A non-intrusive method and system for arc risk assessment in vacuum switches

Abstract

This application provides a non-invasive method and system for assessing arc risk in vacuum switches. The method includes: acquiring voltage and current during the opening and closing process of the vacuum switch in a visible environment with an observation window, and classifying and labeling the arc voltage and current based on the arc morphology observed from the window; performing window-level segmentation and other necessary data preprocessing on the data; using the preprocessed arc voltage and current signals as input to an arc identification model to complete model pre-training; acquiring voltage and current signals in real time in an engineering environment without visibility, performing the same data preprocessing process as in the visible environment, and inputting the data into the model to obtain the arc pattern probability distribution of the window-level data; and obtaining the arc pattern through an integrated voting decision mechanism based on the arc pattern probability distribution, thereby enabling accurate identification of the arc pattern of the vacuum switch in engineering environments where visibility is difficult to achieve.

Description

Technical Field

[0001] This application relates to the field of electrical equipment condition monitoring and fault diagnosis technology, and in particular to a non-invasive method and system for assessing arc risk in vacuum switches. Background Technology

[0002] As a core switching device in medium-voltage power distribution systems, the performance and lifespan of vacuum switches directly depend on the intensely burning vacuum arc within the vacuum interrupter. Accurately identifying the risk of arcing, especially the timely recognition of high-risk arc states such as "anode spot arcs" indicating severe contact erosion and "point spot arcs" that can evolve into anode spots, is crucial for assessing switch status, predicting its remaining electrical life, and implementing early warning maintenance. This is of great significance for improving the reliability of the switching system and avoiding catastrophic failures.

[0003] Currently, the most direct and mainstream method for identifying arc modes in vacuum switches is visual observation based on high-speed photography. However, this method is difficult to use in practical power systems, and its main drawbacks in engineering applications are as follows: First, capturing arc images requires destructive structural modifications to the vacuum interrupter to create observation windows and relies on expensive high-speed camera systems, which are practically impossible in actual vacuum switches. Second, the high cost and complexity of arc imaging systems make it difficult to widely promote and popularize.

[0004] It is evident that existing technologies lack a method that does not rely on optical signals and can perform online detection and risk assessment of electric arc conditions in practical engineering applications. Summary of the Invention

[0005] This application provides a non-invasive method and system for assessing arc risk in vacuum switches, which solves the problem of reliance on visual observation and low timeliness in the prior art. It does not require any physical modification to the vacuum switch and can be directly applied to real-time monitoring and diagnosis of arc risk in actual operating vacuum switch equipment.

[0006] To achieve the above objectives, the technical solution of this application embodiment is as follows: In a first aspect, embodiments of this application provide a non-invasive method for assessing arc risk in vacuum switches, the method comprising: In a vacuum interrupter experimental platform equipped with an observation window, arc voltage and arc current signals are simultaneously acquired during the interruption process. Based on the arc morphology observed through the observation window, arc mode classification and labeling are performed to establish an arc voltage and current dataset with arc mode labels. The arc voltage and current dataset is preprocessed by window segmentation and standardization, and the window-level data is input into a preset classification model for training to obtain an arc recognition model. In engineering environments without visibility, the target arc voltage and current signals are non-invasively acquired using voltage and current sensors, and the same preprocessing procedure is performed before model training. The preprocessed target arc voltage signal and target arc current signal are used as inputs to the arc identification model to obtain the arc mode probability distribution output by the arc identification model. Based on the probability distribution of the electric arc mode, the electric arc mode is obtained through an integrated voting decision-making mechanism; When the arc mode is a dotted spot arc, a risk warning message is output; the risk warning message is used to alert the user that the vacuum switch poses a safety risk. When the arc mode is an anode spot arc, an alarm message is output; the alarm message is used to trigger an alarm signal.

[0007] In one possible implementation, the arc voltage signal and arc current signal are preprocessed by window segmentation and normalization in the following manner: extracting the effective arc segment from the arc voltage signal and arc current signal; slicing the arc voltage signal and arc current signal using a sliding window; and normalizing the arc voltage signal and arc current signal using a normalization method.

[0008] In one possible implementation, the arc identification model is a model that processes time-series data.

