Wideband oscillation risk prediction method and system based on wide-area broadband measurement

Abstract

The invention discloses a broadband oscillation risk prediction method and system based on wide-area broadband measurement, and relates to the technical field of wide-area broadband measurement. The method comprises the following steps: firstly, collecting grid-connected point high-frequency synchronous phasor data by using a PMU device at a photovoltaic field station side, and performing filtering and time mark calibration processing to eliminate noise and ensure time consistency; performing spectrum correlation analysis on wide-area multi-station PMU data based on time mark synchronization, screening broadband oscillation candidate features, and transmitting the broadband oscillation candidate features to a scheduling master station prediction network; and finally, inputting the real-time operation parameters of the power grid and historical oscillation sample data into the network, carrying out risk assessment and trend prediction, and assisting broadband oscillation early warning and prevention and control decision. According to the method, efficient fusion of multi-area equipment data is realized, a full-system broadband dynamic monitoring view angle is formed, the oscillation dynamic state is captured in real time, the risk is pre-judged, the broadband oscillation risk caused by new energy power fluctuation and power electronic equipment transient response can be dealt with, and the dynamic monitoring requirement of a novel power system is met.

Description

Technical Field

[0001] This invention relates to the field of wide-area broadband measurement technology, and in particular to a method and system for predicting broadband oscillation risk based on wide-area broadband measurement. Background Technology

[0002] With the large-scale grid connection of new energy sources (such as energy storage power stations and wind farms), the new power system is characterized by a high proportion of power electronic devices and large power fluctuations, which can easily trigger wideband oscillations of 0.1Hz-10kHz, posing a serious threat to grid security. Against this backdrop, wide-area broadband measurement technology has become a key support, with its core equipment being the phasor measurement unit (PMU). This PMU can achieve high-frequency synchronous acquisition of dynamic data from the entire system (sampling rate up to kilohertz), and the data carries a unified time scale, enabling precise capture of broadband oscillation processes.

[0003] Current power system monitoring primarily relies on Supervisory Control and Data Acquisition (SCADA) systems and fault recording systems: SCADA monitors the power grid by acquiring steady-state data from Remote Terminal Units (RTUs); fault recording systems acquire local transient waveforms during faults, lasting only a few seconds. Through steady-state monitoring and post-fault local tracing, both systems enable daily monitoring of the power grid's stable operation and, combined with historical fault data, facilitate analysis to predict basic power grid fault risks.

[0004] For example, Chinese invention patent CN117630541A discloses a broadband oscillation monitoring method and device, which includes: segmenting and preprocessing electrical measurement data in the range of 0 to 2500 Hz that are collected in real time; extracting features of the signal within the sliding window based on the results of the segmented preprocessing; and using a fully connected network (FC) to predict the broadband oscillation state based on the extracted feature values, thereby judging the monitored electrical measurement data.

[0005] However, in new power system scenarios with large-scale grid connection of renewable energy and dense integration of power electronic equipment, broadband oscillations are easily triggered due to the large fluctuations in renewable energy output power and the high proportion of power electronic equipment. Existing monitoring methods are insufficient to meet the dynamic broadband monitoring requirements. On the one hand, the data interaction methods between PMU devices and dispatch master stations, as well as the analysis methods for broadband measurement data, have not yet formed a unified standard. This makes it difficult to efficiently integrate equipment data from different regions and manufacturers, and thus cannot form a unified broadband dynamic monitoring perspective for the entire system. On the other hand, existing systems mostly focus on data acquisition and post-event analysis, making it difficult to capture the dynamic process of broadband oscillations from their inception to development in real time, let alone predict risks in advance. In addition, the randomness of renewable energy output power fluctuations and the rapid transient response of power electronic equipment make it difficult to respond promptly to the broadband oscillation risks caused by high grid connection ratios. Summary of the Invention

[0006] To address the technical problems in the prior art, this invention provides a method and system for predicting broadband oscillation risk based on wide-area broadband measurement. The technical solution is as follows: On the one hand, a broadband oscillation risk prediction method based on wide-area broadband measurement is provided. The method includes: S1, at the photovoltaic power station side, high-frequency synchronous phasor data of the grid-connected points of the photovoltaic power station side are collected in real time through the deployed PMU device, and filtered and time-calibrated to eliminate noise interference and ensure data time consistency. The high-frequency synchronous phasor data is used to reflect the synchronous dynamic impact of photovoltaic power station output fluctuations and load changes on the voltage, current and power of each grid-connected point; S2, based on the PMU device after filtering and time-calibration, the wide-area multi-station PMU data is used to perform spectral correlation analysis to screen out candidate oscillation features, and the candidate oscillation features are transmitted to the oscillation trend prediction network of the dispatch master station. The wide-area multi-station PMU data is used to reflect the grid operation status of the grid-connected points of the photovoltaic power station side in different geographical areas, and the candidate oscillation features are used to reflect the broadband domain signal coordinated fluctuation law of the grid-connected points of the photovoltaic power station in different geographical areas; S3, real-time grid operation parameters and historical oscillation sample data are acquired and input into the oscillation trend prediction network for risk assessment and trend prediction, so as to realize the early warning of broadband oscillations caused by photovoltaic grid connection in the power system.

[0007] On the other hand, a broadband oscillation risk prediction system based on wide-area broadband measurement is provided. This system includes: a high-frequency synchronous phasor data acquisition and processing module, a candidate oscillation feature screening module, and a risk assessment and trend prediction module. The high-frequency synchronous phasor data acquisition and processing module is used to acquire high-frequency synchronous phasor data from the grid-connected points of photovoltaic power plants in real time through deployed PMU devices, and performs filtering and time-scale calibration to eliminate noise interference. The candidate oscillation feature screening module is used to perform spectral correlation analysis on the wide-area multi-station PMU data acquired by the filtered and time-scale-calibrated PMU devices to screen out candidate oscillation features, and transmit the candidate oscillation features to the oscillation trend prediction network of the dispatch master station. The risk assessment and trend prediction module is used to acquire real-time operating parameters of the power grid and historical oscillation sample data, and input them into the oscillation trend prediction network for risk assessment and trend prediction, so as to realize early warning of broadband oscillations caused by photovoltaic grid connection in the power system.

[0008] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This invention first deploys PMU devices at photovoltaic power plants to collect high-frequency synchronous phasor data from grid-connected points in real time. Noise is eliminated through filtering and time-scale calibration to ensure time consistency. This step standardizes data preprocessing, resolving the difficulty in integrating PMU data from different manufacturers due to differences in time scales and noise, thus laying a consistent data foundation for subsequent system-wide analysis. Next, spectral correlation analysis is performed based on time-scaled, multi-plant PMU data from a wide area, screening broadband oscillation candidate features and transmitting them to the dispatch master station's prediction network. This process integrates photovoltaic power plant operating status data from different geographical regions, forming a unified broadband dynamic monitoring perspective for the entire system. Finally, real-time grid operating parameters and historical oscillation sample data are input into the prediction network for risk assessment and trend prediction, assisting in early warning and prevention decisions. This process captures oscillation dynamics in real time and predicts risks in advance, addressing the broadband oscillation risks caused by the randomness of new energy power fluctuations and the rapid transient response of power electronic equipment, meeting the dynamic monitoring needs of new power systems.

