A fusion method and system of an automatic reclosing protector of a distribution box

Abstract

This application relates to the technical field of automatic reclosing protectors for distribution boxes. A fusion method for an automatic reclosing protector for a distribution box includes acquiring initial multi-source raw data and target multi-source feature data; inputting the target multi-source feature data into a target tunnel scenario fault identification model to obtain model judgment results; matching preset fault rules based on the initial multi-source raw data and model judgment results to obtain fault type judgment results and fault parameters; when the fault type judgment result is a temporary fault, the protector triggers a temporary reclosing strategy, obtaining a reclosing action command from the protector; when the fault type judgment result is a permanent fault, the protector triggers a permanent reclosing strategy and generates fault alarm information, obtaining a blocking signal from the protector and permanent fault alarm data.

Description

Technical Field

[0001] This application relates to the technical field of automatic reclosing protectors for distribution boxes, and more specifically, to a fusion method and system for automatic reclosing protectors for distribution boxes. Background Technology

[0002] As a core hub of rail transit, the power distribution boxes in high-speed railway tunnels bear the responsibility of power distribution and protection for critical loads such as tunnel lighting, emergency fans, and signal communication equipment, directly affecting train operation safety and power supply continuity. However, the special environment of tunnels—enclosed and humid, with strong electromagnetic interference and frequent train vibrations—leads to frequent temporary faults such as lightning-induced surges, stray current interference, and momentary poor terminal contact, accounting for more than 80% of all tripping events.

[0003] In existing technologies, tunnel power distribution boxes mostly adopt a separate design of conventional protectors and external reclosing modules. The two are only linked by simple electrical signals and have no underlying data communication. Schemes without reclosing function require manual on-site reset, which is limited by maintenance windows and takes 15-30 minutes to restore power. Schemes with reclosing function only close the circuit with a fixed delay, which cannot distinguish between temporary and permanent faults. Blindly reclosing can easily lead to the risk of equipment burnout.

[0004] The aforementioned problems result in low power supply reliability, high operation and maintenance costs, and prominent safety hazards in tunnel power distribution systems. There is an urgent need for a method that adapts to tunnel scenarios and achieves deep integration of protection and reclosing to solve the existing technical problems. Summary of the Invention

[0005] The main objective of this application is to provide a method and system for integrating an automatic reclosing protection device for a distribution box, so as to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, in a first aspect, this application provides a method for integrating an automatic reclosing protection device for a distribution box, comprising: Acquire current, voltage, leakage current, and surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals in the tunnel environment, to form initial multi-source raw acquisition data, and extract features from the initial multi-source raw acquisition data to obtain target multi-source feature data; The target multi-source feature data is input into the target tunnel scene fault identification model to obtain the model judgment result, wherein the model judgment result includes at least the fault type probability and the comprehensive confidence level. Based on the initial multi-source raw data and the model judgment results, the preset fault rules are matched to obtain the fault type judgment results and fault parameters. When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy and obtains the reclosing action command of the protector. When the protector triggers a tripping strategy, the fault status is verified by the target tunnel scenario fault identification model before each reclosing. When the fault type determination result is a permanent fault, the protector triggers a permanent reclosing strategy and generates a fault alarm information, obtaining the protector's blocking signal and permanent fault alarm data, wherein the fault alarm data includes at least the fault parameters; Based on the reclosing action command of the protector a preset number of times, the temporary reclosing strategy of the protector is adjusted, and the fault type determination result and the fault parameters are used as training data for the fault identification model of the target tunnel scenario.

[0007] In some feasible methods, the steps of acquiring current, voltage, leakage current, surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals in the tunnel environment to form initial multi-source raw acquisition data, and extracting features from the initial multi-source raw acquisition data to obtain target multi-source feature data, include: Using multiple sensors, the current, voltage, leakage current, surge current parameters of the power distribution circuit, as well as electromagnetic interference signals and vibration signals of the tunnel environment are collected in real time to obtain initial multi-source raw data with timestamps. Feature extraction is performed on the initial multi-source raw data with timestamps to obtain target multi-source feature data with timestamps.

[0008] In some feasible methods, the step of inputting the target multi-source feature data into the target tunnel scene fault identification model to obtain the model judgment result includes: Historical and simulated fault data in tunnel scenarios are split into three categories based on data type: the first category, the second category, and the third category. The first category includes current, voltage, leakage current, and surge current parameter data under the energized condition of the distribution box. The second category includes electromagnetic interference and vibration signal data in the tunnel environment. The third category includes response data of the fault circuit to the safe small current injected by the test excitation module under the reclosing standby condition after the power supply circuit is disconnected. Under the condition of live fault detection, a first lightweight convolutional neural network model is trained using the first type of data as training samples. This model is used to process the electrical parameter characteristics of the distribution box under the condition of live fault detection to obtain a first result. The first result includes the probability of temporary fault tendency, the probability of permanent fault tendency, and the prediction confidence level. The sum of the probability of temporary fault tendency and the probability of permanent fault tendency is 1. Using the second type of data as training samples, a gated recurrent unit network model is trained to process environmental temporal features and obtain a second result, wherein the second result includes the probability of false faults caused by environmental interference and the probability of environmental signals assisting in the confirmation of real faults. Under the reclosing standby condition after the power supply circuit is disconnected, a second lightweight convolutional neural network model is trained using the third type of data as training samples. This model is used to process the small current response characteristics and obtain a third result, which includes the probability that the fault has been eliminated and the probability that the fault has not been eliminated. Using the outputs of the first lightweight convolutional neural network model, the second lightweight convolutional neural network model, and the gated recurrent unit network model as training samples, a multi-feature fusion judgment model is trained to couple multi-dimensional results and output the model judgment result. The model judgment result includes at least the probability of temporary failure, the probability of permanent failure, and the comprehensive confidence level. The first lightweight convolutional neural network model, the gated recurrent unit network model, the second lightweight convolutional neural network model, and the multi-feature fusion judgment model are combined to form a target tunnel scene fault identification model.

[0009] In some feasible methods, the step of matching preset fault rules based on the initial multi-source raw data and the model determination result to obtain the fault type determination result and fault parameters includes: A preset fault tripping threshold benchmark value is established, which includes a current tripping threshold, a voltage tripping threshold, and a remaining leakage current tripping threshold. Under the condition of live fault detection, the fault amplitude in the initial multi-source raw data is monitored in real time. When the fault amplitude is detected to reach or exceed the preset fault tripping threshold benchmark value, a protector tripping command is generated. When the trip command of the protector is generated, the first lightweight convolutional neural network model is used to process the current first type of data to obtain the first result; In the reclosing standby condition after the power supply circuit is disconnected, a safe small current is injected into the fault circuit that has been tripped and disconnected. The gated cyclic unit network model is used to process the current second type of data to obtain the current second result, and the second lightweight convolutional neural network model is used to process the current third type of data to obtain the current third result. The first result, the current second result, and the current third result are input into the multi-feature fusion judgment model for processing to obtain the fault type judgment result and fault parameters.

[0010] In some feasible implementations, the step of triggering a temporary reclosing strategy and obtaining a reclosing action command from the protector when the fault type determination result is a temporary fault includes: A temporary reclosing strategy is constructed, wherein the temporary reclosing strategy automatically performs reclosing preparation actions according to a preset time point, performs fault status verification before each reclosing, and performs reclosing at the preset time point after the verification is passed. When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy; Based on the temporary reclosing strategy, before automatic reclosing, the target tunnel scenario fault identification model is used to process the current second type of data, the current third type of data, and the first type of data at the moment before the trip to obtain the current fault type determination result. The current fault type determination result is verified, and based on the verification result, the next action time point in the temporary reclosing strategy is determined, and whether the protector performs a reclosing action is determined. The verification includes whether the current fault type determination result is a temporary fault, comparing the fault type probability with a preset probability value, comparing the comprehensive confidence level with a preset comprehensive confidence level, and comparing the probability of a false fault caused by environmental interference with a preset interference probability threshold.

[0011] In some feasible implementation methods, the step of triggering a permanent reclosing strategy and generating fault alarm information when the fault type determination result is a permanent fault, and obtaining the protection device's blocking signal and permanent fault alarm data, includes: The permanent reclosing strategy means that after the protector is reclosed, it will always be in a locked state unless it receives operation from the host computer or staff. The fault alarm data consists of the fault parameters, the fault occurrence time, and the fault circuit number.

[0012] In some feasible methods, the steps of adjusting the temporary reclosing strategy of the protector according to the reclosing action command of the protector a preset number of times, and using the fault type determination result and the fault parameters as training data for the fault identification model of the target tunnel scenario, include: Based on the reclosing action commands of the protector and the corresponding associated data of a preset number of times, statistical data is generated, wherein the associated data includes at least the execution result of each reclosing action and the corresponding fault parameters; Based on the statistical data, the execution effect of the temporary reclosing strategy is analyzed, analysis results are generated, and the temporary reclosing strategy is adjusted based on the analysis results. Based on the fault type determination result and the fault parameters, an incremental training dataset is formed to obtain the training data for the fault identification model of the target tunnel scene.

