Edge computing supports power electronics and methods for monitoring and suppressing grid oscillations.

By constructing a three-level edge computing architecture, real-time local monitoring and rapid suppression of oscillation signals in the power electronic grid are realized, solving the problems of insufficient response time and lack of collaborative upgrades in existing technologies, and achieving rapid response and globally coordinated oscillation suppression effects.

CN123121576APending Publication Date: 2026-11-17KAIFENG POWER SUPPLY COMPANY STATE GRID HENAN ELECTRIC POWER
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202611468432.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, in the oscillation monitoring and suppression schemes of power electronic power grids, the edge side only undertakes the functions of data acquisition and analysis, while the oscillation suppression decision-making and execution still rely on centralized processing in the cloud, resulting in insufficient response timeliness; the edge-side local suppression scheme lacks a cloud-based collaborative upgrade mechanism, and cannot obtain wide-area coordinated support in a timely manner when oscillations occur in multiple regions; the oscillation monitoring and suppression link lacks a complete closed-loop mechanism.

Method used

A three-tiered edge computing architecture is constructed, comprising an edge sensing layer, an edge computing layer, and a cloud collaboration layer, to achieve real-time local monitoring and rapid suppression of oscillation signals. The edge sensing layer collects and preprocesses electrical signals; the edge computing layer performs multimodal synchronous phasor calculation and adaptive threshold filtering; the edge computing layer performs local adaptive oscillation suppression; and the cloud collaboration layer performs wide-area situational awareness and collaborative optimization.

Benefits of technology

It achieves rapid response in oscillation monitoring and suppression, reduces the burden on communication networks, shortens response time, improves the efficiency of oscillation suppression, balances the needs of rapid response and global coordination, and reduces the consumption of cloud computing resources and communication bandwidth.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121576A_ABST
    Figure CN123121576A_ABST
Patent Text Reader

Abstract

This invention discloses a method for monitoring and suppressing oscillations in a power electronic grid supported by edge computing, relating to the field of power system stability control technology. The method constructs a three-level collaborative architecture consisting of an end-side sensing layer, an edge computing layer, and a cloud-based collaborative layer. Electrical quantities are collected and preprocessed at the end-side before being uploaded. The edge computing layer obtains the time-frequency matrix through full-wave FFT and sliding-window FFT, filters oscillation frequency points based on adaptive thresholds, calculates the damping ratio, and clusters to determine oscillation sources. Based on the damping ratio and oscillation amplitude, a graded evaluation is performed; when a level two warning or higher is reached, adaptive suppression is implemented locally at the edge. After suppression, the damping ratio recovery rate is calculated. If suppression fails, a collaborative suppression request is sent to the cloud. The cloud constructs a wide-area oscillation situation and generates collaborative control commands for execution. This invention achieves rapid completion of the edge-side monitoring and suppression closed loop and solves the collaborative suppression problem in scenarios with multiple oscillation sources operating in parallel by actively requesting cloud collaboration after suppression failure.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power system stability control technology, specifically to a method for monitoring and suppressing power grid oscillations supported by edge computing in power electronics. Background Technology

[0002] With the large-scale grid connection of renewable energy and the formation of high-voltage direct current (HVDC) transmission networks, the proportion of power electronic equipment in power systems continues to increase, and power systems are gradually showing a trend towards power electronics. The interactions between power electronic equipment and the power grid, as well as among power electronic equipment themselves, can trigger broadband oscillations in the frequency range of several hertz to several kilohertz, including subsynchronous oscillations, supersynchronous oscillations, and medium-to-high frequency oscillations. Broadband oscillations can damage power equipment and cause renewable energy units to shut down, becoming one of the important factors restricting the efficient absorption of renewable energy.

[0003] In the prior art, several technical solutions have been proposed for the monitoring and suppression of oscillations in power electronic grids. For example, Chinese patent application CN121983980A discloses a "Method for Online Monitoring and Analysis of Broadband Oscillations in New Energy and a Cloud-Edge Architecture System." This system employs a cloud-edge-device architecture for data processing. Intelligent terminals on the edge side collect data from new energy generating units and grid-connected points, while edge servers perform online analysis of broadband oscillations. Full-wave FFT and sliding-window FFT techniques are used to screen oscillation frequencies, calculate damping ratios, and perform clustering. The cloud server provides centralized display and multi-station comparative analysis. Although this solution achieves online monitoring and analysis of oscillations, the edge side only undertakes monitoring and analysis functions and does not involve closed-loop control for oscillation suppression. Oscillation suppression still relies on cloud-based decision-making and command issuance, resulting in insufficient response time.

[0004] Chinese patent application CN110571830A discloses an emergency control method for an external power system stabilizer. This method employs an edge computing processor to identify the frequency and damping ratio of oscillation modes online. The damping ratio is used to determine emergency control conditions. When the conditions are met, a drive selection control switch is activated to switch to the PSS emergency control output channel. Edge computing is then used to adjust the emergency control parameters in real time. While this scheme achieves local suppression at the edge, its suppression methods rely on the external stabilizer hardware and do not involve a collaborative suppression mechanism between multiple edge nodes.

[0005] Chinese patent application CN116960990B discloses a method for suppressing broadband oscillations in power systems. This method identifies the oscillation frequency and amplitude of broadband oscillations online based on the difference between the actual and desired output parameters of the power electronic controller. Then, it adjusts the parameters of a target oscillation suppression element in real time based on the oscillation frequency and amplitude, enabling the target oscillation suppression element to adaptively suppress broadband oscillations. This solution addresses the adaptive suppression problem under conditions of system parameter uncertainty, but it does not address multi-node collaboration and edge-cloud hierarchical collaboration mechanisms under edge computing architectures.

