Power grid frequency balance intelligent control system based on distributed architecture

Abstract

The invention discloses a power grid frequency balance intelligent control system based on a distributed architecture, relates to the technical field of power grid control, and is used for solving the problems that the existing WLS / UKF and other methods are generally assumed to be good in synchronization, are insensitive to heteroscedasticity time delay and missing measurement, and are difficult to ensure robustness under real communication and equipment conditions. Through a through link, under the real clock misalignment, time delay jitter and packet loss conditions, measurement of a unified time axis and quality quantification are realized, and frequency estimation bias and uncertain transmission are significantly reduced. Reconstruction, filtering and triggering are penetrated with the same confidence degree, and abnormal measurement suppression and effective information utilization are improved; the fusion efficiency and timeliness of asynchronous arrival data are improved by utilizing graph prior and information domain incremental updating; and regional measurement abnormity and real power imbalance are discriminated through double-model causality, and error event reporting is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid control, more particularly, to a power grid frequency balance intelligent control system based on a distributed architecture. BACKGROUND

[0002] With the complication of modern power systems and the improvement of interconnection, the stable operation of power grid frequency is facing severe challenges. Frequency is a key indicator of power system operation state, and its balance is directly related to the safety and reliability of the power grid. However, in actual operation, the accurate perception and rapid control of power grid frequency are restricted by various factors.

[0003] The prior art has the following disadvantages: Under the distributed power grid frequency balance scenario, the time stamp inconsistency and spatial observation fragmentation caused by the coexistence of clock misalignment, asynchronous time delay, packet loss and topology disturbance of PMU / RTU and other heterogeneous measurements, cause systematic bias in frequency and RoCoF estimation and distortion in the transmission of upper-layer frequency regulation link, and further induce AGC false triggering, over / under regulation and oscillation and other chain risks. The existing WLS / UKF method generally assumes good synchronization, is not sensitive to heteroscedastic time delay and missing data, and is difficult to ensure robustness under real communication and equipment conditions. The core difficulty is: under uncontrollable network and equipment quality fluctuations, realize unified time axis and measurement reliability quantification, structure reconstruction of missing data, robust state estimation with asynchronous and graph prior constraints, and distinguish between measurement anomalies and real power imbalance, and form a closed loop of data quality to event triggering and control threshold.

[0004] In view of the above problems, the present application provides a solution. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a power grid frequency balance intelligent control system based on a distributed architecture to solve the problems raised in the background art.

[0006] To achieve the above object, the present application provides the following technical scheme: The power grid frequency balance intelligent control system based on a distributed architecture comprises: The acquisition and time correction module is configured to perform joint estimation of clock and time delay on edge measurements, generate corrected time stamps, jitter statistics and node identifiers, establish a time reference table, and provide the time reference table for unified time axis mapping and indexing; The confidence calculation module is configured to calculate the comprehensive confidence Ci according to the clock quality, time delay statistics, residual consistency, graph space coherence and packet loss risk, form a weight table containing version number and time decay parameters, and provide the weight table for reconstruction and filtering; a reconstruction module configured to perform weighted sparse reconstruction on the Ci against fragmented measurements based on the power grid graph model and the frequency domain basis representation, to obtain reconstructed data of frequency sequences of each node, and to generate a reconstructed cache and an index pointer for retrieval by the regional agent; a filtering module configured to perform state estimation on the reconstructed data and asynchronous measurements in information form at the regional agent side, to maintain an information matrix and an information vector, and to advance [f,ḟ] in small steps while outputting a residual set for subsequent determination and weight table update; a causality determination module configured to recursively determine the residual set and neighborhood consistency in combination with graph coherence and online change point detection, to generate event labels and saliencies, and to store them together with node identities for triggering and reference of master station parameter settings.

[0007] In a preferred embodiment, the collection and time calibration module establishes a time reference table containing node identities, correction timestamps, link jitter, and version numbers; synthesizes delay and jitter from unidirectional and bidirectional timestamp packets; and provides the time reference table for unified time axis mapping and subsequent module indexing.

