An air switch control method, system, device and medium with intelligent anti-misjump and conditional automatic reclosing functions

Abstract

This invention discloses a control method, system, device, and medium for air switches with intelligent anti-false tripping and conditional automatic reclosing functions. Belonging to the field of air switch control technology, the invention includes: collecting multi-source heterogeneous data from the air switch; processing the collected data to obtain a fused feature vector and quality score; performing feature extraction and fusion operations on the fused feature vector to obtain a dimensionality-reduced fused state feature vector; inputting the dimensionality-reduced fused state feature vector into a discrimination model to determine the tripping cause, and performing a reclosing operation after determining the delay time according to a delay mechanism; performing multi-source verification before executing the reclosing operation; initiating a delayed retry mechanism when the reclosing conditions are not yet met; and entering a blocking state when the number of consecutive retries reaches a threshold and reclosing is still unsuccessful; and recording the entire process data to construct a knowledge base. This invention significantly reduces the risk of false closing and improves the safety and continuity of power restoration.

Description

Technical Field

[0001] This invention relates to the field of air switch control technology, specifically to an air switch control method, system, device, and medium with intelligent anti-false tripping and conditional automatic reclosing functions. Background Technology

[0002] In the construction of substations and distribution automation systems, the intelligent transformation of air switches has become an important direction for improving operational reliability. Existing solutions mainly achieve remote operation and status monitoring by adding remote control modules, communication interfaces, or reclosing functions, which to some extent reduces the burden of manual operation and improves power supply continuity.

[0003] However, existing technologies still have the following shortcomings when applied to secondary circuit air switches: First, there is a lack of accurate identification of the causes of false trips. Most existing devices can only distinguish whether a trip has occurred based on the trip signal or simple status variables, and cannot further determine whether the trip is caused by human operation, actual fault, or non-faulty false trip. This leads to situations where, in false trip scenarios, either the system cannot automatically restore power supply, or blind reclosing is performed without eliminating the risks, thereby causing damage to secondary equipment or malfunction of protection systems.

[0004] Second, the conditional reclosing logic is incomplete. Traditional reclosing functions mostly rely on fixed delay and number of reclosing operations, lacking differentiated strategies that take into account the characteristics of the secondary circuit. Without combining multi-dimensional conditions such as alarm contacts, circuit status, voltage, and current for judgment, simple reclosing often carries potential risks, failing to effectively prevent false closing and making it difficult to promptly recover from interruptions caused by false tripping.

[0005] Third, there is insufficient utilization of multi-source states. Existing devices still mainly rely on single state quantities for signal acquisition, lacking joint analysis of information such as alarm contacts, circuit electrical quantities, mechanical positions, and operation records, which can easily lead to inaccurate judgments or delayed responses.

[0006] Fourth, the system's collaborative capabilities are limited. Although some products can connect with one-click sequential control and remote monitoring systems, they often lack rapid local interlocking and self-recovery mechanisms in false trip scenarios, still requiring manual intervention, which affects the overall safety and timeliness of operation.

[0007] In summary, although existing technologies have made some progress in remote operation and automatic reclosing, they still generally suffer from problems such as inaccurate false trip detection, insufficient conditional reclosing strategies, inadequate utilization of multi-source information, and imperfect coordination with the system. They are still unable to meet the requirements of high reliability and high security for secondary circuit operation. Summary of the Invention

[0008] In view of the above-mentioned problems, the present invention provides an air switch control method, system, device and medium with intelligent anti-false tripping and conditional automatic reclosing functions.

[0009] Therefore, the technical problem solved by this invention is: how to solve the problems that the existing technology still generally has, such as inaccurate false jump judgment, insufficient conditional coincidence strategy, insufficient utilization of multi-source information and imperfect coordination with the system, and cannot meet the requirements of high reliability and high security for secondary circuit operation.

[0010] To address the aforementioned technical problems, this invention provides the following technical solution: a control method for an air switch with intelligent anti-misoperation and conditional automatic reclosing functions, comprising: collecting multi-source heterogeneous data from the air switch; performing data processing operations on the collected data to obtain a fused feature vector and a quality score; performing feature extraction and feature fusion operations on the fused feature vector to obtain a dimensionality-reduced fused state feature vector; inputting the dimensionality-reduced fused state feature vector into a discriminant model constructed based on ensemble learning and attention mechanisms to determine the tripping cause, obtaining a probability distribution of the tripping type; determining whether to activate a conservative strategy based on the quality score; performing reclosing determination based on the probability distribution and constraints, and executing the reclosing operation after determining the delay time according to the delay mechanism; performing multi-source verification operations before executing the reclosing operation; initiating a delay retry mechanism when the reclosing conditions are not yet met; entering a blocking state when the number of consecutive retries reaches a threshold and reclosing is still unsuccessful; recording the entire process data to construct a knowledge base; performing cross-device collaborative operations based on the knowledge base; and optimizing and updating the strategy parameters.

[0011] As a preferred embodiment of the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions described in this invention, the method involves: collecting multi-source heterogeneous data from the air switch and performing data processing operations on the collected data to obtain a fused feature vector and a quality score value. This includes: performing field standardization and time synchronization processing on the collected multi-source heterogeneous data to obtain a standardized feature vector; performing denoising processing and feature extraction on the standardized feature vector to obtain multi-scale features; performing anomaly detection and missing value completion processing on the multi-scale features; and performing quality scoring on the processed data to obtain a fused feature vector and a quality score value.

[0012] As a preferred embodiment of the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions described in this invention, the step of performing feature extraction and feature fusion operations on the fused feature vector to obtain a dimensionality-reduced fused state feature vector includes: performing multimodal signal analysis on the fused feature vector to extract time-domain and frequency-domain features; constructing an event sequence network to extract event features; performing feature modulation on environmental parameters and trajectory modeling on historical data; using a gated attention fusion mechanism to dynamically weight and fuse various features to obtain a fused state feature vector; and performing supervised dimensionality reduction processing on the fused state feature vector.

[0013] As a preferred embodiment of the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions described in this invention, the following steps are included: inputting the dimensionality-reduced fused state feature vector into a discriminant model constructed based on ensemble learning and attention mechanisms to determine the tripping cause, obtaining the probability distribution of the tripping type, determining whether to enable a conservative strategy based on the quality score, determining reclosing based on the probability distribution and constraints, and performing the reclosing operation after determining the delay time according to the delay mechanism. This includes: constructing a discriminant model containing a feature recalibration module and a multi-expert ensemble classification module; inputting the dimensionality-reduced fused state feature vector into the discriminant model and outputting the probability distribution of the tripping type; determining whether to enable a conservative strategy based on the quality score; determining the tripping type based on the probability distribution and performing the reclosing operation in conjunction with electrical constraints, fault signal constraints, event logic constraints, and equipment health constraints; and performing the reclosing operation after determining the dynamic delay time based on the risk score.

[0014] As a preferred embodiment of the air switch control method with intelligent anti-misoperation and conditional automatic reclosing functions described in this invention, the step of anomaly detection and missing value completion processing for multi-scale features includes: using the absolute deviation of the median for anomaly detection, calculating the deviation ratio of each feature value in the multi-scale features to the median, and marking it as an anomaly when the deviation ratio exceeds the anomaly threshold; using time-series interpolation to complete short missing segments in the multi-scale features; using similarity-weighted filling based on proximity terminals to complete long missing segments in the multi-scale features, where the similarity weight is calculated based on the power grid topology distance and model difference; and labeling the confidence level of key signals in the completed multi-scale features.

[0015] This preferred scheme uses the median absolute deviation for anomaly detection, which has a stronger ability to resist extreme value interference. It adopts a hierarchical completion strategy for missing segments of different lengths and calculates similarity weights by combining the power grid topology distance and model difference, so that the completed data is closer to the actual operating state. The confidence labeling provides a basis for data credibility for subsequent decision-making to reduce the risk of misjudgment.

