frequency points with amplitudes exceeding 0.6 amps in the frequency range of 2000 Hz to 30,000 Hz. 0.6 amps is a fixed threshold value, which comes from 3% of the fundamental amplitude of 20 amps. It is the lower limit standard for identifying effective high-frequency disturbances, ensuring that all frequency components included in the processing have observable disturbance characteristics. If the above frequency points exist, the maximum amplitude frequency is used as the center frequency of the bandpass filter, the bandwidth is fixed at 1000 Hz, a second-order Butterworth bandpass filter is constructed, and the frequency band is extracted to obtain the filtered harmonic sequence.

[0054] The filtered harmonic sequence is used as the input signal for time-frequency analysis, and a short-time Fourier transform is used to decompose it in both time and frequency dimensions. The parameters are set to a window width of 4096 points, a sliding step of 512 points, and a Hamming window function type to ensure a balance between mainlobe width and sidelobe suppression performance. A discrete Fourier transform is performed within each sliding window, and the results are spliced ​​into a time-frequency distribution diagram. The energy trajectory of each frequency point over time is scanned in the diagram, and each trajectory is integrated to calculate the energy value of the frequency point within each time window in amperes per second. If the energy value of a frequency point exceeds 10 amperes per second in each of three or more consecutive time windows, the frequency trajectory and the corresponding time period are extracted as a time-frequency feature sequence. 10 amperes per second is a fixed threshold for disturbance energy judgment. It is the minimum response activation threshold derived from analysis of wind farm current disturbance data and ensures that the identified frequency has a practical control impact.

[0055] The extracted time-frequency feature sequence is input into a classification model. This model is a rule-based structure with three logical judgments: first, whether the disturbance energy exceeds 10 amperes per second squared; second, whether the disturbance duration exceeds 30 milliseconds, that is, whether it spans at least 1500 sampling points; and third, whether the disturbance occurs within a critical time window, such as within 200 milliseconds before and after grid connection. If any two or more of these three judgments are met, the grid status is output as abnormal; otherwise, it is output as normal. The status assessment results, along with the disturbance parameters, are input into the filter control module.

[0056] When the evaluation result is abnormal, the least mean square filtering algorithm is called again to perform a secondary filtering on the previously output filtered harmonic sequence. The initial value of the filter weight is set to the final weight value of the first filter, and the step size is still 0.01. Repeat the aforementioned point-by-point error iterative calculation process to obtain a new stable current sequence. This sequence has significantly weakened the intensity of high-frequency disturbances in the frequency component. Next, the inverse Fourier transform is used to reconstruct it from the frequency domain into a time domain signal. The inverse Fourier transform uses the same window length of 2048 points as the original Fourier transform and a sampling frequency of 50,000 Hz to restore each spectrum segment to the original time series point, which is then spliced ​​together to form the final optimized current sequence.

[0057] When the state assessment data received from the classification model indicates an abnormal grid operating state, the second-stage intervention process is immediately initiated, performing secondary adaptive filtering on the output filtered harmonic sequence to further eliminate residual high-frequency disturbances. This process uses the least mean square adaptive filtering algorithm. The algorithm initialization step is as follows: the final weight coefficients of the first filtering output are used as the initial weight coefficients of this filter. At the same time, the filtered harmonic sequence is input as the main input signal and is input into the filter structure again. The step size parameter is maintained at the fixed value of 0.01 set in the first setting. The step size value is determined based on experimental results and is the optimal constant value after balancing convergence speed and stability. It can ensure a rapid response to high-frequency disturbances while avoiding overregulation in the stable range.

[0058] The filtering process continues point by point. For each input sampling point, the filtered output under the current weight is calculated, and an error signal is calculated between the output and the compensation signal from the previous stage. This error signal is used to correct the current weight. The correction formula is: the current weight plus the step size multiplied by the error multiplied by the input value. The result serves as the basis for updating the weight for the next sampling point. This calculation logic continues until all sampling points have been updated. The final filtered output is a stable current sequence. This stable current sequence retains the main frequency and normal harmonics in the frequency components, while effectively suppressing high-amplitude high-frequency disturbance components. Through iterative optimization, the entire current waveform remains continuous, smooth, and free of jumps. The stable current sequence is then subjected to an inverse Fourier transform (IFT) to reconstruct it from the frequency domain into a time domain signal, forming the final optimized current sequence. The IFT operating parameters remain the same as those for the forward transform, with a window length of 2048 points and a non-overlapping splicing method. Each spectral segment is processed individually, and a real-valued inverse Fourier operation is performed to restore it to a sequence of equally spaced time points. All time domain segments are then sequentially spliced ​​together to construct a complete time series signal. The optimized current sequence has the same basic form as the original current waveform, keeping the main frequency signal unchanged while significantly reducing the high-frequency disturbance content, meeting the power quality standards, and can be directly transmitted to the virtual synchronous generator main control for feedback adjustment, thereby achieving response, correction and dynamic stability control of power grid disturbances.

[0059] S5 includes extracting phase adjustment parameters and amplitude scaling parameters from the compensation current sequence, decomposing the signal using fast Fourier transform, obtaining phase and amplitude characteristics, and obtaining phase adjustment parameters and amplitude scaling parameters; calculating the phase difference and amplitude ratio between the compensation current sequence and the original digital current sequence based on the phase adjustment parameters and amplitude scaling parameters, and evaluating the matching degree between the two using the cosine similarity algorithm to obtain a matching value; if the matching value is higher than a preset threshold, the compensation current sequence is fused into the main control loop through a weighted signal superposition operation to obtain a fused current signal; for the fused current signal, a filter is used to remove the noise component to obtain a denoised current signal; based on the denoised current signal, the time domain statistical characteristics of the signal are calculated, and the signal stability is judged using a support vector machine classification model to obtain a stability evaluation result; if the stability evaluation result indicates that the signal is unstable, the denoised current signal is secondary adjusted through a least mean square adaptive filtering algorithm to obtain a stable current signal; based on the stable current signal, an inverse fast Fourier transform is used to reconstruct the time domain signal to obtain an optimized current signal.

[0060] In one possible implementation, a fast Fourier transform (FFT) operation is first performed on the compensation current sequence. The Fourier transform window length is set to 2048 points and the sampling frequency is 50,000 Hz. This configuration ensures a frequency resolution of approximately 24.41 Hz, allowing the amplitude and phase characteristics of each frequency component to be clearly distinguished. After the transformation, the phase and amplitude values ​​of the current sequence are extracted from the main frequency point as the main phase adjustment parameters and amplitude scaling parameters of the sequence. Next, the same Fourier transform process is applied to the original digital current sequence to extract the phase and amplitude values ​​of its main frequency point. The phase adjustment parameter is obtained by calculating the phase difference between the two main frequency points as the compensation current phase value minus the original current phase value. The amplitude scaling parameter is the amplitude of the compensation current at the main frequency point divided by the amplitude of the original current at the main frequency point, obtaining the relative ratio of the two sequences in the amplitude direction.

