Intelligent reclosing circuit breaker modular protection system

Abstract

The invention discloses an intelligent reclosing circuit breaker modular protection system, which relates to the technical field of circuit breaker intelligent management, and comprises an acquisition module, a judgment module, a reclosing control module and an optimization module. Voltage holes caused by fault tripping and non-fault disturbance are effectively distinguished, and false triggering of reclosing operation is avoided; meanwhile, the optimization module introduces a machine learning algorithm to carry out continuous filing and strategy updating on misjudgment events, and self-adaptive discrimination logic optimization is realized, so that reclosing judgment accuracy and power supply stability of the power grid system under a complex load condition are remarkably improved, and system impact and equipment damage are effectively prevented.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent management of circuit breakers, and in particular to a modular protection system for intelligent reclosing circuit breakers. Background Art

[0002] Intelligent reclosing circuit breaker modular protection refers to the integration of circuit breakers with intelligent reclosing functions into the power system through modular design, which is used to achieve automatic protection and rapid recovery of the circuit. When a circuit fails and trips, the system will intelligently determine the type of fault and automatically reclose to try to restore power supply. At the same time, each functional module (such as measurement and control, communication, and protection) operates independently and cooperates with each other, which improves the reliability, maintenance convenience and flexibility of the system.

[0003] The prior art has the following disadvantages: In industrial application scenarios, voltage holes may occur during frequent switching of large inductive loads (such as arc furnaces). The existing intelligent reclosing system may misjudge the voltage drop as a tripping condition due to the narrow sampling window or imperfect signal filtering, triggering the reclosing command in advance. In addition, this misjudgment may lead to repeated closing without actual power failure, which in turn causes an impact on the parallel power supply system, causing damage to motor equipment, loss of control system, and other problems, significantly reducing the stability of the power supply system and the service life of the equipment. Summary of the invention

[0004] The object of the present invention is to provide an intelligent reclosing circuit breaker modular protection system to solve the deficiencies in the background technology.

[0005] In order to achieve the above object, the present invention provides the following technical solutions: an intelligent reclosing circuit breaker modular protection system, comprising a collection module, a judgment module, a reclosing control module and an optimization module; The acquisition module is used to collect the voltage, current, switch status and load operation parameters in the power system in real time, and pre-process the collected data; A judgment module, configured to comprehensively judge whether the trip is caused by a fault based on the data provided by the acquisition module and on the basis of a first analysis path and a second analysis path, wherein the first analysis path includes analyzing the time domain characteristics of the voltage waveform to judge whether the voltage drop conforms to the transient fault mode, and the second analysis path includes performing a frequency domain analysis on the voltage signal to judge whether the voltage hole is caused by a non-fault disturbance; A reclosing control module, used to perform one or more automatic reclosing operations on the premise that the judgment module determines that the tripping cause is a transient fault and has been eliminated; The optimization module is used to classify and archive tripping events, record misjudgment events, and dynamically optimize the judgment strategy of the judgment module through a machine learning model.

[0006] Preferably, for the first analysis path, after analyzing the time-domain characteristics of the voltage waveform, a voltage sag anomaly index is generated. The generation method is as follows: Continuously collect voltage waveform data from the power system, obtain three-phase voltage signals by equidistant sampling, and record the original voltage sequence for each phase; set a time window with a fixed length or an adaptive length, start from the starting point of the voltage waveform sequence, and intercept the data within the first sliding window as the analysis target; then slide the window point by point along the time axis according to a preset step size until the entire waveform sequence is processed; in each sliding window, calculate the minimum instantaneous value of the voltage curve within the window, compare the minimum voltage value in each window with the rated voltage set by the system, and find the window with the lowest voltage value among all sliding windows as the most severe moment in this sag event; calculate the ratio between the global minimum value and the rated voltage to obtain the voltage sag anomaly index.