[0008] In a preferred embodiment, the confidence calculation module factorizes the comprehensive confidence Ci into clock quality, delay statistics, residual consistency, graph spatial coherence, and packet loss risk; generates them from the time reference table, round-trip delay statistics, previous residuals, neighborhood consistency, and hidden Markov estimation, respectively; and writes them into the weight table.

[0009] In a preferred embodiment, the weight table contains time decay parameters and effective times; performs overwrite or merge update according to node identities and version numbers when new measurements or regional side residuals arrive; and maintains a one-to-one index relationship for the updated entries for calling by the reconstruction module and the filtering module.

[0010] In a preferred embodiment, the reconstruction module constructs frequency domain representation based on the Laplacian eigenbasis of the power grid graph, forms diagonal weighting according to the Ci and the sampling matrix Sk; uses iterative sparse solution and previous time solution as warm start, limits maximum iteration and stop threshold, and outputs the reconstructed cache and index pointers by node and time.

[0011] In a preferred embodiment, the filtering module maintains an information matrix Ω and an information vector ξ, advances in fixed small steps; performs incremental update on arriving measurements or reconstructed data scaled by Ci on Ri; only exchanges incremental difference quantities of boundary nodes when cross-regional collaboration, and sets boundary variable change threshold and round limit.

[0012] In a preferred embodiment, the cause-effect discrimination module adopts a double model online variable point detection, maintains a running length and a posterior for a measurement abnormality model and a power imbalance model respectively; recursively outputs an event label and a significance with a node residual and a graph coherence score as observation quantities, and stores them in association with a node identifier.

[0013] In a preferred embodiment, it further comprises a trigger arrangement module, which triggers according to a joint indicator of an innovation quantity norm, an average comprehensive confidence and the significance compared with a congestion level threshold; when triggered, it packs boundary incremental information quantity, an event label and an index field, and sets a minimum trigger interval and a hysteresis rule.

[0014] In a preferred embodiment, it further comprises a control interface module, which performs boundary consistency on regional estimation at a master station side to generate interface data containing a frequency, a frequency change rate and a covariance; sets a control threshold according to the event label and the Ci, and forms a priori and a threshold parameter set to issue a region and an edge.

[0015] In a preferred embodiment, the message structure between an edge agent and a region agent contains a node identifier, a correction timestamp, a version number, an incremental information quantity, an event label and a significance, a cache pointer and a source identifier; each field is encapsulated according to a unified encoding and checking scheme and transmitted through a secure channel.

[0016] The power grid frequency balance intelligent control system based on a distributed architecture has the following technical effects and advantages: The present application realizes measurement unified time axis and quality quantization under the conditions of real clock misalignment, delay jitter and packet loss through a through link, significantly reduces frequency estimation bias and uncertain transmission; improves abnormal measurement suppression and effective information utilization by using the same confidence to penetrate reconstruction, filtering and triggering; improves the fusion efficiency and timeliness of asynchronously arriving data by using graph priori and information domain incremental update; reduces false event reporting by distinguishing measurement abnormality and real power imbalance through double model cause-effect discrimination; preferentially transmits key increments in congestion conditions by using trigger arrangement based on innovation quantity and significance; and makes estimation, communication and control form a self-consistent closed loop by setting a control threshold according to an event label and a confidence and writing back a priori and a threshold at a master station, so that more robust, low-latency and measurable uncertain perception and frequency regulation decision support is obtained under a distributed architecture. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a layered distributed architecture schematic diagram of a power grid frequency balance intelligent control system according to an embodiment of the present application; Figure 2 is an indivisible function flowchart of a power grid frequency balance intelligent control system according to an embodiment of the present application; Figure 3is a schematic diagram of a module structure of a power grid frequency balance intelligent control device according to an embodiment of the present application. Figure 4 is a schematic diagram of a structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] The present application relates to the technical field of power grid control, in particular to a power grid frequency balance intelligent control system based on a distributed architecture.

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0020] Embodiment, see Figure 1 The present application provides a power grid frequency balance intelligent control system based on a distributed architecture. The system adopts a hierarchical distributed structure, aiming to solve the problem of unbiased, robust and low-delay perception of power grid node frequency and frequency change rate RoCoF, and output the uncertainty to frequency modulation control after quantization.