[0061] Then, based on the above two parameters, the respective signal vectors are constructed and standardized as real vectors with a length of 2048 points. Subsequently, the cosine similarity algorithm is used to perform a matching calculation on the two vectors. The specific calculation steps are as follows: first, the sum of the dot products of the two vectors is calculated, then the modulus of the two vectors is calculated respectively, and finally the dot product value is divided by the product of the modulus lengths to obtain a matching value between 0 and 1. The threshold is set to 0.95. This value comes from the evaluation and analysis of the stability of the signal after fusion in a large number of experimental simulations. It can ensure that the matching degree is considered high enough only when the phase difference is less than 15 degrees and the amplitude ratio is between 0.9 and 1.1, so as to ensure that unstable disturbances are not introduced after the signal superposition.

[0062] If the matching value is higher than 0.95, the signal fusion operation is performed. The fusion method adopts the weighted linear superposition method, in which the weight of the original current signal is 70% and the weight of the compensation current signal is 30%. The weight ratio is determined according to the control stability and response speed simulation data. It not only retains the main body of the original signal, but also appropriately introduces compensation information to correct the disturbance. The superimposed fused current signal immediately enters the filtering processing module. The filter type is a Butterworth low-pass filter. The cutoff frequency of the filter is set to 2500 Hz. This value is the lower limit of the high-frequency disturbance frequency outside the main frequency component of the original current signal. It can effectively filter high-frequency noise and retain the target frequency component. The order of the filter is set to 2nd order to ensure that the phase response is sufficiently flat. After the filtering is completed, the output is a denoised current signal.

[0063] Subsequently, the time-domain statistical characteristics of the denoised current signal were calculated using a sliding window approach, with each window length of 1024 points and a sliding step size of 512 points. Four characteristic metrics, namely the mean, standard deviation, root mean square value, and crest factor, were calculated for the data within each window. These eigenvalues ​​constitute a 4-dimensional feature vector, which is fed into the support vector machine model. The model uses a radial basis kernel function and is pre-trained for supervised learning using normal current and abnormal disturbance current samples. The model's output value is set to 0 as the decision boundary. A signal greater than 0 is considered stable, while a value less than 0 is considered unstable.

[0064] If the support vector machine model determines that the signal is unstable, the least mean square adaptive filtering algorithm is immediately invoked to perform secondary dynamic adjustments on the denoised current signal. The filter's initial weight vector is set to all zeros, and the step coefficient is fixed at 0.01. The step value is a constant value determined by balancing convergence speed and response sensitivity. At each sampling point, the error between the output value and the expected value is predicted based on the current weights. This error is multiplied by the step value and the input signal to form the weight update. After updating the existing weight vector, the next sampling point is processed, ultimately completing the adaptive filtering of the entire signal and outputting a stable current signal.

[0065] Finally, this stable current signal is fed into the inverse fast Fourier transform module. Using the same 2048-point window parameter as previously described, the amplitude and phase information in the frequency domain are restored to a time-domain sequence, yielding the final optimized current signal. This optimized current signal removes high-frequency disturbances from its frequency components while maintaining the consistency of the main frequency. It also exhibits good continuity and smooth response in the time domain, making it suitable as the current control input signal for the virtual synchronous generator, enabling real-time execution of the stable operation control strategy.

[0066] The matching threshold of 0.95 was determined based on a comprehensive analysis of actual grid-connected disturbance data from a large number of wind farms and stability simulation analysis results. This threshold is used to assess the degree of matching between the compensation current sequence and the original digital current sequence in terms of amplitude and phase. The core reason for setting it to 0.95 is that this value ensures a cosine similarity of at least 95% between the two, indicating a high degree of consistency in signal shape and frequency content. This prevents the risk of secondary disturbances caused by phase misalignment or amplitude imbalance during the signal fusion process. Specifically, when the matching degree is lower than 0.95, common errors manifest as phase shifts exceeding 15 degrees or amplitude ratio deviations exceeding 10%. Such deviations can easily cause nonlinear distortion of the current waveform in virtual synchronous generator control, leading to control command conflicts between links. Therefore, 0.95, as the minimum matching threshold for stable fusion, ensures the effectiveness of the fused compensation current while minimizing the probability of introducing new disturbances.

[0067] S6 includes obtaining time domain features from the optimized current signal, using a digital signal processor for real-time calculation to obtain a suppression instruction sequence; using the suppression instruction sequence, using a feedback loop to calculate the deviation from the main control loop to obtain a deviation correction sequence; if the amplitude of the deviation correction sequence exceeds a preset threshold, the Kalman filter algorithm is used to optimize the suppression instruction sequence to obtain an optimized pulse sequence; based on the optimized pulse sequence, the statistical characteristics of the time series are calculated, and the support vector machine classification model is used to judge the stability of the sequence to obtain a stability judgment result; based on the stability judgment result, unstable sequence fragments are extracted, and an adaptive filtering algorithm is used for secondary adjustment to obtain a stable pulse sequence; based on the stable pulse sequence, a directional suppression pulse sequence is generated, and a digital signal processor is used for real-time output to obtain a final pulse sequence; based on the final pulse sequence, the frequency distribution characteristics of the sequence are calculated, and a fast filter is used to classify the sequence. A fast Fourier transform is used to verify frequency consistency and obtain a verification pulse sequence; network transmission data packets are extracted from the directional pulse sequence, and the data packet delay is calculated using timestamp analysis to obtain a delay judgment result; if the delay judgment result is lower than a preset threshold, the network transmission data packet is encrypted using an encryption standard algorithm to obtain an encrypted data packet; based on the encrypted data packet, the transmission control protocol optimization strategy is used to adjust the data packet sending order to generate an optimized data stream; through the optimized data stream, the sliding window protocol is used to analyze the network bandwidth occupancy and obtain the bandwidth allocation parameters; based on the bandwidth allocation parameters, the flow control algorithm is used to adjust the data packet transmission rate to generate an adjusted data stream; remote adjustment parameters are extracted from the adjusted data stream, and a data compression algorithm is used to generate a compression parameter set to obtain a remote adjustment parameter set; through the remote adjustment parameter set, a verification algorithm is used to verify the integrity of the parameter set to obtain a verification parameter set.

