Abnormality identification method, device and equipment of low-voltage electric energy metering device and medium

Abstract

The application discloses an abnormality identification method, device and equipment of a low-voltage electric energy metering device and a medium, and relates to the technical field of equipment monitoring. The method comprises the following steps: synchronously collecting multi-modal information of a target low-voltage electric energy metering device, including appearance image information for appearance abnormality identification, thermal imaging information for heating abnormality identification, asset identification information for asset information checking, environmental parameter information for environmental state evaluation, and electrical parameter information for data quality analysis; performing spatio-temporal alignment processing and fusion analysis on the multi-modal information to generate an abnormality identification result; performing risk prompting on the result, and generating auxiliary research and judgment information based on a local knowledge base and an end-side reasoning model; automatically generating an abnormality problem list and uploading the abnormality problem list to a background system, and automatically generating a maintenance work order. Through the technical scheme, the problems in traditional manual inspection, such as scattered detection means, fragmented multi-source data, inconvenient human-computer interaction, long business closed loop cycle and the like, are solved.

Description

Technical Field

[0001] This invention relates to the field of equipment monitoring technology, and in particular to a method, device, equipment and medium for anomaly identification of low-voltage power metering devices. Background Technology

[0002] Low-voltage electricity metering devices, acting as the "nerve endings" of the power grid, undertake crucial functions such as electricity metering, data acquisition, and electricity billing. Their operational status directly impacts the fairness of electricity trade settlement and the security of power grid operation. These devices mainly include electricity meters, metering boxes, instrument transformers, and secondary circuits, and are numerous, with most installed outdoors, enduring long-term exposure to complex environments and harsh weather conditions. During operation, these devices are prone to problems such as visible damage, overheating, abnormal asset information, and data acquisition anomalies. Traditional operation and maintenance management mainly relies on manual inspections, which suffers from limitations such as limited detection methods, delayed anomaly detection, and long response cycles.

[0003] In recent years, although individual technologies such as computer vision, thermal imaging temperature measurement, RFID (Radio Frequency Identification) asset verification, environmental sensing, and electrical data acquisition have been gradually introduced, shortcomings still exist: First, although there are many sensing dimensions, the fusion depth is insufficient. Each module collects data independently, lacking spatiotemporal alignment and cross-validation, making it difficult to achieve true multimodal fusion analysis. For example, when temperature anomalies and smoke alarms are triggered independently, it is impossible to comprehensively judge the authenticity of fire hazards. Second, intelligent judgment relies on the cloud, limiting on-site real-time performance. Existing AI diagnosis mostly requires uploading to the backend for processing. Due to network limitations, portable equipment cannot obtain real-time and accurate assistance from the edge. Third, human-computer interaction is still mainly based on one-way instructions, lacking knowledge empowerment. When faced with complex faults, operators still need to consult materials or make phone calls for help. Large model technology has not yet been deeply integrated with field equipment, and cannot provide expert-level support with immediate answers. Fourth, the business loop has not been fully opened. Work order generation still involves manual steps. After on-site confirmation, work order information needs to be filled in manually. The time window from anomaly discovery to work order dispatch is still relatively long. Fifth, the equipment integration is low. Multiple independent devices such as portable testers, thermal imagers, RFID readers, and environmental monitoring instruments need to be carried at the same time. The operation is cumbersome, the data is scattered, and it is difficult to form a unified judgment view.

[0004] Therefore, for low-voltage power metering devices, how to achieve comprehensive on-site detection, intelligent analysis, intuitive human-machine interaction, and efficient closed-loop operation are urgent problems to be solved. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, equipment, and medium for anomaly identification in low-voltage electricity metering devices. For low-voltage electricity metering devices, this invention enables comprehensive on-site detection, intelligent analysis, intuitive human-machine interaction, and efficient closed-loop operation. The specific solution is as follows: In a first aspect, this application discloses a method for identifying anomalies in a low-voltage power metering device, including: Simultaneously collect multimodal information of the target low-voltage power metering device; the multimodal information includes: appearance image information for identifying appearance anomalies, thermal imaging information for identifying heating anomalies, asset identification information for asset information verification, environmental parameter information for environmental status assessment, and electrical parameter information for data quality analysis; The multimodal information is spatiotemporally aligned, and fusion analysis is performed based on the aligned multimodal information to generate anomaly identification results. The anomaly identification results are given a risk warning in a preset interaction mode, and auxiliary judgment information is generated based on the local knowledge base and the preset terminal inference model, so as to receive the confirmation operation returned by the operator based on the anomaly identification results and the auxiliary judgment information. An abnormal problem list containing the abnormal identification results confirmed by the aforementioned confirmation operation is automatically generated and uploaded to the backend system so that a repair work order can be automatically generated based on the abnormal problem list.

[0006] Optionally, the synchronous acquisition of multimodal information of the target low-voltage power metering device includes: The camera is used to capture images of the meter readings, wiring terminals, and enclosure of the target low-voltage electricity metering device to obtain visual information. The heat-generating parts of the target low-voltage power metering device are scanned using a thermal imager to obtain thermal imaging information; The target low-voltage electricity metering device is scanned using a radio frequency identification (RFID) interactive device to obtain asset identification information. The environmental sensor module collects real-time information on the internal and / or surrounding environmental parameters of the target low-voltage power metering device. The electrical parameter information inside the target low-voltage power metering device is copied through the data copying interface; the electrical parameter information includes at least one of voltage, current, power, energy, or power factor.

[0007] Optionally, the spatiotemporal alignment processing of the multimodal information includes: Using a linear interpolation algorithm, the time axes of the thermal imaging information, the asset identification information, the environmental parameter information, and the electrical parameter information are aligned to the video frame time axis of the appearance image information for time alignment. Common feature points are extracted from the appearance image information and the thermal imaging information to calculate the affine transformation matrix. Based on the affine transformation matrix, a pixel-level mapping relationship is established between the thermal imaging information and the appearance image information for spatial alignment.

[0008] Optionally, fusion analysis can be performed based on the aligned multimodal information to generate anomaly identification results, including: The appearance image information is analyzed by calling a deep learning-based target detection model to identify appearance defects, and the meter reading is extracted by optical character recognition and compared with the historical electrical parameter information to identify metering anomalies and obtain the first anomaly feature. Based on the thermal imaging information, the temperature difference between the heating part and the current ambient temperature is determined, and the heating anomaly is identified based on the difference. The temperature rise trend is predicted based on the difference and the load current measured through the image to identify potential overheating risks, thus obtaining a second abnormal feature. The asset identification information is compared with the asset ledger information stored in the local knowledge base to detect asset anomalies and obtain a third anomaly feature. The environmental parameter information is subjected to threshold judgment to obtain the fourth abnormal feature, and the electrical parameter information is analyzed to obtain the fifth abnormal feature; The first, second, third, fourth, and fifth abnormal features are subjected to multimodal cross-validation to comprehensively assess the authenticity and severity of the abnormal features, thereby generating the anomaly identification result.

[0009] Optionally, the step of providing risk warnings for the anomaly identification results using a preset interaction mode includes: The thermal imaging information is overlaid with the appearance image information, and based on the anomaly identification result, the abnormal location is marked on the appearance image information using augmented reality technology, and the anomaly identification result is broadcast through voice synthesis; the abnormal location includes heated parts, damaged areas, or abnormal locations of assets.