[0021] The system mainly includes the following three levels: 1. Edge agent ; deployed on the side of measurement devices such as PMU / RTU, responsible for preliminary processing of raw measurement data.

[0022] 2. Regional agent ; deployed in substations / region master stations, responsible for data aggregation, processing and event determination in the region.

[0023] 3. Master station MS; responsible for coordination control and decision-making in the whole network.

[0024] As shown in Figure 2 , the system achieves its goal through the following inseparable functional processes, specifically including: Step S101: time alignment and time axis unification; In the edge agent On the other hand, joint clock and latency estimation is performed on the raw measurement data. Specifically, this process generates a corrected timestamp and link jitter metric, maps the measurements of each node to a unified time axis, and registers the time base and node identifier metadata for subsequent processing. This step solves the problems of incomparability of timestamps between different nodes caused by inaccuracies in PTP / NTP, Precision Time Protocol / Network Time Protocol, and link jitter, ensuring time consistency in subsequent data processing.

[0025] Furthermore, a recursive model incorporating offset and drift states is used for the joint estimation of clock and delay. Combined with one-way and two-way timestamp message parsing, the combined network and device delay is estimated. and its vibration The corrected timestamps, jitter metrics, and node identifiers are written into the edge agent's time base table for reference in subsequent steps.

[0026] Step S102: Calculate the overall confidence level; On the edge agent side, the overall confidence level is calculated based on clock quality, latency statistics, residual consistency, graph spatial coherence, and packet loss risk. .Should The data is written into the edge-side weight table, serving as a weighted input and unified index key for reconstruction, filtering, event triggering, and communication orchestration, thus quantifying and transmitting the uncertainty of measurement data. This step addresses the issue of varying observation quality requiring unified quantification and transmission to subsequent stages.

[0027] Furthermore, for abnormal jumps in local clock skew... A robust cost function is used for suppression to prevent outlier data from interfering with confidence calculation.

[0028] Optionally, after this step is completed, the... Write the weight table on the edge side along with the time decay parameter and version number; when a new measurement arrives or the residual is written back, retrieve the corresponding entry according to the node identifier and version number to overwrite or merge and update it, ensuring that subsequent steps consistently reference the same weight entry.

[0029] Step S103: Weighted graph compressed sensing missing measurement reconstruction; On the edge agent side, based on the power grid graph model and frequency domain basis representation, fragmented measurements on a unified time axis are executed according to the aforementioned... Weighted sparse reconstruction yields reconstructed data of the frequency sequences of each node. This step addresses the problem of spatial fragmentation of measurement data caused by packet loss or late arrival, which affects the integrity of state estimation. The reconstruction process generates a reconstruction cache and index pointers for regional reading.

[0030] Furthermore, this step constructs a frequency domain representation based on the characteristic basis of the Laplace of the power grid graph, using the method described above. Weighted sparse reconstruction strategy. The reconstruction initializes with the solution of the last time step as a warm start, limits the number of iterations and a stop threshold, and generates a reconstruction data cache and index pointers for the regional agent to retrieve by node and time.

[0031] Optionally, the selection of the fragmented measurements is determined by a sampling matrix Indicating that the sampling matrix is multiplied by the weight matrix according to the Diagonal weighting is generated; and a convergence criterion according to the change amount is set for the iteration process, for example, the relative error is less than , and a maximum iteration upper limit, for example, 5-10 iterations, when any condition is met, the reconstruction data of the corresponding node is output, and the cache time stamp and source identification are registered.

[0032] Step S104: asynchronous graph information filtering and causal variable point detection; On the regional agent side, the state estimation of the information form is performed on the asynchronous arriving measurements and the reconstruction data. This step solves the problem of robust state estimation that needs to be performed on the asynchronous arrival and quality of the measurement data.

[0033] Specifically, this step uses the information filtering method, and injects the graph smoothing prior. When the measurement of the node arrives, the incremental update is performed according to , wherein is the measurement noise covariance matrix, which is scaled by . The information matrix , with a dimension of , wherein is the number of nodes, and the information vector , with a dimension of , is obtained, and is advanced according to a fixed small step frequency, for example, 20-50 ms.