[0006] Chinese patent application CN121689310A discloses a wide-area collaborative control method for flexible DC power grids based on cloud-edge collaboration. It constructs a cloud-edge collaborative wide-area control system, builds the system Jacobian matrix based on the power grid topology and operating parameters, establishes state-space equations, and performs eigenvalue analysis and impedance scanning. The subsynchronous oscillation risk is integrated into a wide-area resonance risk quantification factor, forming a dual-drive decision-making mechanism. The decision-making process in this scheme employs a cloud-based active coordination architecture, with the edge side acting only as a data acquisition and execution terminal. The passive response mechanism that allows the edge side to actively request cloud upgrades after local suppression failure is not disclosed.

[0007] Chinese patent application CN121192925A discloses a cloud-edge collaborative power equipment control method. This method receives initial resonant waveform data from multiple inverters, determines the dynamic correlation strength of resonant frequencies between inverters, generates collaborative control parameters, and an edge server calculates a resonance suppression compensation control signal based on real-time resonant fluctuation data to compensate for the inverters' initial control signals. The collaborative control in this scheme is initiated proactively by the cloud; however, the mechanism by which the edge side determines whether to request a cloud upgrade based on its own suppression effect is not disclosed.

[0008] Chinese patent application CN110927442A discloses a multimodal oscillation online monitoring and early warning system based on edge computing, involving a three-level architecture of acquisition layer, edge node layer, and central layer, for online monitoring and early warning of multimodal oscillations. This scheme focuses on oscillation monitoring and early warning, but does not address closed-loop control for oscillation suppression.

[0009] The paper "Construction of a Cloud-Edge Collaborative Prevention and Control System for Broadband Oscillations in a High-Proportion New Energy Power Grid," published by Ren Chong et al. in the 4th issue of *New Power Systems* in 2025, discloses the following: A new energy monitoring terminal is established to acquire oscillation characteristic quantities and receive and execute power plant tripping commands; a regional control terminal is constructed to receive oscillation information from the monitoring terminal, locate oscillating units, and issue corresponding tripping commands; a dispatch cloud is established to monitor the grid frequency, voltage, and total tripping volume in real time, perform grid safety verification, and conduct coordinated power defense based on the verification results. This scheme focuses on tripping defense strategies and does not involve the active damping local suppression mechanism at the edge computing layer.

[0010] In summary, existing technologies have the following shortcomings: First, in oscillation monitoring solutions under a cloud-edge-device architecture, the edge side mostly only undertakes data acquisition and analysis functions, while oscillation suppression decisions and execution still rely on centralized processing in the cloud, and communication delays lead to insufficient response timeliness; Second, in edge-side on-site suppression solutions, there is a lack of a cloud-based collaborative upgrade mechanism after edge suppression failure, and when the edge node's suppression capability is insufficient or the oscillation affects multiple areas, it is impossible to obtain wide-area coordinated support in a timely manner; Third, in existing solutions, oscillation monitoring and suppression belong to different system levels, lacking a complete closed-loop mechanism from detection to suppression and then to collaborative upgrade. Summary of the Invention

[0011] The technical problem to be solved by this invention is to overcome the existing defects and provide a method for monitoring and suppressing oscillations in the power electronic grid supported by edge computing. It can realize real-time local monitoring, rapid suppression and hierarchical collaborative upgrading of oscillation signals by constructing a three-level edge computing architecture of "end-side acquisition - edge analysis and decision-making suppression - cloud collaborative optimization", which can effectively solve the problems in the background technology.

[0012] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring and suppressing grid oscillations supported by edge computing in power electronics, comprising the following steps:

[0013] S1: Construct a three-level edge computing collaborative architecture consisting of an edge perception layer, an edge computing layer, and a cloud collaboration layer.

[0014] The edge sensing layer includes data acquisition terminals deployed at the grid connection points of new energy power plants, substation busbars, and power electronic converter outputs. The edge computing layer is deployed at regional substations or new energy power plant aggregation stations, including edge computing servers and edge intelligent terminals, possessing local data storage, real-time signal processing, and oscillation decision-making capabilities. The cloud-based collaboration layer is deployed at the dispatch center for wide-area situational awareness, multi-regional coordinated optimization, and long-term trend analysis.

[0015] S2: The edge sensing layer synchronously collects electrical quantity signals of power electronic devices and performs preprocessing to generate multimodal synchronous phasor data, and uploads the multimodal synchronous phasor data to the edge computing layer.

[0016] The end-side sensing layer synchronously acquires three-phase voltage, three-phase current, active power, and reactive power signals at a sampling rate of no less than 10kHz. After analog-to-digital conversion, moving average filtering, and adaptive cancellation of power frequency components, transient oscillation components are extracted. Multi-mode synchronous phasor data containing oscillation frequency, amplitude, and phase is generated using multi-mode synchronous phasor calculation. The end-side sensing layer only uploads preprocessed feature data and waveform recording data during abnormal triggering periods, and does not upload the full original sampling data.

[0017] S3: The edge computing layer obtains the time-frequency matrix based on the multimodal synchronous phasor data by combining full-wave FFT and sliding window FFT. Based on the adaptive threshold, it filters potential oscillation frequency points from the time-frequency matrix and calculates the damping ratio of the oscillation components. It then performs clustering processing on similar oscillation frequency points to determine the oscillation frequency, oscillation amplitude, damping ratio, oscillation source location, and oscillation propagation path.

