Adaptive fault arc detection method and system based on cloud edge-end cooperation

Abstract

The application is based on a cloud edge end cooperative adaptive fault arc detection method and system; through real-time high-frequency sampling of bus current by edge side equipment, analysis is carried out by using the locally preset reference method, if it is determined as a known fault, local protection is executed, or the edge side equipment independently completes the fault diagnosis closed loop; if it is determined as an unknown fault or cannot be identified, the extracted fault features are transmitted to the cloud diagnosis model; the cloud diagnosis model realizes the diagnosis of unknown faults through adaptive feature extraction, realizes early fault diagnosis and isolation through zero sample generalization reasoning and fault situation prediction of unknown working conditions, and outputs the diagnosis result to the mobile terminal; the mobile terminal displays the diagnosis result, and also supports feedback of unknown faults to realize cloud diagnosis model updating, cloud diagnosis model parameter synchronous updating of edge side equipment, and adaptive upgrading. The application can flexibly switch multiple cooperative modes according to the scene, and effectively solves the poor adaptability problem of traditional methods.

Description

Technical Field

[0001] This invention relates to the field of electrical fault detection, specifically to an adaptive fault arc detection method and system based on cloud-edge-device collaboration. Background Technology

[0002] With the widespread deployment of technologies such as distributed energy, energy storage systems, and 5G communication in open scenarios such as smart grids and low-voltage DC distribution networks, the complexity of electrical systems and the dynamic nature of their operating environments have increased unprecedentedly. Electric arc faults, due to their high degree of concealment and destructive power, pose a direct threat to life and property safety, highlighting the increasing importance of timely and effective detection and precise location. Traditional electrical fault protection schemes often rely on fixed physical thresholds or single-level discrimination logic, making it difficult to balance response time and diagnostic depth. They also exhibit significant problems of false tripping and failure to trip under dynamic operating conditions.

[0003] Current technological research is evolving from local algorithms to systematic architectures. At the basic identification level, existing solutions utilize random forests and DBSCAN clustering to improve anomaly classification accuracy at the edge. To enhance system support, existing technologies have also proposed cloud-edge-device collaborative architectures for digital twins, strengthening the monitoring framework through virtual-real interaction. In addition, multi-agent collaborative mechanisms have been introduced to enhance the system's elastic scheduling in dynamic environments. There are also explorations of cloud-edge large-scale model collaborative inference, attempting to introduce deep semantic understanding capabilities to improve detection performance. Meanwhile, power system fault detection is gradually evolving from "single-point intelligence" to "collaborative intelligence," aiming to cope with the processing pressure of massive electricity consumption data through architectural upgrades.

[0004] However, the above solutions still have unresolved shortcomings. First, the edge-side model is limited by the nonlinear processing capabilities of the static model, and its analytical power for instantaneous complex signals remains weak. The cloud-edge-device collaborative architecture faces the inherent contradiction between computational offloading and bandwidth limitations, making it difficult to deeply couple the generalization ability of large models with the millisecond-level real-time response requirements of power circuit breakers. Second, existing systems lack the ability to identify features under dynamic operating conditions such as weak fault arcs, load mutations, and frequent start-stops. This can easily lead to misjudging inrush currents or normal arc initiation as fault arcs, resulting in malfunctions, or missing hidden faults such as series arcs, leading to failure to operate and causing frequent power outages or even fires. Furthermore, traditional methods lack generalization and robustness, and cannot adapt to new load access, new operating conditions, and complex scenarios with multiple branches. The accuracy of fault location for newly introduced branches drops sharply, making it difficult to simultaneously meet multiple requirements such as data confidentiality, communication conditions, and differences in computing resources. Finally, the existing system suffers from delayed information feedback and lacks real-time dynamic monitoring capabilities, making it difficult for maintenance personnel to obtain fault information in a timely manner for intervention. This fails to meet the urgent needs of complex power systems for adaptive, real-time detection, and precise positioning. Summary of the Invention