[0013] Secondly, this application provides a fusion system for automatic reclosing protection devices in distribution boxes, applied to the aforementioned fusion method for automatic reclosing protection devices in distribution boxes, the system comprising: The acquisition unit is used to acquire current, voltage, leakage current, surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals in the tunnel environment, to form initial multi-source raw acquisition data, and to extract features from the initial multi-source raw acquisition data to obtain target multi-source feature data. The model processing unit is used to input the target multi-source feature data into the target tunnel scene fault identification model to obtain the model judgment result, wherein the model judgment result includes at least the fault type probability and the comprehensive confidence level. The protector control unit is used to match preset fault rules based on the initial multi-source raw data and the model judgment results to obtain fault type judgment results and fault parameters. When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy and obtains the reclosing action command of the protector. When the protector triggers a tripping strategy, the fault status is verified by the target tunnel scenario fault identification model before each reclosing. When the fault type determination result is a permanent fault, the protector triggers a permanent reclosing strategy and generates a fault alarm information, obtaining the protector's blocking signal and permanent fault alarm data, wherein the fault alarm data includes at least the fault parameters; The result unit is used to adjust the temporary reclosing strategy of the protector according to the reclosing action command of the protector a preset number of times, and to use the fault type determination result and the fault parameters as training data for the fault identification model of the target tunnel scene.

[0014] Thirdly, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the aforementioned method.

[0015] Fourthly, this application provides a computer program that, when executed by a processor, implements the steps of the aforementioned method.

[0016] The technical solutions provided by the embodiments of this application may include the following beneficial effects: In the fusion method of the automatic reclosing protection device for a distribution box proposed in this application, the method involves targeted collection of electromagnetic interference signals, vibration signals, and multi-source electrical parameter data of the distribution box in the tunnel environment, followed by feature extraction. Combined with a dedicated tunnel scenario fault identification model and preset fault rules for dual judgment, it achieves accurate differentiation between temporary and permanent faults. This solves the problems of existing non-reclosing schemes requiring manual on-site reset and excessively long power restoration times due to limitations imposed by maintenance windows, ensuring the continuity of power supply for the high-speed railway tunnel power distribution system. Furthermore, it overcomes the limitations of existing reclosing schemes that only have a fixed delayed closing time and cannot distinguish fault types, thus blindly applying the method. The system eliminates the safety hazard of equipment burnout caused by reclosing, and overcomes the drawbacks of the conventional separate design of protectors and external reclosing modules, which prevents data from being interconnected and results in functional fragmentation. It achieves deep integration of protection and reclosing, and can dynamically optimize the fault identification model and reclosing strategy based on actual fault data and reclosing execution data. This significantly improves the power supply reliability of the high-speed railway tunnel power distribution system, effectively reduces operation and maintenance costs, and fundamentally solves the problem of power distribution protection and reclosing control in high-speed railway tunnel power distribution boxes under special and complex environments, fully ensuring the safety of rail transit operation and the stable power supply of critical loads. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 A flowchart illustrating the integration method of an automatic reclosing protection device for a distribution box provided in this application. Detailed Implementation

[0018] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0020] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0021] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0022] Furthermore, the terms "installation," "setup," "equipped with," "connection," "linked," and "socketing" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0023] The following explanations of some terms used in this application are provided to aid in understanding the technical solution of this application: The Gated Recurrent Unit (GRU) network model is a classic improved variant of the Recurrent Neural Network (RNN) in the field of deep learning. Its design goal is to solve the gradient vanishing and gradient exploding problems that traditional RNNs are prone to when processing long sequence data. At the same time, it simplifies the complex structure of Long Short-Term Memory (LSTM) networks. While ensuring the effect of processing time-series data, it improves the training efficiency and computing speed of the model. It is one of the mainstream deep learning models for processing sequence data.

[0024] like Figure 1 As shown, in a first aspect, this application provides a method for integrating an automatic reclosing protection device in a distribution box, comprising: S100: Acquire current, voltage, leakage current, surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals in the tunnel environment, to form initial multi-source raw acquisition data, and extract features from the initial multi-source raw acquisition data to obtain target multi-source feature data.

[0025] The process involves acquiring current, voltage, leakage current, and surge current parameter data of the distribution box circuit, as well as electromagnetic interference and vibration signals from the tunnel environment, and integrating them to form initial multi-source raw acquisition data. Based on the initial multi-source raw acquisition data, feature extraction processing is performed to obtain target multi-source feature data that reflects the fault state, providing input data support for the subsequent target tunnel scene fault identification model. The feature extraction method is conventional, and this application does not limit the method of feature extraction.

[0026] Step S100 is the foundational data assurance step of the entire fusion method. Its purpose is to construct a multi-dimensional raw data system covering equipment operating status and environmental influencing factors by comprehensively collecting electrical parameters of the power distribution system and interference signals from the tunnel environment. Then, through feature extraction to remove redundant information and strengthen fault correlation features, it ensures that the subsequent fault identification model can accurately capture fault signals, improving the accuracy and reliability of protector fault judgment. All collected data and extracted features carry a unified timestamp to ensure data temporal consistency and avoid judgment errors caused by temporal misalignment.

[0027] Specifically, the obtained target multi-source feature data may include the following steps: S101 uses multiple sensors to collect real-time data on current, voltage, leakage current, surge current parameters of the power distribution circuit, as well as electromagnetic interference signals and vibration signals of the tunnel environment, to obtain initial multi-source raw acquisition data with timestamps.

[0028] Specifically, current sensors and voltage sensors are installed at the incoming and outgoing ends of the power distribution circuit in the distribution box to collect real-time current and voltage data of the circuit; a leakage current sensor is integrated inside the protection device of the distribution box to collect leakage current data of the circuit; a surge sensor is installed next to the surge protection module to collect surge impact electrical parameter data of the circuit; and environmental sensors are installed around the distribution box body and at corresponding positions on the tunnel wall to collect electromagnetic interference signals and vibration signals of the tunnel environment (among which, the electromagnetic interference signal collection covers the commonly used communication frequency band and power interference frequency band in the tunnel, and the vibration signal collection covers the range of equipment operation vibration and external impact vibration).

[0029] Furthermore, all sensors sample according to a preset sampling frequency and are labeled with a timestamp in a uniform format so that the time sequence correspondence of different types of data can be determined using the timestamp.

[0030] It should be noted that in step S101, the core electrical parameters of the power distribution system and the interference factors of the tunnel environment are fully covered by the collaborative acquisition of multiple sensors. The selection and deployment of the sensors are designed based on the characteristics of the tunnel scenario to ensure that the collected data can truly reflect the operating status of the equipment.

[0031] S102, perform feature extraction on the initial multi-source raw acquisition data with timestamps to obtain target multi-source feature data with timestamps.

[0032] Specifically, the initial multi-source raw data with timestamps obtained in S101 is used to extract features according to data type classification, resulting in target multi-source feature data with timestamps.

[0033] Furthermore, the initial multi-source raw data is divided into two categories according to data type: one is electrical parameter data (including current, voltage, leakage current, and surge electrical parameter data), and the other is environmental signal data (including electromagnetic interference signals and vibration signals). Each type of data is sorted in time stamp order to ensure temporal continuity.

[0034] It should be noted that the feature extraction of electrical parameter data involves analyzing the time-varying patterns and numerical fluctuation characteristics of each type of electrical parameter data, and extracting key features that reflect the fault state, including steady-state features (such as mean, peak value, and effective value within a certain time window) and variation features (such as numerical change rate and fluctuation amplitude). Environmental signal feature extraction: For electromagnetic interference signals, features such as peak signal amplitude and duration of interference are extracted; for vibration signals, features such as vibration frequency range and peak vibration amplitude are extracted, focusing on environmental interference features that can affect the operation of power distribution equipment and fault diagnosis.

[0035] Both electrical parameter data feature extraction and environmental signal feature extraction are conventional feature extraction operations, and this application does not limit the method used for feature extraction.

[0036] S200, input the target multi-source feature data into the target tunnel scene fault identification model to obtain the model judgment result.

[0037] The model's determination results include at least the probability of the fault type and the overall confidence level.

[0038] It should be noted that the target multi-source feature data with timestamps obtained in S102 is input into the target tunnel scenario fault identification model according to data type matching. Through the collaborative processing of multi-dimensional features by the target tunnel scenario fault identification model, the model judgment result that can support the triggering of fault strategies is output. The judgment result of the target tunnel scenario fault identification model includes at least the probability of fault type (probability of temporary fault, probability of permanent fault) and comprehensive confidence, which provides the core basis for subsequent temporary / permanent tripping strategy triggering and reclosing action judgment.

[0039] Step S200 is the core of the fault identification process in the entire fusion method. Its purpose is to take the target multi-source feature data extracted in the previous stage and use a dedicated fault identification model adapted to the tunnel scenario to achieve multi-dimensional fusion judgment of electrical parameter features, environmental features, and fault response features, avoiding the limitations of single feature judgment. The model design focuses on the scenario characteristics of the tunnel environment, such as humidity, complex electromagnetic interference, and diverse fault types. Through the collaborative work of multiple sub-models, it ensures accurate calculation of fault type probability and reliable comprehensive confidence, providing scientific decision-making support for the subsequent action strategy of the protector. At the same time, the model output results form a complete logical closed loop with the subsequent steps S300 (fault rule matching) and S400 (strategy triggering).

[0040] Specifically, obtaining the model's determination result may include the following steps: S201, the historical fault data and simulated fault data of the tunnel scene are split according to data type to form the first type of data, the second type of data and the third type of data.