[0018] The full-wave FFT is used to obtain the global distribution of oscillation frequencies within the analysis period, and the sliding window FFT is used to capture the time-varying characteristics of the oscillation frequencies. The adaptive threshold is dynamically adjusted according to the current power electronics penetration rate, short-circuit ratio, and operating conditions of the power grid. Under high power electronics penetration conditions, the threshold is dynamically adjusted to improve the oscillation detection sensitivity.

[0019] The specific adjustment method of the adaptive threshold is as follows: A baseline threshold T0, a baseline short-circuit ratio K0, and a penetration rate influence weighting coefficient β are preset. The current actual short-circuit ratio K_SCR of the power grid, the total capacity of power electronic equipment P_PE, and the total installed capacity of the power grid P_total are obtained. The ratio of the total capacity of power electronic equipment to the total installed capacity of the power grid is used as the power electronic equipment penetration rate P_PE / P_total. The ratio of the baseline short-circuit ratio to the current actual short-circuit ratio of the power grid is multiplied by a first correction term to obtain the adaptive adjustment coefficient. The first correction term is the sum of 1 and the product of the penetration rate influence weighting coefficient and the power electronic equipment penetration rate. The baseline threshold is multiplied by the adaptive adjustment coefficient to obtain the adaptive threshold. When the power electronic equipment penetration rate increases or the short-circuit ratio decreases, the adaptive adjustment coefficient increases, and the adaptive threshold is adjusted accordingly to improve the oscillation detection sensitivity.

[0020] The baseline threshold T0 is determined as follows: Under standard power grid operating conditions, historical oscillation data is analyzed, and the amplitude distribution of the noise spectrum during normal operation is statistically analyzed. The amplitude corresponding to the 95th percentile of the noise amplitude distribution is taken as the baseline threshold to ensure that no false alarms occur under normal operating conditions. The baseline short-circuit ratio K0 is taken as the typical short-circuit ratio under standard power grid operating conditions, with a value range of 2.5 to 4.0. The penetration rate influence weighting coefficient β has a value range of 0.5 to 2.0; when the power electronic equipment in the power grid is mainly grid-connected converters, β takes a higher value of 1.5-2.0 to improve detection sensitivity; when the power electronic equipment in the power grid is mainly grid-connected converters, β takes a lower value of 0.5-1.0 to reduce the false alarm rate.

[0021] When clustering similar oscillation frequency points, the dominant oscillation frequency of each plant node in a wide area is obtained, and several oscillation frequency clusters are obtained. Within the oscillation range of the oscillation frequency clusters, the oscillation propagation path and oscillation source are determined according to the oscillation power amplitude of each electrical branch of each plant node.

[0022] S4: The edge computing layer performs a graded risk assessment based on the damping ratio and the oscillation amplitude. When the risk level reaches level two or above, the edge computing layer generates a suppression command locally and executes local adaptive oscillation suppression. At the same time, it reports the oscillation characteristics and suppression status to the cloud collaboration layer.

[0023] The graded risk assessment includes: when the damping ratio is between 0.05 and 0.1 and the oscillation amplitude is less than 5% of the rated value, it is determined to be a Level 1 concern state, continuously tracking changes in the oscillation mode and reporting at a preset cycle; when the damping ratio is between 0.02 and 0.05 or the oscillation amplitude reaches 5% to 15% of the rated value, it is determined to be a Level 2 early warning state, initiating pre-adjustment of the suppression device and sending early warning information to the cloud collaboration layer; when the damping ratio is less than 0.02 or the oscillation amplitude exceeds 15% of the rated value, it is determined to be a Level 3 emergency state, immediately triggering local suppression actions and reporting the emergency state to the cloud collaboration layer.

[0024] The damping ratio threshold in the above-mentioned classification criteria is set based on the following: through extensive simulation experiments, the system begins to exhibit a significant and continuous oscillation trend when the damping ratio is below 0.05, and the system exhibits a divergent trend when the damping ratio is below 0.02. The oscillation amplitude threshold is set based on the following: an oscillation amplitude of 5% of the rated value has little impact on the normal operation of the power electronic converter, while an oscillation amplitude of 15% of the rated value may cause the converter overcurrent protection to activate.

[0025] The local adaptive oscillation suppression includes: using the deviation between the actual output parameters and the expected output parameters of the power electronic converter as an oscillation identification signal, identifying the oscillation frequency and oscillation amplitude online, and adjusting the transfer function parameters of the active damping controller in real time; superimposing the oscillation suppression signal onto the current control signal of the grid-connected converter to generate a comprehensive control signal, and generating a PWM trigger pulse based on the comprehensive control signal and outputting it to the control terminal of the grid-connected converter.

[0026] S5: After the local adaptive oscillation suppression is executed, the edge computing layer continuously monitors the damping ratio change. If the damping ratio does not recover to above the safety threshold within a preset time or the oscillation affects multiple areas, a collaborative suppression request is sent to the cloud collaboration layer. In response to the collaborative suppression request, the cloud collaboration layer collects the oscillation characteristics and suppression status reported by multiple edge computing layers, constructs a wide-area oscillation situation, generates inter-regional collaborative control instructions, and sends them to the corresponding edge computing layers for execution.

[0027] When the edge computing layer continuously monitors the damping ratio change, if the damping ratio recovers to above the safety threshold within a first preset time after the suppression action is executed, it maintains the current suppression parameters and reports the suppression result to the cloud collaboration layer for recording; if the damping ratio does not recover to above the safety threshold within the first preset time, it sends a collaborative suppression request to the cloud collaboration layer, and the cloud collaboration layer initiates the inter-regional collaborative suppression process.