[0005] To address the problems mentioned in the prior art, this invention proposes an adaptive fault arc detection method and system based on cloud-edge-device collaboration. By introducing a meta-learning module and a deep belief network, the essential features of fault arcs in a limited number of samples are extracted and transferred to learning. This enables the system to effectively identify fault arcs under unseen load types and new operating conditions without the need for retraining for each new scenario, thus overcoming the problems of traditional methods relying on fixed models and insufficient generalization ability.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention relates to an adaptive fault arc detection method based on cloud-edge-device collaboration, comprising the following steps: S1. The bus current is sampled in real time at high frequency by the edge-side equipment to obtain the current signal; the sampled signal is analyzed in real time using the local preset reference method. If it is determined to be a known fault type, trip protection is executed and the user is notified through the mobile terminal; if it is determined to be an unknown fault type or an unidentifiable type, S2 is triggered. S2. The edge device performs time-frequency transformation processing on the current signal, extracts fault features, and transmits the extracted fault features to the cloud diagnostic model. At the same time, it issues an unknown fault warning through the mobile terminal. S3. The cloud-based diagnostic model receives fault characteristics, performs cascade diagnosis through the meta-learning module and deep belief network within the cloud-based diagnostic model, and performs adaptive learning by combining the on-site label information fed back through the mobile terminal. It outputs diagnostic results containing the arc-generating load type and arc-generating branch information of the fault arc, and pushes the diagnostic results to the mobile terminal. S4, the cloud-based diagnostic model synchronously pushes the updated model parameters to the edge devices, enabling adaptive upgrades to the locally preset benchmark methods.

[0007] As a further improvement of the present invention, it also includes switching the working mode based on the application scenario requirements. The working modes include edge-device independent processing mode, cloud-device independent processing mode, and cloud-edge-device collaborative processing mode.

[0008] As a further improvement of the present invention, the edge-side independent processing mode includes the edge-side device independently completing the detection, diagnosis and trip protection of fault arc, and the diagnosis results are not transmitted to the cloud; The cloud-based independent processing mode includes the edge device uploading the collected current signal to the cloud diagnostic model, which then performs fault diagnosis and sends the diagnosis results down. The cloud-edge-device collaborative processing mode includes real-time high-frequency sampling and preliminary judgment by edge devices, in-depth diagnosis and adaptive learning of unknown faults by cloud diagnostic models, and synchronization of updated model parameters to edge circuit breakers.

[0009] As a further improvement of the present invention, the real-time high-frequency sampling of the bus current in S1 is a point-by-point continuous sampling, with a sampling frequency of 200kHz to 1MHz and 8000 to 12000 sampling points in a single analysis period.

[0010] As a further improvement of the present invention, the local preset benchmark method in S1 is based on common fault arc modes and threshold settings. The edge-side device has a built-in local database that stores known load characteristics and corresponding arc modes. After trip protection, the local database is used to search and match the arc-generating load type and arc-generating branch.

[0011] As a further improvement of the present invention, the edge-side device in S2 performs time-frequency transformation processing on the current signal to extract fault features, including: The current signal was decomposed into a time-frequency matrix using Rbio3.1 mother wavelet. Wavelet packet coefficients in frequency bands with obvious fault characteristics are selected for reconstruction, resulting in a reconstructed wavelet packet coefficient sequence with the same sampling length as the original current signal.

[0012] As a further improvement of the present invention, the process of cascading diagnosis through the meta-learning module inside the cloud-based diagnostic model and the deep belief network in S3 includes: The fault features and the original current signal data are normalized and merged into a high-dimensional hybrid feature vector. The embedding layer of the input meta-learning module is encoded as a dense representation. After passing through a multi-head attention layer to capture the global dependencies between features, the meta-features are output. The meta-features are combined with the original current signal data and input into a deep belief network for processing, and the output is a diagnostic result containing the arc-generating load type and arc-generating branch information of the fault arc.