[0016] As a preferred embodiment of the air switch control method with intelligent anti-misoperation and conditional automatic reclosing functions described in this invention, the step of performing multimodal signal analysis on the fused feature vector to extract time-domain and frequency-domain features includes: performing time-frequency analysis on the voltage and current waveforms in the fused feature vector using an adaptive window-length short-time Fourier transform; dynamically adjusting the window length of the window function according to the instantaneous frequency gradient of the signal; extracting the singular value decomposition features of the time-frequency matrix based on the time-frequency representation as descriptors for steady-state and transient behavior; extracting pulse duration, zero-crossing rate, and envelope exponent from transient events in the fused feature vector; and extracting action delay and rebound count from the switch contact position in the fused feature vector to obtain electrical features, event features, environmental features, and historical features.

[0017] This preferred solution employs an adaptive window-length short-time Fourier transform to dynamically adjust the time-frequency resolution based on signal frequency changes. Combined with multi-dimensional extraction of singular value decomposition features, transient event features, and mechanical action features, it can effectively distinguish between electrical faults, mechanical anomalies, and human error, providing reliable feature support for accurately identifying the cause of tripping.

[0018] As a preferred embodiment of the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions described in this invention, the following steps are included: Constructing a discriminant model comprising a feature recalibration module and a multi-expert integrated classification module includes: in the feature recalibration module, performing feature recalibration on the dimensionality-reduced fused state feature vector based on attention weights; in the multi-expert integrated classification module, setting up multiple expert sub-classifier networks; dynamically generating the integration weights of each expert sub-classifier based on the current features through a gating network; and weighting and integrating the outputs of each expert sub-classifier to obtain the discriminant model; the step of determining the tripping type based on probability distribution and performing reclosing determination in conjunction with electrical constraints, fault signal constraints, event logic constraints, and equipment health constraints includes determining whether the probability of non-fault false tripping is greater than... The process involves several steps: determining whether the voltage and current are within a safe and stable range and whether the root mean square error is below a threshold; determining whether there is a continuous fault alarm signal; determining whether there are manual interlocks or maintenance status tags in the operation event sequence; determining whether the equipment health score derived from historical features meets the health threshold; and outputting a result allowing overlap when all conditions are met. The process of determining the dynamic delay time based on the risk score and then performing the overlap operation includes: calculating the risk score by comprehensively considering the current electrical stability, environmental disturbance intensity, and historical malfunction frequency; adding the base delay time to the gain adjustment of the risk score to obtain the dynamic delay time; performing a real-time secondary verification before performing the overlap operation after the dynamic delay time ends to detect the presence of temporary fault signals; and performing the overlap operation after the real-time secondary verification passes.

[0019] This preferred solution improves the accuracy and generalization ability of trip type identification through feature recalibration using an attention mechanism and multi-expert integrated discrimination. It avoids the one-sidedness of single-condition judgment by combining multi-dimensional cross-validation of electrical constraints, fault signal constraints, event logic constraints and equipment health constraints. It dynamically adjusts the delay time through risk scoring and performs real-time secondary verification to achieve rapid self-recovery while ensuring safety.

[0020] This invention provides an air switch control system with intelligent anti-false tripping and conditional automatic reclosing functions.

[0021] To address the aforementioned technical problems, this invention provides the following technical solution: an air switch control system with intelligent anti-misoperation and conditional automatic reclosing functions, comprising: a data acquisition and processing module, a feature extraction and fusion module, a tripping discrimination and decision module, a multi-source verification module, and a knowledge base optimization module; the data acquisition and processing module is used to acquire multi-source heterogeneous data from the air switch, perform data processing operations on the acquired data, and obtain a fused feature vector and a quality score value; the feature extraction and fusion module is used to perform feature extraction and feature fusion operations on the fused feature vector to obtain a dimensionality-reduced fused state feature vector; the tripping discrimination and decision module is used to process the dimensionality-reduced fused state feature vector... The vector input uses a discriminative model built based on ensemble learning and attention mechanisms to determine the cause of the trip, obtaining a probability distribution of the trip type. Based on the quality score, it determines whether to activate a conservative strategy. Based on the probability distribution and constraints, it performs a re-application determination and executes the re-application operation after determining the delay time according to a delay mechanism. The multi-source verification module performs multi-source verification before executing the re-application operation. If the re-application conditions are not yet met, a delay retry mechanism is activated. If the number of consecutive retries reaches a threshold and re-application is still unsuccessful, a locking state is entered. The knowledge base optimization module records the entire process data to build a knowledge base, performs cross-device collaborative operations based on the knowledge base, and optimizes and updates the strategy parameters.

[0022] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions.

[0023] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the air switch control method having intelligent anti-false tripping and conditional automatic reclosing functions.

[0024] The beneficial effects of this invention are as follows: This invention constructs an attention-enhanced integrated learning discrimination model by standardizing and fusing heterogeneous information from multiple sources, such as operating signals, electrical waveforms, alarm contacts, mechanical positions, and environmental disturbances. Compared with traditional methods that rely on a single electrical quantity or a single contact signal, this invention can more accurately distinguish between three scenarios: human operation, fault tripping, and non-faulty false tripping, significantly improving the accuracy and robustness of tripping cause determination.

[0025] This invention introduces a multi-constraint reclosing mechanism, combining constraints from multiple dimensions such as electrical steady-state performance, fault blocking signals, event logic consistency, and equipment health. Reclosing is only performed when safety conditions are met. Simultaneously, through adaptive delay and risk scoring functions, dynamic risk perception and response to different operating conditions are achieved. Compared to traditional fixed-delay or single-threshold reclosing methods, this invention significantly reduces the risk of erroneous closing and improves the safety and continuity of power restoration.

[0026] Before the overlap operation, cross-verification using multiple sources of signals, including alarm contact verification, steady-state review of electrical quantities, and confirmation of mechanical position, ensures consistency in decision-making across multiple parties. By combining a confidence-weighted mechanism with a progressive interlocking strategy, erroneous overlap and repeated impacts can be effectively avoided in cases of sensor anomalies, complex electromagnetic environments, or permanent faults. Compared to existing single-decision control methods, this invention offers significant advantages in reliability and safety protection.

[0027] All judgment and operation processes are recorded and incorporated into a knowledge base. The system continuously optimizes threshold settings, delay strategies, and risk models based on historical operating samples, achieving adaptive evolution. Compared to traditional static parameter configuration methods, this invention possesses long-term self-learning and dynamic optimization capabilities, enabling it to continuously adapt to changes in the power grid operating environment and ensuring that the strategy remains efficient and reasonable at all times.

[0028] When multiple circuit breakers in the same area experience consecutive false trips or abnormal reclosing, this invention can trigger regional-level coordinated control, preventing local policy conflicts from expanding the fault range. This mechanism overcomes the limitations of existing methods that only focus on the self-recovery of individual devices, further enhancing the overall safety and self-healing capability of the power grid secondary system. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1The above is a flowchart of an air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions, provided as an embodiment of the present invention. Detailed Implementation

[0031] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0032] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides an air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions, including: S1. Collect multi-source heterogeneous data from air switches, perform data processing operations on the collected data, and obtain fused feature vectors and quality scores.

[0033] S2. Perform feature extraction and feature fusion operations on the fused feature vector to obtain the dimensionality-reduced fused state feature vector.

[0034] S3. Input the dimensionality-reduced fusion state feature vector into the discrimination model built on ensemble learning and attention mechanism to determine the cause of the trip, obtain the probability distribution of the trip type, determine whether to enable the conservative strategy based on the quality score, make the overlap judgment based on the probability distribution and the constraint conditions, and perform the overlap operation after determining the delay time according to the delay mechanism.

[0035] S4. Before performing the overlap operation, perform multi-source verification. If the overlap condition is not met, start the delayed retry mechanism. If the number of consecutive retries reaches the threshold and overlap is still not successful, enter the blocking state.

[0036] S5. Record data throughout the entire process to build a knowledge base, perform cross-device collaborative operations based on the knowledge base, and optimize and update strategy parameters.

[0037] It should be noted that during operation, the secondary circuit of an air switch is subject to problems such as asynchronous timing, inconsistent dimensions, noise interference, and missing outliers in the multi-source heterogeneous data, including operation signals, trip contact signals, voltage and current waveforms, and mechanical position signals. Directly using unprocessed data will reduce the accuracy and robustness of the discrimination and control algorithms. At the same time, existing methods have insufficient discrimination accuracy, lack flexibility in reclosing strategies, and are not adequately guaranteed for safety when facing complex power grid conditions. They are difficult to accurately distinguish between human operation, fault tripping, and non-fault false tripping, leading to the risk of false reclosing or delay in power restoration.