[0028] The triggering condition for the collaborative suppression request is one of the core distinguishing features of this invention. Specifically, after the suppression action is executed, the edge computing layer continuously calculates the damping ratio recovery rate over a first preset time window. The damping ratio recovery rate is defined as the ratio of the increment of the damping ratio of the k-th sampling window relative to the damping ratio of the (k-1)-th sampling window after the suppression action is executed to the sampling window time. When the damping ratio recovery rate is positive and the cumulative recovery amount reaches a safety threshold within the first preset time, the suppression is deemed effective; when the damping ratio recovery rate is less than or equal to zero or the cumulative recovery amount does not reach the safety threshold within the first preset time, the suppression is deemed to have failed, triggering the sending of a collaborative suppression request to the cloud collaborative layer. The collaborative suppression request carries the edge computing layer identifier, oscillation source location, current damping ratio, oscillation amplitude, and executed suppression parameters, enabling the cloud collaborative layer to quickly locate and respond.

[0029] The first preset time is determined based on the typical decay time constant of the oscillation mode in the power electronic grid. For subsynchronous oscillation mode, the typical decay time constant is 0.3 to 0.8 seconds, and the first preset time is 0.5 seconds; for supersynchronous oscillation mode, the typical decay time constant is 0.2 to 0.5 seconds, and the first preset time is 0.3 seconds. The value of the first preset time can be configured according to the oscillation mode characteristics of the actual power grid.

[0030] The cloud-based collaborative layer spatiotemporally correlates the oscillation frequency, amplitude, damping ratio, and oscillation source location reported by each edge computing node to construct a wide-area oscillation situation map; and constructs a system state-space equation based on power grid topology data, generating a wide-area resonance risk quantification factor through eigenvalue analysis and impedance scanning.

[0031] The wide-area resonance risk quantification factor is generated as follows: A system state-space equation is constructed based on grid topology data. This equation characterizes the dynamic behavior of the grid by relating state variables to their first derivatives. Eigenvalue analysis is performed on the state matrix of the system state-space equation to obtain eigenvalues ​​for each oscillation mode. These eigenvalues ​​include real and imaginary parts, where the real part represents damping and the imaginary part represents the oscillation frequency. Impedance scanning is used to determine the propagation weighting coefficients of each oscillation mode in the grid. The propagation weighting coefficients are determined by injecting a test signal of a specific frequency into each node of the grid, measuring the response amplitude of each node to this frequency signal, and normalizing the response amplitude as the propagation weighting coefficient for that frequency. For the i-th oscillation mode, the propagation weighting coefficient corresponding to its oscillation frequency is taken as the propagation weighting coefficient w for that mode. i For the i-th oscillation mode, the ratio of the absolute value of the real part of its eigenvalue to the magnitude of the eigenvalue is calculated as the damping ratio contribution factor, and the ratio of the oscillation amplitude of this mode to the rated amplitude is calculated as the amplitude contribution factor. The damping ratio contribution factor, the amplitude contribution factor, and the propagation weighting coefficient are multiplied together to obtain the risk quantification factor for this oscillation mode. The maximum value among all oscillation mode risk quantification factors is taken as the wide-area resonance risk quantification factor. This factor ranges from 0 to 1; a larger value indicates a higher resonance risk.

[0032] The cloud-based collaborative layer compares the wide-area resonance risk quantification factor with the preset risk warning value. When the wide-area resonance risk quantification factor exceeds the warning value, a corresponding system status flag is generated, forming a dual-drive decision-making mechanism that combines threshold-triggered rapid response with in-depth risk quantification insight.

[0033] The inter-regional collaborative control commands include coordinated adjustment strategies for damping parameters of grid-connected converters in each region, output adjustment commands for new energy power plants, and / or generator tripping commands for new energy units; the cloud-based collaborative layer also optimizes the oscillation identification model and suppression parameter library of the edge computing layer based on historical oscillation data, and sends the updated model parameters to each edge computing node.

[0034] The triggering mechanism of the collaborative suppression request and the edge-side local suppression function are mutually supportive and interactive: edge-side local suppression provides the basis for judging the collaborative suppression request—the suppression effect can only be evaluated and a trigger signal generated after local suppression is performed at the edge; the collaborative suppression request provides a safety net for edge-side local suppression—when local suppression capability is insufficient, it compensates for the limitations of edge-side computing resources and control range by requesting collaboration from the cloud. Together, they constitute a complete closed-loop control logic of "edge priority, cloud backup," jointly solving the technical problem of "how to achieve both rapid response in oscillation suppression and global coordination."

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows: This edge computing-supported power electronic grid oscillation monitoring and suppression method has the following advantages:

[0036] 1. By preprocessing data at the edge and performing on-site analysis and suppression at the edge computing layer, oscillation monitoring is transformed from "all data being sent to the cloud for centralized processing" to "real-time on-site analysis and suppression at the edge," significantly reducing the burden on the communication network. Simulation results show that the edge selective upload strategy reduces the amount of communication data by approximately 85% compared to the full upload scheme, and the edge computing layer can complete oscillation detection and implement active damping suppression within two fundamental frequency cycles.

[0037] 2. By generating suppression instructions locally at the edge computing layer and executing local adaptive oscillation suppression, the "detection-decision-suppression" closed loop is rapidly completed at the edge. Simulation results show that under the condition of 25Hz subsynchronous oscillation with a damping ratio of 0.015, the oscillation is effectively attenuated within 0.3 seconds, reducing the response time by approximately 60% compared to the traditional centralized solution.