[0034] Further, when cross-regional cooperation, only the incremental difference components of the information matrix and the information vector of the boundary nodes are exchanged, and a consensus algorithm or an alternating direction multiplier method, ADMM, is used to maintain the boundary consistency. This step sets a boundary variable change threshold and an upper limit of the number of rounds, for example, 5-20 iterations, and outputs the regional level estimation after the condition is met and writes the residual set back to the edge agent EA to update the related entries.

[0035] In addition, this step combines the graph coherence and variable point detection to generate event labels and significance, and writes back the residual statistics to update the edge side confidence related data entries. This step solves the problem of distinguishing between “measurement anomaly” and “real power imbalance” to avoid false control.

[0036] Specifically, the cause-effect change point determination adopts a double model Bayesian online change point detection, and utilizes a BOCPD framework to respectively detect measurement abnormalities, such as model and true power imbalance model , and maintain running length and posterior probability. The graph space coherence score and node residual are used as observations to participate in recursion, and output event labels and significance , which are stored together with node identification for subsequent steps. The prior settings of the BOCPD can be determined according to historical data or expert experience. For example, the prior distribution of the change point probability can be set as a Beta distribution, and the parameters thereof can be adjusted according to the expected change point frequency.

[0037] Step S105: event triggering and communication arrangement On the edge agent and regional agent side, event triggering and communication pruning are performed according to the comprehensive analysis quantity. This step solves the problem of ensuring the priority transmission of critical state information and avoiding excessive reporting when communication is congested.

[0038] Specifically, the triggering criterion is obtained by comparing the joint index of the innovation quantity norm , the average comprehensive confidence and the significance with the threshold related to the congestion level. The exponential parameter in the delay quality can be adjusted according to the network jitter characteristics, and the typical value range is 0.1-10, which is used to balance the influence of delay jitter on confidence. When the triggering condition is met, only the boundary incremental information quantity, event label and necessary index field are packaged, and the minimum triggering interval, for example, 50-200 ms, is used for throttling control and repetition suppression with the hysteresis rule.

[0039] Step S106: master station threshold setting and parameter downlink On the master station MS side, the estimation results and event labels are received, interface data such as frequency and frequency change rate RoCoF, and confidence interval are generated. This step solves the problem that uncertainty should be reflected in the control threshold to suppress false triggering and over-regulation.

[0040] Specifically, after the master station performs boundary unification processing on the regional estimation, an interface data structure including system frequency, frequency change rate RoCoF, and covariance is generated. According to the event label and the , the control threshold is set, and a priori and threshold parameter set are downlinked to the regional agent and the edge agent to update the local configuration.

[0041] Further, the priori and threshold parameter set issued by the master station includes graph priori intensity, trigger threshold, time advancement step and weight table reservation strategy.The regional agent updates the priori item and trigger logic of the information matrix after receiving, and the edge agent updates the decay parameter and version number of the weight table after receiving, and records the effective time and configuration source.

[0042] Optional implementation: In addition, the message structure between the edge agent and the regional agent includes node identification, correction timestamp, version number, incremental information amount, event label and significance, cache pointer and source identification; each field is packaged according to a unified encoding and checking scheme, and is transmitted through a secure channel for the step four and the step five to parse according to the index.

[0043] In order to accurately describe and analyze the power grid frequency balance intelligent control system, the power grid graph model and the frequency field state model are constructed.

[0044] Power grid graph model: The power grid is abstracted as a graph model , where represents the nodes in the power grid, such as generators, loads, substations, represents the connections between nodes, such as transmission lines. The edge weight can be defined according to electrical characteristics, for example, proportional to the inverse of the line reactance , or generated according to the inverse of the electrical distance. Based on this graph model, the graph Laplacian matrix can be constructed, where is the degree matrix, is the weight matrix. The graph Laplacian matrix plays a key role in subsequent graph signal processing and state estimation.

[0045] Frequency field state model: At discrete time , the frequency field state of the power grid is defined as a vector containing all node frequencies and rate of change of frequency (RoCoF) : ; where is the number of nodes in the power grid. The evolution of the frequency field state follows a linear dynamic system model: ; where is the state transition matrix, is the control input matrix, is the control input, is the process noise, which is subject to a mean of 0 and a covariance of Gaussian distribution.