[0010] Optionally, the generation of auxiliary judgment information based on the local knowledge base and the preset edge-side reasoning model includes: Receive natural language processing instructions input by operators; The system invokes a pre-defined edge-side inference model deployed on the field equipment after knowledge distillation, model pruning, and quantization compression, as well as a local knowledge base containing metrological technical specifications, typical fault cases, and anomaly handling procedures, to parse and respond to the natural language processing instructions in order to generate auxiliary judgment information. The auxiliary judgment information includes processing suggestions, professional knowledge answers, or data analysis results for the anomaly identification results.

[0011] Optionally, the automatic generation of an anomaly problem list containing the anomaly identification results confirmed by the confirmation operation, and uploading it to the backend system, so as to automatically generate a repair work order based on the anomaly problem list, includes: In response to the confirmation operation, the system automatically packages the anomaly type, anomaly location, severity level, multimodal evidence collected on-site, timestamp, and geographic location information to generate an anomaly list. The list of abnormal issues is encrypted using an encryption algorithm and uploaded to the backend system via a preset communication network. The backend system then parses the list and matches it with a preset work order template to automatically generate a repair work order.

[0012] Secondly, this application discloses an anomaly identification device for a low-voltage power metering device, comprising: The information acquisition module is used to synchronously acquire multimodal information of the target low-voltage power metering device; the multimodal information includes: appearance image information for appearance anomaly identification, thermal imaging information for heating anomaly identification, asset identification information for asset information verification, environmental parameter information for environmental status assessment, and electrical parameter information for data quality analysis. An anomaly detection module is used to perform spatiotemporal alignment processing on the multimodal information and perform fusion analysis based on the aligned multimodal information to generate anomaly detection results; The interaction module is used to provide risk warnings for the anomaly identification results in a preset interaction mode, and to generate auxiliary judgment information based on the local knowledge base and the preset end-side reasoning model, so as to receive confirmation operations returned by the operators based on the anomaly identification results and the auxiliary judgment information. An automated closed-loop module is used to automatically generate a list of abnormal issues containing the abnormal identification results confirmed by the confirmation operation, and upload it to the backend system so that a maintenance work order can be automatically generated based on the list of abnormal issues.

[0013] Thirdly, this application discloses an electronic device including a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the aforementioned abnormal identification method for low-voltage power metering devices.

[0014] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the aforementioned method for identifying anomalies in a low-voltage power metering device.

[0015] The beneficial effects of this application are as follows: First, at the perception and judgment level, this invention constructs a multimodal information perception architecture, fundamentally solving the problems of scattered detection methods and single perception dimensions in existing technologies. Based on this, through spatiotemporal alignment and fusion analysis of multimodal information, precise time synchronization and spatial registration of multimodal information are achieved, enabling cross-verification of the authenticity of anomalies from multiple dimensions. This significantly improves the accuracy and reliability of anomaly identification, effectively avoiding false alarms or missed alarms from a single dimension. Second, at the interaction and efficiency level, the end-side inference model of this invention is a lightweight metrological large model, enabling equipment to possess local intelligent inference capabilities, eliminating dependence on cloud networks. Even in areas with poor signal, it can obtain instant question-and-answer and data analysis support, transforming expert experience into readily accessible digital capabilities on-site, significantly reducing the reliance on personnel experience in on-site operations. Simultaneously, the automatically generated anomaly identification results are output intuitively on-site, allowing operators to directly confirm them without the need for manual recording or returning to the office for system entry. Finally, the automated work order generation and dispatch process is triggered, seamlessly connecting on-site discoveries to back-end processing. This completely changes the cumbersome process of on-site recording, post-event entry, and manual dispatch in the traditional model, significantly shortening the anomaly handling cycle and improving the digitalization and automation level of operation and maintenance management. In summary, this invention constructs an intelligent closed-loop system covering the entire chain of "collection-fusion-analysis-interaction-handling," solving the inherent defects of fragmented perception capabilities, experience-dependent on-site decision-making, and fragmented business processes in the traditional inspection model. Furthermore, it upgrades the operation and maintenance of low-voltage metering devices from the traditional "manually dominated, segmented operation" model to a new digital model of "data-driven, human-machine collaboration, and closed-loop automation," providing a practical and feasible technical path for the efficient and lean management of massive low-voltage metering devices in the context of new power systems.

[0016] Furthermore, the anomaly identification device, equipment, and storage medium for a low-voltage power metering device provided in this application correspond to the aforementioned anomaly identification method for a low-voltage power metering device, and have the same effect. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1This is a flowchart of an anomaly identification method for a low-voltage power metering device disclosed in this application; Figure 2 This is a schematic diagram of an anomaly identification hardware module disclosed in this application; Figure 3 This application discloses a flowchart of an anomaly identification process for a specific low-voltage power metering device. Figure 4 This is a schematic diagram of the anomaly identification device of a low-voltage power metering device disclosed in this application; Figure 5 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Low-voltage electricity metering devices are exposed to complex environments for extended periods, making them prone to problems such as physical damage, overheating, inconsistencies in asset information, and abnormal data acquisition. However, current maintenance technologies still rely primarily on manual inspections and single-dimensional testing, which present the following issues: First, the detection methods are fragmented and lack comprehensive coverage. Currently, visual inspection relies on manual inspection, heat detection requires handheld infrared thermometers, asset verification uses barcode scanners, environmental conditions (temperature, humidity, smoke) rely on independent instruments, and electrical data needs to be queried through a backend system. During on-site operations, staff need to carry multiple independent devices (such as on-site testing instruments, thermal imagers, RFID readers, environmental monitoring instruments, etc.), which is cumbersome and results in fragmented data. It is impossible to obtain complete information on the operating status of metering devices in the same time and space, leading to the inability to detect potential faults in a timely manner.

[0021] Second, fragmented multi-source data leads to weak comprehensive decision-making capabilities. Because the various sensing modules are independent of each other, it is impossible to perform spatiotemporal alignment and correlation analysis of image, temperature, asset, environmental, and electrical data, resulting in the phenomenon of "data silos." This leads to a high false alarm rate, difficulty in accurately identifying real anomalies, and anomaly identification heavily relies on personal experience.

[0022] Third, human-computer interaction is inconvenient and on-site knowledge is lacking. Existing portable devices mainly rely on button operation and simple digital displays, resulting in unintuitive information presentation. Operators still need to manually locate abnormal locations by referring to drawings. When encountering complex malfunctions, staff cannot obtain immediate professional knowledge support and can only consult paper documents or seek expert help online, leading to low on-site handling efficiency.

[0023] Fourth, there is a disconnect between the on-site and back-end operations, resulting in a lengthy closed-loop cycle. In the traditional model, when an anomaly is discovered on-site, it needs to be manually recorded, then entered into the system and a work order filled out manually upon returning to the office. This process involves multiple human interventions. From anomaly confirmation to work order dispatch, it often takes several hours or even longer. Emergency anomalies cannot be responded to in a timely manner, and there is a high risk of record omissions or incorrect work order dispatch, making it difficult to shift from "passive firefighting" to "proactive early warning."