[0038] Therefore, to address the aforementioned data processing and intelligent decision-making issues, steps S1-S5 are used to standardize, denoise, and label the multi-source heterogeneous data to achieve high-quality feature input; a state representation with strong discriminative capabilities is constructed through a gated attention fusion mechanism and supervised dimensionality reduction; accurate tripping cause identification and safe automatic re-coincidence decision-making are achieved by integrating a gated attention network and multi-constraint conditional overlap determination; robustness in complex environments is improved through multi-source cross-validation and adaptive delay retry mechanisms; and continuous improvement of intelligent control is achieved through full-process recording and cross-device collaborative optimization.

[0039] Example 2, an embodiment of the present invention, provides an air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions based on the previous embodiment, including: This invention proposes a control method for air switches with intelligent anti-false tripping and conditional automatic reclosing functions. Combining trip cause identification, conditional reclosing strategy, and multi-source state dual confirmation mechanism, it constructs an intelligent control system with high security, real-time performance, and self-recovery capabilities. Its core processes include trip cause identification, state acquisition and fusion, conditional reclosing determination, and delayed self-recovery and interlocking execution.

[0040] First, this invention establishes a tripping cause discrimination model by collecting and modeling the operation signals of the air switch body, fault alarm contact signals, circuit voltage and current states, and operation records. This model can distinguish between three scenarios: human operation, fault tripping, and non-faulty false tripping, thus providing a basis for subsequent reclosing strategies. Unlike existing single-signal discrimination methods, this invention adopts a multi-source state information fusion method, improving the accuracy and robustness of discrimination through logical judgment and mutual verification mechanisms.

[0041] Next, this invention proposes a conditional automatic reclosing strategy. The system will only trigger reclosing under conditions such as "non-human operation," "no fault alarm signal," and "stable voltage and current." If a fault or human-caused disconnection is detected, reclosing is automatically blocked to prevent equipment damage or protection malfunction due to blind closing. To further improve recovery stability, this patent introduces a delay control mechanism to delay reclosing in false tripping scenarios, thus balancing power supply continuity and operational safety.

[0042] Subsequently, the overlap determination result is verified through a multi-source status dual-confirmation mechanism. This mechanism utilizes information from alarm contacts, circuit status monitoring, and mechanical position detection for cross-verification, ensuring that a single signal anomaly will not lead to a false judgment, thereby improving the reliability of intelligent overlap control. This mechanism can maintain correct action logic even in complex electromagnetic environments or under abnormal equipment signal conditions, reducing the risk of erroneous control.

[0043] Finally, it supports recording and archiving the entire process, including tripping causes, reclosing determination process, and final action results. Managers can perform statistical analysis and trend assessment based on this data to continuously optimize equipment configuration and operating strategies. Through the aforementioned closed-loop control, this invention achieves rapid self-recovery of the secondary circuit air switch in case of erroneous tripping, while also providing reliable interlocking under fault and human operation scenarios, significantly improving the safety and operational efficiency of the power grid secondary system. The process is as follows: Figure 1 As shown.

[0044] In this embodiment, the data processing operation in step S1 includes performing field standardization and time synchronization processing on the collected multi-source heterogeneous data to obtain a standardized feature vector; performing denoising processing and feature extraction on the standardized feature vector to obtain multi-scale features; performing anomaly detection and missing value completion processing on the multi-scale features; and performing quality scoring on the processed data to obtain a fused feature vector and a quality score value.

[0045] In one optional implementation, the data processing operation may further include cleaning the collected multi-source heterogeneous data to remove obvious erroneous data; normalizing the cleaned data; selecting features from the normalized data to filter out key features; and fusing the filtered features to obtain a fused feature vector and a quality score.

[0046] In another optional implementation, the data processing operation may further include format conversion of the collected multi-source heterogeneous data to unify the data format; time alignment of the converted data; filtering of the aligned data to remove high-frequency noise; feature engineering of the filtered data to construct derived features; and quality evaluation of the constructed features to obtain a fused feature vector and a quality score.

[0047] This invention eliminates format and unit differences between different data sources through field standardization and time synchronization, unifying all time-series data to a reference time axis and ensuring the consistency and comparability of event times. By employing bandpass filtering and wavelet threshold-based denoising operators to denoise voltage and current waveforms, it effectively suppresses power frequency interference and retains effective signal characteristics.

[0048] Furthermore, in step S1, multi-source heterogeneous data of the air switch is collected, and data processing operations are performed on the collected data to obtain the fused feature vector and quality score value, including the following steps A1-A4: A1. Perform field standardization and time synchronization processing on the collected multi-source heterogeneous data to obtain standardized feature vectors.

[0049] A2. Denoise and extract features from the standardized feature vectors to obtain multi-scale features.

[0050] A3. Perform anomaly detection and missing value completion processing on multi-scale features.

[0051] A4. Perform quality scoring on the processed data to obtain the fused feature vector and quality score value.

[0052] Furthermore, in step A3, anomaly detection and missing value completion are performed on the multi-scale features, including the following steps A31-A34: A31. Anomaly detection is performed using the absolute deviation of the median. The deviation ratio of each feature value in the multi-scale feature to the median is calculated. When the deviation ratio exceeds the anomaly threshold, it is marked as an anomaly.

[0053] A32. Use temporal interpolation to complete short missing segments in multi-scale features.

[0054] A33. For long missing segments in multi-scale features, similarity-weighted filling based on neighboring terminals is used to complete them. The similarity weight is calculated based on the power grid topology distance and model difference.

[0055] A34. Label the confidence level of key signals in the completed multi-scale features.

[0056] Specifically, in this invention, multi-source data acquisition and preprocessing provide a reliable input basis for trip cause discrimination and conditional overlap decision-making. Since the decisions of the secondary circuit of the air switch rely on various heterogeneous signals such as operating signals (closing / opening commands), trip / alarm contact signals, circuit voltage and current waveforms, mechanical position detection, operating records, communication link quality, and environmental electromagnetic interference, directly using unprocessed data will introduce problems such as timing asynchrony, inconsistent dimensions, noise interference, missing data, and outliers, thereby reducing the accuracy and robustness of the discrimination and control algorithms. Therefore, this invention designs a complete multi-source data preprocessing workflow, including field / timing standardization, synchronization and denoising, feature extraction, anomaly detection and missing data completion, and edge preprocessing and fusion mapping, enabling the upper-layer intelligent module to obtain high-quality, credibility-labeled fusion vector input.

[0057] Let the set of monitored air switchgear be . For each device , The number of devices is represented by the original data vector. The entire original data matrix is The goal of preprocessing is to obtain a normalized and confidence-weighted fusion feature representation matrix. ,in, To fuse feature dimensions, and simultaneously provide a quality assessment for each fused vector. .

[0058] To perform field standardization and time synchronization, and to eliminate format and unit differences between different data sources, a mapping function is first defined. Perform semantic mapping and type standardization on each source field: in, The standardized feature vector, To standardize the feature dimension, logarithmic values ​​are standardized using uniform units and Z-score normalization is performed. in, For standardized numerical quantities, The first eigenvector after standardization Each feature component The first The mean and standard deviation of each feature over a batch or historical window.

[0059] Due to different signal sampling frequencies Since they are different, time alignment is required. Cross-correlation functions are used to achieve alignment between the two channels: in, For optimal time delay, It is a cross-correlation function. For time, for Time signal, For time delay variables, for The sum of moments The signal is blocked.

[0060] And based on Resampling and interpolation are performed to unify all time-series data to the reference time axis. This ensures the consistency and comparability of event timing.

[0061] Signal denoising and feature extraction are performed for time-series waveforms such as current / voltage and mechanical vibration, employing a multi-stage denoising and feature extraction process. Denoising utilizes bandpass filtering to suppress power frequency interference and a wavelet threshold-based denoising operator. : in, The original waveform. This is the denoised waveform. A set of multi-scale features is calculated on the denoised waveform. : in, For frequency range, For multi-scale features, The root mean square value, Peak value, For energy, For the rising time, For the Fast Fourier Transform amplitude spectrum, The starting frequency, For the termination frequency, This represents the number of frequency points.