[0009] In one possible implementation, obtaining the arc mode based on the arc mode probability distribution through an integrated voting decision mechanism includes: The probability distribution of the electric arc mode is voted on using an integrated voting decision mechanism to obtain the voting result; the integrated voting decision mechanism is one or more of majority voting, confidence-weighted voting, and high-confidence voting; The arc mode is determined based on the voting results.

[0010] In one possible implementation, the arc mode is a diffused arc mode, a dotted spot mode, or an anode spot mode.

[0011] In one possible implementation, the signals used for arc pattern recognition are the arc voltage signal and the arc current signal.

[0012] Secondly, embodiments of this application provide a non-invasive arc risk assessment system for vacuum switches. The system is used to perform the method described in the first aspect above. The system includes: a laboratory environment unit, a migration unit, and a field application unit; the laboratory environment unit is connected to the field application unit through the migration unit. The laboratory environment unit is used to simultaneously acquire arc voltage and arc current signals during the interruption process in a vacuum interruption chamber experimental platform equipped with an observation window, and to classify and label arc patterns based on the arc morphology observed through the observation window, thereby establishing an arc voltage and current dataset with arc pattern labels; the arc voltage and current dataset is subjected to window segmentation and standardization preprocessing, and the window-level data is input into a preset classification model for training to obtain an arc recognition model; The migration unit is used to migrate the arc recognition model to the control device in the field application unit; The field application unit is used to non-invasively acquire target arc voltage and current signals using voltage and current sensors in engineering environments without visibility, and performs the same preprocessing procedure as before model training. The preprocessed target arc voltage and current signals are used as input to the arc identification model to obtain the arc pattern probability distribution output by the model. Based on the arc pattern probability distribution, an integrated voting decision mechanism is used to obtain the arc pattern. If the arc pattern is a dotted arc, a risk warning message is output to alert the user of a safety risk posed by the vacuum switch. If the arc pattern is an anode arc, an alarm message is output to trigger an alarm signal.

[0013] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: (1) It can realize non-invasive diagnosis of arc mode without modifying the vacuum interrupter. Diagnosis can be achieved through the ceramic shell. It can be directly applied to existing equipment. Furthermore, it can measure arc voltage and arc current signals through sensors. The equipment is easy to install, the signals are easy to obtain, and it is easy to conduct long-term stable detection. It has strong engineering feasibility. (2) Since the arc pattern may evolve during the opening and closing of the vacuum switch, the evolution of the arc pattern can be addressed by using window-level data segmentation and voting mechanism, so as to avoid the influence of the evolution process on the recognition results and improve the accuracy of the arc pattern recognition results. (3) As readily available electrical signals, arc voltage and arc current signals can be acquired online in real time by sensors without the need for destructive modifications to existing equipment. This allows for timely and accurate acquisition of arc-related signals, resulting in high signal acquisition timeliness and low cost. This avoids the problems of poor signal timeliness and low accuracy caused by offline signal acquisition in existing technologies. (4) By collecting readily available electrical signals through sensors for arc pattern recognition, the dependence on optical signals in the existing technology is eliminated. Moreover, the sensor structure is simple and stable, which is convenient for long-term maintenance and operation, thereby reducing the operation and maintenance cost of the evaluation system. (5) By identifying “anode spot arc” that has caused severe ablation of the contacts and “point spot arc” that is a typical precursor to it, a graded early warning can be achieved. For the former, an immediate alarm is issued and emergency maintenance is recommended. For the latter, the risk is indicated, and it is recommended to pay close attention and strengthen monitoring in subsequent operations, thereby achieving early intervention and predictive maintenance for the risk of the end of the life of the switchgear. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments of this application or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating a non-invasive method for assessing arc risk in a vacuum switch, as provided in this application embodiment; Figure 2 A block diagram of a non-invasive arc risk assessment system for vacuum switches provided in this application embodiment; Figure 3 A block diagram of a vacuum interrupter experimental platform provided in an embodiment of this application; Figure 4 A schematic diagram of the network structure of a hybrid deep learning recognition model provided in an embodiment of this application; Figure 5 This is an example diagram of a voting output provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] In the description of the embodiments of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the embodiments of this application and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. The terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this application according to the specific circumstances.