[0024] In recent years, with the in-depth advancement of the construction of new power systems and the rapid development of artificial intelligence technology, the operation and maintenance management of low-voltage power metering devices is undergoing a profound transformation from traditional manual inspection to intelligent and digital transformation. Significant progress has been made in five perception dimensions: appearance, heat generation, assets, environment, and electrical data. Computer vision can now automatically identify visible defects such as meter damage and loose wiring; thermal imaging technology is gradually being integrated into inspection equipment, enabling real-time capture of abnormal temperatures such as contact overheating and overload burnout; the integrated application of RFID and mobile terminals has greatly improved asset verification efficiency; intelligent monitoring modules integrating temperature sensing, smoke detection, and video surveillance functions have achieved real-time monitoring of temperature and smoke inside the enclosure; and the electricity information collection system has extremely high coverage, achieving minute-level data collection. Regarding human-computer interaction, voice interaction and image display have become increasingly common, and mobile applications support convenient operations such as "pointing, selecting, scanning, and taking pictures." In terms of business closed-loop management, some systems have already implemented anomaly reporting and work order dispatch.

[0025] Currently, the more similar implementation schemes mainly fall into two categories: First, portable high and low voltage metering secondary circuit field integrated testing instruments. This device reads metering device data via wireless communication, has a built-in database of over 300 anomalies and a multi-dimensional algorithm model, and outputs the anomaly type, severity, and location after on-site comparison. However, it focuses on electrical parameter detection and does not integrate thermal imaging, RFID, or environmental sensing functions. Second, a method and system for diagnosing and analyzing operational errors of low-voltage distribution area metering devices. This involves collecting data through accompanying diagnostic equipment and sending it to the control center. After verification through proximity comparison, tasks are assigned to maintenance personnel. However, the on-site and back-end systems are relatively independent, and manual processing is still required after anomaly confirmation. These schemes generally suffer from shortcomings such as a single sensing dimension, lack of multi-source data fusion, disconnect between on-site and back-end systems, lack of intelligent interactive support, and low equipment integration.

[0026] It is evident that the existing technology has the following main drawbacks: 1. The various sensing modules (vision, thermal imaging, RFID, and environmental sensors) operate independently and fail to achieve spatiotemporal alignment and cross-validation of multi-source data. This results in an inability to make comprehensive judgments when facing complex and abnormal scenarios, thus exhibiting a technical defect of low accuracy in anomaly judgment. 2. Intelligent diagnostic functions are mostly deployed in the cloud backend. Data collected on-site needs to be uploaded and processed, which is heavily dependent on network conditions. In on-site environments with poor signal or network interruption, it is impossible to obtain real-time diagnostic results, thus resulting in technical defects such as low on-site handling efficiency. 3. Existing portable devices only provide data display and simple alarm functions, lacking built-in knowledge base and large model question-and-answer capabilities. When on-site personnel encounter complex faults, they need to consult materials or seek help by phone, and cannot obtain immediate professional knowledge support. Therefore, there is a technical defect of weak ability to handle difficult problems. 4. On-site detection and the back-end system are relatively independent. After the anomaly is confirmed, a work order needs to be filled out manually and sent back to the system. There is a manual intervention link in the process. The time window from the discovery of the anomaly to the dispatch of the work order is relatively long. Therefore, there are technical defects such as incomplete business loop and long handling cycle. 5. Currently, multiple independent devices such as portable testing instruments, thermal imagers, RFID readers, and environmental monitoring instruments need to be carried at the same time. These devices are not interconnected, and the data is scattered. On-site operation is cumbersome and it is difficult to form a unified analysis view. Therefore, there are technical defects such as low equipment integration and poor operation convenience.

[0027] To address the aforementioned shortcomings, this application provides an anomaly identification scheme for low-voltage electricity metering devices. This scheme enables deep fusion perception of multi-source data across five dimensions, real-time intelligent assistance from a large-scale end-side model, natural interaction via voice and augmented reality, one-click closed-loop work order initiation, and highly integrated embodied form, thus supporting the operation and maintenance management of low-voltage metering devices.

[0028] This invention discloses a method for identifying anomalies in low-voltage power metering devices. See [link to relevant documentation]. Figure 1 As shown, the method includes: Step S11: Synchronously collect multimodal information of the target low-voltage power metering device; the multimodal information includes: appearance image information for appearance anomaly identification, thermal imaging information for heat generation anomaly identification, asset identification information for asset information verification, environmental parameter information for environmental status assessment, and electrical parameter information for data quality analysis.

[0029] In this embodiment, multi-modal sensing functionality integrated into the same equipment is first used to synchronously collect multi-dimensional information from the target low-voltage power metering device. This multi-modal sensing functionality integrates sensor data for acquiring different physical properties into a single acquisition moment, aiming to provide a comprehensive and synchronous raw data foundation for subsequent spatiotemporal alignment and fusion analysis.

[0030] Specifically, synchronous acquisition includes acquiring at least five types of multimodal information: (1) Acquisition of appearance image information: The meter readings, wiring terminals, and enclosure appearance of the target low-voltage electricity metering device are captured by a camera to obtain appearance image information. This image is used to identify visible defects of the metering device, such as meter damage, loose wiring, enclosure corrosion, and abnormal opening of the enclosure door. To adapt to different lighting environments, an industrial-grade high-definition camera with a resolution of not less than 1920×1080 and supporting autofocus and supplementary lighting functions can be used.

[0031] (2) Acquisition of thermal imaging information: The heat-generating parts of the target low-voltage power metering device are scanned by a thermal imager to collect the heat distribution information of the metering device and form thermal imaging information. This information is used to identify potential heat-related damage hazards such as overheating of key parts such as terminals, circuit breakers, and busbars, as well as abnormal temperatures caused by equipment overload or insulation aging. The thermal imager can be an uncooled infrared focal plane detector with a certain resolution (e.g., resolution not less than 320×240) and a temperature measurement range (e.g., -20℃ to +150℃), and the temperature measurement accuracy can reach ±2℃.

[0032] (3) Asset Identification Information Collection: Asset identification information is obtained by scanning the electronic tag of the target low-voltage power metering device using a radio frequency identification (RFID) interactive device. This information is used for asset verification, seal integrity verification, and determination of the validity of the verification date. The RFID interactive device supports UHF RFID tag reading and writing, with an adjustable reading distance (e.g., 0-3 meters), and complies with the EPC C1 G2 standard protocol. In one feasible implementation, QR code, barcode recognition, or NFC can be used to replace RFID. For tagless scenarios, nameplate information can be directly identified through OCR.

[0033] (4) Collection of environmental parameter information: The environmental sensor module collects the internal and / or surrounding environmental parameter information of the target low-voltage power metering device in real time. The environmental sensor module integrates temperature sensors, humidity sensors, smoke sensors, water immersion sensors and vibration sensors, etc. The collected environmental parameter information includes, but is not limited to, the temperature and humidity inside the box, smoke concentration, water immersion status, vibration and shock, etc. This data provides environmental data support for the comprehensive assessment of abnormal conditions (such as the risk of insulation degradation caused by dampness, fire hazards indicated by smoke, etc.).