[0009] 2. During filtering, the voltage and current signals in the initial high-frequency synchronization phasor data are first subjected to Fourier transform to decompose the frequency component information. Then, the target frequency band signals are screened and interference suppression is performed to retain the effective signals reflecting the dynamic correlation characteristics of the grid connection point caused by changes in photovoltaic output and load. Finally, smoothing is performed to reduce errors. This process unifies the broadband measurement data processing standard, extracts accurate dynamic signals, and solves the problem of data from different manufacturers being difficult to integrate due to different analysis methods. For time calibration, the offset between the original time scale of each PMU and the GPS reference time scale is first calculated. The device to be calibrated is marked and calibrated. The calibration is repeated until it is qualified. If it is not qualified, an early warning is issued, and finally the time-synchronized PMU device is determined. This process ensures that the time of PMU data in different areas is consistent, which helps to form a unified broadband dynamic monitoring perspective for the entire system, avoids misjudgment of oscillation characteristics due to time scale deviation, and lays the foundation for time synchronization for real-time capture of broadband oscillation dynamics and early prediction of risks, so as to better cope with the oscillation risks caused by new energy fluctuations and transient responses of power electronic equipment.

[0010] 3. In spectral correlation analysis, instantaneous electrical parameter data and real-time power data are first compared and correlated moment by moment within a preset monitoring period. The similarity of frequency change trends of broadband signals from different stations is calculated using a dynamic time warping algorithm. The synchronicity of amplitude fluctuation peaks is calculated by marking the peak amplitude moments. Then, a database containing geographical topology and historical features is used to determine coordinated fluctuation characteristics. This process standardizes broadband data correlation analysis, breaks down barriers to data fusion between different regions, forms a system-wide broadband monitoring perspective, and solves the data fragmentation problem caused by inconsistent existing analysis methods. If coordinated fluctuation characteristics exist and the matching degree meets the standard, they are marked as candidate oscillation characteristics; otherwise, anomaly verification is performed. This process captures the dynamics of each stage of oscillation in real time, avoiding the limitations of traditional post-event analysis. It can cope with the randomness of new energy power fluctuations and the rapid transient response of power electronic equipment, reducing risk misjudgments caused by data problems and helping to respond promptly to broadband oscillation risks.

[0011] 4. When obtaining the matching degree evaluation value, feature datasets for effective candidate oscillation types and real-time photovoltaic power output load trends are first constructed separately. Compared with the existing single-dimensional matching method, this approach uses dual-dimensional feature construction to more comprehensively cover the correlation elements between oscillations and operating conditions, laying the foundation for accurate evaluation. Next, frequency fit and period consistency are calculated, and then the weights from the coordinated fluctuation feature database are retrieved for weighted summation. Finally, the results are normalized to obtain the evaluation value. Existing technologies often neglect weight differentiation, while this process combines historical database experience to allocate weights, which can better reflect the correlation characteristics between different oscillation types and operating conditions, reducing subjective errors. Simultaneously, the dual-dimensional quantitative calculation and normalization processing, compared with traditional qualitative judgment, make the evaluation results more objective and comparable, accurately reflecting the degree of fit between oscillation types and grid operating conditions, providing a reliable basis for subsequent adjustment of prediction model parameters.

[0012] 5. Risk assessment and trend prediction are carried out in three steps: The first step, classifying the effective oscillation types output by the classification network, is the core basis for the second step, the prediction network, to determine the parameters of the appropriate operating condition model. That is, only by clearly identifying the oscillation type can the parameters in the prediction model that are suitable for that type be adjusted in a targeted manner to avoid the prediction deviating from the actual oscillation characteristics. The amplitude and frequency change trends obtained in the second step directly provide key input data for the judgment network in the third step. The judgment network needs to extract the amplitude and frequency risk coefficients from the change trend and then classify the risk level through coupling calculation. At the same time, the prevention and control suggestions generated in the third step can also provide feedback reference for optimizing the type identification accuracy in the first step and adjusting the model parameters in the second step. The three form a closed loop of type identification, trend prediction and risk judgment, ensuring the continuity and accuracy of the entire risk assessment and trend prediction process. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating a broadband oscillation risk prediction method based on wide-area broadband measurement, provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the first step of the risk assessment and trend prediction process provided in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the second step of the risk assessment and trend prediction process provided in an embodiment of the present invention. Figure 4 The flowchart for the third step of the risk assessment and trend prediction process provided in this embodiment of the invention; Figure 5 This is a schematic diagram of the structure of a broadband oscillation risk prediction system based on wide-area broadband measurement provided in an embodiment of the present invention; Figure 6 This is one of the homepage interfaces of the power system operation and maintenance platform provided in the embodiments of the present invention; Figure 7 This is the second homepage of the power system operation and maintenance platform provided in this embodiment of the invention; Figure 8 This is an interface for assessing the oscillation risk of a power system operation and maintenance platform provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0016] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0017] With the large-scale grid connection of new energy sources such as photovoltaics, the new power systems, due to the high proportion of power electronic equipment and large power fluctuations, are prone to broadband oscillations. Existing monitoring methods suffer from difficulties in data fusion and real-time risk prediction. To make the technical problems, solutions, and advantages of this invention clearer, a detailed description will be provided below with reference to the accompanying drawings and specific embodiments.

[0018] This invention provides a method for predicting broadband oscillation risk based on wide-area broadband measurement, such as... Figure 1 The flowchart shown is a method for predicting broadband oscillation risk based on wide-area broadband measurement. The processing flow of this method may include the following steps: S1, on the photovoltaic power station side, uses deployed PMU devices to collect high-frequency synchronous phasor data of the grid connection point of the photovoltaic power station in real time, and performs filtering and time-scale calibration to eliminate noise interference and ensure data time consistency. High-frequency synchronous phasor data usually includes the voltage amplitude and phase, current amplitude and phase, active power, reactive power of the grid connection point of the photovoltaic power station, and also covers the grid connection point frequency and frequency change rate, which can comprehensively capture the dynamic changes of electrical quantities at the grid connection point under power output fluctuations and load changes.

[0019] S2, based on the PMU device (i.e., time-scaled synchronous PMU device) after filtering and time-scale calibration, performs spectral correlation analysis on the wide-area multi-site PMU data to screen out candidate oscillation features, and transmits the candidate oscillation features to the oscillation trend prediction network of the dispatch master station. The wide-area multi-site PMU data is used to reflect the grid operation status of the photovoltaic power station grid connection points in different geographical areas. The candidate oscillation features are used to reflect the wide-frequency domain signal coordinated fluctuation law of the photovoltaic power station grid connection points in different geographical areas. The candidate oscillation features usually include the main frequency of the wide-frequency oscillation and the voltage fluctuation amplitude and current fluctuation amplitude corresponding to the main frequency. The wide-area multi-site PMU data includes instantaneous electrical parameter data to reflect the instantaneous changes in voltage phase and current phase at each grid connection point, and real-time power data to reflect the changes in active / reactive power exchange and instantaneous fluctuation trend of photovoltaic output at the grid connection point. The oscillation trend prediction network includes a classification network for oscillation type identification, a prediction network for predicting the evolution of oscillation amplitude and frequency, and a judgment network for risk level determination.