[0018] As discussed in the background section, there is an urgent need in practical products for a detection and assessment method that can determine the risk of arcing in an arc-extinguishing chamber without relying on photographing the arc. In stark contrast to the complex, expensive, and destructive technical approach based on optical observation, the arcing voltage and current signals generated during the opening and closing of a vacuum switch are physical quantities that can be effectively captured by conventional voltage / current sensors (such as voltage dividers and Rogowski coils). These signals can be fully acquired by adding acquisition modules to the existing measurement and control architecture at extremely low marginal cost, offering unparalleled convenience and availability in engineering practice, and providing an ideal signal foundation for achieving online, real-time early warning.

[0019] However, this significant engineering advantage has long failed to be effectively translated into diagnostic and early warning capabilities. Although the signals are readily available, the underlying deep characteristics and complex mapping relationships that precisely correspond to different arc modes (especially high-risk modes) have not been discovered and utilized by traditional analysis methods. This leads to a core contradiction: on the one hand, key electrical characteristics that indicate the end of equipment life (such as specific voltage and current fluctuations corresponding to dot-like patterns) already exist in easily measurable signals; on the other hand, existing technologies cannot directly and accurately identify high-risk modes and trigger early warnings from these simple electrical signals.

[0020] Therefore, there is an urgent need in this field for an innovative technological approach that can completely eliminate reliance on optical observation and instead fully utilize readily available arc voltage and current signals to construct diagnostic models ranging from simple electrical signals to complex arc modes. This model not only needs to achieve convenient, online, and accurate risk assessment without damaging the equipment structure, but its core value lies in its ability to promptly and accurately identify high-risk arc modes such as "dot-like spots," thereby providing crucial decision-making support for predicting the remaining electrical life of vacuum switches and for early warning maintenance. Ultimately, this will bridge the technological gap between "having signals" and "achieving early warning."

[0021] To address the aforementioned issues, this application provides a non-invasive method and system for assessing arc risk in vacuum switches. The method includes: simultaneously acquiring arc voltage and current signals during the switching process in a vacuum interrupter experimental platform with an observation window; classifying and labeling arc patterns based on the arc morphology observed through the observation window to establish an arc voltage and current dataset with arc pattern labels; performing window segmentation and standardization preprocessing on the arc voltage and current dataset; inputting the window-level data into a preset classification model for training to obtain an arc recognition model; and non-invasively assessing arc risk in engineering environments without visual access using voltage and current sensors. The target arc voltage and current signals are collected and processed using the same preprocessing procedure as before model training: the preprocessed target arc voltage and current signals are used as input to the arc identification model to obtain the arc pattern probability distribution output by the model; based on this probability distribution, the arc pattern is determined through an integrated voting decision mechanism; if the arc pattern is a point-like spot arc, a risk warning message is output to alert the user of a safety risk posed by the vacuum switch; if the arc pattern is an anode spot arc, an alarm message is output to trigger an alarm signal. According to the above technical solution, non-invasive diagnosis of arc patterns can be achieved without modifying the vacuum interrupter; diagnosis can be performed through the ceramic shell, allowing direct application to existing equipment. Furthermore, the arc voltage and current signals can be measured via sensors. The equipment is easy to install, the signals are readily available, facilitating long-term stable detection and strong engineering feasibility. Additionally, utilizing readily available electrical signals results in low signal acquisition costs, a simple and stable sensor structure, and ease of long-term maintenance, thereby reducing the overall operating and maintenance costs of the diagnostic system.

[0022] Figure 1 A flowchart illustrating a non-invasive arc risk assessment method for vacuum switches provided in this application embodiment. Figure 1 As shown, the method may include the following steps.

[0023] S101. In the vacuum interrupter experimental platform with an observation window, the arc voltage signal and arc current signal are simultaneously collected during the interruption process. Based on the arc morphology observed by the observation window, the arc mode is classified and labeled to establish an arc voltage and current dataset with arc mode labels.

[0024] S102. Perform window segmentation and standardization preprocessing on the arc voltage and current dataset, and input the window-level data into the preset classification model for training to obtain the arc recognition model.

[0025] For example, the arc recognition model can be a model that processes time-series data. For instance, the arc recognition model can be a hybrid neural network structure combining CNN, LSTM, and GRU, or a Transformer-based classification model, etc.