[0068] In one possible implementation, the optimized current signal is first analyzed in real time. A digital signal processor (DSP) extracts its time-domain features, including the current amplitude, rate of change, mean, and standard deviation, at a fixed sampling frequency of 50,000 Hz. These features are grouped into a sliding window structure with 1024 sampling points, and statistics are calculated for each group to form a multidimensional feature vector. This processing is performed within the DSP using a preset function to avoid latency. These features are then input into the control module and converted into a quantitative suppression command sequence through linear mapping and threshold grading. The command amplitude range is controlled between 0 and 1, with numerical precision to three decimal places. The slope coefficient of the mapping function is obtained from an offline training model to ensure strong response sensitivity under high-incidence disturbances. Next, the suppression command is input into the control feedback loop and compared with the ideal output signal in the main control loop. The difference between the two is calculated at each sampling point. This difference is multiplied by the feedback gain constant of 0.8 to form a deviation correction value. All correction values ​​are sequentially arranged to form a deviation correction sequence. If the absolute value of the deviation at any sampling point exceeds 1.0, the control error is determined to be excessive, and the Kalman filter optimization process is initiated. The Kalman filter's initial prediction error was set to 0.05, the state transition noise to 0.02, and the observation noise to 0.01. These parameters were determined using an offline simulation model under multiple disturbance tests to ensure convergence within 5 milliseconds. The filter predicts and updates the state value at each sampling point according to a standard gain correction process, ultimately outputting an optimized pulse sequence.

[0069] Afterwards, the optimized pulse sequence is statistically processed, with each 20 points forming a sliding window. The mean and standard deviation within the window are calculated to evaluate the stability of the pulse. These values ​​are input into the support vector machine model, which uses a Gaussian radial basis kernel function. The model structure has been trained before deployment. The training sample contains 1,000 groups of normal pulses and 1,000 groups of abnormal pulses. The optimal hyperparameters are selected through five-fold cross-validation. The model output 0 represents instability and 1 represents stability. If the output is 0, the unstable segment is marked and divided into segments. Each segment is dynamically adjusted using the least mean square adaptive filtering algorithm, with an initial weight of zero and a step size of 0.01. The output value of each sample point is calculated and the difference is calculated from the ideal value. The error is multiplied by the step size and the input to form a weight correction value. After updating all sample points in sequence, a stable pulse sequence is formed.

[0070] The stabilization pulse sequence is input into a digital signal processor and output in chronological order, with each output interval of 1 millisecond. This constitutes the final directional suppression pulse sequence and is sent to the execution control unit. Simultaneously, a fast Fourier transform is performed on the pulse sequence, with each 256-point frequency window being a frequency window. The main frequency amplitude and phase are extracted and compared with the frequency information of the original optimized current signal. If the difference is less than 5 Hz and the amplitude ratio is between 0.95 and 1.05, the frequency is marked as consistent and the verification sequence passes. This sequence is then network-encapsulated and formed into data packets according to standard communication protocols. Each packet is timestamped with a 1 millisecond accuracy and transmitted to a remote server. The receiving end compares the timestamp with the current reception time and calculates the packet delay. If it is less than 50 milliseconds, the packet is encrypted using the AES encryption standard.

[0071] The TCP protocol optimization process then performs sequence number reordering and retransmission logic on the encrypted data packets to ensure correct packet order and minimize packet loss. A sliding window protocol is then used to analyze network bandwidth usage, with the window set to five packets. The maximum throughput per second is analyzed and compared with the current total traffic volume to calculate the bandwidth allocation parameter. This parameter is input into the flow control module, which adjusts the number of packets sent per second in the next cycle to achieve stable transmission rate control. Finally, the remote adjustment parameters are extracted from the adjusted data stream, and redundant parameter bits are removed using a standard compression algorithm to form a compressed parameter set. A cyclic redundancy check is then used to verify the integrity of this compressed parameter set. If the checksum is correct, the final verified parameter set is output, which the remote control module uses to perform local parameter updates and synchronize execution.

[0072] A sliding window protocol is used to monitor and analyze bandwidth for optimized data flows generated during network transmission. The window size is set to 5 packets, with a sliding step of 1 packet. Within each window period, the total number of packet bytes and the transmission time interval are recorded. The actual bandwidth utilization is calculated and compared with the network's preset maximum available bandwidth to determine the bandwidth allocation parameter. This bandwidth allocation parameter is a floating-point value set between 0.1 and 1.0, representing the percentage of available network bandwidth currently occupied. This parameter is dynamically determined through multiple experimental measurements at different time periods and under different network load conditions to ensure real-time reflection of network conditions.

[0073] Subsequently, a flow control algorithm is used to adjust the packet transmission rate based on the bandwidth allocation parameter. The control logic is as follows: if the bandwidth allocation parameter is less than 0.6, the transmission rate is proportionally reduced to prevent overload; if the parameter is greater than 0.8, the transmission rate is allowed to increase to improve transmission efficiency; and if the parameter is between 0.6 and 0.8, the rate is maintained at a stable level. The adjusted packets are periodically transmitted at the new rate, forming the adjusted data flow.

[0074] Embedded remote adjustment parameters are extracted from this adjusted data stream. These parameters include feedback adjustment information generated by the aforementioned control module based on grid status analysis, such as filter parameters, adjustment coefficients, or control thresholds. After collecting these parameters, a data compression algorithm is used to compress them. This compression algorithm uses a fixed dictionary table matching method, abbreviating repeated parameter values ​​or structural information and removing redundant bits to improve transmission efficiency. The compression ratio remains stable at over 2x. The compressed output is the remote adjustment parameter set.

[0075] Finally, the remotely adjusted parameter set is verified, using a cyclic redundancy check (CRC) algorithm to generate a 4-byte checksum. This checksum is then appended to the end of the parameter set. Upon receiving the compressed parameter set, the receiver recalculates the CRC value of the received data and compares it with the appended checksum. If they match, indicating the data has not been corrupted, the data set is passed as the verified parameter set to the remote control module, which then updates the control logic and implements synchronous parameter adjustment.

[0076] S7 includes extracting feedback values ​​from a remote adjustment parameter set, separating effective feedback data using a data parsing algorithm, and obtaining a feedback data set; extracting update parameters based on the feedback data set using server response parsing to generate an update parameter set; if the update parameter set meets a preset threshold, adjusting the feedback data set using a proportional integral differential control algorithm to obtain an adjusted data set; detecting data integrity based on the adjusted data set in combination with an error retransmission mechanism to generate a retransmission correction data set; extracting local filtering parameters from the retransmission correction data set, optimizing parameter smoothness using a sliding average algorithm, and obtaining smoothing filtering parameters; suppressing the output sequence through smoothing filtering parameter adjustment, and forming a final output sequence using a sequence generation algorithm; verifying sequence integrity using a verification algorithm based on the final output sequence to obtain a verified output sequence.