[0013] As a further improvement of the present invention, the adaptive learning in step S3, which uses the on-site tag information fed back by the mobile terminal, includes: The on-site tag information is the tagged sample information uploaded by maintenance personnel through mobile terminals, including fault confirmation information of non-operation samples, on-site troubleshooting results of erroneous operation samples, and corresponding waveform data. The adaptive learning is a cloud-based diagnostic model that adjusts its internal parameters and optimizes the fault discrimination boundary based on the labeled sample information.

[0014] As a further improvement of the present invention, the adaptive upgrade of the locally preset benchmark method in S4 specifically involves the edge device updating the local fault detection, line selection and positioning algorithm logic according to the updated model parameters, and supplementing and updating the locally stored load characteristics and arc mode database.

[0015] This invention proposes an adaptive fault arc detection system based on cloud-edge-device collaboration, including edge-side devices, cloud-based diagnostic models, and mobile terminals; The edge-side device is used to obtain a current signal by performing real-time high-frequency sampling of the bus current; it analyzes the sampled signal in real time using a locally preset benchmark method; if the fault type is determined to be known, it performs trip protection and notifies the user through a mobile terminal; if the fault type is determined to be unknown or unidentifiable, it transmits the fault characteristics to the cloud diagnostic model and issues an unknown fault warning through the mobile terminal. The cloud-based diagnostic model is used to receive fault characteristics, perform cascaded diagnosis through the meta-learning module and deep belief network within the cloud-based diagnostic model, and perform adaptive learning by combining the field label information fed back through the mobile terminal. It outputs diagnostic results including the arc-generating load type and arc-generating branch information of the fault arc, and pushes the diagnostic results to the mobile terminal. It is also used to synchronously push the updated model parameters to the edge device to achieve adaptive upgrade of the locally preset benchmark method.

[0016] Compared with the prior art, the present invention achieves the following technical effects: This invention first uses a locally preset benchmark method to perform millisecond-level tripping protection for known fault types, ensuring the power system's real-time response capability to common faults. Simultaneously, when encountering unknown or unidentifiable fault types, subsequent steps trigger a cloud-based collaborative process, enabling in-depth diagnosis and model evolution for complex and unknown faults. This overcomes the contradiction in existing technologies where a single architecture struggles to balance response time and diagnostic depth through a division of labor between fast side-circuit breaker response and cloud-based diagnostic model. Furthermore, this invention employs a cloud-based diagnostic model composed of a meta-learning module (Transformer) and a cascaded deep belief network. The meta-learning module extracts patterns of fault features from limited samples and achieves zero-sample generalization inference through cross-load feature alignment. This allows for effective identification of fault arcs without retraining when facing new load types or operating conditions, solving the problem of traditional detection methods relying on fixed rules or limited datasets for training and exhibiting poor adaptability to new scenarios.

[0017] This invention enables maintenance personnel to use on-site tag information fed back by mobile terminals to drive cloud-based diagnostic models to perform adaptive learning. By continuously accumulating false operation samples (arc-like interference) and failure to operate samples (weak and hidden arcs), the cloud-based diagnostic model can dynamically adjust its decision boundaries, accurately distinguish between real fault arcs and interference signals such as load start-up / shutdown and current fluctuations, while improving the detection sensitivity to weak and hidden arcs, thereby significantly reducing false operation and failure to operate rates. In addition, this invention synchronously pushes the updated model parameters of the cloud-based diagnostic model to edge devices, realizing adaptive upgrades to local preset benchmark methods. After the upgrade, new loads and new operating conditions that originally required the cloud-based diagnostic model to identify can be independently identified and tripped by edge devices. It can continuously self-optimize as the operating time increases, fundamentally solving the drawback of traditional models that cannot be upgraded. Finally, through the flexible design of the cloud-edge-device collaborative architecture, it can adapt to the differentiated needs of different application scenarios. For privacy-sensitive scenarios, independent local processing can be achieved through configuration to ensure that data is not leaked. For scenarios with weak computing power but high accuracy requirements, cloud computing power can be fully utilized to achieve high-precision diagnosis. For complex and open scenarios, cloud-edge-device collaborative processing can be achieved, taking into account both real-time response and adaptability to complex environments, and achieving a high degree of unity between system reliability and response flexibility. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a confusion matrix diagram of the deep belief network before the meta-learning module was used in this invention to perform fault detection under eight typical load conditions. Figure 3 The final confusion matrix diagram for fault detection of eight typical loads after introducing a meta-learning module to improve the deep belief network in this invention; Figure 4 A bar chart showing the detection accuracy after introducing the meta-learning module in this invention; Figure 5 This is a schematic diagram of the normal state of the present invention; Figure 6 This is a diagram illustrating the normal functional implementation of the present invention; Figure 7 This is a schematic diagram illustrating the independent edge processing of the present invention; Figure 8 This is a diagram illustrating the implementation of the edge-specific processing function of the present invention; Figure 9 This is a schematic diagram illustrating the cloud-based independent processing of the present invention; Figure 10 This diagram illustrates the implementation of the cloud-based independent processing function of the present invention. Figure 11 This is a schematic diagram of the cloud-edge-device collaborative processing of the present invention; Figure 12 This diagram illustrates the implementation of the cloud-edge-device collaborative processing function of the present invention. Detailed Implementation