[0041] The first type of data includes current, voltage, leakage current and surge current parameter data of the distribution box under energized conditions; the second type of data includes electromagnetic interference signal and vibration signal data of the tunnel environment; and the third type of data includes the response data of the fault circuit to the safe small current injected by the test excitation module under the reclosing standby condition after the power supply circuit is disconnected.

[0042] Specifically, historical fault data related to tunnel scenarios are collected and divided into three types of training data according to data type and working condition attributes, providing suitable samples for subsequent training of each sub-model. The specific steps are as follows: 1) Data Collection and Preprocessing: Two core data sets are collected: first, historical operational fault data of the tunnel power distribution box (derived from on-site maintenance records and protector fault logs, including electrical parameters, environmental signals, and fault handling results at the time of the fault); second, simulated fault data (through the construction of a tunnel power distribution simulation scenario in the laboratory, typical faults such as overload, short circuit, leakage, surge, and environmental interference are artificially simulated, and corresponding data are collected); the collected data undergoes preliminary preprocessing to remove invalid and redundant data (based on data integrity and temporal continuity judgment, such as data with missing timestamps or empty key parameters), and is labeled with a unified timestamp and fault type label.

[0043] 2) Segmentation by type and operating condition: Based on the corresponding operating conditions and data attributes, the preprocessed data is divided into three categories: The first type of data: screening fault and normal operation data of the distribution box under the live operation state (no tripping, no reclosing operation), which only includes current, voltage, leakage current and surge electrical parameter data, and is adapted to the electrical parameter feature identification requirements under the live fault detection working condition. The second type of data: Screening of tunnel environmental electromagnetic interference signals and vibration signals under all operating conditions (energized operation, tripped, and reclosing standby), including environmental signals under fault and normal operating conditions, to meet the requirements of environmental interference feature identification and false fault differentiation. The third type of data: Data that has been disconnected (after the protector trips) and is in the reclosing standby condition. It only includes the response data (such as current response amplitude, response time, etc.) fed back by the circuit after the test excitation module injects a safe small current into the fault circuit (a safe small current means that the current amplitude is within a safe range and will not trigger equipment damage or safety risks), which is adapted to the fault clearance status judgment requirements before reclosing.

[0044] 3) Data organization and archiving: Organize the three types of data according to the format of "data type-operating condition label-time stamp-value-unit" and store them as independent training sample sets to ensure that the data in each sample set is consistent in time sequence and clear in label, laying the foundation for subsequent sub-model training.

[0045] It should be noted that the principle of step S201 is the construction of scenario-adaptive samples. Tunnel scenario faults are affected by both abnormal electrical parameters and environmental interference, and the fault characteristics vary under different operating conditions (energized operation, reclosing standby). If mixed data is used to train the model, it will lead to a decrease in model recognition accuracy. By splitting the data into three categories according to type and operating condition, it can be ensured that each subsequent sub-model is trained only for specific scenarios and specific features, improving the model's ability to capture target features. At the same time, the three categories of data cover the entire fault recognition process, providing comprehensive sample support for multi-dimensional fusion judgment.

[0046] S202, Under the condition of live fault detection, a first lightweight convolutional neural network model is trained using the first type of data as training samples, which is used to process the electrical parameter characteristics of the distribution box under the condition of live operation and obtain the first result.

[0047] The first result includes a temporary failure tendency probability, a permanent failure tendency probability, and a prediction confidence level, wherein the sum of the temporary failure tendency probability and the permanent failure tendency probability is 1.

[0048] Specifically, under the condition of live fault detection, a first lightweight convolutional neural network model is trained using the first type of data as dedicated training samples. This model is used to process the electrical parameter features under the live condition in real time and output the first result. The specific steps are as follows: 1) Training sample preparation: Select samples of fault state and normal operation state from the first type of dataset, and divide them into training set, validation set and test set according to a preset ratio (e.g. 7:2:1); standardize the samples.

[0049] 2) Model Training and Optimization: Based on a lightweight convolutional neural network architecture (simplifying network layers, reducing computational complexity, and adapting to the embedded operating environment of the distribution box), the model is trained with the first type of data as input and "probability of temporary fault tendency, probability of permanent fault tendency, and prediction confidence" as output targets. During training, the core parameters of the model are adjusted in real time through the validation set to ensure that the sum of the two types of fault tendency probabilities output by the model is always 1 (i.e., probability of temporary fault tendency + probability of permanent fault tendency = 1), and that the prediction confidence can reflect the reliability of the output results (the higher the confidence, the stronger the credibility of the result).

[0050] 3) Model Deployment and Result Output: The first lightweight convolutional neural network model, which has been trained and verified through the test set, is deployed to the protector control unit. Under the condition of live fault detection, the multi-source feature data of electrical parameter targets extracted by S102 are received in real time. After being processed by the model, the first result (including the probability of temporary fault tendency, the probability of permanent fault tendency, and the prediction confidence) is output, and the timestamp is marked synchronously for subsequent fusion processing.

[0051] It should be noted that the core function of the first lightweight convolutional neural network model trained in step S202 is to capture abnormal characteristics of electrical parameters under energized conditions (such as continuously exceeding the current amplitude under overload, instantaneous surge of current under short circuit, etc.) and quantify the fault tendency type. This first lightweight convolutional neural network model adopts a lightweight architecture design. Its principle is to quickly capture key anomalies in the temporal changes of electrical parameters through a simplified feature extraction layer, achieving efficient processing without complex algorithms, thus adapting to the real-time response requirements of the distribution box. In the first output result, the temporary fault tendency probability reflects the likelihood that the abnormal electrical parameters are caused by temporary factors (such as instantaneous overload, instantaneous surge), while the permanent fault tendency probability reflects the likelihood that they are caused by permanent factors (such as line damage, equipment failure). The sum of the two is 1, based on the principle of mutual exclusion of fault types (under the same energized condition, the fault is either temporary or permanent). The prediction confidence score is used to measure the reliability of this tendency judgment.

[0052] S203, using the second type of data as training samples, a gated recurrent unit network model is trained to process the temporal features of the environment and obtain the second result.

[0053] The second result includes the probability of false faults caused by environmental interference and the probability of real faults being confirmed by environmental signals.

[0054] Specifically, using the second type of data as dedicated training samples, a gated recurrent unit network model is trained to continuously process the temporal features of the tunnel environment and output the second result. The specific steps are as follows: 1) Training sample preparation: Select samples with different environmental interference intensities and different operating states (normal and fault) from the second type of dataset, and divide them into training set, validation set and test set according to the preset ratio; retain complete time series information of the samples (arranged continuously based on timestamps) to ensure that the model can capture the time series change pattern of environmental signals.

[0055] 2) Model Training and Optimization: Based on the gated recurrent unit network architecture, the model is trained with the second type of data (electromagnetic interference signals and vibration signals) as input and "the probability of false faults caused by environmental interference and the probability of real faults confirmed by environmental signals" as output targets. During training, the model's ability to capture time-series signals is optimized to ensure that the model can distinguish between environmental interference (such as tunnel construction vibration and electromagnetic interference of communication equipment) and environmental signals associated with real faults (such as abnormal vibrations accompanying equipment failures and electromagnetic leakage accompanying line damage).

[0056] 3) Model Deployment and Result Output: Deploy the trained and validated gated recurrent unit network model to the protector control unit and set it to continuous operation mode; receive multi-source feature data of environmental signal targets extracted by S102 in real time, continuously process and output the second result (including the probability of false faults caused by environmental interference and the probability of real faults confirmed by environmental signals), and update the results in real time as the environmental signals change, and simultaneously label the timestamp.

[0057] It should be noted that the gated recurrent unit network model trained in step S203 is based on its strong ability to capture time-series signals, continuously analyze the changing patterns of tunnel environmental signals, and distinguish the correlation characteristics between environmental interference and real faults. Tunnel scenarios are characterized by complex and continuous electromagnetic interference and vibration signals. Some strong interference signals may cause abnormal electrical parameters (such as instantaneous voltage fluctuations caused by electromagnetic interference), leading to false fault judgments. The core function of this model is to avoid such misjudgments. The output probability of environmental interference causing false faults reflects the likelihood that the current environmental signal is sufficient to interfere with electrical parameter detection and cause false fault judgments. The probability of environmental signals assisting in confirming real faults reflects the likelihood that the current environmental signal (such as abnormal vibrations accompanying a fault) can corroborate that abnormal electrical parameters indicate a real fault. Together, they provide environmental dimension support for subsequent fault authenticity judgments. The model is designed for continuous operation to ensure real-time adaptation to environmental changes and improve the accuracy of fault identification throughout the entire time period.

[0058] S204, under the reclosing standby condition after the power supply circuit is disconnected, a second lightweight convolutional neural network model is trained using the third type of data as training samples to process the small current response characteristics and obtain the third result.

[0059] The third result includes the probability that the fault has been eliminated and the probability that the fault has not been eliminated.

[0060] Specifically, under the reclosing standby condition after the power supply circuit is disconnected, a second lightweight convolutional neural network model is trained using the third type of data as dedicated training samples to process small current response features and output the third result. The specific steps are as follows: 1) Training sample preparation: Select samples from the third type of dataset for two scenarios: fault eliminated and fault not eliminated (fault eliminated samples: the circuit response is normal after injecting a safe small current; fault not eliminated samples: the circuit response is abnormal after injecting a safe small current), and divide them into training set, validation set and test set according to a preset ratio; the core of the samples includes safe small current injection parameters (such as injection amplitude and frequency) and circuit response data.