[0026] For example, the preprocessed arc voltage signal and arc current signal can be multiple window segments, which can be used as inputs to the arc identification model; and the arc mode probability distribution output by the arc identification model can be the arc mode probabilities corresponding to the multiple window segments, that is, each window segment corresponds to a mode probability.

[0027] S103. In engineering environments where visibility is limited, non-invasively acquire the target arc voltage signal and target arc current signal using voltage and current sensors, and perform the same preprocessing procedure as before model training.

[0028] For example, the lack of visibility could be an engineering site where signals cannot be directly observed or collected visually.

[0029] S104. The preprocessed target arc voltage signal and target arc current signal are used as inputs to the arc identification model to obtain the arc mode probability distribution output by the arc identification model.

[0030] S105. Based on the probability distribution of the electric arc mode, the electric arc mode is obtained through an integrated voting decision-making mechanism.

[0031] In some embodiments, S105 may include: voting on the probability distribution of the arc mode through an integrated voting decision mechanism to obtain a voting result; the integrated voting decision mechanism may be one or more of majority voting, confidence-weighted voting, and high-confidence voting; and determining the arc mode based on the voting result.

[0032] For example, this integrated voting decision-making mechanism can also be a method for integrated decision-making based on the probability of multiple window segments, etc., which is not limited here.

[0033] S106. When the arc mode is a dotted spot arc, output a risk warning message; the risk warning message is used to remind the user that the vacuum switch has a safety risk.

[0034] S107. When the arc mode is an anode spot arc, an alarm message is output; the alarm message is used to trigger an alarm signal.

[0035] According to the above technical solution, non-invasive diagnosis of arc mode can be achieved without modifying the vacuum interrupter. Diagnosis can be achieved through the ceramic shell, which can be directly applied to existing equipment. Furthermore, the arc voltage and arc current signals can be measured by sensors. The equipment is easy to install, the signals are easy to acquire, and it is convenient for long-term stable detection. It has strong engineering feasibility. In addition, the advantage of using readily available electrical signals is that the signal acquisition cost is low, the sensor structure is simple and stable, and it is convenient for long-term maintenance and operation, thereby reducing the overall operation and maintenance cost of the diagnostic system.

[0036] It should be noted that in actual production applications, vacuum interrupters are expensive and can cause structural damage to equipment. They generally lack observation windows, making it difficult to directly observe the arcing generated by the vacuum switch and determine the corresponding pattern. To address these issues, engineers devised a method to collect data and images in a laboratory using a vacuum interrupter with an observation window. Based on this data and images, an arc recognition model was trained. This model was then transferred to actual production equipment to perform pattern recognition of the arcing patterns generated by the equipment, improving the efficiency of arcing pattern recognition. This method solves the problems of reliance on visual observation and low timeliness in existing technologies, and requires no physical modification to the vacuum switch.

[0037] In one possible implementation, the arc voltage signal and arc current signal are preprocessed by windowing and standardization in the following manner: extracting the effective arc segment from the arc voltage signal and arc current signal; slicing the arc voltage signal and arc current signal using a sliding window; and normalizing the arc voltage signal and arc current signal using a standardization method.

[0038] Since the different modes of a vacuum arc are determined by their underlying physical mechanisms (such as cathode spot behavior and plasma characteristics), the differences in these physical processes will inevitably be reflected in the voltage and current waveforms of the external circuit (such as fluctuation frequency, amplitude statistics, and zero-crossing characteristics). However, these differences are often subtle and nonlinear, making them difficult to extract using traditional methods. Deep learning models possess powerful automatic feature learning and complex pattern recognition capabilities, enabling them to mine these abstract features deeply correlated with arc modes from high-dimensional time-series data. This allows for the establishment of an accurate mapping from electrical signals to arc modes. Therefore, arc recognition models based on deep learning models can accurately identify different arc modes.

[0039] In one possible implementation, 70% of the data in the arc voltage and current dataset is used as a training set, 10% as a test set, and 20% as a validation set; a preset classification model is trained using the training set, and the trained preset classification model is used as the arc recognition model.

