Metering box remote monitoring method, system and equipment based on artificial intelligence and medium

Abstract

The invention relates to the technical field of power grids, in particular to a metering box remote monitoring method, system and device based on artificial intelligence and a medium. Acquiring electrical parameter data and environmental data of the metering box; performing time domain and frequency domain feature extraction on the data to obtain key features; inputting the processed data into a preset abnormal state detection model; identifying an abnormal state of the metering box based on a model output result; and when the abnormity is identified, alarm information is generated and sent to a monitoring terminal through a remote communication network. Multi-dimensional monitoring data are collected in real time through multiple sensors, feature extraction and anomaly detection are performed by applying an artificial intelligence technology, a model is continuously optimized in combination with an online learning mechanism, and comprehensive sensing of the running state of the metering box and accurate recognition of the abnormal state are achieved. Compared with a traditional monitoring system, the method improves the detection accuracy, reduces the false alarm rate and the missing report rate, improves the fault processing efficiency through the automatic generation of the maintenance suggestions, and provides a powerful guarantee for the operation of a power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power grids, and in particular to a remote monitoring method, system, device and medium for metering boxes based on artificial intelligence. Background Art

[0002] With the continuous advancement of the modernization of the power system and the construction of the smart grid, as a key metering and control device in the power system, the operation status monitoring of metering boxes has received increasing attention. Traditional metering box monitoring systems mainly rely on regular manual inspections or simple data collection terminals, and can only obtain basic power parameters and power-off status information, unable to achieve a comprehensive perception of the operation status of metering boxes and timely identification of abnormal conditions. In recent years, with the development of Internet of Things technology, sensor technology and communication technology, remote monitoring systems have been applied in metering box management, realizing a certain degree of remote data collection and status monitoring. However, most of these systems use fixed threshold judgment or simple rule reasoning methods for anomaly detection, which are difficult to adapt to complex and changing operating environments, and have limited ability to perform correlation analysis on multi-dimensional data, resulting in prominent problems of high false alarm rate and high missed alarm rate. At the same time, they are unable to achieve early warning of potential faults and accurate classification of abnormal states.

[0003] Currently, the remote monitoring system for metering boxes still faces many technical challenges in practical applications. First, the electrical parameters and environmental data of metering boxes have characteristics such as multi-source heterogeneity, high dimensionality, and time-series correlation, and traditional data processing methods are difficult to effectively extract the complex patterns and implicit relationships contained therein. Second, the abnormal states of metering boxes are diverse and complex, including not only obvious abnormalities such as overload and overvoltage, but also hidden abnormalities such as harmonic interference and three-phase imbalance. These abnormal states are often the result of the combined action of multiple factors and are difficult to accurately identify with simple thresholds or rules. Third, the operating environment and load characteristics of metering boxes will dynamically adjust over time and seasons, and static monitoring models cannot adapt to this change, resulting in a decrease in monitoring effectiveness over time. In addition, existing systems generally lack the ability of intelligent alarm and maintenance suggestions, and are unable to provide targeted treatment plans according to the type and severity of the abnormality, increasing the workload and response time of maintenance personnel. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to solve the technical problems of the existing remote monitoring system for metering boxes, such as low accuracy in detecting abnormal states, poor adaptability, and weak early warning ability. By introducing artificial intelligence technology, efficient feature extraction of multi-dimensional monitoring data of metering boxes and accurate identification of abnormal states are realized, and the detection ability of the system for abnormal conditions in complex environments is improved. At the same time, through the online learning mechanism, the monitoring system can adapt to the dynamic changes of the operating characteristics of metering boxes and continuously optimize the abnormal state detection model. In addition, the present invention also solves the deficiencies of the existing system in alarm handling and maintenance suggestions. By intelligently analyzing the types and severity of abnormal states, targeted maintenance suggestions are automatically generated to improve the efficiency and accuracy of fault handling of metering boxes.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: Obtain the first monitoring data of the first target metering box, where the first monitoring data includes metering box electrical parameter data and surrounding environment data; Extract features from the first monitoring data to obtain second monitoring data; Input the second monitoring data into a preset abnormal state detection model; Based on the output result of the abnormal state detection model, identify the abnormal state of the first target metering box; When an abnormal state is identified, generate an alarm message and send it to the monitoring terminal through a remote communication network.