[0019] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0020] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" 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 this invention and 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 invention.

[0021] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0022] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a communication connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0023] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0024] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0025] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0026] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0028] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0029] Example 1 like Figure 1 As shown, this embodiment proposes an adaptive fault arc detection method based on cloud-edge-device collaboration, including the following steps: S1. The bus current is sampled in real time at high frequency by the edge-side equipment to obtain the current signal; the sampled signal is analyzed in real time using the local preset reference method. If it is determined to be a known fault type, trip protection is executed and the user is notified through the mobile terminal; if it is determined to be an unknown fault type or an unidentifiable type, S2 is triggered. S2. The edge device performs time-frequency transformation processing on the current signal, extracts fault features, and transmits the extracted fault features to the cloud diagnostic model. At the same time, it issues an unknown fault warning through the mobile terminal. S3. The cloud-based diagnostic model receives fault characteristics, performs cascade diagnosis through the meta-learning module and deep belief network within the cloud-based diagnostic model, and performs adaptive learning by combining the on-site label information fed back through the mobile terminal. It outputs diagnostic results containing the arc-generating load type and arc-generating branch information of the fault arc, and pushes the diagnostic results to the mobile terminal. S4, the cloud-based diagnostic model synchronously pushes the updated model parameters to the edge devices, enabling adaptive upgrades to the locally preset benchmark methods.

[0030] Step 1: In this embodiment, the chip built into the edge device performs real-time, continuous point-by-point sampling of the bus current of the low-voltage DC distribution network system. The sampling frequency fs is preferably set to 1MHz. When the number of sampling points q of the acquired current signal reaches the preset analysis period requirement (e.g., q=8000), the chip built into the edge device will immediately perform preliminary analysis of the current signal in that period to detect whether there is a fault arc.

[0031] This preliminary analysis relies on a locally preset benchmark method stored on the edge device, based on common fault modes and threshold settings. If the built-in chip detects a fault arc using this method, the edge device will immediately trip to protect the circuit and prevent further escalation of the fault. After tripping, the edge device chip will immediately search its stored database to attempt to determine the type of load and faulty branch where the fault occurred. This database contains known, common load characteristics and corresponding arc patterns.

[0032] If the local database cannot identify the load type (i.e., it is determined to be a new load type, or its characteristics do not match known patterns, suggesting a possible fault under a new operating condition, new scenario, or new branch), the process proceeds to step two, initiating a deeper analysis and cloud-based collaborative processing flow. If the local database can identify the load type, the circuit breaker chip directly notifies the user of the load type and fault branch information via a mobile terminal, and waits for maintenance personnel to handle it. In this embodiment, the mobile terminal is preferably a mobile app. In this case, the cloud-based collaborative learning process is not initiated to ensure rapid response to common faults and efficient system operation.