[0038] 3. A passive response mechanism, in which the edge computing layer sends a collaborative suppression request to the cloud after edge suppression fails, and the cloud responds to the request to initiate collaboration, achieves layered collaborative control with edge priority and cloud backup. Simulation verification shows that when the oscillation affects more than two regions and the on-site suppression by the edge node fails to effectively control the oscillation, the edge computing layer sends a collaborative suppression request to the cloud after 0.5 seconds. The cloud completes wide-area situational analysis and collaborative command generation within 200ms, and the oscillation is effectively suppressed within 1.2 seconds. Compared with the existing cloud-active coordination scheme, this mechanism reduces the consumption of cloud computing resources and communication bandwidth by about 70%, while ensuring collaborative suppression capabilities in complex scenarios.

[0039] 4. Through tiered risk assessment and differentiated response strategies, the system can flexibly select between continuous tracking, on-site suppression, or cloud-based collaboration based on the oscillation risk level, balancing the needs of rapid response and global optimization. Simulation verification shows that when the damping ratio is between 0.05 and 0.1 and the oscillation amplitude is less than 5% of the rated value, the first-level attention state is used for continuous tracking without initiating suppression actions, avoiding unnecessary control intervention. When the damping ratio is between 0.02 and 0.05 or the oscillation amplitude reaches 5% to 15% of the rated value, the second-level early warning state is activated to pre-adjust the suppression device, effectively preventing further deterioration of the oscillation.

[0040] 5. By combining full-wave FFT and sliding-window FFT with time-frequency analysis and oscillation frequency clustering, a complex scenario involving multiple oscillation sources occurring simultaneously in a power electronic grid can be effectively addressed. Simulation results demonstrate that in a scenario where three oscillation sources of different frequencies occur simultaneously, the clustering method can accurately separate each oscillation source, achieving a source localization accuracy of over 95%. Attached Figure Description

[0041] Figure 1 This is a flowchart of the workflow of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Please see Figure 1 The present invention provides the following embodiments:

[0044] Example 1

[0045] Taking a power electronic grid in a certain region with a high proportion of wind power and photovoltaic power as an example, the region has 3 new energy collection stations, 2 flexible DC converter stations and a number of power electronic converters. Figure 1 This is a flowchart of the overall process of the method of the present invention, which is described below in conjunction with... Figure 1 The complete process of this embodiment will be described in detail.

[0046] I. Deployment and Operation of the Edge Sensing Layer

[0047] A total of 56 voltage and current acquisition terminals with a sampling rate of 20kHz were deployed at the grid connection points and converter outlets of various new energy power plants. Each node is equipped with a local FPGA preprocessing unit to perform moving average filtering and adaptive cancellation of power frequency components to extract transient oscillation components. A multi-mode synchronization phasor calculation module was used to generate multi-mode synchronization phasor data containing oscillation frequency, amplitude, and phase. Each node uploads the preprocessed feature data to its respective edge computing server at 10ms intervals, uploading only the waveform recording data of the triggered period when abnormal oscillations are detected; under normal operating conditions, the full amount of raw sampled data is not uploaded. Statistical analysis shows that this selective uploading strategy reduces the amount of communication data by approximately 85% compared to the full uploading scheme.

[0048] II. Deployment and Operation of the Edge Computing Layer

[0049] One edge computing server is deployed at each of the three new energy aggregation stations, equipped with a high-performance GPU for real-time FFT calculations. The edge servers execute the following process:

[0050] (1) Receive multimodal synchronization phasor data uploaded by each end node in this region, and perform full-wave FFT and sliding window FFT analysis every 100ms to obtain the time-frequency matrix. Full-wave FFT is used to obtain the global distribution of oscillation frequency within the analysis period, and sliding window FFT is used to capture the time-varying characteristics of oscillation frequency.

[0051] (2) Screening potential oscillation frequency points from the time-frequency matrix based on an adaptive threshold. In this embodiment, the baseline threshold T0 is 0.02 (per unit), the baseline short-circuit ratio K0 is 3.0, and the penetration rate influence weighting coefficient β is 1.0. When the penetration rate of power electronic equipment P_PE / P_total in the region is 0.65 and the current short-circuit ratio K_SCR is 2.0, the ratio of the baseline short-circuit ratio to the current short-circuit ratio is multiplied by the first correction term. The first correction term is the sum of 1 and the product of the penetration rate influence weighting coefficient and the penetration rate of power electronic equipment, resulting in an adaptive adjustment coefficient of approximately 2.475. The baseline threshold of 0.02 is multiplied by the adaptive adjustment coefficient to obtain an adaptive threshold of approximately 0.05.

[0052] (3) Calculate the damping ratio of the oscillation component for the selected potential oscillation frequency points, perform clustering processing on similar oscillation frequency points, obtain the dominant oscillation frequency of each plant node in the wide area, and obtain several oscillation frequency clusters. Within the oscillation range of the oscillation frequency clusters, determine the oscillation propagation path and oscillation source based on the oscillation power amplitude of each electrical branch of each plant node.

[0053] (4) Risk assessment is conducted based on damping ratio and oscillation amplitude. When the damping ratio is between 0.05 and 0.1 and the oscillation amplitude is less than 5% of the rated value, it is determined to be a Level 1 concern state; when the damping ratio is between 0.02 and 0.05 or the oscillation amplitude reaches 5% to 15% of the rated value, it is determined to be a Level 2 warning state; when the damping ratio is less than 0.02 or the oscillation amplitude exceeds 15% of the rated value, it is determined to be a Level 3 emergency state.