[0040] In one possible implementation, the Adam optimizer can be used, combined with cross-entropy loss function, gradient clipping, and L2 regularization to train the preset classification model. During training, the cross-entropy loss function and accuracy curve are monitored, and training is terminated early when the performance on the validation set no longer improves, in order to obtain the optimal model parameters.

[0041] In one possible implementation, the arc mode is a diffused arc mode, a dotted spot mode, or an anodic spot mode. For example, the arc mode may also include a strong arc mode or an anodic plume mode, etc., which can be classified by the user according to the actual situation, and is not limited here.

[0042] In one possible implementation, the signals used for arc pattern recognition are the arc voltage signal and the arc current signal.

[0043] Figure 2 This is a block diagram of a non-invasive arc risk assessment system for vacuum switches provided in an embodiment of this application. The system 200 is used to perform the aforementioned non-invasive arc risk assessment method for vacuum switches. The system 200 may include: a laboratory environment unit 210, a migration unit 220, and a field application unit 230; the laboratory environment unit 210 is connected to the field application unit 230 through the migration unit 220.

[0044] The laboratory environment unit 210 is used to simultaneously acquire arc voltage and arc current signals during the interruption process in a vacuum interrupter experimental platform with an observation window. Based on the arc morphology observed through the observation window, arc mode classification and labeling are performed to establish an arc voltage and current dataset with arc mode labels. The arc voltage and current dataset is then subjected to window segmentation and standardization preprocessing, and the window-level data is input into a preset classification model for training to obtain an arc recognition model.

[0045] The migration unit 220 is used to migrate the arc identification model to the control equipment in the field application unit.

[0046] The field application unit 230 is used to non-invasively acquire target arc voltage and target arc current signals using voltage and current sensors in engineering environments without visibility, and performs the same preprocessing procedure as before model training. The preprocessed target arc voltage and current signals are used as input to the arc identification model to obtain the arc pattern probability distribution output by the model. Based on this arc pattern probability distribution, an integrated voting decision mechanism is used to obtain the arc pattern. If the arc pattern is a dotted arc, a risk warning message is output to alert the user that the vacuum switch poses a safety risk. If the arc pattern is an anode arc, an alarm message is output to trigger an alarm signal.

[0047] According to the above technical solution, non-invasive diagnosis of arc mode can be achieved without modifying the vacuum interrupter. Diagnosis can be achieved through the ceramic shell, which can be directly applied to existing equipment. Furthermore, the arc voltage and arc current signals can be measured by sensors. The equipment is easy to install, the signals are easy to acquire, and it is convenient for long-term stable detection. It has strong engineering feasibility. In addition, the advantage of using readily available electrical signals is that the signal acquisition cost is low, the sensor structure is simple and stable, and it is convenient for long-term maintenance and operation, thereby reducing the overall operation and maintenance cost of the diagnostic system.

[0048] The following example illustrates the workflow of a non-invasive arc risk assessment method for vacuum switches, which may include the following steps.

[0049] 1. Establish a vacuum arc experimental platform and construct a dataset.

[0050] In a laboratory environment, as reference Figure 3A vacuum interrupter with an observation window was used as the main experimental component. This chamber included a transparent glass window, a ceramic shell, and contacts, connected to a differential high-voltage probe (voltage sensor) and a Rogowski coil (i.e., a current sensor) to observe the vacuum arc generated during the opening of the vacuum switch. The differential high-voltage probe was used for voltage measurement, with a bandwidth of at least 50MHz and a common-mode rejection ratio of at least 80dB; the Rogowski coil was used for current measurement, with a bandwidth of at least 1MHz. Data acquisition employed a multi-channel synchronous acquisition card with a sampling rate set at at least 500 kHz. An oscilloscope was connected to both the differential high-voltage probe and the Rogowski coil to display voltage and current waveforms. A high-speed camera was used as a synchronous observation device to capture the dynamic physical characteristics of the arcing process in real time. A trigger control system ensured that all measuring devices were activated by a synchronous trigger. Based on this equipment, a complete non-invasive online diagnostic system for vacuum arc modes, from data preparation to online deployment, was constructed, achieving for the first time a "transparent" diagnostic of the arc state inside a sealed commercial vacuum interrupter.