[0033] Step 2: When Step 1 determines that the fault is a new load type or an unknown fault mode, the built-in chip of the edge device will perform time-frequency analysis on the acquired current signal to extract deeper fault characteristics. The current signal is decomposed into multiple wavelet packets based on the Rbio3.1 mother wavelet, generating wavelet packet coefficients at each node, thus obtaining the transformed time-frequency matrix. This decomposition process can effectively separate the energy distribution of the signal in different frequency ranges, and is particularly helpful in capturing the high-frequency transient components generated by the fault arc.

[0034] To focus on the significant information of fault arcs, instead of specific frequency bands, wavelet packet coefficients in frequency bands with obvious fault characteristics are specifically selected for reconstruction. This results in a reconstructed wavelet packet coefficient sequence with the same length as the original data (the number of sampling points is q). This reconstructed sequence serves as the basis for subsequent feature extraction, aiming to highlight information within this frequency band that might be overlooked by traditional methods but is crucial for arc identification. Simultaneously, during time-frequency analysis performed by the chip built into the edge device, a fault alert is sent to the user via a mobile app, informing them of the potential presence of an unknown type of fault arc, allowing the user to pay attention promptly and providing early warning for subsequent maintenance intervention.

[0035] This embodiment extracts features from the reconstructed wavelet packet coefficients, specifically by constructing a one-dimensional feature vector using a many-to-one mapping method, and then using this one-dimensional feature vector as the fault feature:

[0036] In the formula: d j,k For the wavelet packet coefficients at the corresponding wavelet packet decomposition tree nodes; Φ j,k These are wavelet packet basis functions; j As a scale indicator; k t represents the location indicator; t represents the sampling time.

[0037] Step 3: The fault features extracted in Step 2 are synchronously transmitted to the cloud-based diagnostic model. In this embodiment, the cloud-based diagnostic model is based on a meta-learning module constructed from a Transformer model and a Deep Belief Network (DBN) to obtain and implement deep learning and diagnosis. After receiving the fault features, the cloud-based diagnostic model enters a crucial feature combination input stage, where the time-frequency domain features extracted by wavelet packet decomposition are normalized and then merged with the original current sequence data to form a high-dimensional hybrid feature vector. Subsequently, the high-dimensional hybrid feature vector is input into the embedding layer of the Transformer, encoding it into a dense representation. Next, the dense representation passes through a multi-head attention layer group of the Transformer. In the multi-head attention layer group, the cloud-based diagnostic model can capture the global dependencies and long-distance correlations within the feature data in parallel from different "attention heads," thereby better understanding the complex patterns and contextual information in the fault arc signal. This is crucial for identifying weak or novel fault features that are difficult to capture using traditional methods.

[0038] After processing by the meta-learning module, the data enters the Deep Belief Network (DBN) for dynamic feature injection. The meta-features output by the meta-learning module are combined with the original current sequence data (referring to wavelet packet decomposition features or part of the original normalized sequence) and input together into the DBN as enhanced input to the Deep Belief Network. This is used to learn deeper feature representations and nonlinear mapping relationships, further improving the discriminative power of the features and the robustness of the model.

[0039] In the example implementation, for non-operational conditions, when edge-side devices fail to identify fault arc characteristics due to weak or concealed current signals and transmit them to the cloud, the cloud-based diagnostic model combines this with on-site confirmation information uploaded by maintenance personnel via a mobile app, treating it as negative samples or difficult-to-identify samples with true labels for learning. The meta-learning module's learning capability enables it to quickly identify and generalize the unique patterns of weak arcs from a small number of such samples, optimizing its ability to distinguish low signal-to-noise ratio signals. For maloperational conditions, when edge-side devices are misjudged as faults and tripped due to arc-like interference such as new load startup or fluctuations in normal operating conditions, maintenance personnel upload on-site troubleshooting results and corresponding waveform data via a mobile app. The cloud-based diagnostic model incorporates these maloperational samples and their "non-fault" labels into its training. The meta-learning module adjusts its decision boundaries through learning, enabling it to more accurately distinguish between real arcs and various arc-like interference signals, enhancing the model's anti-interference capability. DBN assists the cloud-based diagnostic model in better understanding these normal but complex signal characteristics, avoiding misjudging them as faults. The cloud-based diagnostic model adaptively learns and iteratively optimizes based on these new fault characteristics, no-operation / false-operation samples, and its massive existing historical data, continuously adjusting its internal parameters and pattern recognition logic, ultimately providing diagnostic results that include the arc-generating load type and arc-generating branch information of the fault arc.