[0046] The measurement model considers asynchrony / delay / missing; Node At local time Generate observations . Considering the asynchrony, delay and missing of measurements in real power grid, the measurement model is represented as: ; Wherein represents the frequency value of node At real time , is the network and device combined delay, is the measurement noise, which is subject to Gaussian distribution with mean value 0 and variance .

[0047] Local clock model: The local clock of each node has offset and drift from global real time : ; Wherein is the clock offset, is the clock drift.

[0048] The missing of measurement data is indicated by the sampling matrix , which is a diagonal matrix. The diagonal elements are 1, indicating that the corresponding node has measurement, and 0, indicating that there is no measurement.

[0049] These mathematical models provide a theoretical basis for subsequent time alignment, confidence calculation, missing reconstruction, state estimation and event judgment of the application.

[0050] The application constructs a comprehensive confidence for each observation, which is used to quantify the quality and reliability of the observation data throughout the various levels of the system and guide the subsequent processing process. The calculation formula of the comprehensive confidence is as follows: ; Wherein: Clock quality : This index reflects the accuracy of the local clock. It is based on the edge agent, EA, side to the Kalman estimation variance of local clock offset and drift , and maps the variance to interval. The smaller the variance, the higher the clock quality, closer to 1.

[0051] Further, the mapping employs a piecewise linear function to avoid threshold chattering, ensuring smooth variation.

[0052] Latency quality : This index reflects the stability and predictability of network transmission latency. It is mapped from the form of round-trip time, RTT, and jitter statistics . Wherein represents the latency jitter, the larger the jitter, the smaller, indicating the worse latency quality. The parameter can be adjusted according to the network jitter characteristics, and the typical value range is 0.1-10, used to balance the influence of latency jitter on confidence.

[0053] Residual consistency : This index reflects the degree of agreement between the current observation and the system state estimation. It is mapped from the form of Mahalanobis distance of the last filtering residual . The larger the Mahalanobis distance, the larger the residual, indicating the greater deviation between the observation and the estimation, the smaller.

[0054] Graph spatial coherence : This index reflects the consistency of the current node frequency with the frequency of the electrical neighborhood nodes. It is obtained by calculating the frequency consistency score of the current node and the electrical neighborhood , and then normalized after temperature scaling. When the node frequency is highly consistent with the frequency of its neighborhood, the higher, indicating that the observation has better spatial coherence.

[0055] Optionally, when part of the nodes lack neighborhood data, it can be degraded to a local consistency estimate weighted by the electrical distance.

[0056] Packet loss risk : This index reflects the probability of packet loss of the measurement data at the next moment. It is obtained by online estimation of good or congestion in a two-state hidden Markov chain. The higher the packet loss risk, the smaller, thereby reducing the comprehensive confidence.

[0057] Comprehensive confidence has multiple purposes in the entire system, ensuring consistent transmission of evidence, including: Weighted reconstruction: In the missing measurement reconstruction module, it is used for weighted sparse reconstruction of fragmented measurements, so that high-quality measurement data plays a greater role in reconstruction.

[0058] Measurement Covariance Scaling: In the regional estimation event decision module, used to scale the measurement covariance so that high-quality measurement data has a higher weight in the information filtering.

[0059] Event Trigger Threshold Calculation: In the trigger orchestration module, as one of the core inputs of the event trigger criterion, it affects the dynamic adjustment of the trigger threshold.

[0060] Through this comprehensive confidence system, the invention can comprehensively and dynamically evaluate the quality of each measurement data, and effectively transmit this uncertainty information to each processing link of the system, thereby improving the robustness and accuracy of the entire system.

[0061] Referring to Figure 3 , the embodiment also provides a power grid frequency balance intelligent control system, which can include but is not limited to the following modules: Acquisition and time correction module: used for joint estimation of clock and time delay of edge measurement, generation of corrected time stamp and link jitter index, and output of corrected time and clock quality , time delay quality . This module maps each node measurement to a unified time axis, and records time reference and node identification metadata for subsequent processing reference.