[0077] In one possible implementation, feedback values ​​are first extracted from a remote adjustment parameter set. This parameter set is generated by a host server after a comprehensive analysis of the operating status of field devices, the level of grid disturbances, and the response delay of control signals. The parameter set contains multiple feedback fields, including filter adjustment gain, current balance deviation, delay compensation coefficient, and phase synchronization parameters. Using an embedded data parsing algorithm, each record in the parameter set is split into a key-value pair structure according to a preset protocol format. Records with format errors, missing fields, or data overflow are filtered out, retaining only feedback fields with complete structures and valid values. These fields are combined to generate a feedback data set and stored in a local register module. After extraction, each feedback field is immediately converted to a unit format compatible with the local controller. For example, filter gain is recorded as a real number with a unit of 1; current deviation is recorded in amperes; delay compensation is recorded in milliseconds; and phase synchronization is recorded as an angle with a unit of degree. These units are clearly defined in the protocol, and upper and lower limits are configured for each parameter. For example, the effective range of filter gain is 0.1 to 10, the current deviation is allowed to fluctuate within ±5 amps, the delay compensation limit is 100 milliseconds, and the angle synchronization range is within 0 to 360 degrees. These ranges are determined through preliminary engineering tests combined with actual grid fluctuations to ensure that they can respond to disturbances while avoiding oscillations caused by excessive control.

[0078] The feedback dataset is then parsed using the server response. The feedback fields are sorted and sorted based on the timestamp, priority, and modification flags included in the feedback. Based on the update strategy, the latest batch of unsynchronized control parameters is extracted as the updated parameter set. Each field is validated against a preset threshold range. For example, the filter gain fluctuation must not exceed 0.5, the current deviation adjustment must not exceed 3 amps, the delay correction error must be less than 20 milliseconds, and the phase angle update step must not exceed 15 degrees. If the update amplitudes for all fields fall within the corresponding thresholds, the updated parameter set is deemed valid and the proportional-integral-derivative (PID) control process begins. The PID control module first multiplies each feedback error value by a proportional coefficient set to 0.8 based on historical disturbance response stability. The integral component multiplies the continuous error by an integral coefficient of 0.05. The differential component multiplies the error increment by a differential coefficient of 0.02. These three values ​​are summed to form the adjusted output, forming the adjusted dataset.

[0079] To ensure that no data is lost or corrupted in the feedback link, the adjusted data set is transmitted to the retransmission detection module, which verifies the integrity of each packet using a hash checksum and frame number confirmation mechanism. If a packet does not receive an acknowledgment response within the specified time, or if the checksum is inconsistent, the packet is added to the retransmission queue and resent, ensuring that all critical data is successfully delivered. Once all fields are verified to be correct, the retransmitted complete data set is integrated to form a retransmission correction data set.

[0080] After extracting all the local filter parameters from the data set, the sliding average processing module is called in sequence, with every 5 consecutive data as a window, and the filter gain mean within the window is calculated, and this is used as the filter parameter output value at the current moment. The sliding average process slides in chronological order with a step size of 1 to ensure that each filter update action has a smooth transition and avoids control oscillation caused by parameter mutations. The processed smoothing filter parameters are input into the control module to correct the current suppression output sequence, that is, the amplitude, response speed and stability index of the output current are adjusted according to the new parameter settings of the filter. The sequence generation algorithm is used to construct the final output sequence of the corrected control output according to the preset output interval and sequence length. The output interval is set to 1 millisecond, and the total length of the sequence is 1024 sampling points to ensure that the requirements of high-speed response and stable control are met.

[0081] Finally, to ensure the reliability of the final output sequence during subsequent processing or transmission, a cyclic redundancy check is performed on the sequence. A checksum (4 bytes) is calculated for the complete sequence and appended to the end of the output sequence to form a verified output sequence. When parsing the sequence, the receiver uses the same algorithm to calculate the checksum and compares it with the appended value. If they match, the sequence has not been corrupted during transmission or caching and can be used to control actuators, update parameters, or perform grid regulation. This process ensures the closed-loop integrity, real-time performance, and anti-interference capabilities of the control link, ensuring the stable execution of the grid harmonic mitigation process.

[0082] Once the final output sequence is constructed by the sequence generation module, the verification phase immediately begins. This phase ensures that the sequence has not been lost, tampered with, or misplaced during transmission, caching, or execution. This process begins by invoking a cyclic redundancy check (CRC) algorithm, which calculates a checksum for the entire final output sequence according to a fixed polynomial generation rule. During this calculation, the final output sequence is read byte by byte and logically divided based on a preset generator polynomial. Each operation is XORed, ultimately generating a 4-byte checksum. This checksum is a unique summary of the sequence data content, used to identify data integrity.

[0083] After the checksum calculation is completed, the 4-byte checksum is appended to the end of the final output sequence to form a verification output sequence with integrity protection. At this point, the sequence contains both complete data for control and additional codes for integrity verification. During the subsequent transmission process, the receiving end or local execution module will recalculate the cyclic redundancy check code for the received data portion and compare the calculation result with the attached checksum byte by byte. If the comparison is completely consistent, it means that no errors have occurred in the output sequence during the entire processing and transmission process, that is, the verification output sequence is confirmed to be valid and allowed to be used for the execution of the current suppression control instruction; if the verification fails, the sequence is immediately discarded and a request is made to the data source to regenerate the final output sequence to ensure the accuracy and security of the operation.

[0084] S8 includes extracting grid stability characteristics from the suppressed output sequence, separating intermittent disturbance data using a feature extraction algorithm, and obtaining a disturbance feature set; if the disturbance feature set is within the deviation range of a preset threshold, verifying the stability index using a threshold comparison algorithm and generating a verification result set; based on the verification result set, re-collecting data from the digital current sequence using a cyclic sampling mechanism to obtain an updated sampling data set; extracting time series characteristics from the updated sampling data set, analyzing the continuity of the sequence using a sliding window algorithm, and obtaining a continuity analysis result; if the continuity analysis result meets the preset continuity condition, merging the updated sampling data set and the disturbance feature set using a data integration algorithm to generate an integrated monitoring data set; based on the integrated monitoring data set, using a state monitoring algorithm to judge the operation state of the grid and obtain a state assessment result; extracting abnormal fluctuation characteristics from the state assessment result, and using an anomaly detection algorithm to generate an abnormal marking sequence.