[0062] For transient events (such as tripping impact waveforms), the pulse duration also needs to be extracted. Zero crossover rate, envelope index, and other time-domain and frequency-domain characteristics are used to distinguish between transient faults and mechanical tripping errors. Action delay is extracted from the switch contact position. These characteristics, such as the time difference between the closing or opening command and the stabilization of the mechanical position, as well as the number of rebounds, are crucial for identifying human error and equipment malfunctions.

[0063] To improve the reliability of subsequent discrimination, this invention employs robust statistics and a similarity-based imputation strategy for anomaly detection and missing value completion. Median absolute deviation (MAD) is used for anomaly detection. in, For the first The first of the equipment Anomaly detection coefficients for each feature, For the first The median absolute deviation of each feature For the first The median of the features.

[0064] like If it is, then it is marked as abnormal, where the threshold is... Determined by experience or an adaptive strategy. Temporal interpolation is used for short missing segments; for longer or structural missing segments, similarity-weighted imputation based on neighboring terminals is employed. in, The filled-in eigenvalues, For equipment The collection of nearby devices, For equipment The nearest device index, This represents the similarity weight.

[0065] The similarity weights use an exponential decay method: in, For the power grid topology distance, For the overall distance, For model differences, These are adjustable parameters, set based on historical data. Confidence levels must be labeled for key signals after filling in the gaps. This allows the upper-level decision-making module to adopt protective strategies based on the confidence level.

[0066] Edge preprocessing, compression coding, and quality scoring are performed. Considering the high real-time requirements of the secondary loop and the limited communication bandwidth, this invention performs first-layer preprocessing and compression coding at the edge intelligent node. The edge node calculates its local feature vector. And through dimensionality reduction encoding Obtain the transmission representation: in, For transmission representation, The dimension is the dimension after dimensionality reduction.

[0067] After being transmitted to the central fusion unit, it is then processed by the mapping function. Restore or further extract fusion features: in, For the first The integration characteristics of the devices are represented. This includes non-time-series metadata, such as model number, installation location, and last maintenance date. Additionally, a quality score is calculated for each fused vector. Taking into account sensor confidence, data imputation amount, time alignment error, and communication packet loss rate, it can be formalized as follows: in, For Sigmoid mapping , The health status reported by the sensor itself. The missing percentage, For time alignment deviation, , , These are weighting coefficients, set based on historical data.

[0068] After performing preliminary mapping and output criteria for multi-source fusion, and completing the above processing, the various features are weighted according to semantic categories and concatenated or obtained through learned mapping to obtain the final fusion vector: in, Represents the set of signal sources. This represents the multi-scale features of the source. These are the source weights. The final fused feature representation matrix will be obtained as follows: And accompanied by a quality score vector .Should and This will serve as the direct input to the trip cause determination and conditional decision-making module: when a certain device of When the threshold is lowered, a conservative strategy will be activated at the decision-making level.

[0069] In this embodiment, the feature fusion operation in step S2 includes performing multimodal signal analysis on the fused feature vector to extract time-domain and frequency-domain features; constructing an event sequence network to extract event features; performing feature modulation on environmental parameters and trajectory modeling on historical data; using a gated attention fusion mechanism to dynamically weight and fuse various features to obtain a fused state feature vector; and performing supervised dimensionality reduction on the fused state feature vector.

[0070] In one optional implementation, the feature fusion operation may further include extracting features from the fused feature vector to obtain electrical features, event features, environmental features, and historical features; performing weighted average fusion on each type of feature, with the weights preset according to the importance of the features; and performing principal component analysis to reduce the dimensionality of the fused feature vector to obtain the dimensionality-reduced fused state feature vector.

[0071] In another optional implementation, the feature fusion operation may further include extracting classification features from the fused feature vector, extracting electrical features, mechanical features and environmental features respectively; concatenating the various features in a concatenated manner to form a high-dimensional feature vector; and performing dimensionality reduction on the high-dimensional feature vector using an autoencoder to obtain the dimensionality-reduced fused state feature vector.

[0072] This invention performs time-frequency analysis on voltage and current waveforms using short-time Fourier transform with an adaptive window length. The window length is dynamically adjusted according to the instantaneous frequency gradient of the signal. When the signal frequency changes drastically, the window length is shortened to improve time resolution, and when the signal frequency is stable, the window length is extended to improve frequency resolution.

[0073] Furthermore, in step S2, feature extraction and feature fusion operations are performed on the fused feature vector to obtain the dimensionality-reduced fused state feature vector, including the following steps B1-B5: B1. Perform multimodal signal analysis on the fused feature vectors to extract time-domain and frequency-domain features.

[0074] B2. Construct an event sequence network and extract event features.

[0075] B3. Modulate environmental parameters using features and model trajectories from historical data.

[0076] B4. A gating attention fusion mechanism is used to dynamically weight and fuse various features to obtain a fused state feature vector.

[0077] B5. Perform supervised dimensionality reduction on the fused state feature vector.

[0078] In this embodiment, the multimodal signal analysis in step B1 includes performing time-frequency analysis on the voltage and current waveforms in the fused feature vector using a short-time Fourier transform with an adaptive window length, dynamically adjusting the window length of the window function according to the instantaneous frequency gradient of the signal; extracting the singular value decomposition features of the time-frequency matrix based on the time-frequency representation as descriptors for steady-state and transient behavior; extracting pulse duration, zero-crossing rate, and envelope exponent from transient events in the fused feature vector; and extracting action delay and rebound count from the switch contact position in the fused feature vector to obtain electrical features, event features, environmental features, and historical features.

[0079] In one alternative implementation, multimodal signal analysis may further include performing a fast Fourier transform on the voltage and current waveforms in the fused feature vector to extract frequency domain amplitude and phase features; performing time-domain statistical analysis on the waveforms to extract mean, variance, and peak-to-peak value; performing waveform matching and identification on transient events to determine the event type; and performing vibration signal analysis on the switching mechanical state to extract vibration frequency features, thereby obtaining electrical features, event features, environmental features, and historical features.

[0080] In another alternative implementation, multimodal signal analysis may further include performing wavelet transform on the voltage and current waveforms in the fused feature vector to extract multi-scale time-frequency features; extracting the envelope of the waveforms to analyze signal energy changes; performing edge detection on transient events to identify abrupt change moments; and performing state transition analysis on the switch position signals to extract switching time features, thereby obtaining electrical features, event features, environmental features, and historical features.

[0081] This invention employs an adaptive window length short-time Fourier transform to dynamically adjust the time-frequency resolution based on signal frequency changes. During transient processes with drastic signal frequency changes, the window length is shortened to improve the ability to capture rapid changes, while during steady-state operation with stable signal frequency, the window length is extended to improve frequency recognition accuracy.

[0082] Furthermore, in step B1, multimodal signal analysis is performed on the fused feature vector to extract time-domain and frequency-domain features, including the following steps B11-B14: B11. Short-time Fourier transform with adaptive window length is used to perform time-frequency analysis on the voltage and current waveforms in the fused feature vector, and the window length of the window function is dynamically adjusted according to the instantaneous frequency gradient of the signal.

[0083] B12. Extract singular value decomposition features of the time-frequency matrix based on time-frequency representation as descriptors for steady-state and transient behavior.

[0084] B13. Extract the pulse duration, zero crossover rate, and envelope index from the transient events in the fused feature vector.

[0085] B14. Extract the action delay and rebound count from the switch contact position in the fused feature vector to obtain electrical features, event features, environmental features and historical features.

[0086] Specifically, after completing the acquisition and preprocessing of multi-source operating data, this invention proposes a multi-level feature extraction and adaptive fusion method for air switch state assessment. Traditional methods typically rely on single-type signal features or simple feature splicing, failing to fully consider the dynamic correlation between different signal sources and the temporal dependence of state changes, resulting in insufficient discrimination accuracy in complex operating environments. To address this, this invention performs deep feature mining from four dimensions: electrical signals, event sequences, environmental parameters, and historical behavior, and introduces an attention-based feature fusion strategy to construct a state feature representation with strong discriminative power and high interpretability.