[0054] (5) When the risk level reaches Level II warning or above, local adaptive oscillation suppression is implemented: the deviation between the actual output parameters and the expected output parameters of the power electronic converter is used as the oscillation identification signal to identify the oscillation frequency and oscillation amplitude online, and the transfer function parameters of the active damping controller are adjusted in real time; the oscillation suppression signal is superimposed on the current control signal of the grid-connected converter to generate a comprehensive control signal, and a PWM trigger pulse is generated according to the comprehensive control signal and output to the control terminal of the grid-connected converter.

[0055] (6) After the suppression action is executed, the damping ratio change is continuously monitored. In this embodiment, the first preset time is 0.5 seconds. Within the 0.5-second time window, the damping ratio recovery rate is continuously calculated with a sampling period of 10ms. If the damping ratio recovers to above the safety threshold (0.05 in this embodiment) within 0.5 seconds, the current suppression parameters are maintained and the suppression result is reported to the cloud for recording; if the damping ratio does not recover to above the safety threshold within 0.5 seconds, a collaborative suppression request is sent to the cloud. The collaborative suppression request carries the edge computing layer identifier, oscillation source location, current damping ratio (0.018 in this embodiment), oscillation amplitude, and the suppression parameters that have been executed.

[0056] III. Deployment and Operation of the Cloud Collaboration Layer

[0057] Deployed in the cloud at the provincial dispatch center, it collects oscillation characteristic data reported by various edge nodes. When a collaborative suppression request from the edge computing layer is received, the following process is executed:

[0058] (1) The oscillation frequency, amplitude, damping ratio and oscillation source location reported by each edge computing node are spatiotemporally correlated to construct a wide-area oscillation situation map and identify the regional impact range and propagation trend of the oscillation.

[0059] (2) Based on the power grid topology data, the system state-space equation is constructed, and eigenvalue analysis is performed on the state matrix to obtain the eigenvalues ​​of each oscillation mode. The eigenvalues ​​include real and imaginary parts. The propagation weight coefficient of each mode in the power grid is determined by impedance scanning. For each oscillation mode, the ratio of the absolute value of the real part of its eigenvalue to the magnitude of the eigenvalue is calculated as the damping ratio contribution factor, and the ratio of the oscillation amplitude of the mode to the rated amplitude is calculated as the amplitude contribution factor. The damping ratio contribution factor, the amplitude contribution factor, and the propagation weight coefficient are multiplied together to obtain the risk quantification factor of the mode. The maximum value among all mode risk quantification factors is taken as the wide-area resonance risk quantification factor. When the factor exceeds the preset warning value (0.6 in this embodiment), the corresponding system state flag is generated, forming a dual-drive decision-making mechanism.

[0060] (3) Generate inter-regional collaborative control instructions through distributed optimization, including the coordinated adjustment strategy of damping parameters of grid-connected converters in each region, the output adjustment instructions of new energy power plants and / or the tripping instructions of new energy units, and send them to each edge computing node for execution.

[0061] (4) Continuously optimize the oscillation identification model and suppression parameter library of the edge computing layer based on historical oscillation data, and send the updated model parameters to each edge computing node.

[0062] IV. Simulation Verification Results

[0063] Simulation results show that, under the condition of a 25Hz subsynchronous oscillation with a damping ratio of 0.015, the edge computing layer can complete oscillation detection and activate active damping suppression within two fundamental cycles, effectively attenuating the oscillation within 0.3 seconds, reducing the response time by approximately 60% compared to the traditional centralized solution. In another simulation scenario, when the oscillation affects more than two regions and the on-site suppression at the edge nodes fails to effectively control the oscillation, the edge computing layer sends a collaborative suppression request to the cloud after 0.5 seconds. The cloud completes wide-area situational analysis and collaborative command generation within 200ms. Through coordinated adjustment of damping parameters between regions and adjustment of power output from renewable energy stations, the oscillation is effectively suppressed within 1.2 seconds. Statistical analysis shows that, in this embodiment, the on-site suppression success rate at the edge reaches 92%, and only 8% of oscillation events require triggering cloud-based collaborative upgrades. Compared to a solution where all oscillation events are handled by the cloud, this reduces the consumption of cloud computing resources and communication bandwidth by approximately 70%.

[0064] V. Experimental Verification of the Threshold for the Grading Criterion

[0065] Under the condition of a damping ratio of 0.08 and an oscillation amplitude of 3% of the rated value (Level 1 Attention Status), the system oscillation shows a decreasing trend, and no suppression action is required. Under the condition of a damping ratio of 0.03 and an oscillation amplitude of 10% of the rated value (Level 2 Warning Status), the system oscillation shows a continuous constant amplitude oscillation trend, and the oscillation amplitude gradually decreases after the suppression device is activated for pre-adjustment. Under the condition of a damping ratio of 0.015 and an oscillation amplitude of 20% of the rated value (Level 3 Emergency Status), the system oscillation shows a diverging trend, and if not suppressed in time, it will cause the converter overcurrent protection to activate.

[0066] Example 2

[0067] The difference from Example 1 is that the edge computing layer also deploys a forced oscillation detection model based on an adaptive pulse recurrent neural network.

[0068] The model comprises an encoding layer, a spiking recurrent neural network layer, and a decoding layer. The encoding layer encodes continuous data acquired by the synchronization phasor measurement unit into a sparse pulse sequence; the spiking recurrent neural network layer employs adaptive spiking neurons, and the pulse sequence carrying feature information is input into this layer for feature extraction; the decoding layer outputs the forced oscillation detection result. The model aims to minimize the cross-entropy loss function, uses a Gaussian gradient substitution function, is trained using a backpropagation algorithm over time, and is experimentally validated using simulation and measured data.