[0034] (5) Acquisition of electrical parameter information: The electrical parameter information inside the target low-voltage electricity metering device is acquired through the data transceiver interface; the electrical parameter information includes at least one of voltage, current, power, energy consumption, or power factor, and may also include historical data from the most recent several days (e.g., 7 days). The data transceiver interface supports multiple communication protocols such as DL / T645 and Modbus, and can communicate with the acquisition terminal via power line carrier, low-power wireless, or RS485, for subsequent judgment of data acquisition quality and metering data accuracy. In a feasible implementation, electrical parameter information can also be obtained by directly reading the electricity meter register via Bluetooth or Wi-Fi, without relying on the data transceiver interface.

[0035] All five types of data are synchronously triggered and recorded in the same acquisition task, with the acquisition time and acquisition location remaining consistent, thus forming a multi-dimensional, synchronized raw data set, providing input with a unified time and space reference for subsequent spatiotemporal alignment processing.

[0036] It should be noted that, in this embodiment, through integrated hardware design, the camera, thermal imager, RFID interactive device, environmental sensor module (including temperature sensor, humidity sensor, and smoke sensor), and data transmission interface are highly integrated into a single device, forming a five-in-one multimodal sensing integrated architecture. This changes the traditional single or scattered detection mode, realizing real-time synchronous acquisition of five dimensions of information: appearance image, thermal imaging temperature distribution, RFID asset information, environmental status parameters (such as temperature, humidity, and smoke), and electrical data (voltage, current, power, and electricity). This provides a complete hardware foundation for a comprehensive and three-dimensional understanding of the operating status of the metering device.

[0037] Step S12: Perform spatiotemporal alignment processing on the multimodal information, and perform fusion analysis based on the aligned multimodal information to generate anomaly identification results.

[0038] After multimodal information acquisition, the raw data collected by different sensors (such as cameras, thermal imagers, RFID readers, environmental sensor modules, and data transfer interfaces) have different sampling frequencies, triggering times, and physical coordinate systems, resulting in deviations in both time and spatial location, making direct correlation analysis impossible. Therefore, this step first performs spatiotemporal alignment processing on the multimodal information collected by different sensors, unifying it to the same time reference in the time dimension and mapping the data from different sensors to the same coordinate space in the spatial dimension. This ensures that data from the same time and spatial location can be correlated.

[0039] After completing spatiotemporal alignment, the sensing data from each dimension are matched in time and space, forming a complete information description of the target low-voltage electricity metering device at the same time and spatial location. Furthermore, based on this aligned multimodal information, independent anomaly identification is performed from five dimensions: appearance, heat generation, assets, environment, and electrical systems. Anomaly features are extracted, and the identification results from each dimension are fused through multimodal cross-validation. This allows for accurate identification of the abnormal state of the target low-voltage electricity metering device, generating anomaly identification results that include information such as anomaly type, location, and severity.

[0040] Step S13: Provide risk warnings for the anomaly identification results using a preset interactive mode, and generate auxiliary judgment information based on the local knowledge base and the preset edge-side reasoning model, so as to receive confirmation operations returned by the operators based on the anomaly identification results and the auxiliary judgment information.

[0041] After completing the multimodal fusion analysis and generating anomaly identification results, this step uses a human-computer interaction module to present the anomaly identification results to the on-site operators in a pre-defined interaction mode, and allows the operators to confirm the authenticity of the anomaly on-site.

[0042] Specifically, the human-computer interaction module integrates an image display unit, an augmented reality annotation unit, and a voice interaction unit. First, the image display unit displays real-time images of the target low-voltage electricity metering device captured by the camera. Simultaneously, the augmented reality annotation unit constructs a multi-dimensional, intuitive interactive system using augmented reality technology. It precisely overlays the thermal imaging information collected by the thermal imager with the appearance image information collected by the camera, visually presenting the temperature distribution on the visible light image, making hot spots immediately apparent. Building upon this, augmented reality technology overlays anomaly identification results (such as hot spots, damaged areas, and abnormal asset locations) onto the real-time image in a visual annotation format (such as colored annotation boxes, temperature labels, and warning symbols), providing a WYSIWYG risk warning. For example, a red high-temperature area is overlaid at the location of a hot spot, and a yellow border is overlaid at a damaged area to intuitively understand the location of the anomaly. Furthermore, the voice interaction unit uses speech synthesis technology to convert the anomaly recognition results into voice broadcasts, issuing prompts to operators, such as "The temperature of the A-phase terminal is too high, the current temperature is 65℃, please check." The full-process voice interaction allows operators to complete the entire process of equipment operation, anomaly confirmation, and knowledge query through voice commands, completely freeing their hands and significantly improving the convenience and safety of on-site operations.

[0043] In one feasible implementation, augmented reality interaction can also be achieved by projecting directly onto a physical surface using a micro-projector, or by separating the display function into head-mounted AR glasses.

[0044] In a preferred embodiment of the present invention, before outputting the anomaly identification results to the operators, a local knowledge base and a preset edge-side inference model are invoked to generate auxiliary judgment information. The local knowledge base contains a built-in metrology professional knowledge base, including metrology technical specifications, typical fault cases, anomaly handling procedures, etc. The preset edge-side inference model, through model lightweighting technology, first uses knowledge distillation to transfer the knowledge of the large cloud model to the lightweight student model, compressing it to a scale suitable for edge deployment, and then combines pruning and INT8 quantization compression to generate a preset edge-side inference model with a model size of approximately 100MB. It supports real-time model loading and on-site inference, enabling instant question answering and auxiliary judgment without relying on the cloud, thus completely eliminating dependence on cloud networks.

[0045] Operators can input natural language processing (NLP) commands via voice or text, which are then parsed and responded to, building NLP interaction and knowledge query capabilities to generate auxiliary judgment information. This auxiliary judgment information includes processing suggestions for the anomaly identification results (e.g., check the tightness of the wiring terminals), professional knowledge explanations (e.g., analysis of common overheating causes of this type of electricity meter), or data analysis results (e.g., electricity consumption trend analysis). For example, if an operator asks "How should this anomaly be handled?", the system provides step-by-step processing suggestions based on the anomaly type and processing flow in the knowledge base, transforming expert experience into readily accessible digital capabilities on-site. This auxiliary judgment information is also output to operators through augmented reality annotation units and voice interaction units, providing immediate expert-level support for on-site decision-making. After on-site verification based on the anomaly identification results and auxiliary judgment information, operators can confirm the anomaly via voice commands (e.g., "Confirm anomaly") or touch operation. This confirmation marks the final on-site determination of the anomaly's authenticity, providing a trigger for a one-click business loop in subsequent steps.

[0046] Step S14: Automatically generate an abnormal problem list containing the abnormal identification results confirmed by the confirmation operation, and upload it to the background system so that a repair work order can be automatically generated based on the abnormal problem list.

[0047] After outputting the anomaly identification results and receiving confirmation from the operator, this embodiment executes the anomaly confirmation and business closure triggering steps. Specifically, in response to the confirmation operation returned by the operator in a preset interaction mode, the equipped equipment automatically generates a standardized list of anomalies and uploads the list to the backend system through a preset communication network, thereby triggering the subsequent automated work order generation and dispatch process, achieving seamless connection from on-site discovery to backend handling.