[0087] This invention performs multimodal signal analysis and dynamic feature extraction. For electrical signals, it extracts not only conventional time-domain features (such as RMS value, peak value, and waveform distortion rate) and frequency-domain features (such as fundamental and harmonic amplitudes, and total harmonic distortion rate), but also pays special attention to subtle differences in transient processes. To better distinguish between different types of faults and interference, this invention proposes an improved time-frequency analysis scheme by introducing an adaptive window-length Short-time Fourier Transform (STFT) to dynamically adjust the time-frequency resolution. in, For time-frequency resolution, The original signal, The window function is adaptively adjusted based on the local stationarity of the signal, and the window length parameter is... It is determined by the instantaneous frequency gradient of the signal. Based on this time-frequency representation, the singular value decomposition (SVD) features of the time-frequency matrix are extracted as supplementary descriptors for steady-state and transient behavior, thereby enhancing the feature stability under non-stationary conditions.

[0088] This invention proposes a method for constructing a time-constrained event network for event sequence modeling and temporal causality mining, targeting discrete events such as operation signals and alarm contacts. This method considers not only the transition probabilities between events but also introduces time delay constraints and logical condition constraints between events. Specifically, a weighted directed graph is constructed. ,in, For a set of nodes, For edge set, For a time-constrained set, each edge Additional time interval conditions and logical conditions For example, the time from "issuing the closing command" to "feedback of the mechanism position" exceeds the preset range. At that time, candidate features for "mechanical anomaly" are generated.

[0089] To further enhance the ability to distinguish event sequences, this invention also proposes an event trigger feature pooling mechanism: aggregating electrical and environmental quantities (such as mean, variance, and extreme values) within the time window before and after a specific key event (such as "trip alarm") to form event context features, thereby modeling the correlation between discrete events and continuously changing signals.

[0090] This invention employs a dynamic weighted integration of environmental and historical characteristics. Environmental parameters, such as temperature, humidity, and vibration, after standardization, not only serve as independent features but also modulate the effective range of electrical characteristics. The invention introduces an environmental adaptive feature scaling mechanism to perform weighted adjustments of electrical characteristics under environmental factors. in, The electrical characteristics are weighted and adjusted. Electrical characteristics, These are the current environmental parameter values. The mean of the environmental parameters. The standard deviation of environmental parameters, It is a learnable environmental modulation coefficient used to amplify or reduce the contribution of the current environment to electrical characteristics deviating from normal conditions, and is set based on historical data.

[0091] In addition to statistical indicators, historical operating characteristics also incorporate state degradation trajectory modeling. A Hidden Markov Model (HMM) is used to model historical trip sequences and operation counts, extracting hidden state probabilities as long-term reliability assessment features to capture the gradual changing trends of equipment performance.

[0092] This invention performs attention-based feature fusion and interpretable dimensionality reduction. In the feature fusion stage, it abandons the traditional splicing or fixed-weighted fusion method and proposes a gated attention fusion mechanism to dynamically calculate the fusion weights of different feature components. in, For equipment The fused feature vector, For the first A set of features (such as electrical, event, environment, history). For feature group index, For the first Attention weights for each feature group and These are learnable parameters. This mechanism can adaptively adjust the contribution levels of various features based on the current state, enhancing the model's discriminative ability under abnormal conditions.

[0093] To alleviate the computational burden and overfitting risk associated with high-dimensional features, a supervised dimensionality reduction method is employed: supervised principal component analysis (SPCA) based on tripping discriminant labels. This method maximizes the correlation between features and labels during dimensionality reduction while retaining the most discriminative low-dimensional representation. in, For equipment The reduced-dimensional feature vectors The projection matrix is ​​learned by optimizing the discriminant loss function. The final output is a low-dimensional fused feature vector and its corresponding feature importance ranking, providing highly reliable input for the subsequent discrimination module.

[0094] Furthermore, in step S3, the dimensionality-reduced fused state feature vector is input into the discriminant model constructed based on ensemble learning and attention mechanisms to determine the cause of the trip, thereby obtaining the probability distribution of the trip type. Based on the quality score, it is determined whether a conservative strategy should be activated. Based on the probability distribution and constraints, overlap determination is performed, and after determining the delay time according to the delay mechanism, the overlap operation is executed, including the following steps C1-C5: C1. Construct a discriminant model that includes a feature recalibration module and a multi-expert integrated classification module.

[0095] C2. Input the dimensionality-reduced fused state feature vector into the discrimination model and output the probability distribution of the tripping type.

[0096] C3. Determine whether to activate the conservative strategy based on the quality score.

[0097] C4. Determine the tripping type based on probability distribution, and combine electrical constraints, fault signal constraints, event logic constraints and equipment health constraints to determine overlap.

[0098] C5. After determining the dynamic delay time based on the risk score, perform the overlap operation.

[0099] In this embodiment of the application, the discrimination model in step C1 includes a feature recalibration module and a multi-expert ensemble classification module. In the feature recalibration module, the feature vector of the fusion state after dimensionality reduction is recalibrated based on attention weights. In the multi-expert ensemble classification module, multiple expert sub-classifier networks are set up. The ensemble weights of each expert sub-classifier are dynamically generated by a gating network according to the recalibrated fusion state feature vector. The outputs of each expert sub-classifier are weighted and integrated to output the probability distribution of human operation, fault tripping and non-fault false tripping.

[0100] In an alternative implementation, the discriminant model may further include a single deep neural network classifier that inputs the dimensionality-reduced fused state feature vector into a multi-layer fully connected neural network, passes it through an activation function and a softmax layer, and outputs the probability distribution of human operation, fault tripping, and non-fault false tripping.

[0101] In another alternative implementation, the discriminant model may also include a combination of a support vector machine classifier and a decision tree classifier. The dimensionality-reduced fused state feature vector is input into the support vector machine and the decision tree respectively for classification. The output results of the two classifiers are fused by voting to output the probability distribution of human operation, fault tripping and non-fault false tripping.

[0102] This invention recalibrates the fused state feature vector based on attention weights in the feature recalibration module, which can dynamically adjust the importance of different feature dimensions, enhance the response capability to the most relevant features of the current tripping state, and make the model focus on the feature dimensions with the most discriminative value. By setting up multiple expert sub-classifier networks, different expert networks can focus on the discriminative learning of different types of tripping scenarios.

[0103] Furthermore, in step C1, a discriminative model is constructed that includes a feature recalibration module and a multi-expert ensemble classification module, comprising the following steps C11-C14: C11. In the feature recalibration module, the feature vector of the fusion state after dimensionality reduction is recalibrated based on attention weights.

[0104] C12. In the multi-expert ensemble classification module, set up multiple expert sub-classifier networks.

[0105] C13. The ensemble weights of each expert subclassifier are dynamically generated based on the current features through a gating network.

[0106] C14. The outputs of each expert sub-classifier are weighted and integrated to obtain the discriminant model.

[0107] Furthermore, in step C4, the tripping type is determined based on the probability distribution, and overlap determination is performed in conjunction with electrical constraints, fault signal constraints, event logic constraints, and equipment health constraints, including the following steps C41-C46: C41. Determine whether the probability of a non-fault false trip is greater than the probability threshold.

[0108] C42. Determine whether the voltage and current are within a safe and stable range and whether the root mean square error is below the threshold value.

[0109] C43. Determine if there is a continuous fault alarm signal.

[0110] C44. Determine whether there are manual interlocking or maintenance status labels in the operation event sequence.

[0111] C45. Determine whether the equipment health score derived from historical characteristics meets the health threshold.

[0112] C46. When all judgment conditions are met, output the judgment result that allows overlap.

[0113] Furthermore, in step C5, after determining the dynamic delay time based on the risk score, the overlap operation is performed, including the following steps C51-C54: C51. Calculate the risk score by taking into account the current electrical stability, the intensity of environmental disturbances, and the frequency of historical malfunctions.

[0114] C52. Add the base delay time to the gain adjustment of the risk score to obtain the dynamic delay time.

[0115] C53. Before performing the overlapping action after the dynamic delay time ends, perform a real-time secondary verification to detect whether there is a temporary fault signal.

[0116] C54. After the real-time secondary verification passes, the overlap operation is performed.