[0007] Preferably, for the second analysis path, after performing frequency-domain analysis on the voltage signal, a harmonic energy ratio fluctuation index is generated. The generation method is as follows: Collect continuous voltage waveform data from the power system, apply a window function to the collected voltage signal for edge processing, convert the time-domain voltage signal within each time sequence window into a frequency-domain signal to obtain its amplitude spectrum or power spectrum; divide the frequency-domain data into multiple frequency bands according to the harmonic order, integrate the frequency-domain amplitude or power values within each harmonic frequency band to estimate the total harmonic energy of the frequency band; then calculate the ratio of the energy of each frequency band to the fundamental wave energy to obtain the harmonic energy ratio of each frequency band; repeat the calculation of the energy ratio of each frequency band for multiple time windows in sequence, and record the harmonic energy ratios of each frequency band within each time period to form an energy ratio sequence that changes with time; compare the change amplitude and change direction of the energy ratios of each frequency band in consecutive time windows, and calculate the standard deviation of the energy ratio sequence to obtain the harmonic energy ratio fluctuation index.

[0008] Preferably, the voltage sag anomaly index and the harmonic energy ratio fluctuation index are respectively normalized so that their numerical ranges are both limited to the interval [0, 1]. According to a preset weighting coefficient, the two normalized indices are weighted and summed to form a comprehensive disturbance evaluation index.

[0009] Preferably, the comprehensive disturbance evaluation index is compared with a preset standard threshold to determine the disturbance attribute of the current tripping event. Specifically, when the comprehensive disturbance evaluation index is greater than or equal to the preset standard threshold, it is determined that the event belongs to a voltage cavity or transient anomaly caused by a non-fault disturbance. At this time, the reclosing operation will be delayed or prohibited from being executed, and a disturbance event record will be generated. When the comprehensive disturbance evaluation index is less than the preset standard threshold, it is determined that the tripping event is caused by a real instantaneous fault. At this time, the reclosing logic will enter the standby state and perform one or more automatic reclosing attempts after confirming that the fault condition has been lifted.

[0010] Preferably, the optimization module adopts a graph neural network algorithm, including: Each tripping event is defined as a node in the graph. When constructing the graph structure, edge connections are established according to the similarity between event features to form a tripping event graph G=(V, E, X), where V is the set of event nodes; E is the set of edges, representing the similarity between events; X is the node feature matrix, and each row corresponds to an event. The graph convolutional neural network is used to update the node representation, and the expression is: ; where: is the node representation matrix of the l-th layer, initially X; =A + I is the adjacency matrix with self-loops added; is the corresponding degree matrix; is the trainable weight matrix of the l-th layer; σ is the activation function; Through the output of the last layer of the graph neural network , the tripping event classification is performed, and the output label is obtained, where: = 1: indicates a misjudged event; = 0: indicates that the judgment result is reliable; The model training objective is to minimize the classification loss function, and the expression is: ; where is the true label, is the labeled training set.

[0011] Preferably, according to the classification result, optimization actions are performed: if a node is classified as a misjudged event, extract its corresponding feature dimension, and automatically adjust the weighting factor in the judgment module; adjust the graph neural network weights through the backpropagation of the loss function , to achieve the online learning ability; update the event graph database and gradually construct a complete knowledge graph of tripping misjudgments.

[0012] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. The present invention realizes the accurate identification and intelligent response to tripping events by constructing a multi-level system composed of a collection module, a judgment module, a reclosing control module, and an optimization module. The system introduces a dual-path analysis mechanism of "time domain and frequency domain" in the judgment module, calculates the voltage sag anomaly index and the harmonic energy ratio fluctuation index respectively, and forms a comprehensive disturbance evaluation index through normalization and weighted fusion, which can effectively distinguish the tripping caused by instantaneous faults from the abnormal voltage sags caused by non-fault disturbances, thus avoiding mis-triggering of reclosing operations and improving power supply stability and equipment safety.