[0048] In this embodiment, for anomalies confirmed on-site, the system automatically generates an anomaly list that includes at least the following standardized fields: anomaly type (e.g., abnormal heating, visual damage, asset discrepancies), anomaly location (e.g., A-phase terminal block), severity level (e.g., general, severe, emergency), multimodal evidence collected on-site (including anomaly images, thermal images, RFID information, environmental parameter snapshots, and transcribed electrical data), timestamps, and geographic location information (e.g., GPS coordinates provided by the equipment's built-in positioning module). To ensure data transmission security, the system encrypts the generated anomaly list using the AES-256 encryption algorithm and then transmits it back to the backend system with a single click via a preset communication network (e.g., 4G / 5G long-distance wireless communication), ensuring secure data transmission.

[0049] After receiving the list of abnormal issues, the backend system automatically matches the corresponding work order template based on the type and severity level of the abnormality recorded in the list, generates a maintenance work order, and assigns it to the corresponding maintenance team. For abnormalities with an emergency severity level (such as smoke alarms, water immersion alarms, or combined abnormalities of high temperature accompanied by smoke), the backend system can further trigger SMS or instant message notifications to achieve immediate response to emergency abnormalities. Thus, a complete business loop is formed from on-site abnormality discovery, confirmation, evidence collection, list generation, encrypted uploading, to automatic work order dispatch in the backend, eliminating intermediate steps such as manual recording, post-event entry, and manual form filling in the traditional model, significantly shortening the abnormality handling cycle.

[0050] The beneficial effects of this application are as follows: First, at the perception and judgment level, this invention constructs a multimodal information perception architecture, fundamentally solving the problems of scattered detection methods and single perception dimensions in existing technologies. Based on this, through spatiotemporal alignment and fusion analysis of multimodal information, precise time synchronization and spatial registration of multimodal information are achieved, enabling cross-verification of the authenticity of anomalies from multiple dimensions. This significantly improves the accuracy and reliability of anomaly identification, effectively avoiding false alarms or missed alarms from a single dimension. Second, at the interaction and efficiency level, the end-side inference model of this invention is a lightweight metrological large model, enabling equipment to possess local intelligent inference capabilities, eliminating dependence on cloud networks. Even in areas with poor signal, it can obtain instant question-and-answer and data analysis support, transforming expert experience into readily accessible digital capabilities on-site, significantly reducing the reliance on personnel experience in on-site operations. Simultaneously, the automatically generated anomaly identification results are output intuitively on-site, allowing operators to directly confirm them without the need for manual recording or returning to the office for system entry. Finally, the automated work order generation and dispatch process is triggered, seamlessly connecting on-site discoveries to back-end processing. This completely changes the cumbersome process of on-site recording, post-event entry, and manual dispatch in the traditional model, significantly shortening the anomaly handling cycle and improving the digitalization and automation level of operation and maintenance management. In summary, this invention constructs an intelligent closed-loop system covering the entire chain of "collection-fusion-analysis-interaction-handling," solving the inherent defects of fragmented perception capabilities, experience-dependent on-site decision-making, and fragmented business processes in the traditional inspection model. Furthermore, it upgrades the operation and maintenance of low-voltage metering devices from the traditional "manually dominated, segmented operation" model to a new digital model of "data-driven, human-machine collaboration, and closed-loop automation," providing a practical and feasible technical path for the efficient and lean management of massive low-voltage metering devices in the context of new power systems.

[0051] Based on the above embodiments, in one specific embodiment, after the multimodal information acquisition is completed, the acquired data from each channel is first preprocessed to remove noise and outliers; then, the data acquired by different sensors are spatiotemporally aligned. Specifically, the spatiotemporal alignment process in step S12 includes two sub-steps: time alignment and spatial alignment.

[0052] (1) Time alignment: Using a linear interpolation algorithm, the time axes of the thermal imaging information, the asset identification information, the environmental parameter information and the electrical parameter information are aligned to the video frame time axis of the appearance image information to perform time alignment.

[0053] Time alignment is the process of aligning various data streams to a unified time reference. Since the data sampling frequency of thermal imagers, RFID interactive devices, environmental sensor modules, and data transceiver interfaces is usually lower than the camera video frame rate, this embodiment uses a linear interpolation method to align low-frequency data such as thermal imaging, RFID, environmental parameters, and electrical parameters to the time axis of the appearance image information video frames.

[0054] (2) Spatial alignment: Extract common feature points from the appearance image information and the thermal imaging information to calculate the affine transformation matrix, and establish a pixel-level mapping relationship between the thermal imaging information and the appearance image information based on the affine transformation matrix to perform spatial alignment.

[0055] Spatial alignment involves spatially registering the thermal imaging information acquired by the thermal imager with the appearance image information acquired by the camera, enabling precise overlay of temperature field data onto the visible light image. This embodiment employs a feature-point-based image registration algorithm for spatial alignment. First, common feature points are extracted from both the appearance image and thermal imaging information, such as meter edges, screw locations, cabinet corners, and terminals—areas with clearly defined geometric features. Then, by matching these feature point pairs, an affine transformation matrix (including transformation parameters such as rotation, translation, and scaling) is calculated. Finally, this affine transformation matrix is ​​used to map each pixel of the thermal image to the corresponding pixel in the visible light image, establishing a pixel-level mapping relationship between the thermal imaging information and the appearance image information, achieving pixel-level alignment. After spatial alignment, each pixel in the visible light image corresponds to a precise temperature value, allowing for intuitive viewing of the temperature distribution of any location on the visible light image. In a feasible implementation, a multispectral camera can be used instead of the independent configuration of a camera and thermal imager, allowing a single device to simultaneously acquire visible light images and infrared temperature distributions.

[0056] Understandably, for other dimensions of perceived data (such as environmental parameters and electrical parameters), since they do not involve spatial coordinate information, only temporal alignment is required for subsequent fusion analysis. Spatiotemporal alignment can also be achieved by using hardware synchronous triggering instead of post-processing interpolation, and by using real-time feature point registration or deep learning networks instead of pre-calibration.

[0057] Furthermore, in one specific embodiment of the present invention, the fusion analysis based on the aligned multimodal information is performed sequentially according to the following six sub-steps: Step 1: Call the deep learning-based target detection model to analyze the appearance image information to identify appearance defects, and extract the meter reading through optical character recognition and compare it with the historical electrical parameter information to identify metering anomalies and obtain the first anomaly feature; In this step, the appearance images captured by the camera are used to identify defects such as meter damage, loose wiring, cabinet corrosion, and open doors by employing deep learning-based target detection models (e.g., YOLO, SSD). Simultaneously, optical character recognition (OCR) technology is used to read the meter readings, which are then compared with historical data to determine if any metering anomalies exist, such as sudden increases in power consumption, reverse readings, or prolonged periods of inactivity. The identified appearance defects and metering anomalies are combined to determine the primary anomaly characteristic.

[0058] Step 2: Based on the thermal imaging information, determine the difference between the temperature value of the heating part and the current ambient temperature, identify the heating anomaly based on the difference, and predict the temperature rise trend based on the difference and the load current through the transceiver to identify potential overheating risks, thereby obtaining the second abnormal feature.