[0020] S3 acquires real-time operating parameters and historical oscillation sample data of the power grid and inputs them into the oscillation trend prediction network for risk assessment and trend prediction. This enables early warning and decision support for prevention and control strategies against broadband oscillations caused by photovoltaic grid connection in the power system. The real-time operating parameters of the power grid include the real-time voltage amplitude / phase value, real-time current amplitude / phase value, active / reactive power exchange value, real-time output of photovoltaic arrays, frequency deviation of grid connection points, and key parameters such as line transmission power, node voltage level, and system equivalent impedance of the wide-area power grid. These parameters can comprehensively reflect the dynamic operating status of the power grid after photovoltaic grid connection and provide real-time operating condition data for the oscillation trend prediction network.

[0021] Taking a 200MW photovoltaic power plant grid-connected scenario as an example, this method is applied to predict broadband oscillation risk: First, PMU devices are deployed at the grid connection point of the photovoltaic power station and three surrounding related stations to collect high-frequency synchronous phasor data such as voltage amplitude / phase, current amplitude / phase, and active power in real time. After filtering, signal noise caused by sandstorms is eliminated, and time deviation of data from each station is controlled within 5ms through time-scale calibration to ensure data consistency.

[0022] Next, based on the calibrated data, a spectrum correlation analysis was performed, and the similarity of the frequency change trend of the wideband signal of each station was calculated to be 0.85, and the amplitude peak synchronization was 0.92. Combining the power grid geographical topology and historical oscillation data, candidate oscillation characteristics with a main frequency of 120Hz and a voltage fluctuation amplitude of 0.03pu were selected and transmitted to the dispatch master station prediction network.

[0023] Finally, by inputting the real-time output of the photovoltaic array (150MW) and the frequency deviation at the grid connection point (0.02Hz), the network was predicted to be at risk of supersynchronous oscillation. The trend that the amplitude would rise to 0.05pu in the next 10 minutes was output, and the network was classified as a level two risk. The network was then given a prevention and control suggestion to reduce the photovoltaic output by 10% and to put a static var generator (SVG) into operation for reactive power compensation, which effectively prevented the oscillation from worsening.

[0024] The aforementioned data (such as frequency change trend similarity of 0.85, time scale deviation of 5ms, and risk prediction duration of 10min) are derived from statistical analysis of actual photovoltaic power plant grid-connected operation parameters, grid topology characteristics, and historical oscillation cases, ensuring close alignment with actual engineering conditions. In high-altitude, windy photovoltaic power plant scenarios, the PMU data acquisition frequency (e.g., increasing from 250Hz to 500Hz) and filtering algorithm threshold can be fine-tuned based on the characteristics of sudden changes in photovoltaic module output caused by strong winds and the interference parameters of high altitude on signal transmission. Simultaneously, the reactive power compensation response speed of the SVG can be improved, further enhancing the adaptability of broadband oscillation risk prediction and control.

[0025] Further filtering and time-scale calibration are performed, specifically including: performing Fourier transforms on the voltage and current signals collected from the initial high-frequency synchronous phasor data that reflect the output fluctuations and load changes of the photovoltaic power plant, and decomposing them to obtain the amplitude and phase information of each frequency component; acquiring the target frequency band signal corresponding to the current grid broadband oscillation monitoring requirements; performing interference suppression processing on the target frequency band signal to retain the effective dynamic signal in the target frequency band signal that reflects the dynamic correlation characteristics of grid-connected voltage, current, and power in the broadband domain caused by the output fluctuations and load changes of the photovoltaic power plant; and smoothing the effective dynamic signal to reduce signal fluctuation errors, forming high-frequency synchronous phasor data that can accurately reflect the synchronous dynamic correlation of grid-connected voltage, current, and power.

[0026] The original acquisition timescales of each PMU device are obtained from the smoothed high-frequency synchronization phasor data, and the difference between the timescales and the GPS reference timescales is calculated to obtain the timescale offset of each PMU device. PMU devices whose absolute timescale offset is greater than the reference allowable offset are marked as devices to be calibrated, and the rest are marked as timescale qualified devices. Timing calibration is performed on the high-frequency synchronization phasor data acquired by the devices to be calibrated using the timescale offsets to ensure time synchronization of data from different sites. After timing calibration, if the timescale offset of the corresponding device to be calibrated is not greater than the reference allowable offset, then the corresponding device to be calibrated is... If a device is deemed calibrated as qualified, it is recorded as qualified; otherwise, the corresponding device to be calibrated is recorded as unqualified and the calibration process is re-executed. If the unqualified device is still unqualified after recalibration, a time synchronization anomaly warning is issued. This involves sending a warning message to the maintenance terminal through the scheduling system, including the device number, current offset, and duration of the anomaly, prompting maintenance personnel to check the corresponding hardware circuit to eliminate communication or equipment failures. Otherwise, the unqualified device is recorded as qualified. All qualified and time-calibrated devices are recorded as time-synchronization PMU devices.

[0027] Specifically, a Discrete Fourier Transform (DFT) is performed on the voltage and current signals to convert the time-domain signals into frequency-domain signals, obtaining the complex form (real part + imaginary part) of the signal at different frequency points. The amplitude of each frequency component is obtained by taking the roots of the squares of the real and imaginary parts. The phase of each frequency component is obtained by calculating the arctangent of the ratio of the imaginary to the real part, thus decomposing the amplitude and phase information of the entire frequency band. Based on the current requirements for monitoring broadband oscillations in the power grid (e.g., focusing on oscillations in the 20-200Hz band), the signal components corresponding to all frequency points within the target frequency band are selected from the frequency domain results obtained from the Fourier Transform, integrated to form the target frequency band signal, and irrelevant signals outside the frequency band are eliminated to focus on the core monitoring object.

[0028] To address unwanted signals such as electromagnetic interference and equipment noise within the target frequency band, adaptive filtering algorithms (such as Kalman filtering) or wavelet threshold denoising methods are employed to identify and suppress anomalous signal components unrelated to photovoltaic output fluctuations and load changes, retaining effective signals that reflect the dynamic correlation characteristics of electrical quantities at the grid connection point. The target frequency band signal after interference suppression is processed using either a moving average method or an exponential smoothing method. The moving average method replaces the original data points by calculating the mean of N adjacent data points, while the exponential smoothing method assigns higher weights to recent data to calculate the smoothing value, reducing random fluctuation errors and making the signal more closely resemble the actual dynamic change trend.

[0029] Each PMU device records the original acquisition time stamp (such as the timestamp generated by the device's local clock) when acquiring data. It then obtains the standard GPS reference time stamp through a Global Positioning System (GPS) receiver module. A positive time stamp offset indicates local time stamp lag, while a negative offset indicates lead. Based on power grid dispatching specifications (such as the "Application Protocol for Real-time Data Communication in Power Systems") and the accuracy requirements for time synchronization in broadband oscillation analysis (typically ≤10ms), and combined with the PMU device's hardware performance parameters, a reference allowable offset is determined (5ms in this case) to ensure that the data synchronization accuracy meets the requirements of subsequent correlation analysis. For devices whose time stamp offset exceeds the allowable limit, the timestamps of their acquired data are adjusted according to the offset. If the offset is +3ms, the timestamps of all data from that device are delayed by 3ms; if it is -2ms, they are advanced by 2ms, aligning the timestamps of each device's data with the GPS reference time stamp.