[0013] 2. The present invention introduces a graph neural network algorithm through the optimization module to perform structured modeling and misjudgment classification on tripping events, and adaptively adjusts the weighted parameters in the judgment module based on the misjudgment node characteristics. By continuously optimizing the misjudgment recognition accuracy and the strategy threshold setting, the system realizes the dynamic update and knowledge accumulation of the tripping recognition model, has the ability of online learning and strategy self-repair, and significantly improves the robustness, adaptability and intelligence level of the intelligent reclosing system in a high-disturbance industrial environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0015] Figure 1 It is the system module mind map of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0017] Embodiment, please refer to Figure 1 As shown, the intelligent reclosing circuit breaker modular protection system described in this embodiment includes a collection module, a judgment module, a reclosing control module, and an optimization module; The collection module is used to collect the voltage, current, switch status and load operation parameters in the power system in real time, and preprocess the collected data; A judgment module, configured to comprehensively judge whether a trip is caused by a fault based on the data provided by the acquisition module, on the basis of a first analysis path and a second analysis path. The first analysis path includes analyzing the time-domain characteristics of the voltage waveform to judge whether the voltage dip conforms to the instantaneous fault mode, and the second analysis path includes performing a frequency-domain analysis on the voltage signal to judge whether it is a voltage cavity caused by a non-fault disturbance; A reclosing control module, configured to perform one or more automatic reclosing operations on the premise that the judgment module determines that the cause of the trip is an instantaneous fault and has been eliminated; An optimization module, configured to classify and file trip events, record misjudgment events, and dynamically optimize the discrimination strategy of the judgment module through a machine learning model.

[0018] In the present invention, the acquisition module is the data source basis of the entire system, and is used to realize the real-time acquisition, preliminary processing and synchronous transmission of key operating parameters in the power system, so as to ensure that the subsequent judgment module and control module can accurately and stably perform fault identification and decision control.

[0019] The acquisition module includes: a voltage acquisition unit, a current acquisition unit, a switch state detection unit, a load state monitoring unit, a signal processing unit and a synchronous control unit.

[0020] The voltage acquisition unit adopts a high-precision isolated voltage sensor (such as a voltage transformer or a precision resistor voltage divider) to continuously sample the bus voltage of the three-phase power grid line. The sampling frequency can be configured in the range of 2~10 kHz to meet the capture ability for short-time disturbances (such as voltage cavities, transient overvoltages).

[0021] The signal is converted into a digital signal through an analog-to-digital converter (ADC). The precision of the ADC is not less than 12 bits, and the conversion error is better than ±0.5%. The converted data is subjected to transient extraction and waveform shaping by a DSP processing chip, and supports processing logics such as RMS calculation, peak value identification, and change rate calculation.

[0022] The current acquisition unit obtains branch or bus current information through a through-hole current transformer (CT) or an open-loop Hall element, and is especially used to detect the overcurrent amplitude, waveform distortion and unbalance characteristics at the fault moment.

[0023] The sampling frequency is synchronized with the voltage acquisition unit. The acquired current data is filtered by a filter module to eliminate high-frequency interference and sampling jitter, and supports the function of automatically intercepting the peak value of the short-circuit current, providing a complete current curve for subsequent fault judgment.

[0024] This unit supports the auxiliary acquisition of zero-sequence current or differential current for quickly distinguishing between ground faults and phase-to-phase faults.

[0025] The switch status detection unit collects the opening and closing states of the circuit breaker through a non-contact Hall sensor or a travel switch, and is provided with a tripping signal detection channel.

[0026] In some embodiments, in combination with the electromagnetic operating circuit of the circuit breaker, the actual closing of the contacts is judged by sampling the DC voltage, enhancing the reliability of the state discrimination.

[0027] This unit is in real-time linkage with the reclosing control logic to ensure that the reclosing operation is not started when the mechanical state is not ready.

[0028] The load status monitoring unit obtains the operating conditions by communicating with the intelligent terminal on the load side or industrial controllers (such as PLC, DCS systems), including load types (motors, rectifiers, electric arc furnaces, etc.), switching cycles, power fluctuation curves, etc.

[0029] For loads with strong periodic disturbances (such as electric arc furnaces), this system can mark according to the operating mode to assist the judgment module to identify whether it is a "normal disturbance".

[0030] This unit supports multiple industrial communication protocols such as Modbus RTU, CAN, and RS485.

[0031] A signal processing unit is arranged inside the acquisition module, including an anti-aliasing filter, a sample-and-hold circuit, a digital filtering algorithm module, etc.

[0032] In the implementation process, a dual-channel filtering strategy is used: on the one hand, the full-bandwidth signal is retained for transient fault detection, and on the other hand, intermediate-frequency sampling is used to construct a stable trend curve for trend analysis.

[0033] At the same time, the FFT processing function is introduced to extract spectral features (such as high-frequency noise, low-frequency harmonics) as one of the subsequent misjudgment identification bases.