[0062] Confidence calculation module: used for calculating comprehensive confidence according to clock quality, time delay statistics, residual consistency, graph space coherence and packet loss risk. This module writes the into the edge side weight table as the weighted input and unified index key of reconstruction, filtering, event triggering and communication orchestration.

[0063] Reconstruction module: used for performing sparse reconstruction of fragmented measurements on the unified time axis based on the power grid graph model and frequency domain basis representation, obtaining reconstruction data of each node frequency sequence, and generating reconstruction cache and index pointer for regional side reading.

[0064] Filtering module: used for performing asynchronous graph information filtering, performing information form state estimation on asynchronously arrived measurement and reconstruction data, and outputting estimated frequency , rate of change of frequency RoCoF, and residual set.

[0065] Causal discrimination module: used for analyzing the residual output by the filtering module in combination with graph coherence and change point detection, generating event label and significance .

[0066] ​Trigger orchestration module: used for event triggering and communication orchestration according to comprehensive analysis quantities such as innovation quantity norm, average comprehensive confidence and salience, to realize intelligent orchestration of communication.

[0067] Control interface module: used for providing estimated quantities, confidence intervals and adaptive thresholds to primary / secondary frequency modulation and energy storage scheduling, and receiving threshold feedback from the master station to update the local configuration.

[0068] The above modules can be independent hardware units, such as application-specific integrated circuits, ASICs, field-programmable gate arrays, FPGAs, etc., or can be implemented by software / firmware instructions of a processor. In different embodiments, the functions of the modules can be combined or split according to actual needs, for example, in some scenarios, the regional agent, RA, can be deployed to an in-station industrial computing platform, and the edge agent, EA, and the regional agent, RA, can be combined in the same physical device to reduce link delay.

[0069] Referring to Figure 4 The present embodiment also provides an electronic device which can be used to implement the functions of the edge agent, regional agent or master station in the power grid frequency balance intelligent control system. The electronic device includes one or more processors, memories, communication interfaces and user interfaces.

[0070] Processor: the control center of the electronic device, responsible for executing program instructions in the memory to implement the functions described in the present invention. The processor can be a central processing unit, CPU, a graphics processing unit, GPU, a digital signal processor, DSP, or any other programmable logic device.

[0071] Memory: used for storing program instructions, data and intermediate results generated during program execution. The memory can be a random access memory, RAM, a read-only memory, ROM, a flash memory, Flash Memory, or any other type of storage medium. When the program instructions stored in the memory are executed by the processor, the electronic device performs one or more steps of any embodiment, thereby implementing the functions of the power grid frequency balance intelligent control system.

[0072] Communication interface: used for data exchange between the electronic device and other devices or networks. For example, the edge agent is connected with the PMU / RTU and communicates with the regional agent through the communication interface; the regional agent communicates with the edge agent and the master station through the communication interface. The communication interface can be an Ethernet interface, a wireless communication module such as Wi-Fi, 5G, or any other wired or wireless communication interface.

[0073] User interface: used for user interaction with the electronic device, such as displaying running status, configuration parameters or receiving user instructions. The user interface can be a display screen, keyboard, mouse, touch screen or any other input / output device.

[0074] The embodiment also provides a computer readable storage medium having program instructions stored thereon. The program instructions realize the flow of the embodiment when running on the electronic device. The computer readable storage medium can be a memory or a removable storage medium such as a U disk, an SD card, an optical disc, etc. By storing the program instructions on the computer readable storage medium, the technical solution of the application can be conveniently deployed and updated.

[0075] In order to ensure the actual operation effect and engineering feasibility of the system described in the application, some key parameter settings and engineering suggestions are provided in the embodiment. Mapping of PTP quality level and The mapping is performed by using a piecewise linear function to avoid sharp fluctuations of at specific thresholds, ensuring its smoothness and stability.

[0076] Adaptive adjustment of graph prior strength According to the size of the frequency change rate RoCoF, the strength of the graph smooth prior is adaptively adjusted. When the power grid disturbance is large, the strength of the graph smooth prior is weakened to better preserve the boundary information and transient characteristics; when the power grid runs smoothly, the smooth prior is enhanced to improve the smoothness of the estimation.