[0061] 2) Model Training and Optimization: Based on a lightweight convolutional neural network architecture (compatible with the S202 model architecture for easy unified deployment), the model is trained with the third type of data as input and "probability of fault elimination and probability of fault not elimination" as output targets. During training, the model's recognition accuracy of small current response features is optimized to ensure that it can accurately distinguish the response differences between the fault elimination and non-elimination states (e.g., after the fault is eliminated, the loop impedance is normal and the response current amplitude is stable; when the fault is not eliminated, the loop impedance is abnormal and the response current amplitude is distorted).

[0062] 3) Model Deployment and Result Output: The trained and validated second lightweight convolutional neural network model is deployed to the protector control unit. Under the reclosing standby condition, when the test excitation module injects a safe small current into the fault circuit, the model receives the corresponding small current response feature data, processes it, and outputs the third result (including the probability that the fault has been eliminated and the probability that the fault has not been eliminated), and simultaneously marks the timestamp for subsequent fusion processing.

[0063] It should be noted that the second lightweight convolutional neural network model trained in step S204 is adapted to the fault verification requirements of the reclosing standby condition. Its principle is to analyze the response characteristics of the faulty circuit to a safe small current to determine whether the fault has been self-cleared (e.g., for a temporary overload fault, the circuit returns to normal after the fault is cleared, and the response to a small current is normal; for a permanent short circuit fault, the fault has not been cleared, and the response to a small current is distorted). The model adopts a lightweight architecture to ensure rapid data processing and output of results during the reclosing preparation phase, without affecting the timeliness of the reclosing action. The output probability of fault clearance and the probability of fault not clearance are based on the mutual exclusivity of fault states (the fault is either cleared or not cleared), and their sum is implicitly 1. This provides core data support for safety verification before reclosing, avoiding secondary faults caused by blind reclosing.

[0064] S205, using the output results of the first lightweight convolutional neural network model, the second lightweight convolutional neural network model, and the gated recurrent unit network model as training samples, a multi-feature fusion judgment model is trained to couple multi-dimensional results and output the model judgment result.

[0065] The model's determination results include at least the probability of temporary failure, the probability of permanent failure, and the overall confidence level.

[0066] Specifically, the outputs of three sub-models (the first lightweight convolutional neural network model, the gated recurrent unit network model, and the second lightweight convolutional neural network model) are used as joint training samples to train a multi-feature fusion judgment model, which is used to couple and process multi-dimensional results and output the final model judgment result. The specific steps are as follows: 1) Training sample preparation: Collect the output results of the three sub-models (first result, second result, and third result), align them by timestamp, and combine them with the corresponding actual fault types (labeled as temporary faults and permanent faults) to form a joint training sample set; divide it into training set, validation set, and test set according to a preset ratio.

[0067] 2) Model Training and Optimization: Based on a multi-feature fusion architecture, the model is trained with the output results of three sub-models as input and "probability of temporary failure, probability of permanent failure, and comprehensive confidence" as output targets. The training principle is to comprehensively judge the fault type by coupling the results of three dimensions: electrical parameter fault tendency, environmental interference influence, and fault elimination state, thereby avoiding the limitations of single sub-model judgment. During the training process, the weight allocation of the results of each dimension is optimized (based on the degree of influence of different results on fault judgment, such as the fault elimination state having a higher weight for the judgment of temporary faults).

[0068] 3) Model Deployment and Result Output: The trained and validated multi-feature fusion judgment model is deployed to the protector control unit. It adaptively receives and processes the output results of the corresponding sub-models according to the operating conditions. The specific process is as follows: Under live fault detection conditions, the first result output by the first lightweight convolutional neural network model and the second result output by the gated recurrent unit network model are received in real time, and temporarily stored after being aligned by timestamps. When the power supply circuit is disconnected and enters the reclosing standby condition, the reception of new first results stops (at this time, live operating condition electrical parameter data cannot be generated), and only the last set of valid first results before the disconnection is retained. Simultaneously, the gated recurrent unit network model is received in real time. The second result continuously output by the ring unit network model and the third result output by the second lightweight convolutional neural network model are used to match the retained first result with the second and third results under the current working condition according to the timestamp (the first result takes the timestamp before disconnection, and the second and third results take the current timestamp, and are aligned based on the fault handling time sequence). After multi-dimensional coupling processing, the model judgment result is output (including at least the probability of temporary failure, the probability of permanent failure, and the comprehensive confidence score). The calculation of the comprehensive confidence score is based on the weighted fusion of the confidence scores output by each sub-model. The weights can be set as needed, and this application does not limit them.

[0069] 4) Result Verification: After outputting the model's judgment result, a reasonableness verification is performed simultaneously (e.g., the sum of the probability of temporary failure and the probability of permanent failure is close to 1, and the overall confidence level is not lower than a preset threshold to ensure the validity of the result). After the verification passes, the result is stored and transmitted to subsequent steps. For example, if the overall confidence level is not lower than a preset threshold (e.g., ≥90%, which can be adjusted according to the rigor requirements of the fault judgment), and the sum of the probability of temporary failure and the probability of permanent failure is within the range of [0.95, 1.05] (allowing for minor calculation errors), the result is stored and transmitted to subsequent steps after the verification passes.

[0070] It should also be noted that the multi-feature fusion judgment model trained in step S205 achieves multi-dimensional complementary judgment. A single sub-model can only reflect one dimension of the fault characteristics (e.g., the first sub-model only reflects the electrical parameter tendency, and the second sub-model only reflects the environmental impact). By fusing the output results of the three sub-models, a more comprehensive and accurate fault judgment can be achieved. To address the issue of operating condition timing adaptation, the multi-feature fusion judgment model employs a result retention + operating condition adaptive reception mechanism during deployment to resolve the logical contradiction that the first result cannot be generated after the power supply circuit is disconnected: Under energized conditions, the first and second results are received and temporarily stored in real time; after disconnection, no new first results are received, only the last set of valid first results before disconnection (which reflects the core electrical parameter characteristics at the time of the fault) is reused. Simultaneously, the second result corresponding to the environmental signal (the gated cyclic unit network model continues to run) and the third result corresponding to the small current response are received in real time. Data alignment is then completed through timing correlation to ensure the rationality of the coupled processing. The output model judgment results provide the core decision-making basis for subsequent S300 fault rule matching and tripping strategy triggering.

[0071] S206, the first lightweight convolutional neural network model, the gated recurrent unit network model, the second lightweight convolutional neural network model, and the multi-feature fusion judgment model are combined to form a target tunnel scene fault identification model.

[0072] Specifically, the first lightweight convolutional neural network model, the gated recurrent unit network model, the second lightweight convolutional neural network model, and the multi-feature fusion judgment model, which have been trained and validated, are integrated according to a preset logic to form a complete target tunnel scene fault recognition model. The specific steps are as follows: Operating condition adaptation configuration: Configure the model with operating condition adaptive triggering logic. Under the live fault detection condition, the first lightweight convolutional neural network model and the gated recurrent unit network model are automatically started, and the first result and the second result are continuously output. Under the reclosing standby condition, the second lightweight convolutional neural network model is automatically started, and the third result is output. The multi-feature fusion judgment model runs throughout the process, and receives and processes the output results of the three sub-models in real time.

[0073] The target tunnel scenario fault identification model constructed in step S206, through the integration of four models, achieves full coverage of two major operating conditions: live fault detection and reclosing standby. This solves the problem that a single model is difficult to adapt to complex tunnel scenarios. The core logic of the model combination is that each sub-model focuses on feature processing of specific dimensions and specific operating conditions. The multi-feature fusion judgment model realizes comprehensive decision-making based on multi-dimensional results. At the same time, through the operating condition adaptive triggering and linkage optimization mechanism, it is ensured that the target tunnel scenario fault identification model can not only quickly respond to real-time data, but also continuously adapt to changes in the tunnel environment and equipment operating status. Ultimately, it achieves accurate identification of distribution box faults in tunnel scenarios, laying a core foundation for the efficient and safe execution of the automatic reclosing strategy of the protector.

[0074] S300: Based on the initial multi-source raw data and the model judgment result, perform matching of preset fault rules to obtain fault type judgment result and fault parameters.

[0075] In particular, the initial multi-source raw acquisition data (including electrical parameter data and environmental signal data) obtained in step S101, and the model judgment results (including various probabilities and comprehensive confidence) output by S200 are matched and verified against the preset fault rules. Finally, the fault type judgment result (temporary fault / permanent fault) and complete fault parameters are determined, providing a direct decision basis for the subsequent triggering of temporary / permanent tripping strategies.

[0076] Step S300 is the core hub connecting model identification and policy triggering. Its core principle is a dual verification logic of hardware threshold fallback and model judgment corroboration. The preset fault tripping threshold benchmark value is a hard condition for providing rapid response and preventing fault escalation. The model judgment result provides accurate basis for fault type differentiation, avoiding the limitations of single threshold judgment. Through the collaborative matching of initial multi-source raw data (objective operating data) and model judgment results (feature analysis results), the fault type judgment result is ensured to be true and reliable, and the fault parameters are complete and comprehensive. It connects the model output of S200 with the subsequent tripping / reclosing policy triggering, forming a complete logical chain of data acquisition, model identification, rule matching, and policy execution.