[0069] The encoding method of the encoding layer is as follows: the continuous voltage and current data collected by the synchronous phasor measurement unit is divided into multiple segments according to the time window. For each segment, the difference between its instantaneous amplitude and the amplitude at the previous moment is calculated. When the difference exceeds the preset encoding threshold, a pulse signal "1" is output; otherwise, "0" is output. The preset encoding threshold is set according to the noise level of the data collected by the synchronous phasor measurement unit, taking 2 to 3 times the standard deviation of the noise amplitude as the encoding threshold.

[0070] The spiking recurrent neural network layer uses adaptive spiking neurons. The dynamic process of the adaptive spiking neuron is as follows: the neuron membrane potential increases according to the accumulation of input pulses. When the membrane potential exceeds the firing threshold, the neuron fires a pulse and resets the membrane potential. The firing threshold is adaptively adjusted according to the neuron's recent firing history. When the neuron's recent firing frequency is high, the firing threshold is increased to reduce the firing frequency. When the neuron's recent firing frequency is low, the firing threshold is decreased to increase sensitivity.

[0071] The model is trained by substituting the non-differentiable firing function of the spiking neuron with a Gaussian gradient substitution function, aiming to minimize the cross-entropy loss function, and then training based on the backpropagation algorithm over time. The training dataset includes simulation data and measured data. The simulation data is generated by power system simulation software under different oscillation frequencies and damping ratios, and the measured data comes from oscillation event data recorded by the synchronization phasor measurement unit in the actual power grid.

[0072] Example 3

[0073] The difference from Example 1 is that when the cloud-based collaborative layer generates inter-regional collaborative control commands, it prioritizes the coordinated adjustment strategy of damping parameters of grid-connected converters in each region and the output adjustment command of new energy power plants. Only when the above adjustment strategy fails to restore the damping ratio to above the safety threshold within a preset time will the new energy unit trip command be generated, so as to minimize the power generation loss caused by tripping.

[0074] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for monitoring and suppressing grid oscillations supported by edge computing in power electronics, characterized in that, Includes the following steps: S1: Construct a three-level edge computing collaborative architecture consisting of an end-side perception layer, an edge computing layer, and a cloud-based collaboration layer. The edge computing layer is deployed in regional substations or new energy power station aggregation stations and has local data storage, real-time signal processing, and oscillation decision-making capabilities. S2: The end-side sensing layer synchronously collects electrical quantity signals from power electronic devices and preprocesses them to generate multimodal synchronous phasor data, and uploads the multimodal synchronous phasor data to the edge computing layer; S3: The edge computing layer obtains the time-frequency matrix based on the multimodal synchronous phasor data by combining full-wave FFT and sliding window FFT, filters potential oscillation frequency points from the time-frequency matrix based on an adaptive threshold and calculates the damping ratio of the oscillation components, performs clustering processing on similar oscillation frequency points, and determines the oscillation frequency, oscillation amplitude, damping ratio, oscillation source location and oscillation propagation path. S4: The edge computing layer performs a graded risk assessment based on the damping ratio and the oscillation amplitude. When the risk level reaches level two or above, the edge computing layer generates a suppression command locally and executes local adaptive oscillation suppression, while reporting the oscillation characteristics and suppression status to the cloud collaboration layer. S5: After the local adaptive oscillation suppression is executed, the edge computing layer continuously monitors the damping ratio change. If the damping ratio does not recover to above the safety threshold within a preset time or the oscillation affects multiple areas, a collaborative suppression request is sent to the cloud collaboration layer. In response to the collaborative suppression request, the cloud collaboration layer collects the oscillation characteristics and suppression status reported by multiple edge computing layers, constructs a wide-area oscillation situation, generates inter-regional collaborative control instructions, and sends them to the corresponding edge computing layers for execution.

2. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 1, characterized in that, In step S1, the end-side sensing layer includes acquisition terminals deployed at the grid connection point of new energy power plants, substation busbars, and power electronic converter outlets; the edge computing layer includes edge computing servers and edge intelligent terminals; and the cloud-to-cloud collaboration layer is deployed in the dispatch center for wide-area situational awareness, multi-regional coordination optimization, and long-term trend analysis.

3. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 1, characterized in that, In step S2, the end-side sensing layer synchronously acquires three-phase voltage, three-phase current, active power, and reactive power signals at a sampling rate of not less than 10kHz. After analog-to-digital conversion, moving average filtering, and adaptive cancellation of power frequency components, transient oscillation components are extracted. Multi-mode synchronous phasor data containing oscillation frequency, amplitude, and phase is generated using multi-mode synchronous phasor calculation. The end-side sensing layer only uploads the preprocessed feature data and waveform recording data during abnormal triggering periods, and does not upload the full original sampling data.

4. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 1, characterized in that, In step S3, the full-wave FFT is used to obtain the global distribution of oscillation frequency within the analysis period, and the sliding window FFT is used to capture the time-varying characteristics of oscillation frequency; the adaptive threshold is dynamically adjusted according to the current power grid power electronic equipment penetration rate, short-circuit ratio, and operating conditions; the adaptive threshold is determined as follows: a baseline threshold, a baseline short-circuit ratio, and a penetration rate influence weighting coefficient are preset; the ratio of the total capacity of power electronic equipment to the total installed capacity of the power grid is used as the power electronic equipment penetration rate; the ratio of the baseline short-circuit ratio to the current actual short-circuit ratio of the power grid is multiplied by a first correction term to obtain the adaptive adjustment coefficient, where the first correction term is the sum of 1 and the product of the penetration rate influence weighting coefficient and the power electronic equipment penetration rate; The adaptive threshold is obtained by multiplying the baseline threshold by the adaptive adjustment coefficient, so as to dynamically adjust the threshold and improve the oscillation detection sensitivity under high power electronics penetration conditions.

5. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 4, characterized in that, In step S3, when clustering similar oscillation frequency points, the dominant oscillation frequency of each plant node in a wide area is obtained, and several oscillation frequency clusters are obtained. Within the oscillation range of the oscillation frequency clusters, the oscillation propagation path and oscillation source are determined according to the oscillation power amplitude of each electrical branch of each plant node.

6. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 1, characterized in that, In step S4, the graded risk assessment includes: when the damping ratio is between 0.05 and 0.1 and the oscillation amplitude is less than 5% of the rated value, it is determined to be a Level 1 concern state, continuously tracking changes in the oscillation mode and reporting according to a preset cycle; when the damping ratio is between 0.02 and 0.05 or the oscillation amplitude reaches 5% to 15% of the rated value, it is determined to be a Level 2 early warning state, initiating pre-adjustment of the suppression device and sending early warning information to the cloud collaboration layer; when the damping ratio is less than 0.02 or the oscillation amplitude exceeds 15% of the rated value, it is determined to be a Level 3 emergency state, immediately triggering local suppression action and reporting the emergency state to the cloud collaboration layer.

7. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 1, characterized in that, In step S4, the local adaptive oscillation suppression includes: using the deviation between the actual output parameters and the expected output parameters of the power electronic converter as an oscillation identification signal, identifying the oscillation frequency and oscillation amplitude online, and adjusting the transfer function parameters of the active damping controller in real time; superimposing the oscillation suppression signal onto the current control signal of the grid-connected converter to generate a comprehensive control signal, and generating a PWM trigger pulse based on the comprehensive control signal and outputting it to the control terminal of the grid-connected converter.

8. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 7, characterized in that, In step S5, after the suppression action is executed, the edge computing layer continuously calculates the damping ratio recovery rate with a first preset time window. The damping ratio recovery rate is defined as the ratio of the increment of the damping ratio of the k-th sampling window relative to the damping ratio of the (k-1)-th sampling window after the suppression action is executed to the sampling window time. When the damping ratio recovery rate is positive and the cumulative recovery amount reaches a safety threshold within the first preset time, the suppression is determined to be effective, the current suppression parameters are maintained, and the suppression result is reported to the cloud collaboration layer for recording. When the damping ratio recovery rate is less than or equal to zero or the cumulative recovery amount does not reach the safety threshold within a first preset time, the suppression is determined to have failed, and a collaborative suppression request is sent to the cloud collaboration layer. The collaborative suppression request carries the edge computing layer identifier, oscillation source location, current damping ratio, oscillation amplitude, and the suppression parameters that have been executed. The cloud collaboration layer then initiates the inter-regional collaborative suppression process.

9. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 1, characterized in that, In step S5, the cloud-based collaborative layer spatiotemporally correlates the oscillation frequency, amplitude, damping ratio, and oscillation source location reported by each edge computing node to construct a wide-area oscillation situation map; and constructs a system state-space equation based on power grid topology data, performs eigenvalue analysis on the state matrix to obtain eigenvalues ​​of each oscillation mode, wherein the eigenvalues ​​include real and imaginary parts, the real part representing damping and the imaginary part representing oscillation frequency; determines the propagation weight coefficient of each oscillation mode in the power grid through impedance scanning; for the i-th oscillation mode, calculates the ratio of the absolute value of its eigenvalue real part to its eigenvalue magnitude as a damping ratio contribution factor, calculates the ratio of the oscillation amplitude of this mode to its rated amplitude as an amplitude contribution factor, and multiplies the damping ratio contribution factor, amplitude contribution factor, and propagation weight coefficient to obtain the risk quantification factor of this oscillation mode; The maximum value among all oscillation mode risk quantification factors is taken as the wide-area resonance risk quantification factor. The wide-area resonance risk quantification factor is compared with the preset risk warning value to form a dual-drive decision-making mechanism that combines threshold-triggered rapid response with in-depth risk quantification insight.

10. The edge computing-supported power electronic grid oscillation monitoring and suppression method according to claim 1, characterized in that, The edge computing layer also deploys a forced oscillation detection model based on an adaptive pulse recurrent neural network. The model includes an encoding layer, a pulse recurrent neural network layer, and a decoding layer. The encoding layer encodes the continuous data collected by the synchronous phasor measurement unit into a sparse pulse sequence. The pulse recurrent neural network layer uses adaptive spiking neurons for feature extraction. The decoding layer outputs the forced oscillation detection result. The encoding layer divides the continuous voltage and current data into multiple segments according to a time window, calculates the difference between the instantaneous amplitude of each segment and the amplitude of the previous moment, and outputs a pulse signal when the difference exceeds a preset encoding threshold. The membrane potential of the adaptive spiking neuron increases as the input pulses accumulate. When the membrane potential exceeds the firing threshold, a pulse is fired and the membrane potential is reset. The firing threshold is adaptively adjusted based on the neuron's recent firing history.

Citation Information

Patent Citations

  • Emergency control method and device for external power system stabilizer

    CN110571830A

  • Multi-modal oscillation online monitoring and early warning system based on edge calculation

    CN110927442A

  • Methods, apparatus, systems and storage media for suppressing broadband oscillations in power systems

    CN116960990B

  • Power equipment control method based on cloud side cooperation and electronic equipment

    CN121192925A

  • Flexible DC power grid wide area cooperative control method and system based on cloud edge cooperation

    CN121689310A