[0085] First, grid stability features are accurately extracted from the suppression output sequence, which consists of 1024 consecutive current values ​​sampled at 1-millisecond intervals. A feature extraction algorithm is applied to each sampling point in sequence. With the current point as the center, 10 sampling points before and after it are taken to form a 21-point local time segment. Three features are calculated for this segment: maximum amplitude jump, RMS value, and local rate of change. These three metrics are expressed in units of amperes for current amplitude and amperes per millisecond for rate of change. Warning thresholds are set for each feature: amplitude jumps must not exceed 5 amperes, RMS value change rate must not exceed 0.1 amperes per millisecond, and maximum sustained change must not exceed 50 amperes per millisecond (represented by the multiplication of the amplitude and duration). These thresholds are determined by combining 95% statistical results of historical grid fluctuations with measured wind farm data to ensure that the identified disturbances are engineering-relevant. All sampling points are traversed in a time series, and sampling segments that exceed any of the thresholds are classified as disturbance feature sets. The start and end times, characteristic metric values, and duration of each disturbance segment are recorded.

[0086] Then, each disturbance record in the disturbance feature set is retrieved and checked item by item using the threshold comparison algorithm: check whether the maximum amplitude jump is less than or equal to 5 amps, check whether the root mean square value change is less than or equal to 0.1 amps per millisecond, and check whether the duration is less than or equal to 50 milliseconds. If all indicators of a record fall within the corresponding threshold range, the disturbance is determined to be allowed and the record is included in the verification result set. Otherwise, it is considered an anomaly and removed from the verification result set.

[0087] After the verification result set is generated, the cyclic sampling mechanism is triggered, restarting sampling from the current digital current sensor at a fixed interval of 1 millisecond, continuously collecting 1024 sampling points to form an updated sampling data set. The sampling process is controlled by a digital signal processor, with no delay jumps, ensuring sampling continuity and data integrity.

[0088] A sliding window continuity analysis was performed on the updated sample data set, with a window length of 64 points and a step size of 1 point. The current mean and slope of change were calculated for each window. The continuity conditions were set as follows: the window mean change rate did not exceed 0.1 amps per millisecond, and the mean change rate between adjacent windows did not exceed 0.05 amps per millisecond. If all sliding windows met these conditions, the sequence was considered to have good continuity.

[0089] When the continuity condition is met, a data integration algorithm is used to merge the updated sampling dataset with the previously recorded disturbance signature set in timestamp order. During the fusion process, updated sampling data is prioritized for duplicate data within the same time period and the source of the update is marked. Each data entry is assigned a "source tag" and a "time priority" and converted into a structured monitoring record in a unified format to form an integrated monitoring dataset. The integrated monitoring dataset is then processed using a condition monitoring algorithm. The algorithm calculates the fluctuation amplitude, frequency offset, and phase jump for each record and compares them to preset operational thresholds: the fluctuation amplitude must not exceed 3% of the fundamental wave average value, the frequency offset must not exceed 0.5 Hz / ms, and the phase jump must not exceed 10 degrees. If all indicators are within the thresholds, the record is deemed stable; otherwise, it is considered abnormal. The evaluation results of all records are combined to generate a final condition assessment result. Based on the condition assessment results, all records identified as "abnormal" are extracted, and the time interval, current level, and frequency offset within them are analyzed as abnormal fluctuation features. These feature segments are input into the anomaly detection algorithm, which generates start and end time nodes for each anomaly marker and outputs an anomaly marker sequence, which is used to indicate the existence of unstable fluctuations in a specific period of time and can be used for subsequent intervention by control logic or operation and maintenance personnel.

[0090] When the sliding window continuity analysis results meet the preset continuity conditions—that is, the mean change rate within the window is less than 0.1 amperes per millisecond and the variance does not exceed 0.5 amperes squared—the data integration algorithm performs a time-domain merge between the currently updated sampling dataset and the previously extracted disturbance feature set. This merging process uses the timestamp of each record as the primary key, prioritizing the retention of data from the time period covered by the updated sampling dataset. For overlapping data, higher-quality data is selected and formatted uniformly using a data source identification and integrity priority mechanism. The merged results are stored as a structured integrated monitoring dataset. This dataset includes the timestamp, current amplitude, rate of change, disturbance feature identifier, and source label of each data item, ensuring traceability and data consistency for each record.

[0091] Next, a condition monitoring algorithm is invoked to assess the status of the integrated monitoring dataset. The algorithm analyzes each record for peak values, frequency fluctuations, phase offsets, and short-term amplitude variations, comparing each record against the following set standard values: peak deviation must not exceed 3% of the fundamental wave average value, frequency variation must not exceed 0.5 Hz / ms, and phase jumps must not exceed 10 degrees. Based on these rules, each record in the integrated dataset is assigned a "normal" or "abnormal" status label. Subsequently, a grid operation status assessment is generated based on the overall record status. If more than 90% of the records are "normal," the overall grid is assessed as stable; if more than 10% of the records are "abnormal," there is a risk of unstable operation.

[0092] Finally, all recorded intervals marked as "abnormal" are extracted from the state assessment results, and their specific time ranges, current waveforms, and frequency offsets are identified to form abnormal fluctuation characteristic data. The anomaly detection algorithm is then called to process this data. Based on the principle of change mutation point detection, the algorithm analyzes the start time, peak change rate, mutation slope, and duration of each abnormal record. If these indicators reach the set change rate threshold for multiple consecutive sampling points, the waveform segment is marked as an abnormal segment. Ultimately, all abnormal segments are organized into an abnormal marking sequence in chronological order, recording their start and end times, current change characteristics, frequency and phase disturbances, and other information. This information serves as a reference for subsequent control strategy execution or external intervention.