[0007] In a first aspect, an embodiment of the present invention provides a remote monitoring method for a metering box based on artificial intelligence, which includes, As a preferred solution of the remote monitoring method for a metering box based on artificial intelligence according to the present invention, wherein: the first monitoring data includes: Current data, voltage data, power data, power factor data, temperature data, and vibration data; The first monitoring data is obtained by real-time collection through a plurality of sensors arranged inside and outside the metering box.

[0008] As a preferred solution of the remote monitoring method for a metering box based on artificial intelligence according to the present invention, wherein: the feature extraction includes: Extract features of the first monitoring data from two dimensions of time domain and frequency domain; Based on the importance score of the features, select key features to form the second monitoring data.

[0009] As a preferred solution of the remote monitoring method for a metering box based on artificial intelligence according to the present invention, wherein: the construction method of the abnormal state detection model includes: Collect normal operation data and various types of abnormal operation data of historical metering boxes to construct a training data set; Train an abnormal state detection model using the training data set. Use the validation data set to evaluate and optimize the performance of the trained abnormal state detection model.

[0010] As a preferred solution of the method for remote monitoring of a metering box based on artificial intelligence according to the present invention, wherein: the abnormal states include: Overload state, underload state, overvoltage state, undervoltage state, three-phase unbalance state, harmonic anomaly state, leakage anomaly state, and temperature anomaly state; The abnormal state detection model outputs probability values of various abnormal states and determines whether there is an abnormality according to a preset threshold.

[0011] As a preferred solution of the method for remote monitoring of a metering box based on artificial intelligence according to the present invention, wherein: it further includes: Based on historical abnormal state data and equipment maintenance records, perform online learning and update on the abnormal state detection model; The online learning and update are implemented through an incremental learning method to enable the model to adapt to changes in the operating environment and aging characteristics of the metering box.

[0012] As a preferred solution of the method for remote monitoring of a metering box based on artificial intelligence according to the present invention, wherein: based on the type and severity of the abnormal state, automatically generate maintenance suggestions; According to the set alarm level, select the corresponding notification method to send the alarm information and maintenance suggestions to relevant maintenance personnel and management systems.

[0013] In a second aspect, an embodiment of the present invention provides a remote monitoring system for a metering box based on artificial intelligence, which includes a data acquisition module for acquiring first monitoring data of a first target metering box, where the first monitoring data includes metering box electrical parameter data and surrounding environment data; A feature processing module for extracting features from the first monitoring data to obtain second monitoring data; An abnormal detection module for inputting the second monitoring data into a preset abnormal state detection model and identifying the abnormal state of the first target metering box based on the output result of the abnormal state detection model; An alarm processing module for generating an alarm information and sending it to a monitoring terminal through a remote communication network when an abnormal state is identified.

[0014] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and wherein: when the computer program instructions are executed by the processor, the steps of the method for remote monitoring of a metering box based on artificial intelligence as described in the first aspect of the present invention are implemented.

[0015] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program instructions are executed by a processor, the steps of the method for remote monitoring of a metering box based on artificial intelligence as described in the first aspect of the present invention are implemented.