[0077] Specifically, obtaining the fault type determination result and fault parameters may include the following steps: S301, Construct a preset fault tripping threshold benchmark value, which includes a current tripping threshold, a voltage tripping threshold, and a remaining leakage current tripping threshold.

[0078] Specifically, by combining the parameters of power distribution equipment in the tunnel scenario, safety operation specifications, and historical fault data, a preset fault tripping threshold benchmark system adapted to this scenario is constructed. This system includes three core threshold categories, and the specific steps are as follows: 1) Threshold construction basis: Based on national power distribution safety standards and the rated parameters of tunnel power distribution box equipment (such as rated current and rated voltage), combined with the fault characteristics of tunnel scenarios (such as high incidence of overload, short circuit and leakage faults) and historical operation and maintenance data, the benchmark range of various thresholds is determined to ensure that the thresholds meet the safety protection requirements and avoid false triggering (such as avoiding misjudgment of instantaneous parameter fluctuations caused by tunnel environmental interference).

[0079] 2) Definition and construction of core thresholds: Current tripping threshold: There are two categories: overload current threshold and short circuit current threshold. The overload current threshold is set based on the long-term safe operation carrying capacity of the equipment (e.g., if the rated current of the equipment is 100A, the overload current threshold is set to 120A). The short circuit current threshold is set based on the insulation protection limit of the equipment (e.g., the short circuit current threshold is set to 500A). Voltage tripping threshold: There are two types of thresholds: overvoltage threshold and undervoltage threshold. The overvoltage threshold is set to 115% of the rated voltage of the equipment (e.g., if the rated voltage is 380V, the overvoltage threshold is set to 437V), and the undervoltage threshold is set to 85% of the rated voltage of the equipment (e.g., if the undervoltage threshold is set to 323V), to meet the voltage stability requirements of tunnel power distribution. Residual leakage current tripping threshold: set based on human safety current and equipment leakage protection requirements (e.g., set to 30mA). When the residual leakage current in the circuit reaches this threshold, the leakage protection trips.

[0080] 3) Threshold storage and adjustability: The three types of preset fault tripping threshold benchmark values ​​are stored in the protector control unit. Manual fine-tuning is supported according to the operation and maintenance needs of the tunnel scenario (such as equipment aging and load changes). After fine-tuning, the values ​​are automatically synchronized to the monitoring module.

[0081] S302, under the condition of live fault detection, monitors the fault amplitude in the initial multi-source raw data in real time, and when the fault amplitude is detected to reach or exceed the preset fault tripping threshold reference value, a protector tripping command is generated.

[0082] Specifically, when the distribution box is in the live fault detection mode (normal operation, no tripping triggered), the fault amplitude in the initial multi-source raw data is monitored in real time, and a tripping command is triggered based on the preset fault tripping threshold benchmark value. The specific steps are as follows: 1) Real-time monitoring and data extraction: The protector control unit extracts the current amplitude, voltage amplitude, and residual leakage current amplitude (i.e., fault amplitude, which refers to the core parameters reflecting the severity of the fault, such as peak current, effective voltage value, and instantaneous leakage current value) from the initial multi-source raw data in real time according to the preset monitoring frequency (which can be consistent with the S101 sampling frequency, such as 50Hz). It also synchronously associates the timestamp to ensure the temporal continuity of the monitoring data.

[0083] 2) Amplitude and threshold comparison: The amplitude of various faults extracted in real time is compared with the preset fault tripping threshold benchmark value. If any of the following conditions are met, it is determined that a tripping is required: ① The current amplitude reaches or exceeds the current tripping threshold (overload or short circuit); ② The voltage amplitude reaches or exceeds the voltage tripping threshold (overvoltage or undervoltage); ③ The residual leakage current amplitude reaches or exceeds the residual leakage tripping threshold.

[0084] 3) Trip command generation and recording: When the tripping conditions are met, a tripping command is immediately generated for the protector (the command includes a tripping type label, such as overload tripping or leakage tripping). The tripping trigger timestamp, the fault amplitude at the time of triggering (such as the current value and voltage value at the time of triggering the trip), the fault circuit number, and other information are recorded and stored in the local fault log. At the same time, the tripping command is sent to the protector execution unit to prepare for the execution of subsequent tripping actions.

[0085] The principle behind step S302 is threshold-priority triggering. When a fault occurs, the circuit is quickly cut off based on a clearly defined threshold standard, and then the fault type is refined through subsequent model determination. In tunnel scenarios, severe faults such as short circuits and leakage require millisecond-level responses. Relying on model determination (which is accurate but time-consuming) may delay fault handling. Therefore, this step adopts a threshold-triggered tripping logic combined with subsequent model refinement, balancing response speed and determination accuracy. Specifically, the monitoring of fault amplitude uses a real-time synchronization mechanism to ensure that parameter changes at the moment of fault occurrence (such as a sudden surge in current during a short circuit) are captured, preventing fault escalation due to monitoring delays. Relevant information is recorded synchronously when the tripping command is generated. The core purpose is to provide complete data support for subsequent fault tracing and model training, forming a closed loop with the training data from step S201.

[0086] S303, when the trip command of the protector is generated, the first lightweight convolutional neural network model is used to process the current first type of data to obtain the first result.

[0087] Specifically, while S302 generates the protector trip command, it calls the first lightweight convolutional neural network model and uses the first result output at the previous moment. The specific steps are as follows: 1) Extraction of the first type of data: Based on the timestamp triggered by the trip command, extract the first type of data (i.e., current, voltage, leakage current, and surge current parameter data under energized conditions) within a preset time window before and after the timestamp (e.g., from 0.5 seconds before the trip to the moment of the trip). Perform preliminary validity verification on the data (remove extreme abnormal values ​​caused by fault impact and judge based on data continuity) to ensure the reliability of the input data.

[0088] 2) Model invocation and data processing: The first lightweight convolutional neural network model is used to process the first type of data and output the first result.

[0089] The core function of the first result is to provide electrical parameter dimensions to support subsequent fault type determination. For example, if the tripping trigger originates from current overload, and the first result shows a higher probability of a temporary fault (possibly due to instantaneous load fluctuations), then the subsequent fusion determination is more inclined to classify it as a temporary fault. If the probability of a permanent fault is higher (possibly due to continuous overload caused by line damage), then it is more inclined to classify it as a permanent fault. The connection between this step and S302 ensures safety and lays the foundation for subsequent refined determination.

[0090] S304, under the reclosing standby condition after the power supply circuit is disconnected, a safe small current is injected into the fault circuit that has been tripped and disconnected, the current second type of data is processed using the gated cyclic unit network model to obtain the current second result, and the current third type of data is processed using the second lightweight convolutional neural network model to obtain the current third result.

[0091] Specifically, after the power supply circuit disconnects in response to the trip command, it enters the reclosing standby mode. By injecting a small safe current and calling the corresponding model to process the data, the current second and third results are obtained. The specific steps are as follows: 1) Operating condition switching and safe small current injection: After the power supply circuit is completely disconnected, the protector automatically switches to the reclosing standby condition, starts the test excitation module, and injects a safe small current into the tripped fault circuit (the injection current amplitude is set to the equipment's safe range, such as 5-10A, which is far below the short-circuit current threshold to ensure that the equipment will not be damaged or safety risks will be caused; the injection frequency is compatible with the S101 sampling frequency, such as 50Hz).

[0092] 2) Current second type of data processing and second result output: The gated loop unit network model keeps running continuously (without interruption, adapting to full working condition environmental signal monitoring), extracts the current second type of data (i.e., tunnel environmental electromagnetic interference signal and vibration signal data) in real time, processes it based on the principle of environmental feature recognition, and outputs the current second result (including the probability of false faults caused by environmental interference and the probability of environmental signals assisting in confirming real faults), and simultaneously marks the current timestamp.

[0093] 3) Current third type of data processing and third result output: Synchronously collect the response data of the fault circuit to the safe small current (i.e., the current third type of data, including core parameters such as response current amplitude and response time). After the data is initially processed (removing the acquisition noise and judging the validity based on the response pattern), it is input into the second lightweight convolutional neural network model. After the model is processed based on the principle of small current response feature recognition, the current third result (including the probability that the fault has been eliminated and the probability that the fault has not been eliminated) is output, and the current timestamp is marked synchronously.

[0094] 4) Result association storage: The current second result and the current third result are associated by timestamp, and the trip instruction information of S302 and the first result of S303 are also associated and stored in the local cache to prepare for subsequent fusion judgment.

[0095] The principle of step S304 is to complete the state before reclosing. After the power supply circuit is disconnected, it is impossible to determine whether the fault has been eliminated based solely on the electrical parameter characteristics at the moment of tripping (the first result) (for example, a temporary overload fault may recover on its own after disconnection, while a permanent short circuit fault will persist). Therefore, by injecting a safe small current and monitoring environmental signals and small current response signals, the current fault state information is completed. The core design of the safe small current injection is safety and identifiability: the amplitude is within the equipment's safe range, and the difference in circuit response can distinguish whether the fault has been eliminated (if the fault has been eliminated, the circuit impedance is normal and the response current is stable; if the fault has not been eliminated, the circuit impedance is abnormal and the response current is distorted). The gated cyclic unit network model runs continuously, primarily to capture changes in environmental interference under reclosing standby conditions (such as tunnel construction vibration and enhanced electromagnetic interference), avoiding misjudgments of the fault state caused by environmental factors. The acquisition of the current second and third results, together with the first result of S303, forms a complete data chain of historical features and current state, providing comprehensive support for accurate fault type determination.