[0034] Synchronization control unit: Each acquisition unit and the system main control module adopt a unified clock source (such as GPS timing or IEEE1588 PTP network timing) to ensure data synchronization between different channels.

[0035] This unit is provided with a timestamp recording mechanism, and each sampling data is attached with a high-precision time mark, which is convenient for retrospective analysis and algorithm training later.

[0036] When any channel detects that the abnormal parameter change exceeds the set threshold (such as voltage drop > 20%, duration > 30 ms), the acquisition module immediately caches and uploads the full-channel data to the main control unit, and starts the judgment module for rapid evaluation.

[0037] It supports caching data within 200 ms of the front and back fault windows continuously to form a complete waveform map, improving the accuracy of subsequent analysis.

[0038] In the present invention, the judgment module is the core unit for realizing fault identification and reclosing condition determination. It is responsible for diagnosing the tripping event from different analysis dimensions based on the multi-source data provided by the acquisition module to determine whether it is a recoverable instantaneous fault or a non-fault disturbance that does not require reclosing, ensuring that the reclosing operation is only executed under appropriate conditions and avoiding misclosing.

[0039] Among them, the time-domain feature analysis unit (the first analysis path) is used to analyze the time-domain features of the voltage signal to determine whether the voltage dip conforms to the typical instantaneous fault (such as lightning strike, tree fault flashover) mode.

[0040] After analyzing the time-domain features of the voltage waveform, a voltage dip anomaly index is generated. The generation method is as follows: Continuously collect voltage waveform data from the power system. Usually, three-phase voltage signals are obtained by equidistant sampling, and the original voltage sequence is recorded for each phase.

[0041] Set a time window with a fixed length or an adaptive length (such as 20 ms, 40 ms or dynamically adjusted according to the power grid frequency) for local voltage waveform analysis. The window sliding step size can be set as the interval between one sampling point or multiple sampling points.

[0042] Starting from the starting point of the voltage waveform sequence, intercept the data within the first sliding window as the analysis target; then slide the window point by point along the time axis according to the preset step size until the entire waveform sequence is processed.

[0043] In each sliding window, calculate the minimum instantaneous value of the voltage curve within the window. This value represents the local lowest level of the voltage during this time period.

[0044] Compare the minimum voltage value in each window with the rated voltage set by the system as the voltage dip evaluation benchmark for this window; At the same time, record the time point and its duration when the minimum value occurs for subsequent dip event positioning and statistics.

[0045] Among all the sliding windows, find the window with the lowest voltage value as the most severe moment in this dip event; the minimum value extracted from this window is used to calculate the severity of this dip.

[0046] Calculate the ratio between the global minimum value and the rated voltage to obtain the voltage dip anomaly index; this value can be used to determine whether it belongs to a recoverable instantaneous dip or whether the reclosing protection logic needs to be triggered.

[0047] The voltage sag anomaly index is usually defined as the ratio of the minimum voltage value during the voltage sag process to the rated voltage value, and its numerical range is usually between 0 and 1. The smaller the index, the more severe the voltage sag, that is, the minimum voltage is much lower than the rated value; the larger the index, the smaller the voltage fluctuation and the lighter the sag degree.

[0048] In the instantaneous fault mode, it is usually manifested as the voltage rapidly dropping to a lower level (smaller index) within a very short time and quickly recovering within dozens of milliseconds. Therefore, a severe but short-lived drop (low index + short duration) is more in line with the characteristics of an instantaneous fault; if the drop index is larger or the duration is longer, it is more likely to be caused by load disturbances, voltage fluctuations, or permanent faults, and further analysis and judgment are required.

[0049] Among them, the frequency domain and disturbance behavior analysis unit (the second analysis path). This unit judges possible non-fault factors such as voltage holes, system disturbances, and harmonic injections by using frequency domain analysis and system context behavior, specifically including: After performing frequency domain analysis on the voltage signal, a harmonic energy ratio fluctuation index is generated. The generation method is as follows: Collect continuous voltage waveform data from the power system, usually a sample with a window duration of several cycles (such as 200 ms or 10 power frequency cycles), and multiple time series windows can be formed by segmented sampling.

[0050] Apply a window function (such as a Hanning window or a Blackman window) to the collected voltage signal for edge processing to reduce the interference of spectral leakage on the calculation results of high-order harmonics.