[0117] Specifically, based on the completion of multi-source feature extraction and fusion modeling, this invention proposes a tripping cause discrimination mechanism that integrates multi-dimensional state information and logical reasoning, and designs a conditional and adaptive automatic reclosing control strategy accordingly. This scheme aims to solve the problems of insufficient discrimination accuracy, lack of flexibility in reclosing strategies, and inadequate safety guarantees in existing methods when facing complex power grid conditions, significantly improving the intelligent decision-making level and operational reliability of air switches.

[0118] The tripping cause discrimination model based on ensemble learning and attention enhancement adopts a two-stage architecture design. The first stage is a feature recalibration module, and the second stage is a multi-expert ensemble classification module. In the feature recalibration module, the dimensionality-reduced fusion state feature vector is first input into the attention weight calculation unit. This unit transforms the feature vector through a learnable parameter matrix and vector, and calculates the attention weight vector with the same dimension as the feature vector. Then, the attention weight vector is multiplied element-wise with the original fusion state feature vector to obtain the recalibrated feature vector. This process realizes the dynamic adjustment of the importance of different feature dimensions. In the multi-expert ensemble classification module, multiple parallel expert sub-classifier networks are set up, each with an independent neural network structure. A gating network is also included, taking the recalibrated feature vector as input and dynamically generating ensemble weights for each expert sub-classifier based on the current features. The recalibrated feature vector is then simultaneously input into each expert sub-classifier, with each sub-classifier outputting a preliminary judgment result for the three tripping categories. Finally, the outputs of each expert sub-classifier are weighted and summed according to the ensemble weights generated by the gating network to obtain the final probability distributions for human operation, fault tripping, and non-fault false tripping. During model training, a focus loss function combined with class balancing weights is used as the objective function. Adjustable focus parameters reduce the weight of easily classified samples, making the model more attentive to difficult examples and imbalanced sample conditions.

[0119] The fusion and dimensionality reduction feature vectors extracted in the previous stage As input, a high-precision, highly generalized multi-class classification model is constructed. Unlike traditional single neural network or support vector machine approaches, this invention proposes an integrated gated attention network, whose structure includes a feature recalibration module and a multi-expert ensemble classification module. in, For equipment The predicted output; For the discriminant model function; These are all the learnable parameters of the model; An index for expert networks; The number of expert networks; This is a feature recalibration vector based on attention weights; This represents element-wise multiplication; For the first A network of experts (subclassifiers); The weights are dynamically generated by the gating network based on the current features. The model outputs a three-class probability distribution: in, Human operation, Tripping due to a fault. This is a non-faulty false bounce. The model training uses an objective function that combines focal loss with class balancing weights: in, The loss function; Indexed by category; The model predicts the category to which the sample belongs. The probability of; This is a real label; For category Balance weights; This is an adjustable focusing parameter used to reduce the weight of easily classified samples, making the model pay more attention to difficult examples and imbalanced samples.

[0120] The multi-constraint overlap determination mechanism, after obtaining the probability discrimination result of the tripping type, does not simply determine whether there is overlap based on the maximum probability, but introduces a multi-dimensional, hierarchical conditional determination function: in, for Overlap determination; The function is an indicator that outputs 1 if all conditions are true, and 0 otherwise. The non-fault false bounce probability predicted by the model; This is the probability threshold; For logical AND operation; The number of constraints; For the first The constraints mainly include electrical steady-state constraints, fault signal blocking constraints, event logic consistency constraints, and equipment health constraints.

[0121] Electrical steady-state constraints require that voltage and current remain within a safe and stable range. And the mean squared error was below the threshold value for a period of time, where, This is the current voltage value. For reference voltage value, For the allowable voltage deviation, This is the current current value. For reference current value, This refers to the allowable range for current deviation.

[0122] The fault signal blocking constraint is that there are no continuous fault alarms (such as overcurrent, grounding, and insulation alarms).

[0123] The event logic consistency constraint is that the "manual lockout" or "maintenance status" label does not appear in the operation event sequence.

[0124] The equipment health constraint is an equipment health score derived from historical characteristics. ,in, Rate the health of the equipment. This represents the minimum health threshold.

[0125] It supports user-defined constraint combination strategies and can update constraint conditions and their thresholds in real time to adapt to different operating scenarios.

[0126] To further reduce the risk of false overlap, this invention proposes a state-aware dynamic delay mechanism, whose delay time is jointly determined by a base delay and a dynamic adjustment amount: Adaptive delay strategy and dynamic risk control. in, To overlap the delay time, Based on the base delay time, The risk scoring function comprehensively considers factors such as current electrical stability, intensity of environmental disturbance, and frequency of historical malfunctions, and outputs a standardized risk coefficient. The adjustable gain is used to control the level of risk sensitivity. When a higher risk is detected, the delay time is automatically extended to allow time for maintenance personnel to intervene or for further status confirmation.

[0127] The final overlapping action must be triggered simultaneously if the following conditions are met: in, Execute instructions for overlapping actions; This is an indicator function that outputs 1 when all conditions are true, and 0 otherwise. This is a rapid secondary check performed instantaneously before the action, such as detecting whether any temporary fault signals are transmitted.

[0128] The closed-loop verification and safety interlocking mechanism prioritizes low-voltage trial closing technology before performing reclosing operations. By applying low voltage to detect the status of the detection circuit and the load response, it determines in real time whether the closing conditions are met. After formal closing, it enters a continuous monitoring state. If an abnormality or tripping signal is detected again within a set time, the interlocking procedure is immediately initiated to prohibit subsequent automatic reclosing and to send a "reclosing failure" alarm.

[0129] All judgment, decision-making, and operational processes are recorded and used to update the historical feature database and optimize the model, forming a closed-loop learning mechanism. It also has cross-device policy coordination capabilities. When multiple switches in the same area trip err consecutively, the regional protection coordination strategy can be activated to prevent the reclosing process from causing the fault range to expand.

[0130] Furthermore, in step S4, a multi-source verification operation is performed before the overlap operation. If the overlap condition is not met temporarily, a delayed retry mechanism is initiated. If the number of consecutive retries reaches a threshold and overlap is still unsuccessful, the system enters a locked state, including the following steps: Building upon conditional coincidence determination, this invention proposes a multi-source dual-confirmation and time-delay self-recovery mechanism to further enhance decision-making reliability in uncertain scenarios such as complex electromagnetic environments, sporadic sensor anomalies, or intermittent failures. This mechanism effectively prevents false coincidences and reduces the risk of repeated impacts through multi-dimensional information cross-validation and intelligent delay strategies, while also possessing continuous self-optimization capabilities, significantly improving overall robustness and security.

[0131] The hierarchical multi-source cross-validation and confidence fusion mechanism, before issuing the overlap instruction, does not solely rely on the classification model output. Instead, it introduces a hierarchical multi-source cross-validation strategy to perform real-time verification of key state variables, ensuring multi-party consistency in decision-making. This mechanism includes the following three core verification layers: Alarm contact status verification: Real-time acquisition and verification of all fault-related alarm contact signals (such as overcurrent, short circuit, grounding, insulation alarms, etc.). Even if the judgment model outputs "false tripping", as long as any alarm contact is in an active state, re-closing is prohibited to avoid missing real faults.

[0132] Steady-state verification of electrical quantities: Before determining reclosing, perform a short-time steady-state reassessment of voltage and current to determine whether they remain within the safe range. Avoid misjudgments caused by instantaneous recovery or measurement noise.

[0133] Mechanical position and mechanism status confirmation: Through position sensor and mechanism status feedback, confirm that the switch body is in the closed ready state and there are no abnormalities such as jamming or lack of energy storage.

[0134] The above multi-source confirmation conditions are integrated through the following formal logic: in, To confirm the conditions, For the first A Boolean function for validation conditions. It also supports a confidence-weighted validation mechanism, which assigns weights to different validation sources and allows for temporary deviations in individual non-critical validation items to a certain extent.

[0135] To improve self-recovery capability after temporary anomalies, this invention designs a state-aware adaptive delay retry mechanism with a progressive blocking strategy. This mechanism is activated when the overlap condition is not yet met, and dynamically adjusts the retry waiting time based on real-time risk assessment. in, This is the waiting time for retrying. For risk scoring functions based on feature input, This is an adaptive coefficient. If the initial reconnection fails due to a temporary disturbance (such as a voltage drop or transient interference), it automatically enters a delayed retry state, and the multi-source confirmation process is re-executed after the delay ends.