[0096] S305, the first result, the current second result, and the current third result are input into the multi-feature fusion judgment model for processing to obtain the fault type judgment result and fault parameters.

[0097] Specifically, the first result obtained in S303, the current second result obtained in S304, and the current third result are input into the multi-feature fusion judgment model for coupling processing, and finally output the fault type judgment result and complete fault parameters. The specific steps are as follows: 1) Model Coupling Processing: The aligned three types of results are input into the multi-feature fusion judgment model. The multi-feature fusion judgment model performs coupling processing based on the principle of multi-dimensional complementary verification. It combines the electrical parameter fault tendency of the first result, the environmental interference influence of the current second result, and the fault elimination status of the current third result to comprehensively judge the type of fault (temporary fault / permanent fault) and calculate the reliability of the fault type judgment (i.e., comprehensive confidence).

[0098] 2) Fault type determination: Based on the model coupling processing results, the fault type determination results are clarified: if the comprehensive determination is "the fault can be eliminated by itself and there is no continuous safety risk" (such as temporary overload, instantaneous surge), it is determined to be a temporary fault; if the comprehensive determination is "the fault cannot be eliminated by itself and there is a continuous safety risk" (such as line damage and short circuit, equipment leakage fault), it is determined to be a permanent fault; the determination results are simultaneously labeled with the comprehensive confidence level.

[0099] 3) Result output and storage: The fault type determination result (temporary fault / permanent fault) and complete fault parameters are synchronously output to the subsequent policy triggering stage, and stored in the local fault log of the protector and the upper-level monitoring system to ensure that the fault data is traceable.

[0100] The core principle of multi-feature fusion judgment is to avoid the limitations of single-dimensional judgment and achieve accurate and reliable fault type judgment. A single first result can only reflect the electrical parameter trend, a single second result can only reflect the environmental influence, and a single third result can only reflect the fault elimination status. By coupling and processing the three types of results, the fault type that best fits the reality can be obtained by comprehensively considering various factors.

[0101] The core objective of this step is to determine the fault type and output the fault parameters: the fault type directly determines whether to trigger a temporary reclosing strategy (temporary fault) or a permanent tripping strategy (permanent fault), while the fault parameters provide complete data support for fault tracing, operation and maintenance, and model iteration training, forming a closed loop with the model training mentioned earlier and the strategy triggering mentioned later, ensuring the integrity and practicality of the entire fusion method.

[0102] It should be noted that after obtaining the fault type determination result and fault parameters in step S300, the corresponding operation needs to be performed according to the corresponding fault type, as follows: When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy and obtains the reclosing action command of the protector.

[0103] Specifically, when the protector triggers the tripping strategy, the fault status is verified by the target tunnel scenario fault identification model before each reclosing of the protector.

[0104] When the fault type determination result is a permanent fault, the protector triggers a permanent reclosing strategy and generates a fault alarm message, thereby obtaining the protector's blocking signal and permanent fault alarm data.

[0105] The fault alarm data includes at least the fault parameters.

[0106] Specifically, obtaining the reclosing operation command of the protector may include the following steps: Step 1: Construct a temporary reclosing strategy.

[0107] The temporary reclosing strategy involves automatically performing reclosing preparation actions based on a preset time point. Before each reclosing, a fault status verification is completed, and after the verification is passed, the reclosing is performed at the preset time point.

[0108] Specifically, based on the operational requirements of the power distribution system in the tunnel scenario, equipment safety specifications, and experience in handling temporary faults, a temporary reclosing strategy adapted to this scenario is constructed.

[0109] Preset time points: These can be set to 5 seconds, 1 minute, 5 minutes, etc., with a maximum of 8 preset time points. When the preset time point is reached, a reclosing operation is performed to restore power in case the temporary fault is cleared. Before reclosing, a fault status verification must be completed. Only after the verification passes will the protector initiate the reclosing action. In other words, reclosing can be performed after verification; if the verification fails, reclosing at the current preset time point is blocked, awaiting verification before the next preset time point.

[0110] For example, multiple sets of reclosing action time points can be preset (such as the first reclosing time point T1 = 5 seconds after tripping, the second T2 = 60 seconds, the third T3 = 300 seconds, the fourth T4 = 600 seconds... up to 8 times).

[0111] The verification is performed within 0-5 seconds of T1. If the verification passes, the circuit breaker is reclosed; otherwise, the reclosing is skipped. The verification is performed again within 5-60 seconds. If the verification passes, the circuit breaker is reclosed; otherwise, the reclosing at time T2 is skipped, and so on.

[0112] Step 2: When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy.

[0113] Specifically, when the protector receives and confirms that the fault type determination result output by S305 is a temporary fault, it immediately triggers the established temporary reclosing strategy and enters the reclosing execution process. The specific steps are as follows: Confirmation of judgment result: The protector control unit receives the fault type judgment result (temporary fault) and the corresponding comprehensive confidence level output by S305. First, it verifies the validity of the judgment result. Only when the comprehensive confidence level is ≥ the preset confidence threshold (e.g., 0.9, consistent with the judgment validity threshold of S305) is the judgment result confirmed to be reliable and the strategy triggering process is started.

[0114] Strategy Triggering and State Switching: After confirming the validity of the judgment result, the temporary reclosing strategy is triggered, and the protector switches from "reclosing standby mode" to "temporary reclosing execution mode". The strategy trigger timestamp is recorded synchronously (associated with the trip timestamp for easy timing traceability).

[0115] Step 3: Based on the temporary reclosing strategy, before automatic reclosing, the target tunnel scenario fault identification model is used to process the current second type of data, the current third type of data, and the first type of data at the moment before the trip to obtain the current fault type determination result.

[0116] Specifically, based on the temporary reclosing strategy, before the reclosing action is executed at each preset time point, the target tunnel scenario fault identification model is invoked to process the specified data and output the current fault type determination result. The specific steps are as follows: 1) Data extraction and preprocessing: Extract the first type of data at the moment before the trip: Extract the first type of data (current, voltage, leakage current, and surge current parameters under energized conditions) stored in S303 at the moment of trip (before the power supply circuit is disconnected) from the local cache of the protector. This data has been validated and can be directly used for model processing (at this time, the power supply circuit is disconnected and new first type of data cannot be collected, so the data stored before the trip is reused).

[0117] Collect and extract the current second type of data: collect electromagnetic interference signals and vibration signals of the current tunnel environment in real time through environmental sensors (i.e., the current second type of data). The collection frequency is consistent with S101 (e.g., 50Hz). Perform preliminary preprocessing on the collected data (remove collection noise and remove invalid data based on signal baseline values) to ensure that the data can reflect the true state of the current environment.

[0118] Collect and extract the current third type of data: Inject a safe small current into the faulty circuit through the test excitation module (consistent with the S304 injection parameters, such as 5-10A, to ensure safety), collect the circuit's response data to the safe small current in real time (i.e., the current third type of data, including core parameters such as response current amplitude and response time), and perform preliminary data processing (remove extreme outliers and verify validity based on response patterns).

[0119] 2) Data timing alignment: The first type of data, the current second type of data, and the current third type of data at the moment before the trip are time-series correlated and aligned to ensure that all three types of data correspond to the current temporary fault, and to avoid timing misalignment leading to model processing errors.

[0120] 3) Model collaborative processing: The aligned three types of data are input into the target tunnel scene fault identification model, and the model works collaboratively according to the preset logic.

[0121] Step four: Verify the current fault type determination result, and based on the verification result, determine the next action time point in the temporary reclosing strategy and whether the protector should perform the reclosing action.

[0122] The verification includes determining whether the current fault type is a temporary fault, comparing the fault type probability with a preset probability value, comparing the overall confidence level with a preset overall confidence level, and comparing the tunnel environment electromagnetic interference signal and vibration signal with a preset interference threshold.

[0123] Specifically, the current fault type determination result and related parameters output in step 3 are verified in multiple dimensions. Based on the verification results, it is determined whether the protector will perform a reclosing action at the next action time point of the temporary reclosing strategy. The specific steps are as follows: 1) Verification Content: Based on the preset verification benchmark, four core verifications can be performed. The verification content and principles of each verification are as follows: Verification item 1: Whether the current fault type determination result is a temporary fault. Principle: Only when the current determination result is still a temporary fault is the basic condition for reclosing met (if it is determined to be a permanent fault, reclosing is directly blocked).

[0124] Verification Item 2: Comparison of fault type probability with preset probability value. Principle: Extract the temporary fault probability from the current fault type determination result and compare it with the preset temporary fault probability threshold (e.g., 0.9). Only when the temporary fault probability is ≥ the preset threshold is it confirmed that the current fault still has the characteristic of self-elimination.

[0125] Verification Item 3: Comparison of overall confidence level with preset overall confidence level. Principle: Extract the overall confidence level of the current fault type determination result and compare it with the preset confidence threshold (e.g., 0.9) to ensure the reliability of the current determination result and avoid safety risks caused by misjudgment due to low confidence level.