[0030] In this embodiment, the aforementioned frequency band range, smoothing parameters, allowable offset, and other values ​​are set based on power grid dispatch specifications (such as the "Application Protocol for Real-time Data Communication of Power Systems") and statistical analysis of historical oscillation data in a large number of photovoltaic grid-connected scenarios. This meets the basic requirements of broadband oscillation monitoring for data accuracy and synchronization. In complex scenarios such as high altitude with multiple interferences and large photovoltaic power station clusters connected to the grid, the values ​​can be fine-tuned according to the actual oscillation characteristics of the power grid, equipment accuracy, and monitoring needs to ensure that data processing is adapted to the on-site working conditions.

[0031] This example focuses on core monitoring signals through precise Fourier transform and frequency band selection; interference suppression and smoothing improve the data signal-to-noise ratio and stability; combined with GPS time-stamp calibration, it ensures time synchronization of data from multiple stations, providing a high-quality and highly consistent data foundation for subsequent spectrum correlation analysis, effectively reducing misjudgments of oscillation characteristics caused by data errors, and improving the accuracy of broadband oscillation risk prediction.

[0032] Further, spectral correlation analysis is conducted. The specific process is as follows: within a preset monitoring period, instantaneous electrical parameter data and real-time power data are compared and correlated moment by moment. The similarity of frequency change trends and the synchronization of amplitude fluctuation peaks of broadband signals from different stations within the same time window are calculated. The similarity of frequency change trends is used to quantify the similarity of frequency change patterns of broadband signals from different stations, while the synchronization of amplitude fluctuation peaks is used to determine the consistency of the time when the amplitude of broadband signals from different stations reaches its peak value. It is determined whether there is a consistent rise and fall pattern in the frequency dimension and whether there are synchronous increase and decrease characteristics in the amplitude dimension of the broadband signals from each station, which are considered as cooperative fluctuation characteristics. At the same time, factors such as the geographical topology of multiple stations and historical oscillation signal characteristics are introduced. A database of coordinated fluctuation features, including parameter propagation delay, is used to improve the accuracy and reliability of coordinated fluctuation feature determination. If the determination result shows the existence of coordinated fluctuation features and the feature matching degree with the historical broadband oscillation features in the coordinated fluctuation feature database exceeds the preset feature matching degree, then the coordinated fluctuation feature is marked as a candidate oscillation feature and pushed to the oscillation trend analysis stage. The preset feature matching degree is represented by the sum and average of the historical feature matching degrees under historical coordinated fluctuation conditions in the coordinated fluctuation feature database. If no coordinated fluctuation features exist or the feature matching degree does not exceed the preset feature matching degree, then anomaly verification of coordinated fluctuation features is performed to improve the accuracy of coordinated fluctuation feature identification and the effectiveness of anomaly investigation.

[0033] Specifically, the verification of anomalies in coordinated fluctuation characteristics involves the following steps: If at least one of the frequency difference or amplitude difference between broadband signals from different stations exceeds the corresponding allowable range in the coordinated fluctuation characteristic database, a signal difference exists. The signal from the station with the largest signal difference is selected as the abnormal signal sample to prompt correction of the wide-area multi-station PMU data acquisition process. The remaining station signals are marked as signals to be verified, prompting designated personnel to check them one by one. If the data acquisition duration deviation of broadband signals from different stations within the preset monitoring period is outside the allowable range, a signal is missing. First, the amplitude variation trend and frequency fluctuation pattern of the synchronous signals from adjacent stations in the same area are extracted. Then, combined with the line impedance and power transmission characteristics in the power grid topology, the signal value of the missing period is calculated as a reference value for signal completion to ensure... The temporal continuity and data integrity of wideband signals from multiple stations are assessed. If the frequency or amplitude difference between wideband signals from different stations is within the allowable range corresponding to the coordinated fluctuation feature database, or if the data acquisition duration deviation of wideband signals from different stations within the preset monitoring period is within the allowable range of data acquisition duration deviation, it indicates that the data acquisition of wideband signals from each station is synchronized, and the current multi-station signals are retained for coordinated fluctuation feature determination. After signal correction and completion, the coordinated fluctuation feature determination of the multi-station signals is re-performed. If coordinated fluctuation features exist, the main frequency of the wideband oscillation and the corresponding voltage / current fluctuation amplitude are finally selected to form candidate oscillation features. Otherwise, the corresponding multi-station wideband signals are marked as ineffective oscillation signals and fed back to the data acquisition stage to reacquire wideband multi-station PMU data.

[0034] Specifically, the method for obtaining frequency change trend similarity is as follows: First, extract the frequency change slope sequence of the broadband signal of each station within a unit time. Then, use the dynamic time warping algorithm to calculate the matching distance between the slope sequences of different stations. That is, in the dynamic time warping algorithm, by stretching and compressing the time axis, the frequency change slope sequences of different stations are aligned, and the sum of the differences of the corresponding data points is used to quantify the sequence similarity. Finally, the matching distance is normalized, and the result is the frequency change trend similarity. The method for obtaining amplitude fluctuation peak synchronization is as follows: First, mark the amplitude peak time of the broadband signal of each station within the preset monitoring period. Then, calculate the time difference between the corresponding amplitude peak times of different stations. Finally, take the reciprocal of the time difference, and the result is the amplitude fluctuation peak synchronization.

[0035] The Cooperative Fluctuation Feature Database is a database built before designing a broadband oscillation risk prediction method based on wide-area broadband measurement. It stores multi-dimensional broadband oscillation correlation data, including but not limited to the geographical topology of multiple photovoltaic power plants (such as plant spacing and connecting line impedance) and the baseline characteristics of broadband signals from multiple plants (such as amplitude fluctuation range). For example, the corresponding allowable ranges include the allowable range of frequency difference (e.g., ±0.5Hz) and the allowable range of amplitude difference (e.g., ±0.02pu). These ranges are determined based on the statistical values ​​of frequency / amplitude differences under historical cooperative fluctuation conditions in the Cooperative Fluctuation Feature Database, superimposed with 1.5 times the standard deviation, ensuring that normal fluctuations do not trigger anomaly detection.

[0036] Some of these data are directly set by technicians in conjunction with power grid dispatching specifications and PMU device performance, while others are obtained through statistical analysis of historical oscillation processes (such as calculating the mean of characteristic values ​​under heterogeneous scenarios). These data can be fine-tuned in the future according to actual monitoring needs and changes in power grid operating conditions, providing reliable data support for the determination of coordinated fluctuation characteristics.

[0037] Calculate the instantaneous frequency difference of broadband signals from different stations within the same time window. The instantaneous frequency difference represents the absolute value of the difference between the frequencies of station A and station B, and the amplitude difference represents the absolute value of the difference between the amplitudes of station A and station B. The maximum difference within the monitoring period is used as the criterion. For stations corresponding to abnormal signal samples, check whether the sampling frequency of the PMU device meets the standard and whether the signal transmission link is interfered with. Recalibrate the device clock synchronization, replace the faulty sensor, and re-acquire data after correction to eliminate the deviation. The data acquisition duration deviation represents the difference between the actual data acquisition duration and the reference data acquisition duration of a station within a preset monitoring period. The reference data acquisition duration is represented by the summation and averaging of historical data acquisition durations under historical coordinated fluctuation conditions in the coordinated fluctuation characteristic database. The allowable range of the data acquisition duration deviation represents the closed interval corresponding to the maximum and minimum values ​​of the historical data acquisition durations under each historical coordinated fluctuation condition in the coordinated fluctuation characteristic database.