[0077] FISTA warm start for EA reconstruction: the missing data reconstruction of the edge agent, EA, uses the fast iterative shrinkage threshold algorithm, FISTA, and the solution of the previous moment is used as the warm start initialization. Experience shows that 5-10 iterations can meet the real-time requirements in typical cases, and the convergence criterion can be set to a relative error of less than , effectively balancing the calculation efficiency and reconstruction accuracy.

[0078] Small step frequency advancement of RA information filtering: the information filtering of the regional agent, RA, is advanced in small steps of 20-50 ms, and the measurement data is updated as it comes. This high-frequency update mechanism ensures the system's rapid response capability to changes in the power grid state.

[0079] Event-triggered minimum interval: the minimum interval of event triggering is recommended to be set to 50-200 ms and dynamically adjusted according to the communication congestion level. When the communication congestion is serious, the triggering interval is appropriately lengthened to avoid excessive communication leading to network paralysis; when the communication conditions are good, the triggering interval can be shortened to improve the system response speed.

[0080] ​Upper limit of ADMM rounds: When using ADMM algorithm for cross-region collaboration, the typical upper limit of iteration number can be set to 5-20 times to balance the convergence speed and communication overhead.

[0081] BOCPD prior: The prior setting of Bayesian Online Change Point Detection, BOCPD, has an important impact on detection performance. For example, the prior distribution of change point probability can be set as Beta distribution, and its parameters can be adjusted according to the expected change point frequency, such as one change point per hour, to guide the algorithm to balance between detection sensitivity and false alarm rate.

[0082] In order to evaluate the performance and effect of the system described in the present application, a series of verifiability indicators are proposed in this embodiment: Estimation error: Mean absolute error of frequency estimation.

[0083] Root mean square error of RoCoF estimation.

[0084] Unbiasedness: In the undisturbed window, the expected deviation between the system frequency estimation value and the reference frequency tends to 0, indicating the accuracy of the system estimation.

[0085] Trigger quality: False trigger rate: The proportion of non-events judged as events by the system.

[0086] Miss trigger rate: The proportion of real events that the system fails to detect.

[0087] These indicators can be based on event labels and real labels generated by manual annotation or simulation.

[0088] Communication load: Compare the communication bandwidth occupancy before and after event triggering to evaluate the improvement effect of event triggering and communication arrangement mechanism on communication efficiency.

[0089] Control benefit: Over / under adjustment ratio: Evaluate the suppression effect of the control system on frequency deviation.

[0090] AGC / energy storage action times and equivalent life consumption: Evaluate the economy of control action and its impact on device life.

[0091] The present application also includes some optional implementations to adapt to different application scenarios and needs: ​RA is down and EA / RA is set: in some scenarios, the function of the regional agent RA can be down to the station industrial computing platform, and even the edge agent EA and the regional agent RA are set in the same physical device to further reduce the link delay and improve the system response speed.

[0092] Degradation processing of : When some nodes lack neighborhood data, the graph space coherence can be degraded to a local consistency estimate weighted by electrical distance to ensure the effectiveness of the index in the case of incomplete data.

[0093] Large-scale system cross-zone consensus optimization: for large-scale systems, cross-zone consensus can use an incremental transmission strategy by boundary blocks, only exchanging the information increment of boundary nodes, reducing redundant exchange, and thus reducing communication load.

[0094] The above embodiments constitute a continuous and indivisible chain from time calibration, confidence calculation, missing data reconstruction, asynchronous graph information filtering, causal variable point discrimination, event triggering and communication arrangement to master station threshold setting. The comprehensive analysis results are connected through weights, covariances, thresholds and other forms at each link, so that the system can still provide reliable, robust and low-latency state estimation and decision basis for distributed frequency balance control under the constraints of real communication and hardware. Those skilled in the art can make modifications and substitutions to the above embodiments without departing from the spirit of the present application, which shall fall within the protection scope of the present application.