[0126] Verification Item 4: Comparison of tunnel environmental electromagnetic interference signals and vibration signals with preset interference thresholds. Principle: Extract the "probability of environmental interference causing false faults" output from the gated cyclic unit network model and compare it with a preset interference probability threshold (e.g., 0.1) to ensure that the current environmental interference will not lead to false fault judgments or affect the stable operation of the equipment after reclosing. It should be noted that the second result includes the probability of environmental interference causing false faults and the probability of environmental signals assisting in confirming real faults. Focusing on the probability of environmental interference causing false faults, compare it with the preset interference probability threshold (e.g., 0.1). When the probability of environmental interference causing false faults is ≤ the preset interference probability threshold, it is confirmed that the current environmental interference will not lead to false fault judgments or affect the stable operation of the equipment after reclosing (Logical adaptation explanation: The two probabilities of the second result are complementary and correlated. The lower the probability of environmental interference causing false faults, the higher the probability of environmental signals assisting in confirming real faults. There is no need to verify the latter separately; the former alone is sufficient to effectively determine the impact of environmental interference).

[0127] 2) Verification result determination: All four verification items must be satisfied simultaneously (i.e. all pass) for the verification to be considered passed; if any one of the verification items fails, the verification is considered failed.

[0128] 3) Reclosing action decision: If the verification passes: at the next preset action time point of the temporary reclosing strategy (e.g., T1 = 5 seconds after tripping), generate and output the reclosing action command of the protector, and synchronously record data such as verification pass information and reclosing time point; the command is sent to the protector execution unit to prepare for the reclosing action.

[0129] If the verification fails: the reclosing action at time T1 is blocked, and the power supply circuit remains disconnected; the process proceeds to T1 and T2 (the second time point, forming a time window for verification). If the verification passes: the closing action is executed, and the process ends; if the verification still fails, the process proceeds to T2 and T3 (the third time point, forming a time window for verification). At this time, the reclosing action at time T2 is blocked until the maximum number of attempts is reached. If the process still fails, it is determined to be a permanent fault, and the protector will no longer attempt reclosing.

[0130] Verification item 1 ensures that the fault type has not fundamentally changed (it remains a temporary fault); verification item 2 ensures that the probability of the fault clearing itself is high enough; verification item 3 ensures that the judgment result is reliable; and verification item 4 ensures that environmental interference will not affect the safety of reclosing and the stable operation of the equipment. The core of this design, which requires all four verifications to be met simultaneously, is to improve the safety of reclosing operations.

[0131] In addition, obtaining the lockout signal and permanent fault alarm data of the protector may include the following steps: Step 1: The permanent reclosing strategy means that after the protector is reclosed, it will always be in a locked state unless it receives operation from the host computer or staff.

[0132] Step two, the fault alarm data consists of the fault parameters, the fault occurrence time, and the fault circuit number.

[0133] Specifically, when the temporary reclosing strategy terminates and a permanent fault is determined, the protector automatically triggers the permanent fault blocking mechanism and generates alarm data. The blocking signal and permanent fault alarm data are obtained through the following steps to ensure stable fault state management and traceable operation and maintenance: The core of the permanent fault blocking strategy is that when the protector determines a permanent fault, it will always be in the reclosing blocking state unless it receives remote operation instructions from the host computer or on-site operation instructions from the staff.

[0134] Alarm data: The fault parameters are extracted from the protector's local cache. The complete fault parameters for this fault include: basic fault triggering parameters (fault type label that triggers tripping, fault amplitude, tripping timestamp), model output related parameters (core values ​​of the first result, the current second result, and the current third result), and fault severity parameters.

[0135] Fault occurrence time confirmation: The timestamp of the trip command generated by S302 is used as the "fault occurrence reference time". At the same time, the "permanent fault determination timestamp" and "blocking signal generation timestamp" are supplemented to form a complete time chain. The time format is uniformly "year-month-day hour:minute:second.millisecond" and consistent with the time recording standard of the previous steps, which facilitates time sequence traceability.

[0136] Fault circuit number confirmation: Extract the circuit number associated with this fault. This number is a unique identifier preset by the protector (such as L1, L2, L3 power distribution circuit, or tunnel zone power distribution circuit number). After extraction, compare it with the circuit information database to confirm the physical circuit location corresponding to the number (such as "tunnel zone 1 lighting power distribution circuit"). Add it to the alarm data to facilitate maintenance personnel to quickly locate the fault location.

[0137] 2) Alarm data formatting and organization: The above three types of core data are organized according to the preset format to form structured permanent fault alarm data. The data format is adapted to the receiving specifications of the host computer monitoring system and includes a data header (identified as "permanent fault alarm"), core data segments (fault parameters, time chain, loop information), and check code (to ensure that the data transmission process is not tampered with); at the same time, local visual alarm information is generated (adapted to the local display screen of the protector).

[0138] Fault parameters help maintenance personnel determine the severity and type of faults (such as leakage faults and short circuit faults), the time chain can trace the fault development process (tripping-reclosing verification-permanent determination), and the fault circuit number can quickly locate the physical fault location. The combination of these three features can significantly improve the efficiency of maintenance and repair. The design of data formatting and multi-channel output is suitable for the dual management mode of remote monitoring and on-site maintenance in tunnel scenarios, ensuring that no alarm information is missed.

[0139] S400, adjust the temporary reclosing strategy of the protector according to the reclosing action command of the protector a preset number of times, and use the fault type determination result and the fault parameters as training data for the fault identification model of the target tunnel scene.

[0140] The S400 step operates on a data-driven closed-loop optimization principle. The fault handling processes in the preceding steps (steps 1-5) generate a large amount of real-world scenario data (reclosing action commands, execution results, fault parameters, judgment results, etc.). This data contains the actual characteristics of tunnel scenario faults and the adaptation patterns of strategy operation. Through data analysis in the S400 step, insufficient adaptation of temporary reclosing strategies can be accurately identified (such as unreasonable preset time points or poor adaptability of verification thresholds), and the handling effect can be optimized through strategy adjustments. Simultaneously, converting real fault data into model training data allows the target tunnel scenario fault identification model to continuously learn new features and patterns of faults within the scenario, preventing a decrease in judgment accuracy due to scenario changes (such as increased environmental interference or equipment aging), and ensuring the long-term reliability and adaptability of the entire fault handling solution. The two core objectives of this step (strategy adjustment and model training data preparation) are independent yet synergistically supportive, jointly improving the intelligence level of fault handling in tunnel power distribution systems.

[0141] Specifically, step S400 may include the following steps: S401, statistical data is generated based on the reclosing operation commands of the protector and the corresponding associated data of the preset number of times.

[0142] The associated data includes at least the execution result of each reclosing action and the corresponding fault parameters.

[0143] Specifically, statistical data is generated by statistically analyzing the reclosing operation commands of the protector and the corresponding associated data a preset number of times. In addition, based on the real data, an appropriate amount of simulated fault data is derived. The simulated fault data is used as an incremental statistical data and merged with the real statistical data to form a statistical data set.

[0144] S402, based on the statistical data, analyze the execution effect of the temporary reclosing strategy, form an analysis result, and adjust the temporary reclosing strategy according to the analysis result.

[0145] The effectiveness of the temporary reclosing strategy is analyzed based on the statistical data set, which may specifically include: The closing success rate is the percentage of successful closing after the reclosing action command is executed within a preset number of counts, reflecting the strategy's ability to quickly restore power supply. Temporary fault misjudgment rate: The proportion of faults judged as temporary faults but actually permanent faults (tripping immediately after closing) within a preset number of counts reflects the accuracy of the strategy judgment. The timeliness of the strategy response is measured by the average time from the occurrence of a fault to successful closing within a preset number of counts, which reflects the response efficiency of the strategy. The rationality of the number of overlaps is determined by statistically analyzing the distribution ratio of successful closing at each preset time point (T1-T8) and during the supplementary verification stage within the preset number of overlaps, reflecting the compatibility between the preset time points and the supplementary verification interval. Equipment safety adaptability is measured by the percentage of times the equipment operates without abnormality after successful closure within a preset number of counts, reflecting the strategy's ability to ensure equipment safety.

[0146] The specific analysis method can use conventional statistical analysis methods in mathematics, and this application does not limit the statistical analysis method.

[0147] Based on the analysis results obtained from statistical analysis methods, the temporary reclosing strategy was fine-tuned and then simulated to verify the feasibility of the fine-tuned temporary reclosing strategy.

[0148] S403, Based on the fault type determination result and the fault parameters, an incremental training dataset is formed to obtain the training data for the target tunnel scene fault identification model.

[0149] Specifically, based on the fault type determination results and the fault parameters that can be applied to the fault identification model of the target tunnel scene, an incremental training dataset is formed as the training data of the fault identification model of the target tunnel scene, so that the output results of the fault identification model of the target tunnel scene are more accurate.

[0150] Secondly, this application provides a fusion system for automatic reclosing protection devices in distribution boxes, applied to the aforementioned fusion method for automatic reclosing protection devices in distribution boxes, the system comprising: The acquisition unit is used to acquire current, voltage, leakage current, surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals in the tunnel environment, to form initial multi-source raw acquisition data, and to extract features from the initial multi-source raw acquisition data to obtain target multi-source feature data. The model processing unit is used to input the target multi-source feature data into the target tunnel scene fault identification model to obtain the model judgment result, wherein the model judgment result includes at least the fault type probability and the comprehensive confidence level. The protector control unit is used to match preset fault rules based on the initial multi-source raw data and the model judgment results to obtain fault type judgment results and fault parameters. When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy and obtains the reclosing action command of the protector. When the protector triggers a tripping strategy, the fault status is verified by the target tunnel scenario fault identification model before each reclosing. When the fault type determination result is a permanent fault, the protector triggers a permanent reclosing strategy and generates a fault alarm information, obtaining the protector's blocking signal and permanent fault alarm data, wherein the fault alarm data includes at least the fault parameters; The result unit is used to adjust the temporary reclosing strategy of the protector according to the reclosing action command of the protector a preset number of times, and to use the fault type determination result and the fault parameters as training data for the fault identification model of the target tunnel scene.