[0040] Step 4: After the cloud-based diagnostic model completes the algorithm update and identifies new fault types, the updated algorithm or model parameters will be synchronously pushed to the edge devices, enabling adaptive upgrades for the edge devices. This update not only improves the accuracy of known fault mode recognition but also optimizes rejection actions (by improving the sensitivity of weak arc detection) and false actions (by enhancing the ability to distinguish arc-like interference).

[0041] Specifically, for the update of fault arc detection and line selection functions under new operating conditions, new scenarios (e.g., new energy storage systems, high-frequency switching power supply access), and newly introduced branches (i.e., new branch conditions): After the cloud diagnostic model successfully learns and identifies the fault arc characteristics and their corresponding branch information under these new environments through step four, its internal detection, line selection, and location algorithm logic will be updated accordingly. The cloud diagnostic model will generate new algorithm model parameters containing this updated logic and push these updates synchronously to all relevant edge devices through the algorithm update module.

[0042] When edge devices receive an update, their local detection, line selection, and location programs are updated in real time. This allows them to directly identify and accurately select the corresponding faulty branch locally when facing the same new operating conditions, new scenarios, or faults in newly introduced branches, without relying on deep learning in the cloud again. Simultaneously, the cloud-based diagnostic model synchronizes the identified arc-generating load branches and load type information to the mobile app, allowing maintenance personnel to view, handle, and clear the fault promptly. This cloud-edge collaborative adaptive upgrade mechanism ensures that the system can continuously learn and adapt to constantly changing load environments and new fault modes, effectively solving key problems faced by traditional detection methods in practical applications, such as failure to operate, false operation, and insufficient adaptability, thereby ensuring the safe and stable operation of the electrical system.

[0043] To verify the technical advantages of this invention, a comparative experiment was conducted in this embodiment, and the results are presented using a confusion matrix and a bar chart. For example... Figure 2 As shown, the detection confusion matrix of traditional DBN under eight typical load operating conditions is presented without using the meta-learning module. It can be seen that the model has a high false alarm rate among some loads with similar current characteristics, and the values ​​outside the diagonal of the matrix are large, reflecting its insufficient generalization ability to unknown loads.

[0044] See Figure 3 and Figure 4 When a meta-learning strategy is introduced: confusion matrix optimization, such as... Figure 3 As shown, the detection results for eight typical loads are highly concentrated on the diagonal, with a significant reduction in false positives and false negatives, demonstrating a substantial improvement in the recognition accuracy of the cloud-based diagnostic model under complex load interference; accuracy improvement comparison; as shown. Figure 4 As shown in the bar chart, compared with the previous method, the method of the present invention has significantly improved the average detection accuracy under all test conditions, especially its excellent robustness in dynamic environments.

[0045] Example 2 This embodiment is basically the same as embodiment 1, except that this embodiment provides an adaptive fault arc detection system based on cloud-edge-device collaboration, including edge-side devices, cloud-based diagnostic models, and mobile terminals; The edge-side device is used to obtain a current signal by performing real-time high-frequency sampling of the bus current; it analyzes the sampled signal in real time using a locally preset benchmark method; if the fault type is determined to be known, it performs trip protection and notifies the user through a mobile terminal; if the fault type is determined to be unknown or unidentifiable, it transmits the fault characteristics to the cloud diagnostic model and issues an unknown fault warning through the mobile terminal. The cloud-based diagnostic model is used to receive fault characteristics, perform cascaded diagnosis through the meta-learning module and deep belief network within the cloud-based diagnostic model, and perform adaptive learning by combining the field label information fed back through the mobile terminal. It outputs diagnostic results including the arc-generating load type and arc-generating branch information of the fault arc, and pushes the diagnostic results to the mobile terminal. It is also used to synchronously push the updated model parameters to the edge device to achieve adaptive upgrade of the locally preset benchmark method.