[0136] To avoid multiple invalid coincidences due to permanent faults, a progressive locking strategy is introduced: the number of consecutive retries is recorded. .when If the system still fails to re-align successfully, a persistent fault is detected, and the system automatically enters a locked state, prohibiting further automatic operation and sending a "locked alarm" signal. The locked state can only be released through manual reset or remote command.

[0137] This process can be expressed by the following recovery function: in, The recovery function is used for retry determination. This is a dynamic risk threshold that increases with the number of retries, becoming increasingly cautious with each retry.

[0138] A closed-loop learning and strategy optimization mechanism records all overlapping operations, confirmation results, delay parameters, and final execution status, forming an overlapping operation knowledge base. Based on this knowledge base, the following self-learning and optimization are performed periodically: Recalibrate the thresholds for each confirmation condition based on historical data (e.g.) This makes it more closely suited to the actual operating environment.

[0139] Optimize base latency using reinforcement learning algorithms and adaptive coefficients To minimize mean recovery time and overlap failure rate.

[0140] The risk scoring model is continuously updated using actual overlapping success / failure samples. This will improve the accuracy of its predictions.

[0141] Through the aforementioned closed-loop optimization mechanism, it is possible to continuously adapt to changes in the on-site operating environment, gradually improve the accuracy and reliability of intelligent overlap, and form intelligent decision-making with continuous improvement capabilities.

[0142] Building upon the aforementioned multi-source verification and self-recovery mechanisms, this invention proposes a complete process recording standard and cross-device collaborative optimization architecture to achieve traceability, evaluability, and continuous evolution capabilities for behavior. This part not only realizes the digital archiving of single-device control processes but also significantly improves overall reliability and adaptability to complex power grid environments through regional collaboration and data-driven optimization.

[0143] Furthermore, in step S5, the entire process data is recorded to build a knowledge base. Based on the knowledge base, cross-device collaborative operations are performed, and the strategy parameters are optimized and updated, including the following steps: Building upon the aforementioned multi-source confirmation and self-recovery mechanisms, this invention proposes a complete process recording standard and cross-device collaborative optimization architecture to achieve traceability, assessability, and continuous evolution capabilities of behavior. This part not only realizes the digital archiving of single-device control processes but also significantly improves overall reliability and adaptability to complex power grid environments through regional collaboration and data-driven optimization.

[0144] The entire process is recorded and a knowledge base for overlap is constructed. During each tripping event and subsequent judgment, decision-making, and overlap processes, comprehensive and high-granular data is recorded to form a traceable operational trajectory. The recorded information mainly includes: Data for the discrimination stage: Input feature vectors , probability distribution of the discriminant model output Real-time status and satisfaction of various constraints (electrical, event, environmental).

[0145] Confirmation and Delay Process: The verification results of each source in multi-source cross-validation and the delay parameters used. Dynamic risk scoring Curve showing how it changes over time.

[0146] Execution and feedback information: whether reclosing was ultimately performed, the status of the equipment after closing, whether it entered the interlock, and the final result of this operation (success / failure).

[0147] After being anonymized and standardized, this data is stored in the "overlapping process knowledge base." This knowledge base uses a time-series data model for storage, supports multi-dimensional queries and aggregation analysis, and provides a data foundation for subsequent strategy evaluation and model optimization.

[0148] Cross-device collaboration and intelligent coordination for regional protection: To achieve regional-level security cooperation and strategy optimization, it supports information exchange and strategy coordination among multiple circuit breakers through centralized coordination nodes or distributed communication networks. Specifically, this includes: Regional event correlation analysis: Based on event timestamps and power grid topology information, identify whether there are correlated tripping events. Through time-series correlation rule mining and graph structure analysis, determine whether multiple switch trips are caused by the same fault event, thereby distinguishing between local faults and system-level disturbances.

[0149] Dynamic reclosing sequence strategy: When a regional voltage sag or frequency fluctuation is detected, the higher-level coordinator can generate an ordered sequence of reclosing commands to avoid the superposition of inrush currents or secondary faults caused by multiple switches reclosing simultaneously. For example, reclosing can be performed step by step according to line level or electrical distance, from near to far, to maintain grid stability.

[0150] Area interlocking and risk isolation: If a switch fails to reclose multiple times or faults occur frequently in the area, the coordinator can issue an area interlocking command to suspend the automatic reclosing function of adjacent switches, limit the propagation range of electrical disturbances, and prevent the fault from escalating.

[0151] Based on machine learning, a dynamic evolution mechanism for control strategies is established, leveraging a large amount of operational records and regional collaborative scenarios accumulated in a knowledge base. This mechanism enables dynamic optimization of strategy parameters, achieving an evolution from "single-time control" to "continuous optimization." Guided by a multi-objective loss function that minimizes mean recovery time, number of erroneous closings, and fault impact range, the control strategy parameters are updated periodically. in, This is a set of strategy parameters, including various discrimination thresholds, delay bases, risk weights, etc. The loss function is obtained by aggregating historical data. This is a regularization term used to avoid overfitting; These are hyperparameters. The optimization process can be implemented using gradient-based or gradient-free methods such as Bayesian optimization and reinforcement learning (e.g., DQN, PPO).

[0152] The optimized strategy parameters can be verified and deployed to field devices in batches. Alternatively, through federated learning technology, cross-site strategy fusion and model evolution can be achieved while protecting data privacy, thereby forming intelligent overlapping control with networked learning capabilities.

[0153] In summary, this invention proposes a control method for air switches with intelligent anti-false tripping and conditional automatic reclosing functions. It constructs a full-process intelligent control system encompassing multi-source data acquisition and preprocessing, state feature extraction and fusion modeling, intelligent tripping cause determination and conditional reclosing strategy, multi-source dual confirmation and delayed self-recovery mechanism, and full-process operation recording and cross-device collaborative optimization. First, through the collaboration of edge nodes and the central fusion unit, standardization, denoising, and confidence labeling of multi-source heterogeneous data such as operating signals, electrical waveforms, alarm contacts, mechanical positions, and environmental parameters are achieved, ensuring the accuracy and reliability of input features. Second, multi-modal feature modeling combining electrical signals, event sequences, environmental disturbances, and historical degradation trajectories is combined, and a state representation with strong discriminative ability and high interpretability is constructed through a gating attention fusion mechanism and supervised dimensionality reduction technology. Based on this, this invention introduces an integrated attention discrimination model and a multi-constraint conditional reclosing determination method, through adaptive delay control and dynamic... Risk assessment enables rapid self-recovery in non-faulty tripping scenarios, while strict interlocking under fault and human operation conditions significantly reduces the risk of accidental closing. Furthermore, through hierarchical multi-source cross-verification, confidence fusion confirmation, and adaptive delay retry mechanism, robustness under complex electromagnetic environments and sensor anomaly conditions is effectively improved, and continuous optimization is achieved through progressive interlocking and closed-loop learning. Finally, this invention establishes a full-process operation record and overlap knowledge base, and extends it to cross-device regional level collaboration and strategy evolution, realizing intelligent, adaptive, and networked optimization control of power grid secondary circuit air switches.

[0154] Example 3 is an embodiment of the present invention. This embodiment provides an air switch control system with intelligent anti-false tripping and conditional automatic reclosing functions, including a data acquisition and processing module, a feature extraction and fusion module, a tripping judgment and decision module, a multi-source verification module, and a knowledge base optimization module.

[0155] The data acquisition and processing module is used to collect multi-source heterogeneous data from air switches, perform data processing operations on the collected data, and obtain fused feature vectors and quality scores.

[0156] The feature extraction and fusion module is used to perform feature extraction and feature fusion operations on the fused feature vector to obtain the dimensionality-reduced fused state feature vector.

[0157] The tripping discrimination decision module is used to input the dimensionality-reduced fused state feature vector into the discrimination model built based on ensemble learning and attention mechanism to discriminate the tripping cause, obtain the probability distribution of the tripping type, determine whether to enable the conservative strategy based on the quality score, make overlap judgment based on the probability distribution and the constraints, and perform the overlap operation after determining the delay time according to the delay mechanism.

[0158] The multi-source verification module is used to perform multi-source verification before performing the overlap operation. When the overlap condition is not met temporarily, a delayed retry mechanism is started. When the number of consecutive retries reaches the threshold and overlap is still not successful, it enters the blocking state.