[0051] Convert the time-domain voltage signal within each time series window into a frequency-domain signal to obtain its amplitude spectrum or power spectrum; the frequency-domain components should cover at least the fundamental wave to the 50th harmonic frequency to adapt to the harmonic range of common industrial applications.

[0052] Divide the frequency domain data into multiple frequency bands according to the harmonic order. For example: Band 1 (fundamental wave): 1st order; Band 2 (low-order harmonics): 2 - 5th order; Band 3 (medium-order harmonics): 6 - 15th order; Band 4 (high-order harmonics): 16 - 30th order; Band 5 (ultra-high-order harmonics): 31 - 50th order. This division can be customized according to the application scenario.

[0053] Integrate (i.e., sum) the frequency domain amplitude or power values within each harmonic frequency band to estimate the total harmonic energy of the frequency band; then calculate the ratio of the energy of each frequency band to the fundamental wave energy to obtain the harmonic energy ratio of each frequency band.

[0054] Repeat the calculation of the energy proportion of each frequency band for multiple time windows in sequence, and record the harmonic energy ratios of each frequency band within each time period to form an energy ratio sequence that changes with time; Compare the change amplitude and direction of the energy ratio of each frequency band in consecutive time windows, calculate the standard deviation of the energy ratio sequence, and obtain the harmonic energy ratio fluctuation index. This index can be used to judge whether there is an increasing trend of harmonic pollution caused by non-fault disturbances (such as the start-up of electric arc furnaces, switching of capacitor banks, etc.).

[0055] The harmonic energy ratio fluctuation index reflects the fluctuation trend of the non-fundamental components (especially medium and high-order harmonics) in the voltage signal within a certain time range. The larger the index, the more continuous the fluctuation of harmonic energy, which is usually related to the operation or switching of high-power nonlinear loads (such as electric arc furnaces, frequency converters) in the power system, indicating that the system may be affected by non-fault disturbances, thereby causing voltage sags.

[0056] On the contrary, if the fluctuation index is small or basically stable, it indicates that the harmonic disturbance in the voltage waveform is not obvious, and the voltage drop is more likely to be caused by actual faults such as sudden physical short circuits and groundings. At this time, more attention should be paid to the operation of protective reclosing. Therefore, the harmonic energy ratio fluctuation index can be used as a key basis for assisting in judging whether the source of the voltage sag is a disturbance or a fault.

[0057] In the present invention, the voltage drop anomaly index and the harmonic energy ratio fluctuation index are respectively normalized so that their numerical ranges are both limited to the interval [0, 1]. The normalization process can adopt methods such as linear mapping, maximum-minimum scaling, or Z-score to standardize the original features into a dimensionless unified expression, which is convenient for fusion analysis between physical quantities of different scales.

[0058] The normalized voltage drop anomaly index reflects the severity of the instantaneous voltage drop. The smaller the value, the deeper the drop, and vice versa, indicating that the voltage is relatively stable; the normalized harmonic energy ratio fluctuation index represents the change intensity of the frequency-domain disturbance, and the larger the value, the more significant the harmonic disturbance.

[0059] The system weights and sums the two normalized indices according to a preset weighting coefficient to form a comprehensive disturbance evaluation index for judging the nature of voltage anomaly events. The weighting coefficient can be configured according to the actual application scenario. For example, the weight of harmonic disturbance is higher in the industrial load scenario, while the weight of voltage drop is greater in the transmission line protection scenario. This comprehensive index, as the fusion decision-making basis of the judgment module, can effectively improve the system's ability to accurately identify the type of tripping in a complex power grid environment.

[0060] After completing the normalization of the voltage drop anomaly index and the harmonic energy ratio fluctuation index and obtaining the comprehensive disturbance evaluation index through weighted calculation, the system compares this comprehensive index with a preset standard threshold to determine the disturbance attribute of the current tripping event.

[0061] When the comprehensive disturbance evaluation index is greater than or equal to a preset standard threshold, the system determines that the event has strong harmonic disturbance characteristics and voltage fluctuation characteristics, belonging to voltage holes or transient abnormalities caused by non-fault disturbances. At this time, the reclosing operation will be delayed or prohibited, and a disturbance event record will be generated for subsequent analysis by the optimization module.