[0059] In this step, the thermal imaging information collected by the thermal imager is combined with spatial alignment results to extract the temperature values ​​of key heat-generating components (such as terminals, circuit breakers, busbars, etc.). A temperature difference analysis method is used to calculate the difference ΔT between the temperature value of each heat-generating component and the current ambient temperature. When ΔT exceeds a preset threshold, an abnormal heating phenomenon is identified at that component. Simultaneously, combined with load current data obtained from the data transmission interface, a temperature rise curve fitting method is used to correlate the current temperature value with the load current to determine if overheating anomalies exist. Temperature trends are predicted to provide early warning of potential overheating risks. The identified heating anomalies and predicted potential overheating risks are combined to determine the second anomaly characteristic.

[0060] Step 3: Compare the asset identification information with the asset ledger information stored in the local knowledge base to detect asset anomalies and obtain the third anomaly feature.

[0061] In this step, the asset identification information (including asset number, seal information, inspection date, etc.) read by the RFID interactive device is compared with the asset ledger information stored in the local knowledge base to verify whether the asset number matches, whether the seal is complete, and whether the inspection date has expired. If the asset identification information is found to be inconsistent, it is determined as an asset anomaly, and the third anomaly characteristic is determined accordingly.

[0062] Step 4: Perform threshold judgment on the environmental parameter information to obtain the fourth abnormal feature.

[0063] In this step, the environmental parameter information collected by the environmental sensor module is evaluated. Specifically, a smoke alarm is triggered when the smoke concentration exceeds a threshold (e.g., >0.2 mg / m³); a humidity alarm is triggered when the humidity exceeds a threshold (e.g., >85%RH) and persists for a longer period than a set time (e.g., 10 minutes), indicating a potential risk of insulation degradation; a water immersion alarm is triggered when the water immersion sensor is activated; and a vibration alarm is triggered when the vibration sensor detects abnormal vibration, indicating potential external damage or equipment loosening. The fourth abnormal characteristic is determined based on the combined results of these assessments.

[0064] Step 5: Analyze the electrical parameter information to obtain the fifth abnormal feature.

[0065] In this step, the electrical parameters obtained from the data transmission interface are analyzed. This includes: determining whether the effective values ​​of voltage and current are zero or below the threshold to identify voltage phase loss or current loss; calculating the imbalance of three-phase voltage or current to determine whether three-phase imbalance exists; analyzing power factor changes to determine whether reactive power compensation is normal; and comparing the current electricity consumption with the historical electricity consumption for the same period to determine whether there are abnormal fluctuations in electricity consumption.

[0066] Step 6: Perform multimodal cross-validation on the first abnormal feature, the second abnormal feature, the third abnormal feature, the fourth abnormal feature, and the fifth abnormal feature to comprehensively assess the authenticity and severity of the abnormal features and generate the abnormal identification result.

[0067] In this step, the identification results from the above dimensions are subjected to multimodal cross-validation to comprehensively assess the authenticity and severity of the abnormal features, generating the final anomaly identification result. In one feasible implementation, abnormal heating is correlated with load current: if an increase in temperature is accompanied by an increase in load current, it is considered normal heating; if the temperature increases but the load current is normal, it may indicate poor contact or insulation aging. Abnormal heating is correlated with smoke alarms: if an abnormal temperature is triggered simultaneously with a smoke alarm, the fire hazard level is high and immediate action is required. External damage is correlated with environmental anomalies: if the enclosure is damaged and the internal humidity is abnormal, it suggests a potential risk of rainwater infiltration. Asset anomalies are correlated with unpacking events: if the RFID cannot read the seal information and the camera recently captured unpacking footage, there may be suspicion of electricity theft.

[0068] Through the above spatiotemporal alignment and fusion analysis of multi-source data, this method can output highly accurate anomaly identification results, enabling multi-source data to produce a synergistic effect of "1+1>2", fundamentally solving the problem of accurate identification and efficient closed-loop handling of complex anomalies that cannot be accomplished by a single dimension. First, by establishing a unified timeline based on video frames and employing interpolation or resampling techniques, the system accurately synchronizes multi-source information with different sampling frequencies, such as thermal imaging, RFID, environmental parameters, and electrical parameters, thus solving the problem of time asynchrony in heterogeneous data. Second, by pre-calibrating the internal and external parameters of the camera and thermal imager, a pixel-level mapping relationship between visible light images and thermal images is established, achieving precise spatial registration and overlay of temperature field data and visible light images. Based on this, the identification results across five dimensions undergo deep multimodal cross-validation. For example, abnormal heating is correlated with load current to distinguish between normal heating and poor contact; abnormal temperature is correlated with smoke alarms to assess fire risk levels; and abnormal assets are correlated with unpacking events to determine suspected electricity theft. This enables a comprehensive judgment of the authenticity and severity of anomalies, effectively improving the accuracy and reliability of anomaly identification.

[0069] In one feasible implementation, multimodal fusion analysis can also use graph neural networks to model the relationships between components, and can use a federated learning architecture to achieve edge-cloud collaborative optimization.

[0070] like Figure 2 The diagram shows a schematic of an anomaly identification hardware module based on the above embodiments. It can be applied to scenarios such as on-site inspection, fault diagnosis, anti-theft electricity investigation, and final acceptance of low-voltage electricity metering devices in industrial and mining enterprises, commercial complexes, etc., improving on-site work efficiency and the accuracy of anomaly identification. The core of the equipment includes four main modules: 1. Information Acquisition Module: This module is equipped with the function of collecting and aggregating external data. It mainly includes a camera, thermal imager, RFID interactive device, environmental sensor module, and data transfer interface. 2. Data Processing Module: This module receives and synchronously analyzes multi-source data, consisting of seven units: data receiving, data preprocessing, spatiotemporal alignment, multimodal fusion analysis, local knowledge base, and lightweight metrology large-scale model. The data receiving unit receives various types of raw data from the information acquisition module; the data preprocessing unit performs filtering, noise reduction, and normalization on the raw data; the spatiotemporal alignment unit aligns multi-source data collected by different sensors to ensure that data from the same time and spatial location can be correlated; the multimodal fusion analysis unit performs fusion analysis based on the aligned data, extracts abnormal features, and generates anomaly identification results. The local knowledge base unit contains a built-in metrology professional knowledge base, including technical specifications for metrology devices, typical fault cases, and anomaly handling procedures; the lightweight metrology large-scale model unit, based on knowledge distillation technology, compresses the large cloud model to a scale suitable for edge deployment, supports natural language question answering and data analysis, and can achieve real-time question answering and auxiliary judgment without relying on the cloud. 3. Communication Unit Module: This module possesses various data transmission functions. Due to the complex environment surrounding the low-voltage metering device, communication distance must be considered; therefore, short-range and long-range wireless communication units are deployed separately. The short-range wireless communication unit is used for data interaction with handheld terminals or other field devices, supporting short-range wireless communication such as Bluetooth and Wi-Fi. The long-range wireless communication module is used for data interaction with the main station system, uploading anomaly lists, receiving work order instructions, etc., and supports 4G / 5G mobile communication. 4. Human-Computer Interaction Module: This module enables two-way information interaction, operation control, and status feedback between humans and equipment. It mainly comprises three units: voice interaction, image display, and augmented reality annotation. It includes a microphone and speaker, supports speech recognition and speech synthesis, allows operation of the equipment via voice commands, and enables the equipment to broadcast risk information via voice. 5. Power Management Module: This module supplies power to the above modules and includes a rechargeable battery and power management circuitry, supporting extended field operations. In one feasible implementation, power can also be supplied via wireless charging, solar supplementation, or direct power from a metering device.