[0159] The knowledge base optimization module is used to record data throughout the entire process to build a knowledge base, perform cross-device collaborative operations based on the knowledge base, and optimize and update strategy parameters.

[0160] This embodiment also provides an electronic device applicable to a method for controlling an air switch with intelligent anti-false tripping and conditional automatic reclosing functions, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions as proposed in the above embodiment.

[0161] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements an air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions as proposed in the above embodiment.

[0162] The storage medium proposed in this embodiment and the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0163] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A control method for an air switch with intelligent anti-false tripping and conditional automatic reclosing functions, characterized in that: include, Collect multi-source heterogeneous data from air switches, perform data processing operations on the collected data, and obtain fused feature vectors and quality scores; Perform feature extraction and feature fusion operations on the fused feature vector to obtain the dimensionality-reduced fused state feature vector; The dimensionality-reduced fusion state feature vector is input into the discrimination model built on ensemble learning and attention mechanism to determine the cause of the trip, and the probability distribution of the trip type is obtained. The conservative strategy is then determined based on the quality score. The overlap determination is performed based on the probability distribution and the constraint conditions. The overlap operation is then performed after the delay time is determined according to the delay mechanism. Before performing the overlap operation, a multi-source verification operation is performed. If the overlap condition is not met temporarily, a delayed retry mechanism is started. If the number of consecutive retries reaches the threshold and the overlap is still not successful, the lockout state is entered. Record data throughout the entire process to build a knowledge base, perform cross-device collaborative operations based on the knowledge base, and optimize and update strategy parameters.

2. The control method of air switch with intelligent anti-misoperation and conditional automatic reclosing function according to claim 1, characterized in that: The system collects multi-source heterogeneous data from air switches, performs data processing operations on the collected data, and obtains a fused feature vector and a quality score, including... The collected multi-source heterogeneous data is subjected to field standardization and time synchronization processing to obtain standardized feature vectors; Denoising and feature extraction are performed on the standardized feature vectors to obtain multi-scale features; Anomaly detection and missing value completion processing are performed on multi-scale features; The processed data is scored to obtain the fused feature vector and quality score value.

3. The control method of air switch with intelligent anti-misoperation and conditional automatic reclosing function according to claim 2, characterized in that: The process of performing feature extraction and feature fusion operations on the fused feature vector to obtain a dimension-reduced fused state feature vector includes, Multimodal signal analysis is performed on the fused feature vectors to extract time-domain and frequency-domain features; Construct an event sequence network to extract event features; Feature modulation of environmental parameters and trajectory modeling of historical data; A gated attention fusion mechanism is used to dynamically weight and fuse various features to obtain a fused state feature vector. Supervised dimensionality reduction is performed on the fused state feature vector.

4. The control method of air switch with intelligent anti-misoperation and conditional automatic reclosing function according to claim 3, characterized in that: The process involves inputting the dimensionality-reduced fused state feature vector into a discriminant model constructed based on ensemble learning and attention mechanisms to determine the cause of the trip, obtaining a probability distribution of the trip type, determining whether to activate a conservative strategy based on the quality score, performing overlap determination based on the probability distribution and constraints, and executing the overlap operation after determining the delay time according to the delay mechanism. This includes... Construct a discriminative model that includes a feature recalibration module and a multi-expert ensemble classification module; The dimensionality-reduced fusion state feature vector is input into the discrimination model, and the probability distribution of the tripping type is output. Whether to activate a conservative strategy is determined based on the quality score. The tripping type is determined based on probability distribution, and overlap determination is made by combining electrical constraints, fault signal constraints, event logic constraints and equipment health constraints. After determining the dynamic delay time based on the risk score, the overlap operation is performed.

5. The control method of air switch with intelligent anti-misoperation and conditional automatic reclosing function according to claim 2, characterized in that: The anomaly detection and missing value completion processing for multi-scale features includes, Anomaly detection is performed using the absolute deviation of the median. The ratio of the deviation of each feature value in the multi-scale features to the median is calculated. When the deviation ratio exceeds the anomaly threshold, it is marked as an anomaly. Temporal interpolation is used to complete short missing segments in multi-scale features; Long missing segments in multi-scale features are filled using similarity-weighted filling based on proximity terminals. The similarity weights are calculated based on the power grid topology distance and model differences. The confidence level of key signals in the completed multi-scale features is labeled.

6. The control method of air switch with intelligent anti-misoperation and conditional automatic reclosing function according to claim 3, characterized in that: The process of performing multimodal signal analysis on the fused feature vector to extract time-domain and frequency-domain features includes, A short-time Fourier transform with an adaptive window length is used to perform time-frequency analysis on the voltage and current waveforms in the fused feature vector, and the window length of the window function is dynamically adjusted according to the instantaneous frequency gradient of the signal. Based on time-frequency representation, singular value decomposition features of the time-frequency matrix are extracted as descriptors for steady-state and transient behavior; Extract pulse duration, zero-crossing rate, and envelope exponent from transient events in the fused feature vector; The action delay and rebound count are extracted from the switch contact position in the fused feature vector to obtain electrical features, event features, environmental features and historical features.

7. The control method of air switch with intelligent anti-misoperation and conditional automatic reclosing function according to claim 4, characterized in that: The construction of the discriminant model, which includes a feature recalibration module and a multi-expert ensemble classification module, includes, In the feature recalibration module, the feature vector of the fusion state after dimensionality reduction is recalibrated based on attention weights. In the multi-expert ensemble classification module, multiple expert sub-classifier networks are set up; The gating network dynamically generates the ensemble weights of each expert sub-classifier based on the current features. The outputs of each expert sub-classifier are weighted and integrated to obtain the discriminant model; The method of determining tripping type based on probability distribution and combining electrical constraints, fault signal constraints, event logic constraints, and equipment health constraints for overlap determination includes, Determine whether the probability of a non-fault false trip is greater than a probability threshold; Determine whether the voltage and current are within a safe and stable range and whether the root mean square error is below the threshold value; Determine if a persistent fault alarm signal exists; Determine whether there are manual interlock or maintenance status tags in the sequence of operation events; Determine whether the device health score derived from historical characteristics meets the health threshold; When all the judgment conditions are met, output the judgment result that allows overlap; The step of performing the overlap operation after determining the dynamic delay time based on the risk score includes: A risk score is calculated by taking into account current electrical stability, intensity of environmental disturbances, and historical frequency of malfunctions. The dynamic delay time is obtained by adding the base delay time to the gain adjustment of the risk score. Before performing the overlapping action after the dynamic delay time ends, a real-time secondary verification is performed to detect whether there are any temporary fault signals. Once the real-time secondary verification passes, the overlap operation is performed.

8. An air switch control system with intelligent anti-misoperation and conditional automatic reclosing functions, which applies the air switch control method with intelligent anti-misoperation and conditional automatic reclosing functions according to any one of claims 1-7, characterized in that, include: The module includes a data acquisition and processing module, a feature extraction and fusion module, a tripping judgment and decision module, a multi-source verification module, and a knowledge base optimization module. The data acquisition and processing module is used to acquire multi-source heterogeneous data from air switches, perform data processing operations on the acquired data, and obtain fused feature vectors and quality score values. The feature extraction and fusion module is used to perform feature extraction and feature fusion operations on the fused feature vector to obtain a dimension-reduced fused state feature vector. The tripping discrimination decision module is used to input the dimensionality-reduced fusion state feature vector into the discrimination model constructed based on ensemble learning and attention mechanism to discriminate the tripping cause, obtain the probability distribution of the tripping type, determine whether to enable the conservative strategy based on the quality score, make overlap judgment based on the probability distribution and the constraints, and perform the overlap operation after determining the delay time according to the delay mechanism. The multi-source verification module is used to perform multi-source verification before performing the overlap operation. When the overlap condition is not met, a delayed retry mechanism is started. When the number of consecutive retries reaches the threshold and overlap is still not successful, it enters the blocking state. The knowledge base optimization module is used to record data throughout the entire process to build a knowledge base, perform cross-device collaborative operations based on the knowledge base, and optimize and update strategy parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the air switch control method with intelligent anti-false tripping and conditional automatic reclosing functions as described in any one of claims 1 to 7.

Images