[0051] Within a current range of 1kA to 30kA, dozens to hundreds of breaking experiments are conducted. By combining real-time observation of the stability of arc morphology and brightness with a synchronous observation device, synchronous analysis of electrical waveform characteristics, and supplementing with contact surface condition inspection after the breaking experiments, the arc mode of each breaking experiment is comprehensively judged and labeled, thereby constructing a labeled dataset for model training.

[0052] 2. Establish a hybrid deep learning recognition model.

[0053] Reference Figure 4 A hybrid deep learning recognition model is constructed using a deep learning framework. This model consists of the following sequentially connected modules: a local temporal feature extraction module, a temporal dependency learning module, and a pattern discrimination module.

[0054] The model comprises several modules: a local temporal feature extraction module (composed of a one-dimensional convolutional neural network) for automatically extracting local features from the input signal; a temporal-dependent learning module (composed of a long short-term memory network) for learning the dynamic evolution of these local features; and a pattern discrimination module (using fully connected layers and a softmax function to map the learned features to probability distributions of various arc patterns). Thus, this model synergistically utilizes the spatial feature extraction capabilities of CNNs and the temporal modeling capabilities of LSTMs to achieve accurate end-to-end classification.

[0055] It should be noted that multi-channel time-series data, namely arc voltage signal and arc current signal, can be used as model input, and the input data can be preprocessed. The data preprocessing process can be as follows: preprocessing the original arc voltage signal and arc current signal, including: extracting the effective arc segment; slicing using a sliding window; and normalizing the data using a standardization method.

[0056] 3. Model training and optimization.

[0057] The labeled dataset was divided into training, validation, and test sets. The Adam optimizer, combined with cross-entropy loss, gradient clipping, and L2 regularization, was used to train the model. The training set was used to learn information, the validation set to provide feedback and adjust the learning effect, and the test set to validate model performance. During training, the loss function and accuracy curves were monitored. Training was terminated early when the validation set performance no longer improved to obtain the optimal model parameters.

[0058] For example, to fully utilize the data and obtain robust performance evaluation, k-fold cross-validation can be used instead of a simple single split. The dataset is randomly divided into 5 parts (i.e., k=5), with 4 parts used as the training set and the remaining part as the validation set. Training and testing are repeated cyclically, and the average of the 5 test results is taken as the final performance indicator of the model.

[0059] Reference Figure 5 This is an example of voting output. (a) is the voltage signal, (b) is the current signal, and (c) is the prediction result of the sliding window voting. The window size is 80, the step size is 8, and there are a total of 10 window segments. It can be seen that all 10 windows correctly predict the results (i.e., the point pattern), and the voting consistency is high, which intuitively reflects the effectiveness of the integrated voting decision-making mechanism.

[0060] The loss function and accuracy curves during training clearly demonstrate the model's convergence behavior. Specifically, the training loss and validation loss decrease synchronously and steadily with increasing training epochs, eventually leveling off without significant divergence, indicating that the model has not overfitted and possesses good generalization ability. The training accuracy and validation accuracy increase rapidly with each training epoch, with the training set accuracy quickly reaching over 90% and the test set over 70% (this accuracy only represents the model's classification performance on segments of the complete data after the sliding window; the complete data, after ensemble voting, can achieve over 97% accuracy). This demonstrates that the model can efficiently learn features from the data and quickly converge to a stable state with excellent performance.

[0061] Thus, through the above technical solution, the core idea of ​​"generating supervision signals in the laboratory using visual conditions and training models to solve the problem of no visibility on site" is proposed, namely, multimodal supervision signal migration, which cleverly transforms the identification of vacuum arc mode from relying on direct optical observation to intelligent analysis based on easily measurable electrical quantities (arc current, voltage).

[0062] 4. Online deployment and real-time identification.

[0063] The trained model is deployed to an edge computing device. In practical applications, only the arc voltage and arc current signals of the vacuum switch need to be collected. After preprocessing consistent with the training phase, these signals are input into the model to obtain preliminary probability analysis results for each signal segment. Subsequently, an integrated voting decision mechanism is employed, combining multiple strategies from majority voting, confidence-weighted voting, and high-confidence voting, to statistically integrate the preliminary results of all window segments to determine the final arc mode.

[0064] 5. Diagnosis, early warning, and output.