[0046] Example 3 This embodiment is basically the same as Embodiments 1 and 2, except that it proposes several different deployment modes based on the method of the present invention. It should be noted that this embodiment... Figures 5-12 The legend is explained below: Node illustration: The edge side in the diagram is not limited to circuit breakers; the physical carrier can change according to the scene. Solid arrow: Deterministic command issuance with real-time response; Dashed arrows: Non-real-time, triggered feedback streams (such as maintenance confirmations, incremental model updates); Bidirectional arrows: represent the collaborative learning evolution and prediction between the cloud and the edge (bidirectional flow).

[0047] See Figure 5 and Figure 6 This embodiment provides a normal operating state, where the system remains under normal monitoring in open scenarios with uncertain environments and dynamically changing loads. The key here lies in the cloud's continuous incremental learning of unknown loads; such as... Figure 6 As shown in the flowchart, the sensor collects current signals in real time and uploads them, and the system maintains real-time updates to the local database during stable operation. (See also...) Figure 7 and Figure 8For specialized scenarios with extremely high privacy protection requirements, such as government classified data centers and industrial parks, this invention provides an edge-only processing solution. This solution relies on embedded hardware or edge gateways deployed at the end of the power distribution system to independently complete fault detection; for example... Figure 8 As shown, all data processing, feature extraction, and fault determination are completed locally to ensure that sensitive data is not leaked.

[0048] See Figure 9 and Figure 10 For scenarios with limited on-site computing power but extremely high accuracy requirements, such as unattended substations or large photovoltaic power plants, this invention employs a cloud-based independent processing mode, such as... Figure 10 As shown, the field terminal is only responsible for data acquisition and transmission, making full use of the powerful computing power of the cloud to perform high-dimensional Transformer model calculations, thereby achieving high-precision remote detection and centralized monitoring.

[0049] See Figure 11 and Figure 12 For open scenarios with many branch lines and complex operating conditions, such as urban commercial power distribution and smart homes, this invention adopts a cloud-edge-device collaborative solution. Figure 12 As shown, this mode takes into account both the response speed of the edge (millisecond-level initial judgment) and the environmental adaptability of the cloud (deep learning model updates). By continuously sending updated model parameters through the cloud diagnostic model, the generalization ability of edge devices to unknown operating conditions is improved.

[0050] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0051] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. An adaptive fault arc detection method based on cloud-edge-device collaboration, characterized in that, Includes the following steps: S1. The bus current is sampled in real time at high frequency by the edge-side equipment to obtain the current signal; the sampled signal is analyzed in real time using the local preset reference method. If it is determined to be a known fault type, trip protection is executed and the user is notified through the mobile terminal. If the fault type is determined to be unknown or unidentifiable, then S2 is triggered; S2. The edge device performs time-frequency transformation processing on the current signal, extracts fault features, and transmits the extracted fault features to the cloud diagnostic model. At the same time, it issues an unknown fault warning through the mobile terminal. S3. The cloud-based diagnostic model receives fault characteristics, performs cascade diagnosis through the meta-learning module and deep belief network within the cloud-based diagnostic model, and performs adaptive learning by combining the on-site label information fed back through the mobile terminal. It outputs diagnostic results containing the arc-generating load type and arc-generating branch information of the fault arc, and pushes the diagnostic results to the mobile terminal. S4, the cloud-based diagnostic model synchronously pushes the updated model parameters to the edge devices, enabling adaptive upgrades to the locally preset benchmark methods.

2. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 1, characterized in that, It also includes switching working modes based on application scenario requirements, including edge-device independent processing mode, cloud-device independent processing mode, and cloud-edge-device collaborative processing mode.

3. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 2, characterized in that, The edge-side independent processing mode includes the edge-side equipment independently completing the detection, diagnosis, and tripping protection of fault arcs, and the diagnosis results are not transmitted to the cloud; The cloud-based independent processing mode includes the edge device uploading the collected current signal to the cloud diagnostic model, which then performs fault diagnosis and sends the diagnosis results down. The cloud-edge-device collaborative processing mode includes real-time high-frequency sampling and preliminary judgment by edge devices, in-depth diagnosis and adaptive learning of unknown faults by cloud diagnostic models, and synchronization of updated model parameters to edge circuit breakers.

4. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 1, characterized in that, In S1, the real-time high-frequency sampling of the bus current is a point-by-point continuous sampling, with a sampling frequency of 200kHz to 1MHz and 8000 to 12000 sampling points in a single analysis period.

5. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 1, characterized in that, The local preset benchmark method in S1 is based on common fault arc modes and threshold settings. The edge device has a built-in local database that stores known load characteristics and corresponding arc modes. After trip protection, the local database is used to search and match the arc-generating load type and arc-generating branch.

6. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 1, characterized in that, In S2, the edge-side device performs time-frequency transformation processing on the current signal to extract fault features, including: The current signal was decomposed into a time-frequency matrix using Rbio3.1 mother wavelet. Wavelet packet coefficients in frequency bands with obvious fault characteristics are selected for reconstruction, resulting in a reconstructed wavelet packet coefficient sequence with the same sampling length as the original current signal.

7. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 1, characterized in that, The process of cascading diagnosis in S3 through the meta-learning module within the cloud-based diagnostic model and the deep belief network includes: The fault features and the original current signal data are normalized and merged into a high-dimensional hybrid feature vector. The embedding layer of the input meta-learning module is encoded as a dense representation. After passing through a multi-head attention layer to capture the global dependencies between features, the meta-features are output. The meta-features are combined with the original current signal data and input into a deep belief network for processing, and the output is a diagnostic result containing the arc-generating load type and arc-generating branch information of the fault arc.

8. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 1, characterized in that, The adaptive learning in S3, based on the on-site tag information fed back by the mobile terminal, includes: The on-site tag information is the tagged sample information uploaded by maintenance personnel through mobile terminals, including fault confirmation information of non-operation samples, on-site troubleshooting results of erroneous operation samples, and corresponding waveform data. The adaptive learning is a cloud-based diagnostic model that adjusts its internal parameters and optimizes the fault discrimination boundary based on the labeled sample information.

9. The adaptive fault arc detection method based on cloud-edge-device collaboration according to claim 1, characterized in that, The adaptive upgrade of the locally preset benchmark method in S4 specifically involves the edge device updating the local fault detection, line selection, and location algorithm logic according to the updated model parameters, and supplementing and updating the locally stored load characteristics and arc pattern database.

10. An adaptive fault arc detection system based on cloud-edge-device collaboration, characterized in that, This includes edge devices, cloud-based diagnostic models, and mobile terminals; The edge-side device is used to obtain a current signal by performing real-time high-frequency sampling of the bus current; it analyzes the sampled signal in real time using a locally preset benchmark method; if the fault type is determined to be known, it performs trip protection and notifies the user through a mobile terminal; if the fault type is determined to be unknown or unidentifiable, it transmits the fault characteristics to the cloud diagnostic model and issues an unknown fault warning through the mobile terminal. The cloud-based diagnostic model is used to receive fault characteristics, perform cascade diagnosis through the meta-learning module and deep belief network within the cloud-based diagnostic model, and perform adaptive learning by combining the field label information fed back through the mobile terminal. It outputs diagnostic results containing the arc-generating load type and arc-generating branch information of the fault arc, and pushes the diagnostic results to the mobile terminal. It is also used to synchronously push the updated model parameters to the edge device to achieve adaptive upgrades of the locally preset benchmark method.

Images