[0038] In this embodiment, by comparing electrical parameters and power data time-by-time, and combining frequency similarity and amplitude synchronization quantitative analysis, the coordinated fluctuation characteristics of signals from multiple stations are accurately captured, avoiding misjudgment of data from a single station. A database containing geographical topology and historical oscillation characteristics is introduced, and a preset matching degree calculated by the mean is used to improve the accuracy and reliability of feature determination. The anomaly verification stage corrects and completes the signal differences and missing data respectively. Finally, after a second judgment, valid candidate oscillation features are screened or feedback is provided for re-sampling, providing high-quality data support for subsequent oscillation trend analysis, reducing the risk of broadband oscillation misjudgment due to data problems, and improving prediction accuracy.

[0039] Furthermore, risk assessment and trend prediction are conducted, specifically including: First, the classification network performs oscillation type identification, matching historical samples with real-time power grid parameter fluctuation characteristics to output valid oscillation types; second, the prediction network combines the output valid oscillation types to determine model parameters suitable for the operating conditions, performing oscillation amplitude and frequency evolution prediction and outputting amplitude and frequency change trends; third, the judgment network extracts amplitude and frequency risk coefficients based on the predicted amplitude and frequency change trends, couples them to classify risk levels, and generates corresponding prevention and control recommendations. Specifically, the classification network inputs the fluctuation feature vector of real-time power grid operating parameters and outputs valid oscillation types verified for effectiveness; the prediction network inputs valid oscillation types and outputs amplitude and frequency change trends; and the judgment network inputs amplitude and frequency change trends and outputs risk levels and prevention and control recommendations.

[0040] Further understanding is needed, such as Figure 2 The flowchart shown illustrates the first step of the risk assessment and trend prediction process. Its design logic is as follows: First, the classification network identifies the oscillation type by matching the input real-time power grid parameter fluctuation feature vector with historical oscillation sample features, outputting candidate oscillation types. Next, validity verification is performed. Abnormal types that do not match the current operating conditions are first eliminated, then the similarity value between the candidate types after anomaly elimination and historical samples is calculated, retaining types with similarity higher than the reference value. Finally, valid candidate oscillation types are obtained, ensuring the accuracy and reliability of the identification results and laying the foundation for subsequent oscillation trend prediction and risk assessment.

[0041] The first step, specifically, involves inputting the fluctuation feature vector from the real-time operating parameters of the power grid into a classification network. This network uses a built-in recognizer to match the network with features from historical oscillation samples, outputting the broadband oscillation type (such as subsynchronous, supersynchronous, and high-frequency oscillations) with the highest feature matching degree as candidate oscillation types. The fluctuation feature vector reflects the fluctuation pattern of the broadband signal. The effectiveness of the candidate oscillation types is then verified, including feature similarity filtering and outlier removal. The specific steps are as follows: Based on historical power grid operation simulation results, candidate oscillation types that do not match the current power grid topology and photovoltaic output conditions are identified and removed as abnormal oscillation types. The similarity value between the removed candidate oscillation types and historical sample features is calculated. Candidate oscillation types with similarity values ​​higher than the reference similarity value are retained and designated as valid candidate oscillation types. The reference similarity value is represented by the sum and average of historical similarity values ​​obtained during the historical candidate oscillation type selection process. The similarity value is obtained by: acquiring the main frequency features of the candidate oscillation type and the corresponding features of the historical oscillation samples; calculating the matching degree of their corresponding feature vectors using the cosine similarity algorithm; normalizing the calculation results to obtain a quantized value in the 0-1 range, which is the similarity value used to quantify the matching degree between the candidate oscillation type and the historical oscillation sample features.

[0042] like Figure 3 The flowchart shown illustrates the second step of the risk assessment and trend prediction process. Its design logic is as follows: First, obtain the matching degree assessment result between the effective candidate oscillation type and the load trend, and compare it with the reference value. If the matching degree assessment value is not less than the reference value, directly call the preset prediction model to execute the preset prediction process; if it is less, calculate the matching degree deviation value, use it as the key to query the prediction parameter adjustment set, obtain the time series window length correction value and feature weight coefficient compensation value, update the preset prediction model parameters, and then call the updated model. Finally, output the amplitude frequency change trend, realizing flexible adjustment of prediction parameters according to the matching situation, thereby improving prediction accuracy.

[0043] The second step, specifically, involves obtaining the matching degree evaluation results between the valid candidate oscillation types and the power output load trend. If the matching degree evaluation results show that the obtained matching degree evaluation value is not less than the reference matching degree evaluation value, the preset prediction model is directly called to execute the preset prediction process. The matching degree evaluation value is used to reflect the degree of correlation between the current oscillation type and the power grid operating conditions. The matching degree evaluation results typically include the matching degree evaluation value, frequency adaptation, period matching degree, and the weight coefficients corresponding to the frequency adaptation and period matching degree, which can fully present the calculation basis and final quantitative results of the correlation between the oscillation type and the power grid operating conditions. Otherwise, the numerical results corresponding to the reference matching degree evaluation value and the obtained matching degree evaluation value are used as the matching degree deviation value. The matching degree deviation value is used as the query key value. Based on the queried prediction parameter adjustment set, the preset prediction model is updated with parameters. Then, the preset prediction model with updated parameters is called to output the amplitude and frequency change trends. The reference matching degree evaluation value is represented by the sum and average of the historical matching degree evaluation values ​​in the historical power output load trend prediction process. The prediction parameter adjustment set includes the time series window length correction value and the feature weight coefficient compensation value.

[0044] like Figure 4 The flowchart illustrating the third step of the risk assessment and trend prediction process is designed as follows: First, a judgment triggers the classification of network risk levels. Next, the amplitude risk coefficient and frequency risk coefficient are obtained. Then, these two coefficients are proportionally amplified and coupled through weighting to calculate a comprehensive risk assessment index. Based on the range of this index, it is determined whether it belongs to the first, second, or third risk assessment range, thereby determining the corresponding risk level. Finally, targeted prevention and control recommendations are generated based on different risk levels, providing decision support for the safe and stable operation of the power grid.

[0045] The third step, specifically, involves obtaining an amplitude risk coefficient from the amplitude-frequency variation trend. This coefficient reflects the deviation of the peak oscillation amplitude from the allowable range set by pre-defined personnel. A frequency risk coefficient reflects the deviation of the oscillation frequency from the allowable range set by pre-defined personnel. The amplitude risk coefficient is represented by the normalized result of the ratio of the difference between the peak oscillation amplitude and the maximum allowable range of the peak oscillation amplitude. The maximum allowable range of the peak oscillation amplitude is represented by the average of the sums of the maximum values ​​of historical peak oscillation amplitudes in each historical amplitude-frequency variation trend. The frequency risk coefficient is represented by the normalized result of the difference between the oscillation frequency and the maximum allowable range of the oscillation frequency. The maximum allowable range of the oscillation frequency is represented by the average of the sums of the maximum values ​​of historical oscillation frequencies in each historical amplitude-frequency variation trend. After proportionally amplifying the amplitude risk coefficient and the frequency risk coefficient, they are weighted and coupled with their assigned weights to obtain a risk assessment index used to quantitatively characterize the risk level of broadband oscillations.

[0046] The proportional amplification process refers to setting fixed amplification factors (such as 1.2 and 1.5) for the amplitude risk coefficient and frequency risk coefficient respectively, based on the sensitivity requirements of broadband oscillation risk assessment for amplitude and frequency risks. The normalized coefficient (0-1 range) is then multiplied by the corresponding amplification factor to make the coefficient value more closely match the actual risk impact. This avoids the risk contribution being underestimated due to the original coefficient value being too small, and provides a more accurate quantitative basis for subsequent weighting and coupling.