[0065] The final arc pattern recognition result is output, a process completed within milliseconds. When a "point-like arc" is identified, a risk warning is issued and close attention is recommended; when a preset high-risk pattern such as "anode-like arc" is identified, an early warning signal is immediately triggered and reported to the maintenance system.

[0066] Based on the above technical solution, a mapping model can be established between the externally measurable electrical quantities (arc voltage signal, arc current signal) of the vacuum arc and its internal invisible arc mode. By utilizing the model's powerful feature extraction and pattern recognition capabilities, subtle pattern differences hidden in electrical signals that are difficult for the human eye to perceive can be captured, thereby improving the accuracy of discrimination.

[0067] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. Each embodiment focuses on describing the differences from other embodiments.

[0068] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of this application.

Claims

1. A non-invasive method for assessing arc risk in vacuum switches, characterized in that, The method includes: In a vacuum interrupter experimental platform equipped with an observation window, arc voltage and arc current signals are simultaneously acquired during the interruption process. Based on the arc morphology observed through the observation window, arc mode classification and labeling are performed to establish an arc voltage and current dataset with arc mode labels. The arc voltage and current dataset is preprocessed by window segmentation and standardization, and the window-level data is input into a preset classification model for training to obtain an arc recognition model. In engineering environments without visibility, the target arc voltage and current signals are non-invasively acquired using voltage and current sensors, and the same preprocessing procedure is performed before model training. The preprocessed target arc voltage signal and target arc current signal are used as inputs to the arc identification model to obtain the arc mode probability distribution output by the arc identification model. Based on the probability distribution of the electric arc mode, the electric arc mode is obtained through an integrated voting decision-making mechanism; When the arc mode is a dotted spot arc, a risk warning message is output; the risk warning message is used to alert the user that the vacuum switch poses a safety risk. When the arc mode is an anode spot arc, an alarm message is output; the alarm message is used to trigger an alarm signal.

2. The method according to claim 1, characterized in that, The arc voltage signal and arc current signal are preprocessed by window segmentation and standardization in the following ways: extracting the effective arc segment from the arc voltage signal and arc current signal; slicing the arc voltage signal and arc current signal using a sliding window; and normalizing the arc voltage signal and arc current signal using a standardization method.

3. The method according to claim 1, characterized in that, The arc identification model is a model for processing time-series data.

4. The method according to claim 1, characterized in that, The process of obtaining the arc mode based on the arc mode probability distribution through an integrated voting decision mechanism includes: The probability distribution of the electric arc mode is voted on using an integrated voting decision mechanism to obtain the voting result; the integrated voting decision mechanism is one or more of majority voting, confidence-weighted voting, and high-confidence voting; The arc mode is determined based on the voting results.

5. The method according to claim 1, characterized in that, The arc mode is a diffused arc mode, a dotted spot mode, or an anode spot mode.

6. The method according to claim 1, characterized in that, The signals used for arc pattern recognition are the arc voltage signal and the arc current signal.

7. A non-invasive arc risk assessment system for vacuum switches, characterized in that, The system is used to perform the method according to any one of claims 1-6, and the system includes: a laboratory environment unit, a migration unit, and a field application unit; the laboratory environment unit is connected to the field application unit through the migration unit. The laboratory environment unit is used to simultaneously acquire arc voltage and arc current signals during the interruption process in a vacuum interruption chamber experimental platform equipped with an observation window, and to classify and label arc patterns based on the arc morphology observed through the observation window, thereby establishing an arc voltage and current dataset with arc pattern labels; the arc voltage and current dataset is subjected to window segmentation and standardization preprocessing, and the window-level data is input into a preset classification model for training to obtain an arc recognition model; The migration unit is used to migrate the arc recognition model to the control device in the field application unit; The field application unit is used to non-invasively acquire target arc voltage and current signals using voltage and current sensors in engineering environments without visibility, and performs the same preprocessing procedure as before model training. The preprocessed target arc voltage and current signals are used as input to the arc identification model to obtain the arc pattern probability distribution output by the model. Based on the arc pattern probability distribution, an integrated voting decision mechanism is used to obtain the arc pattern. If the arc pattern is a dotted arc, a risk warning message is output to alert the user of a safety risk posed by the vacuum switch. If the arc pattern is an anode arc, an alarm message is output to trigger an alarm signal.

Images