[0062] When the comprehensive disturbance evaluation index is less than the preset standard threshold, the system determines that the tripping event is more likely caused by a real instantaneous fault, such as line short circuit, flashover, etc. At this time, the reclosing logic will enter the standby state, and one or more automatic reclosing attempts will be executed after confirming that the fault condition has been eliminated to achieve rapid power supply restoration.

[0063] Through this comparison and judgment mechanism, the system can effectively distinguish between disturbance-type tripping and fault-type tripping, avoiding unnecessary closing shocks or protection refusal due to misjudgment, thereby improving the response accuracy and operation stability of the intelligent reclosing system.

[0064] In the present invention, the reclosing control module, as the execution unit for fault handling and power supply restoration, is used to execute one or more automatic reclosing operations according to the set strategy on the premise that the judgment module determines that the tripping event is an instantaneous fault and the fault condition has been eliminated, so as to achieve the rapid self-healing of the system and ensure power supply continuity.

[0065] This module mainly includes the following functional units: Closing logic trigger unit, reclosing times control unit, fault redetection and confirmation unit, closing execution unit, safety protection interlock unit, and status feedback and recording unit.

[0066] The closing logic trigger unit receives the "reclosable" signal output by the judgment module, and this signal needs to meet the following conditions: the tripping reason is an instantaneous fault (such as flashover, lightning strike); the fault has been naturally eliminated within the set time window; the system voltage and current have returned to the normal range.

[0067] Once the above conditions are met, this unit will activate the reclosing control logic and enter the standby closing state.

[0068] The reclosing times control unit is used to prevent secondary faults or mechanical losses caused by repeated reclosing. This unit has the following adjustable parameters: The maximum number of reclosings (usually 1 - 3 times); the reclosing interval time (such as 300 - 500 ms); the failure locking strategy (such as prohibiting reclosing after consecutive failures); the system automatically delays the next attempt after each reclosing failure until the maximum number of attempts is reached or successful closing occurs.

[0069] The fault re-detection and confirmation unit quickly detects the line status before each closing attempt: checks for the presence of short-circuit current, ground current, or abnormal voltage; if a fault is still detected, the reclosing operation is immediately aborted, and a "reclosing failure" flag is sent to the system; if the detection result is normal, the closing process is allowed to continue.

[0070] The closing execution unit sends a closing drive signal to the circuit breaker to control the electromagnetic operating device or the electric mechanism to complete the actual closing operation.

[0071] Meanwhile, the action feedback signal is monitored: the closing success signal, the contact in-place feedback, and the spring energy storage status. If the closing fails, this unit will report the failure status to the reclosing times control unit, initiate the next round of reclosing, or issue a locking command.

[0072] The safety protection and interlock unit is provided with a safety lock control mechanism to prevent misclosing under the following abnormal conditions: insufficient operating power supply voltage; abnormal control circuit (such as control fuse blown); presence of ground or reverse current signal; prohibited closing command issued by the background.

[0073] Once any of the above situations is detected, the system will forcibly interrupt the current reclosing instruction and record the fault event.

[0074] The status feedback and recording unit will record in real time information such as the execution process, result (success / failure), timestamp, and number of attempts of each reclosing action, and upload it to the communication module or local storage. In addition, this unit can also generate status codes or alarm information to prompt on-site or remote personnel about the system operation status.

[0075] In the present invention, the optimization module is used to perform data-driven intelligent classification and discriminant strategy adaptive optimization on tripping events. To overcome the problems of strong dependence on the feature space and high requirements for sample annotation in traditional machine learning, this embodiment adopts the graph neural network algorithm to model the tripping event data as a multi-dimensional graph structure, realizing structured learning and strategy update between fault and disturbance patterns.

[0076] The implementation steps of the graph neural network optimization method include: Each tripping event is defined as a node in the graph, and its features include: voltage sag anomaly index (normalized); harmonic energy ratio increasing index (normalized); tripping time, duration; load type encoding; tripping determination result (initially the output result of the judgment module).

[0077] When constructing the graph structure, edge connections are established according to the similarity between event features (such as Euclidean distance, cosine similarity, etc.) to form the tripping event graph G=(V, E, X), where V is the set of event nodes; E is the set of edges, representing the similarity between events; X is the node feature matrix, with each row corresponding to an event.