[0151] Thirdly, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the aforementioned method.

[0152] Fourthly, this application provides a computer program that, when executed by a processor, implements the steps of the aforementioned method.

[0153] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0154] Obviously, those skilled in the art should understand that the various units or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0155] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method of fusing an automatic reclosing protector of a distribution box, characterized in that, include: Acquire current, voltage, leakage current, and surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals in the tunnel environment, to form initial multi-source raw acquisition data, and extract features from the initial multi-source raw acquisition data to obtain target multi-source feature data; The target multi-source feature data is input into the target tunnel scene fault identification model to obtain the model judgment result, wherein the model judgment result includes at least the fault type probability and the comprehensive confidence level. Based on the initial multi-source raw data and the model judgment results, the preset fault rules are matched to obtain the fault type judgment results and fault parameters. When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy and obtains the reclosing action command of the protector. When the protector triggers a tripping strategy, the fault status is verified by the target tunnel scenario fault identification model before each reclosing. When the fault type determination result is a permanent fault, the protector triggers a permanent reclosing strategy and generates a fault alarm information, obtaining the protector's blocking signal and permanent fault alarm data, wherein the fault alarm data includes at least the fault parameters; Based on the reclosing action command of the protector a preset number of times, the temporary reclosing strategy of the protector is adjusted, and the fault type determination result and the fault parameters are used as training data for the fault identification model of the target tunnel scenario.

2. The method of claim 1, wherein the automatic reclosing protector of the distribution box is a circuit breaker. The steps of acquiring current, voltage, leakage current, and surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals from the tunnel environment to form initial multi-source raw acquisition data, and extracting features from the initial multi-source raw acquisition data to obtain target multi-source feature data, include: Using multiple sensors, the current, voltage, leakage current, surge current parameters of the power distribution circuit, as well as electromagnetic interference signals and vibration signals of the tunnel environment are collected in real time to obtain initial multi-source raw data with timestamps. Feature extraction is performed on the initial multi-source raw data with timestamps to obtain target multi-source feature data with timestamps.

3. The method of claim 1, wherein the power box automatic reclosing protector is a circuit breaker. The step of inputting the target multi-source feature data into the target tunnel scene fault identification model to obtain the model judgment result includes: Historical and simulated fault data in tunnel scenarios are split into three categories based on data type: the first category, the second category, and the third category. The first category includes current, voltage, leakage current, and surge current parameter data under the energized condition of the distribution box. The second category includes electromagnetic interference and vibration signal data in the tunnel environment. The third category includes response data of the fault circuit to the safe small current injected by the test excitation module under the reclosing standby condition after the power supply circuit is disconnected. Under the condition of live fault detection, a first lightweight convolutional neural network model is trained using the first type of data as training samples. This model is used to process the electrical parameter characteristics of the distribution box under the condition of live fault detection to obtain a first result. The first result includes the probability of temporary fault tendency, the probability of permanent fault tendency, and the prediction confidence level. The sum of the probability of temporary fault tendency and the probability of permanent fault tendency is 1. Using the second type of data as training samples, a gated recurrent unit network model is trained to process environmental temporal features and obtain a second result, wherein the second result includes the probability of false faults caused by environmental interference and the probability of environmental signals assisting in the confirmation of real faults. Under the reclosing standby condition after the power supply circuit is disconnected, a second lightweight convolutional neural network model is trained using the third type of data as training samples. This model is used to process the small current response characteristics and obtain a third result, which includes the probability that the fault has been eliminated and the probability that the fault has not been eliminated. Using the outputs of the first lightweight convolutional neural network model, the second lightweight convolutional neural network model, and the gated recurrent unit network model as training samples, a multi-feature fusion judgment model is trained to couple multi-dimensional results and output the model judgment result. The model judgment result includes at least the probability of temporary failure, the probability of permanent failure, and the comprehensive confidence level. The first lightweight convolutional neural network model, the gated recurrent unit network model, the second lightweight convolutional neural network model, and the multi-feature fusion judgment model are combined to form a target tunnel scene fault identification model.

4. The method of claim 3, wherein the power box automatic reclosing protector is a circuit breaker. The step of matching preset fault rules based on the initial multi-source raw data and the model determination result to obtain fault type determination results and fault parameters includes: A preset fault tripping threshold benchmark value is established, which includes a current tripping threshold, a voltage tripping threshold, and a remaining leakage current tripping threshold. Under the condition of live fault detection, the fault amplitude in the initial multi-source raw data is monitored in real time. When the fault amplitude is detected to reach or exceed the preset fault tripping threshold benchmark value, a protector tripping command is generated. When the trip command of the protector is generated, the first lightweight convolutional neural network model is used to process the current first type of data to obtain the first result; In the reclosing standby condition after the power supply circuit is disconnected, a safe small current is injected into the fault circuit that has been tripped and disconnected. The gated cyclic unit network model is used to process the current second type of data to obtain the current second result, and the second lightweight convolutional neural network model is used to process the current third type of data to obtain the current third result. The first result, the current second result, and the current third result are input into the multi-feature fusion judgment model for processing to obtain the fault type judgment result and fault parameters.

5. The integration method of the automatic reclosing protection device in the distribution box as described in claim 4, characterized in that, The step of triggering a temporary reclosing strategy and obtaining a reclosing action command from the protector when the fault type determination result is a temporary fault includes: A temporary reclosing strategy is constructed, wherein the temporary reclosing strategy automatically performs reclosing preparation actions according to a preset time point, performs fault status verification before each reclosing, and performs reclosing at the preset time point after the verification is passed. When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy; Based on the temporary reclosing strategy, before automatic reclosing, the target tunnel scenario fault identification model is used to process the current second type of data, the current third type of data, and the first type of data at the moment before the trip to obtain the current fault type determination result. The current fault type determination result is verified, and based on the verification result, the next action time point in the temporary reclosing strategy is determined, and whether the protector performs a reclosing action is determined. The verification includes whether the current fault type determination result is a temporary fault, comparing the fault type probability with a preset probability value, comparing the comprehensive confidence level with a preset comprehensive confidence level, and comparing the probability of false faults caused by environmental interference with a preset interference probability threshold.

6. The integration method of the automatic reclosing protection device in the distribution box as described in claim 1, characterized in that, The step of triggering a permanent reclosing strategy and generating fault alarm information when the fault type determination result is a permanent fault, and obtaining the protection device's blocking signal and permanent fault alarm data, includes: The permanent reclosing strategy means that after the protector is reclosed, it will always be in a locked state unless it receives operation from the host computer or staff. The fault alarm data consists of the fault parameters, the fault occurrence time, and the fault circuit number.

7. The integration method of the automatic reclosing protection device in the distribution box as described in claim 1, characterized in that, The steps of adjusting the temporary reclosing strategy of the protector according to the reclosing action command of the protector according to a preset number of times, and using the fault type determination result and the fault parameters as training data for the fault identification model of the target tunnel scenario, include: Based on the reclosing action commands of the protector and the corresponding associated data of a preset number of times, statistical data is generated, wherein the associated data includes at least the execution result of each reclosing action and the corresponding fault parameters; Based on the statistical data, the execution effect of the temporary reclosing strategy is analyzed, analysis results are generated, and the temporary reclosing strategy is adjusted based on the analysis results. Based on the fault type determination result and the fault parameters, an incremental training dataset is formed to obtain the training data for the fault identification model of the target tunnel scene.

8. A fusion system for an automatic reclosing protection device in a distribution box, characterized in that, The fusion method applied to the automatic reclosing protection device of the distribution box according to any one of claims 1-7, the system comprising: The acquisition unit is used to acquire current, voltage, leakage current, surge current parameter data in the distribution box, as well as electromagnetic interference signals and vibration signals in the tunnel environment, to form initial multi-source raw acquisition data, and to extract features from the initial multi-source raw acquisition data to obtain target multi-source feature data. The model processing unit is used to input the target multi-source feature data into the target tunnel scene fault identification model to obtain the model judgment result, wherein the model judgment result includes at least the fault type probability and the comprehensive confidence level. The protector control unit is used to match preset fault rules based on the initial multi-source raw data and the model judgment results to obtain fault type judgment results and fault parameters. When the fault type determination result is a temporary fault, the protector triggers a temporary reclosing strategy and obtains the reclosing action command of the protector. When the protector triggers a tripping strategy, the fault status is verified by the target tunnel scenario fault identification model before each reclosing. When the fault type determination result is a permanent fault, the protector triggers a permanent reclosing strategy and generates a fault alarm information, obtaining the protector's blocking signal and permanent fault alarm data, wherein the fault alarm data includes at least the fault parameters; The result unit is used to adjust the temporary reclosing strategy of the protector according to the reclosing action command of the protector a preset number of times, and to use the fault type determination result and the fault parameters as training data for the fault identification model of the target tunnel scene.

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 method according to any one of claims 1 to 7.

10. A computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images