[0095] Those skilled in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application of the technical solution and the constraints of the invention. Professional technicians can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0096] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0097] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which shall be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

[0098] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. A power grid frequency balance intelligent control system based on a distributed architecture, characterized in that, include: The time acquisition module is configured to perform joint estimation of clock and delay for edge measurements, generate correction timestamps, jitter statistics and node identifiers, establish a time reference table, and make the time reference table available for unified time axis mapping and indexing. The confidence calculation module is configured to calculate a comprehensive confidence score Ci based on clock quality, delay statistics, residual consistency, spatial coherence, and packet loss risk, forming a weight table containing version number and time decay parameters, and making the weight table available for reconstruction and filtering. The reconstruction module is configured to perform weighted sparse reconstruction on fragmented measurements based on the power grid diagram model and frequency domain basis representation, according to the comprehensive confidence level Ci, to obtain the reconstruction data of the frequency sequence of each node, and generate reconstruction cache and index pointer for regional agent retrieval. The filtering module is configured to fuse asynchronous measurements and the reconstructed data in the form of information on the regional agent side to perform state estimation, maintain information matrix and information vector, and advance [f,ḟ] at a small step frequency, while outputting a residual set for subsequent judgment and weight table update; The causal discrimination module is configured to combine graph coherence and online change point detection to recursively determine the consistency between the residual set and the neighborhood, generate event labels and saliency, and store them together with node identifiers for triggering and reference in the main station parameter settings.

2. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 1, characterized in that: The time acquisition and synchronization module establishes a time reference table, which includes node identifier, correction timestamp, link jitter and version number; it synthesizes latency and jitter by parsing one-way and two-way timestamp messages, and uses the time reference table for unified time axis mapping and subsequent module indexing.

3. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 1, characterized in that: The confidence calculation module decomposes the comprehensive confidence Ci into clock quality, delay statistics, residual consistency, graph space coherence, and packet loss risk. The data are generated from the time base table, round-trip delay statistics, previous residuals, neighborhood consistency and hidden Markov estimation, respectively, and written into the weight table.

4. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 3, characterized in that... ； The weight table includes time decay parameters and effective time; when a new measurement or regional residual write-back arrives, an overwrite or merge update is performed based on the node identifier and version number; the updated entries maintain a one-to-one index relationship for use by the reconstruction module and the filtering module.

5. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 4, characterized in that: The reconstruction module constructs a frequency domain representation based on the Laplace feature basis of the power grid graph, and forms a diagonal weighting with Ci and the sampling matrix Sk; it adopts iterative sparse solution and starts with deheating at the previous time step, limits the maximum iteration and stopping threshold, and outputs the reconstruction buffer and index pointers by node and time step.

6. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 5, characterized in that, The filtering module maintains the information matrix Ω and the information vector ξ, and advances at a fixed small step frequency; it performs incremental updates on the arriving measurement or reconstructed data by scaling Ri with Ci; in cross-regional collaboration, it only exchanges the incremental difference components of the boundary nodes, and sets the threshold for boundary variable changes and the upper limit of the number of rounds.

7. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 6, characterized in that, The causal discrimination module adopts a dual-model online change point detection, maintaining the run-length and posterior of the measurement anomaly model and the power imbalance model respectively; it uses the node residuals and graph coherence scores as observations for recursion, outputs event labels and saliency, and stores them in association with node identifiers.

8. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 1, characterized in that, It also includes a trigger orchestration module, which triggers the event based on a joint indicator of the innovation norm, average comprehensive confidence, and significance, compared with the congestion level threshold; when triggering, it packages the boundary incremental information, event label, and index field, and sets the minimum trigger interval and hysteresis rules.

9. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 1, characterized in that, It also includes a control interface module, which performs boundary consistency on the region estimation on the main station side, generates interface data including frequency, frequency change rate and covariance; sets control thresholds based on event tags and Ci, and distributes the prior and threshold parameter sets to the region and edge.

10. The intelligent power grid frequency balance control system based on a distributed architecture according to claim 9, characterized in that, The message structure between edge agents and regional agents includes node identifier, correction timestamp, version number, incremental information, event label and salience, cache pointer and source identifier; Each field is encapsulated according to a unified encoding and verification scheme and transmitted through a secure channel.

Images