[0093] When extracting grid stability characteristics, the maximum amplitude jump threshold is set at 5A. This is based on the fact that, in historical current fluctuation data collected at the grid connection point under typical wind farm operating conditions, over 95% of normal operating samples have transient amplitudes below 5A. Therefore, setting 5A as the threshold effectively eliminates normal disturbances while sensitively detecting abnormal high-frequency disturbances, ensuring high identification accuracy and engineering applicability during the screening process. The duration threshold is set at 50 milliseconds. This is based on the fact that, in actual grid-connected stable operation, intermittent disturbances typically last less than 30 milliseconds. Disturbances exceeding 50 milliseconds can have a substantial impact on grid control stability. To avoid misjudgment of transient disturbances while taking into account response delays and oversampling, a 50-millisecond upper limit provides sufficient technical tolerance to ensure that stability monitoring functions are not disrupted by short-term fluctuations. The energy accumulation threshold is based on the energy distribution of historical data during normal operation. The 95th percentile statistical value is set as the upper limit to ensure that the assessment of disturbance intensity is neither overly sensitive nor too insensitive. The energy accumulation value represents the square of the current amplitude multiplied by the duration per unit time, and comprehensively reflects the intensity and persistence of the disturbance. The purpose of setting this threshold is to exclude non-critical disturbances and improve the suppression algorithm's ability to respond to actual threats. The threshold for the mean change rate in the sliding window continuity analysis is set to 0.1 amperes per millisecond. This is a steady-state threshold calculated by calculating the average and maximum values ​​of the fluctuation rate during continuous operation, assuming a sampling period of 1 millisecond. This value ensures the continuity of the current sequence, thereby avoiding false triggering and misjudgment of status due to sudden data changes. The frequency offset is set to no more than 0.5 Hz per millisecond, and the phase jump is set to no more than 10 degrees. These two items are set based on international grid-connected standards and the wind turbine converter control tolerance range, respectively, and are basic constraints for high-frequency disturbance identification.

[0094] like Figure 2As shown, a virtual synchronous generator grid-connected current harmonic suppression system is also provided, which is used to implement the steps of the virtual synchronous generator grid-connected current harmonic suppression method. The system includes a sampling module for digitally sampling intermittent high-frequency disturbances using an analog-to-digital converter through the grid-connected current signal of the wind farm in the power grid to obtain an original digital current sequence; a spectrum analysis module for analyzing the wind load frequency component based on the original digital current sequence using a fast Fourier transform algorithm to determine the amplitude and phase characteristics of the high-frequency harmonic component; a disturbance extraction module for extracting the intermittent disturbance harmonic sequence from the frequency component when the amplitude of the high-frequency harmonic component exceeds a preset threshold to obtain a harmonic subset to be suppressed; a filtering control module for dynamically adjusting the filtering parameters using an adaptive filtering algorithm through the harmonic subset to be suppressed to generate a compensation current sequence; a matching calculation module for obtaining phase matching adjustment and amplitude scaling parameters from the compensation current sequence to determine the matching degree between the compensation current sequence and the original digital current sequence; if the matching degree is higher than the threshold, the signal is superimposed The calculation and noise suppression filtering integrate the compensation current sequence into the main control loop to obtain an optimized current signal; the digital signal processing module is used to use a digital signal processor to calculate the suppression instruction in real time according to the optimized current signal, and at the same time perform loop feedback integration and stability verification checks to generate a directional suppression pulse sequence; the remote transmission module is used to obtain network transmission data packets, perform delay threshold judgment and network bandwidth evaluation, and if the delay is lower than the threshold, it is sent to the remote server using data packet encryption and transmission protocol optimization to obtain a remote adjustment parameter set; the parameter update module is used to extract feedback values ​​from the remote adjustment parameter set, and update the local filter parameters through server response analysis and parameter set download and update using a proportional integral differential control algorithm combined with an error retransmission mechanism and feedback value extraction processing to generate a final suppression output sequence; the stability monitoring module is used to judge whether the intermittent disturbance stability index of the remote wind farm power grid meets the requirements based on the final suppression output sequence. If it meets the requirements, it loops back to the sampling process of the original digital current sequence to maintain a continuous monitoring state.

[0095] The sampling module first acquires current signals from the wind farm's grid connection point and continuously digitizes them using an analog-to-digital converter (ADC). The sampling frequency is set to 10,000 Hz to ensure high-resolution capture of high-frequency disturbances. Each sampling step generates a 1,024-point raw digital current sequence, which is then fed into the spectrum analysis module.

[0096] The spectrum analysis module receives the raw digital current sequence and calculates its frequency domain components using a fast Fourier transform algorithm. It then extracts the amplitude and phase of each frequency component in the high-frequency range above 500 Hz. The system compares the amplitude of each high-frequency component with a preset threshold. If the amplitude of a frequency component exceeds 3 amps, the disturbance extraction module is triggered.

[0097] The disturbance extraction module takes a set of frequency components as input and uses spectral analysis to identify intermittent frequencies. Combining short-time Fourier transforms with wavelet transforms, it calculates the energy distribution of each frequency over time. It then selects a sequence of frequency components with a disturbance duration exceeding 30 milliseconds and transmits this to the filter control module as a subset of harmonics to be suppressed.

[0098] The filter control module receives the subset of harmonics to be suppressed and uses a minimum mean square error adaptive filtering algorithm to dynamically adjust the filter weights, generating a compensating current sequence with opposite phases and equal amplitudes in real time. This sequence is used to suppress the target disturbance in the time domain.

[0099] The matching calculation module then compares the compensated current sequence with the original current sequence, extracting the phase difference and amplitude ratio for each corresponding frequency and calculating the cosine similarity value as a matching indicator. If the matching degree is greater than 0.95, the system performs signal superposition and noise filtering, integrates the compensated current sequence into the main control loop, and outputs the optimized current signal.

[0100] The digital signal processing module optimizes the time-domain characteristics of the current signal, calculates the next suppression command through a built-in digital signal processor, and updates the control strategy in real time based on current changes in the feedback channel. This module also implements stability verification logic and generates a directional suppression pulse train for regulating the power converter.

[0101] The remote transmission module is used to convert the above-mentioned suppression pulse sequence into a control data packet. Before sending, it uses the delay evaluation mechanism to determine whether the network transmission delay is less than 20 milliseconds. If it meets the requirements, the system uses an encryption algorithm to encrypt the data packet and adopts an optimization mechanism based on the transmission control protocol to send the data packet to the remote server.

[0102] The parameter update module receives the adjustment parameter set from the remote server, parses the effective feedback value through the feedback response channel, and uses the proportional integral derivative control algorithm combined with the error retransmission mechanism to correct the parameters. Finally, the updated control parameters are applied to the filter configuration to generate the final suppressed output sequence.

[0103] The stability monitoring module analyzes the final suppressed output sequence, extracting the volatility, frequency offset, and continuity indicators of the grid-connected current sequence to determine whether it meets the set wind farm grid stability requirements. If the evaluation result meets the system's set threshold, the system automatically loops back to the sampling module for the next round of sampling and analysis, thus achieving full-cycle continuous monitoring.

[0104] The system architecture features clear modules, clear functional division of labor, and closed-loop execution process. It can effectively meet the needs of high-frequency harmonic detection, identification, and compensation control in wind power grid-connected scenarios, ensuring grid-connected current quality and grid stability.