[0078] The node representation is updated using a Graph Convolutional Neural Network (GCN), and the expression is as follows: where: is the node representation matrix of the l-th layer, initially X; = A + I is the adjacency matrix with self-loops added; is the corresponding degree matrix; is the trainable weight matrix of the l-th layer; σ is the activation function (such as ReLU); The output of the last layer of the graph neural network is used for tripping event classification, and the output label is obtained, where: = 1: indicates a misjudged event (e.g., a non-fault disturbance is misjudged as a fault); = 0: indicates a reliable judgment result; The training objective of the model is to minimize the classification loss function, and the expression is: where is the true label, is the labeled training set.

[0079] According to the classification result, the system performs the following optimization actions: When the graph neural network classifies a tripping event node as a "misjudged event" (i.e., the judgment module misjudges a non-fault disturbance as an instantaneous fault), the system immediately locates the feature vector information carried by this node, including but not limited to: voltage sag anomaly index; harmonic energy ratio fluctuation index; tripping duration; load disturbance background information; weighted parameters used in the judgment. By analyzing the feature dimensions of this node, the system identifies the key parameter combinations that lead to misjudgment.

[0080] The system dynamically adjusts the following discriminant logic parameters in the judgment module according to the misjudgment feature pattern: Increase the judgment threshold of the voltage sag anomaly index to avoid triggering a "fault" judgment for minor sags; decrease the sensitivity weight of harmonic disturbances so that the system pays more attention to the short-time sag characteristics in the time domain; adjust the judgment boundary of the comprehensive disturbance evaluation index to achieve a classification judgment more suitable for the current system environment; update the weight factor combination table of each parameter for subsequent matching of similar events and experience reuse. This adjustment can be driven by an empirical strategy or guided and optimized by the weights output by the GNN.

[0081] The training process of the graph neural network adopts a supervised learning mechanism to optimize the loss function of the network based on the labeled misjudged event samples. When a new misjudged event is identified, its label will be used to update the network weights in real time. The specific steps are as follows: Calculate the classification error between the current network prediction value and the true label; Use the backpropagation algorithm to perform gradient updates on the weight parameters of each layer of the graph neural network; Incorporate this new sample into the training set to improve the discriminative accuracy of the model for future similar events.

[0082] This online learning ability ensures that the system continuously improves its ability to identify misjudgment boundaries during long-term operation.

[0083] The system incorporates all trip event nodes identified as misjudgments into the trip misjudgment atlas database and records: the original waveform and spectrum data; the judgment result and the GNN classification result; the parameter combination and label; the optimized system policy value. As event samples continue to accumulate, this atlas can be used to periodically retrain the graph neural network model to achieve experience precipitation and policy migration, and construct a more adaptable "knowledge graph".

[0084] The optimization module periodically or in an event-triggered manner feeds back the learning results to the judgment module, enabling the system to gradually establish an adaptive trip judgment model during long-term operation, thereby improving the misjudgment recognition ability and the overall stability of the system.

[0085] The above formulas are all dimensionless and take their numerical calculations. The formula is obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0086] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that includes one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0087] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. An intelligent reclosing circuit breaker modular protection system, characterized in that: It includes a data acquisition module, a judgment module, a reclosing control module, and an optimization module; The data acquisition module is used to continuously collect the voltage, current, switch status, and load operation parameters in the power system, and preprocess the collected data; The judgment module is used to comprehensively judge whether the trip is caused by a fault based on the data provided by the data acquisition module. The first analysis path includes analyzing the time-domain characteristics of the voltage waveform to judge whether the voltage sag conforms to the instantaneous fault mode, and the second analysis path includes performing frequency-domain analysis on the voltage signal to judge whether it is a voltage hole caused by non-fault disturbance; The reclosing control module is used to perform one or more automatic reclosing operations on the premise that the judgment module determines that the trip cause is an instantaneous fault and has been eliminated; The optimization module is used to classify and file the trip events, record the misjudgment events, and dynamically optimize the discrimination strategy of the judgment module through a machine learning model.