[0071] As can be seen, the equipment adopts a highly flexible modular design concept at the hardware level. The equipment is designed with expandable sensor interfaces, which can flexibly configure the type and number of environmental sensors according to actual application scenarios, such as adding gas sensors, light sensors, etc., to meet diverse inspection needs. Based on this modular architecture, the equipment can be designed in various forms such as handheld terminals, wearable devices (such as AR glasses + handheld detectors), or autonomous mobile robots to adapt to the requirements of different operation scenarios and inspection modes. At the same time, it is equipped with a high-capacity rechargeable battery and intelligent power management circuit, which can support long-term continuous field operation and can dynamically adjust the power supply strategy according to the working status of each module, effectively extending the battery life.

[0072] Furthermore, such as Figure 3 The image shown is in Figure 2 This is a specific anomaly detection flowchart implemented based on the device architecture, which includes: 1. Multimodal Information Acquisition: The information acquisition module performs multimodal information acquisition on the target metering device. Specifically, a camera captures images of the metering device's exterior, including meter readings, wiring terminals, enclosure appearance, and sealing status; a thermal imager captures thermal images of the metering device, focusing on heat-prone areas such as wiring terminals, circuit breakers, and busbars; an RFID interactive device scans the RFID tags on the metering device to read asset numbers, sealing information, and calibration dates; an environmental sensor module collects environmental parameters such as temperature, humidity, and smoke concentration inside the enclosure; and a data transmission interface communicates with the metering device to transmit electrical parameters such as current voltage, current, power, and electricity consumption, as well as historical data from the past 7 days. 2. Multi-source data preprocessing + spatiotemporal alignment: The data preprocessing unit of the data processing module first preprocesses the data collected from each source to remove noise and outliers; then, the spatiotemporal alignment unit performs spatiotemporal alignment on the data collected from different sensors. 3. Multimodal Fusion Analysis and Anomaly Identification: The multimodal fusion analysis unit performs fusion analysis based on spatiotemporally aligned multi-source data to extract anomaly features. Further in-depth analysis includes: appearance anomaly identification, heat generation anomaly identification, asset information verification, environmental status assessment, data quality analysis, and multimodal cross-validation. For more detailed procedures of each of the above steps, please refer to the corresponding content disclosed in the foregoing embodiments; further details will not be repeated here. 4. Human-Computer Interaction and Assisted Judgment: The data processing module pushes the analysis results to the human-computer interaction module, which outputs risk warnings to operators through the voice interaction unit and image display unit. Specifically, the voice prompt function uses the voice interaction unit to broadcast abnormal information via synthesized voice, such as "The temperature of phase A terminal is too high, current temperature 65℃, please check"; the AR annotation function overlays abnormal locations onto the image, such as overlaying a red high-temperature area on the location of a hot spot and a yellow border on a damaged area, providing a clear visual understanding of the abnormal location; the knowledge base question-and-answer function calls upon the local knowledge base and lightweight metrology model for real-time question and answer, with options for voice or contact query. 5. On-site anomaly confirmation: The operator performs an on-site review based on the system prompts to confirm whether the anomaly is true; 6. One-click generation of anomaly list: The system automatically generates an anomaly list based on the confirmed anomaly information, including: anomaly type (e.g., abnormal heating), anomaly location (e.g., phase A wiring terminal), severity level (e.g., general / serious / urgent), multimodal evidence collected on site (anomaly images, thermal images, RFID information, environmental parameters, electrical data), timestamps, and geographical location information; 7. Communication module sends back to the main station: Sends a list of abnormal issues back to the main station system with one click via the network; 8. Automatic work order generation and dispatching by the main station: After receiving the list of anomalies, the main station system automatically generates maintenance work orders based on the anomaly type and severity level, and assigns them to the corresponding work teams. 9. Work Order Processing + Result Upload: After receiving a work order, team members go to the site to process it and upload the processing result through the human-computer interaction module, forming a complete business loop.

[0073] Accordingly, this application also discloses an anomaly identification device for a low-voltage power metering device, see [link to relevant documentation]. Figure 4 As shown, the device includes: The information acquisition module 11 is used to synchronously acquire multimodal information of the target low-voltage power metering device; the multimodal information includes: appearance image information for appearance anomaly identification, thermal imaging information for heating anomaly identification, asset identification information for asset information verification, environmental parameter information for environmental status assessment, and electrical parameter information for data quality analysis. The anomaly identification module 12 is used to perform spatiotemporal alignment processing on the multimodal information and perform fusion analysis based on the aligned multimodal information to generate anomaly identification results. The interaction module 13 is used to provide risk warnings for the anomaly identification results in a preset interaction mode, and to generate auxiliary judgment information based on the local knowledge base and the preset end-side reasoning model, so as to receive confirmation operations returned by the operator based on the anomaly identification results and the auxiliary judgment information. The automated closed-loop module 14 is used to automatically generate an abnormal problem list containing the abnormal identification results confirmed by the confirmation operation, and upload it to the background system so as to automatically generate a maintenance work order based on the abnormal problem list.

[0074] For more detailed information on the working process of each of the above modules, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.

[0075] Therefore, through the above-described solution in this embodiment, firstly, at the perception and judgment level, this invention constructs a multimodal information perception architecture, fundamentally solving the problems of scattered detection methods and single perception dimensions in existing technologies. Based on this, through spatiotemporal alignment and fusion analysis of multimodal information, precise time synchronization and spatial registration of multimodal information are achieved, enabling cross-verification of the authenticity of anomalies from multiple dimensions. This significantly improves the accuracy and reliability of anomaly identification, effectively avoiding false alarms or missed alarms from a single dimension. Secondly, at the interaction and efficiency level, the end-side inference model of this invention is a lightweight metrological large model, enabling equipment to possess local intelligent inference capabilities, eliminating dependence on cloud networks. Even in areas with poor signal, it can obtain instant question-and-answer and data analysis support, transforming expert experience into readily accessible digital capabilities on-site, significantly reducing the reliance on personnel experience in on-site operations. Simultaneously, the automatically generated anomaly identification results are output intuitively on-site, allowing operators to directly confirm them without the need for manual recording or returning to the office for system entry. Finally, the automated work order generation and dispatch process is triggered, seamlessly connecting on-site discoveries to back-end processing. This completely changes the cumbersome process of on-site recording, post-event entry, and manual dispatch in the traditional model, significantly shortening the anomaly handling cycle and improving the digitalization and automation level of operation and maintenance management. In summary, this invention constructs an intelligent closed-loop system covering the entire chain of "collection-fusion-analysis-interaction-handling," solving the inherent defects of fragmented perception capabilities, experience-dependent on-site decision-making, and fragmented business processes in the traditional inspection model. Furthermore, it upgrades the operation and maintenance of low-voltage metering devices from the traditional "manually dominated, segmented operation" model to a new digital model of "data-driven, human-machine collaboration, and closed-loop automation," providing a practical and feasible technical path for the efficient and lean management of massive low-voltage metering devices in the context of new power systems.

[0076] Furthermore, embodiments of this application also disclose an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0077] Figure 5 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the anomaly identification method for low-voltage power metering devices disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be a computer.