[0105] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for suppressing current harmonics of a virtual synchronous generator connected to the grid, characterized in that: include: S1. Using the wind farm grid-connected current signal in the power grid, an analog-to-digital converter is used to digitally sample the intermittent high-frequency disturbance to obtain the original digital current sequence; S2. Based on the original digital current sequence, the fast Fourier transform algorithm is used to analyze the wind load frequency component and determine the amplitude and phase characteristics of the high-frequency harmonic component; S3. If the amplitude of the high-frequency harmonic component exceeds a preset threshold, extract the intermittent disturbance harmonic sequence from the frequency component to obtain a harmonic subset to be suppressed; S4. Using the subset of harmonics to be suppressed, an adaptive filtering algorithm is used to dynamically adjust the filtering parameters to generate a compensation current sequence; S5. Obtain phase matching adjustment and amplitude scaling parameters from the compensation current sequence, determine the matching degree between the compensation current sequence and the original digital current sequence, and if the matching degree is higher than a threshold, fuse the compensation current sequence into the main control loop through signal superposition operation and noise suppression filtering to obtain an optimized current signal; S6. Based on the optimized current signal, a digital signal processor is used to calculate the suppression instruction in real time, and loop feedback integration and stability verification checks are performed at the same time to generate a directional suppression pulse sequence. Through the directional suppression pulse sequence, network transmission data packets are obtained, and delay threshold judgment and network bandwidth evaluation are performed. If the delay is lower than the threshold, data packet encryption and transmission protocol optimization are used to transmit to the remote server to obtain a remote adjustment parameter set.

2. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 1, characterized in that: Said S1 comprises: The wind farm grid-connected current signal is digitally sampled through an analog-to-digital converter to generate an original digital current sequence; The fast Fourier transform algorithm is used to perform frequency domain analysis on the original digital current sequence to obtain the frequency components of the current signal; If there is an abnormal high-frequency component higher than a preset threshold in the frequency component, the original digital current sequence is filtered through a bandpass filter to obtain a filtered current sequence; According to the filtered current sequence, the short-time Fourier transform algorithm is used to extract the time-frequency characteristics and generate the time-frequency distribution of the current signal; If the duration of the intermittent high-frequency disturbance detected in the time-frequency distribution exceeds a preset threshold, the filtered current sequence is decomposed using a wavelet transform algorithm to obtain the high-frequency disturbance component; According to the high-frequency disturbance component, calculate its energy distribution characteristics and determine the disturbance intensity and occurrence location; The disturbance intensity and location are analyzed through the preset classification model to determine the operating status of the power grid.

3. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 1, characterized in that: The S2 includes: The fast Fourier transform algorithm is used to perform frequency domain analysis on the original digital current sequence to obtain the amplitude and phase characteristics of the high-frequency harmonic components; According to the amplitude and phase characteristics of the high-frequency harmonic components, the original digital current sequence is processed through a bandpass filter to obtain a filtered current sequence; The short-time Fourier transform algorithm is used to perform time-frequency analysis on the filtered current sequence to obtain the time-frequency distribution characteristics; If the energy concentration of the frequency component of the high-frequency disturbance is detected in the time-frequency distribution characteristics, the filtered current sequence is decomposed by the wavelet transform algorithm to obtain the high-frequency component sequence; According to the high-frequency component sequence, the time energy distribution is calculated and the disturbance time characteristics are determined; If the duration of the disturbance time characteristic exceeds the preset threshold, the high-frequency component sequence is analyzed through the preset classification model to determine the grid operation status; According to the operating status of the power grid, status assessment data is generated to determine the grid-connected stability of the wind farm.

4. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 1, characterized in that: The S3 includes: If the amplitude of the high-frequency harmonic component exceeds the preset threshold, the original current sequence is decomposed in the frequency domain using the fast Fourier transform algorithm to obtain a set of frequency components; According to the frequency component set, the spectrum analysis method is used to extract the intermittent disturbance sequence and generate the harmonic sequence to be suppressed; The harmonic sequence to be suppressed is processed by a bandpass filter to obtain a filtered harmonic sequence; If the energy of the harmonic sequence after filtering is concentrated in the frequency range of high-frequency disturbance, the short-time Fourier transform algorithm is used to perform time-frequency analysis to obtain the time-frequency distribution characteristics. According to the time-frequency distribution characteristics, the energy concentration sequence is extracted to generate the high-frequency disturbance sequence; Analyze high-frequency disturbance sequences through a preset classification model to determine the grid operation status and obtain status assessment data; According to the state assessment data, an adaptive filtering algorithm is used to suppress the high-frequency disturbance sequence and generate a stable current sequence.

5. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 1, characterized in that: The S4 includes: By using the subset of harmonics to be suppressed, the least mean square adaptive filtering algorithm is used to dynamically adjust the filtering parameters and generate a compensation current sequence; According to the compensation current sequence, discrete Fourier transform is used to decompose the signal and obtain a set of frequency components; If there is a component in the high-frequency disturbance frequency range in the frequency component set, the harmonic sequence in the frequency range is extracted through a bandpass filter to obtain a filtered harmonic sequence; According to the filtered harmonic sequence, short-time Fourier transform is used to analyze the time-frequency distribution and obtain the time-frequency characteristic sequence; Through the time-frequency feature sequence, the preset classification model is used to judge the power grid status and obtain the status assessment data; If the state assessment data indicates that the grid state is abnormal, the filtered harmonic sequence is processed twice using the least mean square adaptive filtering algorithm to generate a stable current sequence; According to the stable current sequence, the time domain signal is reconstructed by inverse Fourier transform to obtain the optimized current sequence.

6. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 1, characterized in that: The S5 includes: Extract phase adjustment parameters and amplitude scaling parameters from the compensation current sequence, decompose the signal using fast Fourier transform, obtain phase and amplitude characteristics, and obtain phase adjustment parameters and amplitude scaling parameters; According to the phase adjustment parameters and amplitude scaling parameters, the phase difference and amplitude ratio between the compensated current sequence and the original digital current sequence are calculated, and the cosine similarity algorithm is used to evaluate the matching degree between the two to obtain the matching value; If the matching value is higher than the preset threshold, the compensation current sequence is integrated into the main control loop through weighted signal superposition operation to obtain a fused current signal; For the fused current signal, a filter is used to remove the noise component to obtain a denoised current signal; Based on the denoised current signal, the time domain statistical characteristics of the signal are calculated, and the support vector machine classification model is used to judge the signal stability and obtain the stability evaluation result; If the stability evaluation result indicates that the signal is unstable, the denoised current signal is adjusted twice using the least mean square adaptive filtering algorithm to obtain a stable current signal; According to the stable current signal, the time domain signal is reconstructed by inverse fast Fourier transform to obtain the optimized current signal.

7. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 1, characterized in that: The S6 includes: The time domain characteristics are obtained from the optimized current signal, and a digital signal processor is used for real-time calculation to obtain a suppression instruction sequence; By suppressing the command sequence, the feedback loop is used to calculate the deviation from the main control loop to obtain the deviation correction sequence; If the amplitude of the deviation correction sequence exceeds a preset threshold, the Kalman filter algorithm is used to optimize the suppression instruction sequence to obtain an optimized pulse sequence; According to the optimized pulse sequence, the statistical characteristics of the time series are calculated, and the support vector machine classification model is used to judge the stability of the sequence and obtain the stability judgment result; Based on the stability judgment results, the unstable sequence segments are extracted and the adaptive filtering algorithm is used for secondary adjustment to obtain the stable pulse sequence. According to the stable pulse sequence, a directional suppression pulse sequence is generated, and a digital signal processor is used for real-time output to obtain the final pulse sequence; The frequency distribution characteristics of the final pulse sequence are calculated, and the frequency consistency is verified by fast Fourier transform to obtain the verification pulse sequence; Extract network transmission data packets from the directional pulse sequence, calculate the data packet delay using timestamp analysis, and obtain the delay judgment result; If the delay judgment result is lower than the preset threshold, the network transmission data packet is encrypted using the encryption standard algorithm to obtain an encrypted data packet; According to the encrypted data packets, the transmission control protocol optimization strategy is used to adjust the data packet sending order and generate an optimized data stream; By optimizing data flow and using sliding window protocol to analyze network bandwidth usage, bandwidth allocation parameters are obtained; According to the bandwidth allocation parameters, a flow control algorithm is used to adjust the data packet transmission rate to generate an adjusted data stream; Extracting remote adjustment parameters from the adjusted data stream, generating a compressed parameter set using a data compression algorithm, and obtaining a remote adjustment parameter set; By remotely adjusting the parameter set and using a verification algorithm to verify the integrity of the parameter set, a verified parameter set is obtained.

8. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 1, characterized in that: The process also includes S7, extracting feedback values ​​from the remote adjustment parameter set, parsing the server response and downloading and updating the parameter set, and updating the local filter parameters using a proportional-integral-differential control algorithm combined with an error retransmission mechanism and feedback value extraction processing to generate a final suppression output sequence. Specifically, the process includes: Extract feedback values ​​from the remote adjustment parameter set, use data parsing algorithms to separate effective feedback data, and obtain a feedback data set; According to the feedback data set, the server response is parsed to extract the update parameters and generate the update parameter set; If the updated parameter set meets the preset threshold, the proportional integral derivative control algorithm is used to adjust the feedback data set to obtain an adjusted data set; Based on the adjusted data set, the data integrity is detected in combination with the error retransmission mechanism to generate a retransmission correction data set; Extract local filter parameters from the retransmitted correction data set, and use the sliding average algorithm to optimize the smoothness of the parameters to obtain smooth filter parameters; The output sequence is suppressed by adjusting the smoothing filter parameters, and the final output sequence is formed by using the sequence generation algorithm; According to the final output sequence, a verification algorithm is used to verify the sequence integrity and obtain a verification output sequence.

9. The method for suppressing grid-connected current harmonics of a virtual synchronous generator according to claim 8, characterized in that: The process also includes S8, judging whether the intermittent disturbance stability index of the wind farm power grid meets the requirements based on the final suppressed output sequence. If so, looping back to the sampling process of the original digital current sequence to maintain a continuous monitoring state, specifically including: Extract the grid stability features from the suppressed output sequence, use the feature extraction algorithm to separate the intermittent disturbance data, and obtain the disturbance feature set; If the disturbance feature set is within the deviation range of the preset threshold, the threshold comparison algorithm is used to verify the stability index and generate a verification result set; According to the verification result set, a cyclic sampling mechanism is used to re-collect data from the digital current sequence to obtain an updated sampling data set; Extract time series features from the updated sampling data set, use the sliding window algorithm to analyze the continuity of the sequence, and obtain the continuity analysis results; If the continuity analysis result meets the preset continuity condition, the updated sampling data set and the disturbance feature set are merged and updated through the data integration algorithm to generate an integrated monitoring data set; Based on the integrated monitoring data set, the state monitoring algorithm is used to judge the operation status of the power grid and obtain the state assessment results; Abnormal fluctuation features are extracted from the state assessment results, and anomaly detection algorithms are used to generate anomaly marker sequences.

10. A virtual synchronous generator grid-connected current harmonic suppression system, used to implement the steps of the virtual synchronous generator grid-connected current harmonic suppression method according to any one of claims 1 to 9, characterized in that: The system comprises: The sampling module is used to digitally sample the intermittent high-frequency disturbance using the grid-connected current signal of the wind farm in the power grid by using an analog-to-digital converter to obtain the original digital current sequence; The spectrum analysis module is used to analyze the wind load frequency components based on the original digital current sequence using the fast Fourier transform algorithm to determine the amplitude and phase characteristics of the high-frequency harmonic components; A disturbance extraction module is used to extract the intermittent disturbance harmonic sequence from the frequency component when the amplitude of the high-frequency harmonic component exceeds a preset threshold, so as to obtain a subset of harmonics to be suppressed; A filter control module is used to dynamically adjust filter parameters using an adaptive filter algorithm based on a subset of harmonics to be suppressed to generate a compensation current sequence; The matching calculation module is used to obtain phase matching adjustment and amplitude scaling parameters from the compensation current sequence and determine the matching degree between the compensation current sequence and the original digital current sequence. If the matching degree is higher than the threshold, the compensation current sequence is integrated into the main control loop through signal superposition operation and noise suppression filtering to obtain the optimized current signal. A digital signal processing module is used to calculate the suppression command in real time based on the optimized current signal using a digital signal processor, while performing loop feedback integration and stability verification checks to generate a directional suppression pulse train; The remote transmission module is used to obtain network transmission data packets, perform delay threshold judgment and network bandwidth evaluation, and if the delay is lower than the threshold, the data packet is encrypted and sent to the remote server through transmission protocol optimization to obtain the remote adjustment parameter set; The parameter update module is used to extract feedback values ​​from the remote adjustment parameter set, update the local filter parameters through server response parsing and parameter set download, and adopt the proportional integral derivative control algorithm combined with the error retransmission mechanism and feedback value extraction processing to generate the final suppression output sequence; The stability monitoring module is used to determine whether the intermittent disturbance stability index of the remote wind farm power grid meets the requirements based on the final suppressed output sequence. If so, it loops back to the sampling process of the original digital current sequence to maintain a continuous monitoring state.

Images