[0047] If the risk assessment index is within the first risk determination interval, the risk level is determined to be at the first risk level, and emergency prevention and control recommendations are generated, such as immediately disconnecting some photovoltaic units to reduce grid-connected power. If the risk assessment index is within the second risk determination interval, the risk level is determined to be at the second risk level, and routine prevention and control recommendations are generated, such as adjusting the reactive power output of SVG to stabilize the grid-connected voltage. If the risk assessment index is within the third risk determination interval, the risk level is determined to be at the third risk level, and preventive prevention and control recommendations are generated, such as formulating a photovoltaic power output tiered adjustment plan and regularly verifying the time synchronization accuracy of PMU devices. The first risk determination interval (usually 0.7-1.0), the second risk determination interval (usually 0.3-0.7), and the third risk determination interval (usually 0-0.3) represent mutually exclusive intervals divided from high to low based on the risk assessment index. The first risk level, the second risk level, and the third risk level correspond one-to-one with the above three risk determination intervals, and the risk level decreases sequentially.

[0048] The first, second, and third risk assessment intervals are set by the power grid safety operation procedures and historical broadband oscillation accident statistics. They can be fine-tuned according to the dynamic changes in the scale of photovoltaic grid connection and the power grid topology scenario. For example, in a high-proportion photovoltaic grid connection scenario, the lower limit of the first interval can be adjusted to 0.65 and the upper and lower limits of the second interval can be adjusted to 0.25-0.65 to adapt to changes in the risk characteristics of the operating conditions.

[0049] Specifically, the matching degree evaluation value is obtained through the following methods: A feature dataset of effective candidate oscillation types is constructed, including the main frequency range and amplitude fluctuation period of the effective candidate oscillation signals; a feature dataset of real-time photovoltaic power output and load fluctuation trends is constructed, including the photovoltaic power output fluctuation range and the power output-load change period; the intersection of the main frequency range and the photovoltaic power output fluctuation range is performed, and the frequency fit is obtained based on the ratio of the intersection length to the average of the total lengths of the two ranges; the absolute difference between the amplitude fluctuation period and the power output-load change period is calculated to obtain the period deviation rate, and the result of the complement of the period deviation rate is used as the period fit; firstly, the weight coefficients corresponding to the frequency fit and period fit are extracted from the coordinated fluctuation feature database, and then the two indicators are multiplied by the matching weight coefficients to complete the weighting operation. The results of the two weighting operations are summed, and finally, the summation result is normalized to a value range of 0-1 through data processing. This process yields a quantifiable matching degree evaluation value.

[0050] The pre-defined prediction model is usually a time series prediction model, using a Long Short-Term Memory (LSTM) network. Taking the Memory model as an example, its preset prediction process is as follows: First, the candidate oscillation type features (such as main frequency 120Hz, amplitude 0.08pu) and real-time photovoltaic power output load data (such as the power output fluctuation range [80-100MW] in a certain period) are input into the model. Relying on the model's ability to capture sequence dependence, the amplitude change trend of broadband oscillation in the next 10-30 minutes (such as gradually increasing from 0.08pu to 0.12pu) and frequency evolution trend (such as slightly fluctuating from 120Hz to 125Hz) are output. If the model needs to be adjusted, the time series window length correction value in the prediction parameter adjustment set is extracted first (such as the original window 10s, correction value +2s, updated to 12s) to adapt to the fluctuation features of a longer time dimension. Then, the feature weight coefficient compensation value (such as main frequency weight compensation +0.1) is added to the original feature weight of the model (main frequency 0.6, amplitude 0.4) to obtain the updated weight (main frequency 0.7, amplitude 0.4), thus completing the parameter adjustment and making the model more in line with the current power grid conditions.

[0051] In a specific embodiment, frequency fit is obtained as follows: The main frequency range of the valid candidate oscillation signal (e.g., [110Hz, 130Hz]) and the real-time photovoltaic power output fluctuation range (e.g., [105Hz, 125Hz]) are determined; the intersection of the two ranges (e.g., [110Hz, 125Hz]) is calculated, yielding an intersection length of 15Hz; the mean of the total length of the two ranges is calculated as (20Hz + 20Hz) / 2 = 20Hz; the intersection length is divided by the mean, 15Hz / 20Hz = 0.75, yielding a frequency fit of 0.75. Period deviation rate is obtained as follows: The amplitude fluctuation period of the valid candidate oscillation signal (e.g., 8s) and the power output load change period (e.g., 10s) are extracted; the absolute difference between the two periods is calculated as |8s - 10s| = 2s; the absolute difference is divided by the power output load change period, 2s / 10s = 0.2, yielding a period deviation rate of 0.2.

[0052] The corresponding weighting coefficients include frequency adaptability weighting coefficients and period consistency weighting coefficients, which are mainly obtained in two ways: First, based on the statistical analysis of historical oscillation cases in the collaborative fluctuation characteristic database, the contribution of frequency adaptability and period consistency to the oscillation risk judgment is calculated. That is, if the correlation between frequency adaptability and oscillation occurrence in historical data reaches 0.8 and the correlation between period consistency reaches 0.6, then the initial weighting coefficients are set to 0.6 and 0.4, respectively. Second, combined with grid operation and maintenance experience, technical personnel fine-tune the weighting coefficients according to the actual scenario. For example, in the case of a high proportion of photovoltaic grid connection, since the output fluctuation has a greater impact on the frequency, the frequency adaptability weighting coefficient can be increased to 0.7 and the period consistency weighting coefficient can be decreased to 0.3. Finally, the weights are determined by normalization to ensure that the weighting results are consistent with the actual operating conditions.

[0053] In this embodiment, an oscillation type identification mechanism based on classification network matching and validity verification, combined with cosine similarity and operating condition adaptability screening, accurately eliminates abnormal types, laying a reliable foundation for subsequent predictions. The prediction process dynamically adjusts model parameters based on the matching degree, optimizing the amplitude and frequency trends of the prediction model output through time series windows and feature weights, making them more closely aligned with the current photovoltaic output and load conditions, thus improving prediction accuracy. Risk level classification, combined with regulations and historical data, supports dynamic fine-tuning of intervals based on grid connection scale and topology, and couples amplitude and frequency risk coefficient assessment methods to achieve more accurate risk quantification. The layered prevention and control recommendations are highly targeted, with emergency, routine, and preventative measures covering different risk scenarios, enabling rapid guidance for operation and maintenance, effectively reducing the probability of broadband oscillation accidents, and ensuring the safe and stable operation of the photovoltaic grid.

[0054] This invention provides a broadband oscillation risk prediction system based on wide-area broadband measurement, such as... Figure 5The diagram illustrates a broadband oscillation risk prediction system based on wide-area broadband measurement. This system may include: a high-frequency synchronous phasor data acquisition and processing module, a candidate oscillation feature screening module, and a risk assessment and trend prediction module. The high-frequency synchronous phasor data acquisition and processing module is used to acquire high-frequency synchronous phasor data from the grid-connected points of photovoltaic power plants in real time through deployed PMU devices, and perform filtering and time-scale calibration to eliminate noise interference. The candidate oscillation feature screening module is used to perform spectral correlation analysis on the wide-area multi-station PMU data acquired by the filtered and time-scaled PMU devices to screen out candidate oscillation features, and transmit these features to the oscillation trend prediction network of the dispatch master station. The risk assessment and trend prediction module is used to acquire real-time operating parameters of the power grid and historical oscillation sample data, and input them into the oscillation trend prediction network for risk assessment and trend prediction, thereby achieving early warning of broadband oscillations caused by photovoltaic grid connection in the power system.