[0078] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0079] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored on it can include an operating system 221, computer programs 222, and data 223, etc. The data 223 can include various types of data. The storage method can be temporary storage or permanent storage.

[0080] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the anomaly identification method for a low-voltage power metering device executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0081] Furthermore, this application also discloses a computer-readable storage medium, which includes random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, magnetic disks, optical disks, or any other form of storage medium known in the art. The computer program, when executed by a processor, implements the aforementioned abnormal identification method for low-voltage power metering devices. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0082] Furthermore, embodiments of this application also provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements any of the above-described methods for identifying anomalies in low-voltage power metering devices.

[0083] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0084] The steps of the anomaly identification method or algorithm for low-voltage energy metering devices described in conjunction with the embodiments disclosed herein can be implemented directly using hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art.

[0085] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0086] The above provides a detailed description of the anomaly identification method, device, equipment, and medium of the low-voltage power metering device provided by the present invention. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only intended to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying anomalies in a low-voltage electricity metering device, characterized in that, include: Simultaneously acquire multi-modal information from the target low-voltage power metering device; The multimodal information includes: appearance image information for identifying appearance anomalies, thermal imaging information for identifying heating anomalies, asset identification information for asset information verification, environmental parameter information for environmental status assessment, and electrical parameter information for data quality analysis. The multimodal information is spatiotemporally aligned, and fusion analysis is performed based on the aligned multimodal information to generate anomaly identification results. The anomaly identification results are given a risk warning in a preset interaction mode, and auxiliary judgment information is generated based on the local knowledge base and the preset terminal inference model, so as to receive the confirmation operation returned by the operator based on the anomaly identification results and the auxiliary judgment information. An abnormal problem list containing the abnormal identification results confirmed by the aforementioned confirmation operation is automatically generated and uploaded to the backend system so that a repair work order can be automatically generated based on the abnormal problem list.

2. The anomaly identification method for low-voltage power metering devices according to claim 1, characterized in that, The synchronous acquisition of multimodal information from the target low-voltage power metering device includes: The camera is used to capture images of the meter readings, wiring terminals, and enclosure of the target low-voltage electricity metering device to obtain visual information. The heat-generating parts of the target low-voltage power metering device are scanned using a thermal imager to obtain thermal imaging information; The target low-voltage electricity metering device is scanned using a radio frequency identification (RFID) interactive device to obtain asset identification information. The environmental sensor module collects real-time information on the internal and / or surrounding environmental parameters of the target low-voltage power metering device. The electrical parameter information inside the target low-voltage power metering device is copied through the data copying interface; the electrical parameter information includes at least one of voltage, current, power, energy, or power factor.

3. The anomaly identification method for a low-voltage power metering device according to claim 1, characterized in that, The process of performing spatiotemporal alignment on the multimodal information includes: Using a linear interpolation algorithm, the time axes of the thermal imaging information, the asset identification information, the environmental parameter information, and the electrical parameter information are aligned to the video frame time axis of the appearance image information for time alignment. Common feature points are extracted from the appearance image information and the thermal imaging information to calculate the affine transformation matrix. Based on the affine transformation matrix, a pixel-level mapping relationship is established between the thermal imaging information and the appearance image information for spatial alignment.

4. The anomaly identification method for a low-voltage power metering device according to claim 2, characterized in that, Anomaly detection results are generated by fusion analysis based on aligned multimodal information, including: The appearance image information is analyzed by calling a deep learning-based target detection model to identify appearance defects, and the meter reading is extracted by optical character recognition and compared with the historical electrical parameter information to identify metering anomalies and obtain the first anomaly feature. Based on the thermal imaging information, the temperature difference between the heating part and the current ambient temperature is determined, and the heating anomaly is identified based on the difference. The temperature rise trend is predicted based on the difference and the load current measured through the image to identify potential overheating risks, thus obtaining a second abnormal feature. The asset identification information is compared with the asset ledger information stored in the local knowledge base to detect asset anomalies and obtain a third anomaly feature. The environmental parameter information is subjected to threshold judgment to obtain the fourth abnormal feature, and the electrical parameter information is analyzed to obtain the fifth abnormal feature; The first, second, third, fourth, and fifth abnormal features are subjected to multimodal cross-validation to comprehensively assess the authenticity and severity of the abnormal features, thereby generating the anomaly identification result.

5. The anomaly identification method for a low-voltage power metering device according to claim 1, characterized in that, The step of providing risk alerts for the anomaly identification results using a preset interaction mode includes: The thermal imaging information is overlaid with the appearance image information, and based on the anomaly identification result, the abnormal location is marked on the appearance image information using augmented reality technology, and the anomaly identification result is broadcast through voice synthesis; the abnormal location includes heated parts, damaged areas, or abnormal locations of assets.

6. The anomaly identification method for a low-voltage power metering device according to claim 1, characterized in that, The auxiliary judgment information generated based on the local knowledge base and the preset edge-side reasoning model includes: Receive natural language processing instructions input by operators; The system invokes a pre-defined edge-side inference model deployed on the field equipment after knowledge distillation, model pruning, and quantization compression, as well as a local knowledge base containing metrological technical specifications, typical fault cases, and anomaly handling procedures, to parse and respond to the natural language processing instructions in order to generate auxiliary judgment information. The auxiliary judgment information includes processing suggestions, professional knowledge answers, or data analysis results for the anomaly identification results.

7. The method for anomaly identification of a low-voltage power metering device according to any one of claims 1 to 6, characterized in that, The automatic generation of an anomaly list containing the anomaly identification results confirmed by the confirmation operation is uploaded to the backend system so that a repair work order can be automatically generated based on the anomaly list, including: In response to the confirmation operation, the system automatically packages the anomaly type, anomaly location, severity level, multimodal evidence collected on-site, timestamp, and geographic location information to generate an anomaly list. The list of abnormal issues is encrypted using an encryption algorithm and uploaded to the backend system via a preset communication network. The backend system then parses the list and matches it with a preset work order template to automatically generate a repair work order.

8. An anomaly identification device for a low-voltage electricity metering device, characterized in that, include: The information acquisition module is used to synchronously acquire multimodal information of the target low-voltage power metering device; The multimodal information includes: appearance image information for identifying appearance anomalies, thermal imaging information for identifying heating anomalies, asset identification information for asset information verification, environmental parameter information for environmental status assessment, and electrical parameter information for data quality analysis. An anomaly detection module is used to perform spatiotemporal alignment processing on the multimodal information and perform fusion analysis based on the aligned multimodal information to generate anomaly detection results; The interaction module is used to provide risk warnings for the anomaly identification results in a preset interaction mode, and to generate auxiliary judgment information based on the local knowledge base and the preset end-side reasoning model, so as to receive confirmation operations returned by the operators based on the anomaly identification results and the auxiliary judgment information. An automated closed-loop module is used to automatically generate a list of abnormal issues containing the abnormal identification results confirmed by the confirmation operation, and upload it to the backend system so that a maintenance work order can be automatically generated based on the list of abnormal issues.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the anomaly identification method for a low-voltage power metering device as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein the computer programs, when executed by a processor, implement the anomaly identification method for a low-voltage power metering device as described in any one of claims 1 to 7.

Images