2. The modular protection system of an intelligent reclosing circuit breaker according to claim 1, wherein: The first analysis path generates a voltage sag anomaly index after analyzing the time-domain characteristics of the voltage waveform. The generation method is as follows: Continuously collect voltage waveform data from the power system, obtain three-phase voltage signals by equidistant sampling, and record the original voltage sequence for each phase; set a time window with a fixed length or an adaptive length, and start from the starting point of the voltage waveform sequence, intercept the data within the first sliding window as the analysis target; then slide the window point by point along the time axis according to the preset step size until the entire waveform sequence is processed; in each sliding window, calculate the minimum instantaneous value of the voltage curve within the window, compare the minimum voltage value in each window with the rated voltage set by the system, and find the window with the lowest voltage value among all sliding windows as the most severe moment in this sag event; Calculate the ratio between the global minimum value and the rated voltage to obtain the voltage sag anomaly index.

3. The modular protection system of an intelligent reclosing circuit breaker according to claim 2, characterized in that: The second analysis path generates a harmonic energy ratio fluctuation index after performing frequency-domain analysis on the voltage signal. The generation method is as follows: Collect continuous voltage waveform data from the power system, apply a window function to the collected voltage signal for edge processing, convert the time-domain voltage signal within each time series window into a frequency-domain signal to obtain its amplitude spectrum or power spectrum; divide the frequency-domain data into multiple frequency bands according to the harmonic order, integrate the frequency-domain amplitude or power values within each harmonic frequency band to estimate the total harmonic energy of the frequency band; then calculate the ratio of the energy of each frequency band to the fundamental wave energy to obtain the harmonic energy ratio of each frequency band; Repeat the calculation of the energy ratio of each frequency band for multiple time windows in sequence, and record the harmonic energy ratio of each frequency band within each time period to form an energy ratio sequence that changes with time; Compare the change amplitude and change direction of the energy ratio of each frequency band in consecutive time windows, and calculate the standard deviation of the energy ratio sequence to obtain the harmonic energy ratio fluctuation index.

4. An intelligent reclosing circuit breaker modular protection system according to claim 3, characterized in that: The voltage dip anomaly index and the harmonic energy ratio fluctuation index are respectively normalized so that their numerical ranges are both limited to the interval [0, 1]. According to the preset weighting coefficient, the two normalized indices are weighted and summed to form a comprehensive disturbance evaluation index.

5. An intelligent reclosing circuit breaker modular protection system according to claim 4, characterized in that: The comprehensive disturbance evaluation index is compared with the preset standard threshold to determine the disturbance attribute of the current tripping event. Specifically, when the comprehensive disturbance evaluation index is greater than or equal to the preset standard threshold, it is determined that the event belongs to a voltage cavity or transient anomaly caused by a non-fault disturbance. At this time, the reclosing operation will be delayed or prohibited, and a disturbance event record will be generated. When the comprehensive disturbance evaluation index is less than the preset standard threshold, it is determined that the tripping event is caused by a real instantaneous fault. At this time, the reclosing logic will enter the standby state, and one or more automatic reclosing attempts will be performed after confirming that the fault condition has been eliminated.

6. The modular protection system of an intelligent reclosing circuit breaker according to claim 5, characterized in that: The optimization module adopts a graph neural network algorithm, including: Each tripping event is defined as a node in the graph. When constructing the graph structure, edge connections are established according to the similarity between event features to form a tripping event graph G=(V, E, X), where V is the set of event nodes; E is the set of edges, representing the similarity between events; X is the node feature matrix, and each row corresponds to an event. The node representation is updated using a graph convolutional neural network, and the expression is: where: is the node representation matrix of the l-th layer, initially X; = A + I is the adjacency matrix with self-loops added; is the corresponding degree matrix; is the trainable weight matrix of the l-th layer; σ is the activation function; Output of the last layer of the graph neural network , perform tripping event classification and output labels , where: = 1: indicates a misjudgment event; = 0: indicates that the judgment result is reliable; The model training objective is to minimize the classification loss function, and the expression is: ; where is the true label, is the labeled training set.

7. An intelligent reclosing circuit breaker modular protection system according to claim 6, characterized in that: According to the classification results, perform optimization actions: if a node is classified as a misjudgment event, extract its corresponding feature dimension, and automatically adjust the weighting factor in the judgment module; adjust the weights of the graph neural network through the backpropagation of the loss function , to achieve the ability of online learning; update the event graph database and gradually construct a complete knowledge graph of tripping misjudgment.

Images