[0055] In this embodiment, the three modules work together: first, the high-frequency synchronous phasor data acquisition and processing module acquires high-quality, highly synchronized basic data; then, the candidate oscillation feature screening module accurately extracts core features and eliminates invalid interference; finally, the risk assessment and trend prediction module combines real-time and historical data to complete the early warning, forming a complete closed loop of data processing, feature screening, and risk early warning, which efficiently realizes the prediction of broadband oscillation risks of photovoltaic grid connection and ensures grid stability.

[0056] It should be added that, such as Figure 6 One of the homepage interfaces of the power system operation and maintenance platform shown, such as Figure 7 The second homepage of the power system operation and maintenance platform interface is shown. This interface is an interactive window for a broadband oscillation risk prediction system based on wide-area broadband measurement. The system relies on broadband data collected by wide-area measurement units to present data and perform functional operations on the platform interface. As the platform's homepage, users can view the selected photovoltaic power plants, the status of PMU devices (such as synchronization status and operating status), the number of connected power plants in the dispatch master station monitoring area, the number of normal PMU devices, the number of warnings, and trend curves for the total active power and total voltage distortion rate of the entire network. This helps operation and maintenance personnel understand the overall system operation and oscillation-related basic data, providing data support for broadband oscillation risk prediction.

[0057] like Figure 8The power system operation and maintenance platform's oscillation risk assessment interface, as shown, focuses on data and analysis related to broadband oscillation risks. The interface displays oscillation frequency monitoring curves, allowing observation of frequency trends over time; it also shows key indicators such as oscillation amplitude, providing a clear view of the numerical fluctuations of oscillation-related parameters. Furthermore, it includes various statistical and analytical modules for risk assessment, such as historical oscillation event statistics and risk probability assessment results, helping operation and maintenance personnel to deeply analyze the degree of broadband oscillation risk faced by the system. This provides accurate assessment basis for taking preventative measures in advance and ensuring the stable operation of the power system.

[0058] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0059] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0060] In various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0061] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0062] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting broadband oscillation risk based on wide-area broadband measurement, characterized in that, The method includes: S1. High-frequency synchronous phasor data of the grid-connected points of the photovoltaic power station are collected in real time through the deployed PMU device, and filtered and time-scaled to eliminate noise interference. The high-frequency synchronous phasor data is used to reflect the synchronous dynamic impact of photovoltaic power station output fluctuations and load changes on the voltage, current and power of each grid-connected point. S2, based on the PMU device after filtering and time-scale calibration, the wide-area multi-site PMU data is subjected to spectrum correlation analysis to screen out candidate oscillation features, and the candidate oscillation features are transmitted to the oscillation trend prediction network of the dispatch master station. The wide-area multi-site PMU data is used to reflect the grid operation status of the photovoltaic power station grid connection point in different geographical areas, and the candidate oscillation features are used to reflect the wide-frequency domain signal cooperative fluctuation law of the photovoltaic power station grid connection point in different geographical areas. S3 acquires real-time operating parameters of the power grid and historical oscillation sample data, and inputs them into the oscillation trend prediction network for risk assessment and trend prediction, so as to realize the early warning of broadband oscillations caused by photovoltaic grid connection in the power system.

2. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 1, characterized in that, The filtering and time-scale calibration process specifically includes: Fourier transform is performed on the voltage and current signals that reflect the output fluctuations and load changes of the photovoltaic power station, which are collected from the initial high-frequency synchronous phasor data, to decompose and obtain the amplitude and phase information of each frequency component. Obtain the target frequency band signal corresponding to the current power grid broadband oscillation monitoring requirements; Interference suppression processing is performed on the target frequency band signal to retain the effective dynamic signal in the target frequency band signal that reflects the dynamic correlation characteristics of grid connection point voltage, current and power in the wide frequency domain caused by power output fluctuations and load changes of photovoltaic power plants; The effective dynamic signal is smoothed to reduce signal fluctuation errors and form high-frequency synchronous phasor data that reflects the synchronous dynamic correlation of voltage, current and power at the grid connection point.

3. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 2, characterized in that, The filtering and time-scale calibration process also includes: The original acquisition time scale of each PMU device is obtained by acquiring the smoothed high-frequency synchronization phasor data and calculating the difference with the GPS reference time scale to obtain the time scale offset of each PMU device. PMU devices whose absolute value of time offset is greater than the reference allowable offset are marked as devices to be calibrated, and the rest are marked as devices with qualified time offset. Timing calibration is performed on the high-frequency synchronous phasor data collected by the device to be calibrated by using time scale offset to ensure time synchronization of data from different sites. After timing calibration, if the time offset of the corresponding device to be calibrated is not greater than the reference allowable offset, the corresponding device to be calibrated is recorded as a calibrated qualified device; otherwise, the corresponding device to be calibrated is recorded as a calibrated unqualified device, and the calibration process is re-executed. If the corresponding calibrated device is still unqualified after recalibration, a time synchronization anomaly warning will be issued; otherwise, the corresponding calibrated device will be recorded as a calibrated device. All calibrated and time-calibrated devices are designated as time-calibrated synchronization PMU devices.

4. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 1, characterized in that, The candidate oscillation characteristics include the main frequency of the broadband oscillation and the voltage fluctuation amplitude and current fluctuation amplitude corresponding to the main frequency; The wide-area multi-site PMU data includes instantaneous electrical parameter data reflecting the instantaneous changes in voltage phase and current phase at each grid connection point, as well as real-time power data reflecting the changes in active / reactive power exchange at the grid connection point and the instantaneous fluctuation trend of photovoltaic output. The oscillation trend prediction network includes a classification network for oscillation type identification, a prediction network for predicting the evolution of oscillation amplitude and frequency, and a judgment network for determining risk level. The specific process for performing spectral correlation analysis is as follows: Within a preset monitoring period, instantaneous electrical parameter data and real-time power data are correlated and compared moment by moment to calculate the similarity of frequency change trends and the synchronization of amplitude fluctuation peaks of broadband signals from different stations within the same time window. The frequency change trend similarity is used to quantify the similarity of the frequency change patterns of broadband signals from different stations, and the amplitude fluctuation peak synchronization is used to measure the consistency of the time when the amplitude of broadband signals from different stations reaches the peak value. To determine whether there is a consistent rise and fall pattern in the frequency dimension and whether there is synchronous increase and decrease in the amplitude dimension of the wideband signals of each station, and to introduce a cooperative fluctuation feature database to improve the accuracy and reliability of cooperative fluctuation feature determination. If the judgment result shows that there is a coordinated fluctuation feature, and the feature matching degree with the historical broadband oscillation feature in the coordinated fluctuation feature database exceeds the preset feature matching degree, then the coordinated fluctuation feature is marked as a candidate oscillation feature and pushed to the oscillation trend analysis stage. If no coordinated fluctuation characteristics exist or the characteristic matching degree does not exceed the preset characteristic matching degree, then anomaly verification of coordinated fluctuation characteristics is performed to improve the accuracy of coordinated fluctuation characteristic identification and the effectiveness of anomaly investigation.

5. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 4, characterized in that, The verification of anomalies in the coordinated fluctuation characteristics specifically involves: If at least one of the frequency difference or amplitude difference of the wideband signals from different stations exceeds the allowable range corresponding to the cooperative fluctuation characteristic database, the signal from the station with the largest signal difference is taken as the abnormal signal sample to prompt correction of the wide-area multi-station PMU data acquisition process. If the data acquisition duration deviation of broadband signals from different stations within the preset monitoring period is not within the allowable range of data acquisition duration deviation, then a signal completion reference value is generated by combining the characteristics of broadband signals from adjacent stations in the same area and the missing signals corresponding to the power grid topology coupling relationship to ensure the temporal continuity and data integrity of broadband signals from multiple stations in the wide area. If the frequency difference or amplitude difference of the broadband signals from different stations are within the allowable range corresponding to the cooperative fluctuation feature database, or if the data acquisition duration deviation of the broadband signals from different stations within the preset monitoring period is within the allowable range of data acquisition duration deviation, it indicates that the broadband signal data acquisition of each station is synchronized, and the current multi-station signals are retained for cooperative fluctuation feature determination. After completing signal correction and completion, the coordinated fluctuation characteristics of the signals from multiple stations are re-evaluated. If coordinated fluctuation characteristics exist, the main frequency of the broadband oscillation and the corresponding voltage / current fluctuation amplitude are selected to form candidate oscillation characteristics. Otherwise, the corresponding multi-site wideband signal is marked as an ineffective oscillation signal and fed back to the data acquisition stage to reacquire wide-area multi-site PMU data.

6. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 1, characterized in that, The aforementioned risk assessment and trend forecasting specifically include: The first step is to classify the network to identify the oscillation type by matching historical samples with the real-time fluctuation characteristics of the power grid parameters and output the valid oscillation type. The second step is to use the prediction network to determine the model parameters for the appropriate operating conditions by combining the effective oscillation type of the output, so as to perform oscillation amplitude and frequency evolution prediction and output the amplitude and frequency change trend. The third step is to determine the amplitude and frequency risk coefficients based on the predicted amplitude and frequency change trends of the network, couple them together to classify the risk levels, and generate corresponding prevention and control suggestions. The first step is specifically as follows: The fluctuation feature vector in the real-time operating parameters of the power grid is input into the classification network. The built-in recognizer is matched with the features of historical oscillation samples, and the wideband oscillation type with the highest feature matching degree is output as the candidate oscillation type. The fluctuation feature vector is used to reflect the fluctuation pattern of the wideband signal. The validity of candidate oscillation types is verified, including feature similarity filtering and outlier removal. The specific steps are as follows: Based on historical power grid operation simulation results, candidate oscillation types that do not match the current power grid topology and photovoltaic power output conditions are identified and removed as abnormal oscillation types. Calculate the similarity value between the candidate oscillation types after removing abnormal oscillation types and the features of historical samples. Retain the candidate oscillation types with similarity values ​​higher than the reference similarity value and record them as valid candidate oscillation types. The similarity value is obtained through the following process: obtaining the main frequency features of the candidate oscillation type and the features corresponding to the historical oscillation samples; calculating the matching degree of the corresponding feature vectors of the two through the cosine similarity algorithm; normalizing the calculation results to obtain the similarity value used to quantify the matching degree of the candidate oscillation type and the features of the historical oscillation samples.

7. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 6, characterized in that, The second step is as follows: The matching degree evaluation result of the effective candidate oscillation type and the output load trend is obtained. If the matching degree evaluation result shows that the obtained matching degree evaluation value is not less than the reference matching degree evaluation value, the preset prediction model is directly called to execute the preset prediction process. The matching degree evaluation value is used to reflect the correlation between the current oscillation type and the power grid operating condition. Conversely, the numerical result corresponding to the reference matching degree evaluation value and the obtained matching degree evaluation value is used as the matching degree deviation value. The matching degree deviation value is used as the query key value. Based on the queried prediction parameter adjustment set, the parameters of the preset prediction model are updated, and then the preset prediction model with updated parameters is called to output the amplitude frequency change trend. The set of prediction parameter adjustments includes time series window length correction values ​​and feature weight coefficient compensation values.

8. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 7, characterized in that, The matching degree evaluation value is obtained through the following method: Construct a feature dataset of valid candidate oscillation types, including the main frequency range and amplitude fluctuation period of valid candidate oscillation signals; Construct a feature dataset of real-time photovoltaic power output and load fluctuation trends, including photovoltaic power output fluctuation range and power output load change cycle; The frequency fit is obtained by intersecting the main frequency range with the photovoltaic power output fluctuation range. The absolute difference between the amplitude fluctuation period and the output load change period is calculated to obtain the period deviation rate, and the result of the complement of the period deviation rate is used as the period matching degree. The weight coefficients of frequency fit and period fit are retrieved from the cooperative fluctuation feature database. The frequency fit and period fit are weighted by their respective weight coefficients and then summed. The summation result is then normalized to obtain the matching degree evaluation value.

9. The broadband oscillation risk prediction method based on wide-area broadband measurement as described in claim 7, characterized in that, The third step is as follows: Obtain the amplitude risk coefficient, which reflects the deviation of the peak oscillation amplitude from the allowable range, and the frequency risk coefficient, which reflects the deviation of the oscillation frequency from the allowable range, from the amplitude and frequency variation trends. After proportionally amplifying the amplitude risk coefficient and the frequency risk coefficient, they are weighted and coupled with the corresponding assigned weights to obtain a risk assessment index for quantitatively characterizing the risk level of broadband oscillations. If the risk assessment index is within the first risk determination range, it means that the risk level determination result is the first risk level; If the risk assessment index is within the second risk determination range, it means that the risk level determination result is the second risk level. If the risk assessment index is within the third risk determination range, it means that the risk level determination result is the third risk level; The first risk assessment interval, the second risk assessment interval, and the third risk assessment interval represent mutually exclusive intervals divided from high to low based on the risk assessment index. The first risk level, the second risk level, and the third risk level correspond one-to-one with the above three risk assessment intervals, and the risk level decreases sequentially.

10. A broadband oscillation risk prediction system based on wide-area broadband measurement, employing any one of the broadband oscillation risk prediction methods based on wide-area broadband measurement as described in claims 1-9, comprising: High-frequency synchronous phasor data acquisition and processing module, candidate oscillation feature screening module, risk assessment and trend prediction module; The high-frequency synchronous phasor data acquisition and processing module is used to acquire high-frequency synchronous phasor data of the grid connection point of the photovoltaic power station in real time through the deployed PMU device, and perform filtering and time-scale calibration to eliminate noise interference. The candidate oscillation feature screening module is used to perform spectral correlation analysis on the wide-area multi-site PMU data obtained by the PMU device after filtering and time-scale calibration to screen out candidate oscillation features and transmit the candidate oscillation features to the oscillation trend prediction network of the dispatch master station. The risk assessment and trend prediction module is used to acquire real-time operating parameters of the power grid and historical oscillation sample data, and input them into the oscillation trend prediction network for risk assessment and trend prediction, so as to realize the early warning of broadband oscillations caused by photovoltaic grid connection in the power system.

Images