[0016] The beneficial effects of the present invention are as follows: The method for remote monitoring of a metering box based on artificial intelligence provided by the present invention realizes a comprehensive perception of the operation status of the metering box by collecting multi-dimensional monitoring data in real time through a sensor network, overcoming the defects of low efficiency and narrow coverage of traditional manual inspection methods. Through the two-dimensional feature extraction and screening technology in the time domain and frequency domain, the key information contained in the data is effectively refined, solving the problem of low processing efficiency of traditional data processing methods for multi-source heterogeneous and high-dimensional data. By adopting a hierarchical anomaly detection model structure, accurate identification of various complex abnormal states is realized. Compared with the traditional fixed-threshold judgment method, the detection accuracy is increased by about 10 percentage points, and the false alarm rate is reduced by nearly 5 percentage points. By introducing an environmental adaptability threshold mechanism and an online learning and updating technology, the system can automatically adapt to the changes in the operation environment of the metering box and the characteristics of equipment aging, effectively solving the problem of performance degradation of traditional static monitoring models over time. The intelligent alarm and maintenance recommendation generation mechanism based on the type and severity of anomalies reduces the diagnosis time of maintenance personnel, improves the efficiency of fault handling, and shortens the average fault response time by about 30%. Overall, the present invention realizes the intelligence and self-adaptability of remote monitoring of metering boxes, significantly improves the safety and reliability of the power system, reduces the operation and maintenance costs, and provides a strong guarantee for the stable operation of the power system. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0018] Figure 1 It is a flowchart of the method for remote monitoring of a metering box based on artificial intelligence; Figure 2 It is a computer device diagram of the method for remote monitoring of a metering box based on artificial intelligence. Detailed Embodiments

[0019] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be made in conjunction with the drawings in the specification.

[0020] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0021] Secondly, the so-called "one embodiment" or "embodiment" refers to specific features, structures or characteristics that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or selectively exclusive embodiments from other embodiments.

[0022] Embodiment 1

[0023] Referring to Figures 1 to 2 , which is the first embodiment of the present invention. This embodiment provides a remote monitoring method for a metering box based on artificial intelligence, including: S100: Obtain the first monitoring data of the first target metering box. The first monitoring data includes metering box electrical parameter data and surrounding environment data; S200: Extract features from the first monitoring data to obtain the second monitoring data; S300: Input the second monitoring data into a preset abnormal state detection model; S400: Identify the abnormal state of the first target metering box based on the output result of the abnormal state detection model; S500: When an abnormal state is identified, generate an alarm message and send it to the monitoring terminal through a remote communication network.

[0024] It should be noted that as a key electrical equipment in the power system, the metering box faces many challenges in actual operation. First, the traditional metering box monitoring system relies on regular manual inspections or simple data acquisition terminals, and cannot capture diverse abnormal states in real time. The data of operation parameters of the metering box, such as current, voltage, power, etc., have the characteristics of multi-source heterogeneity, high dimensionality, and real-time change. It is difficult for traditional monitoring methods to effectively extract the complex patterns and fault characteristics hidden therein. Second, the abnormal states of the metering box show diversity and complexity, including various forms such as overload, overvoltage, undervoltage, three-phase imbalance, and harmonic abnormality. These abnormalities are often the result of the coupling of multiple factors, and it is difficult to accurately identify them using simple threshold judgment methods. Moreover, the load characteristics of the metering box change dynamically with time, season, and environmental conditions. The static monitoring model cannot adapt to this change, resulting in a decrease in the monitoring effect over time. In addition, the existing system lacks the ability of intelligent alarm and maintenance suggestions, and cannot provide targeted treatment solutions according to the type and severity of the abnormality, increasing the workload and response time of maintenance personnel. Therefore, through the steps of S100 - S500, realizing remote monitoring of the metering box based on artificial intelligence can effectively solve the above problems.

[0025] Through the technical solution described in Claim 1, a remote monitoring method for a metering box based on artificial intelligence is realized. This method obtains the electrical parameter data and environmental data of the metering box, extracts features from the data, inputs the processed data into a preset abnormal state detection model, identifies the abnormal state of the metering box, and generates an alarm message when an abnormality is identified and sends it to the monitoring terminal through a remote communication network. This technical solution can achieve efficient feature extraction of multi-dimensional monitoring data of the metering box, capture data features from both the time domain and the frequency domain, and accurately identify abnormal conditions in a complex environment through an artificial intelligence model. At the same time, through the online learning and model update mechanism, the monitoring system can adapt to the dynamic changes of the operation characteristics of the metering box and continuously optimize the abnormal state detection model. In addition, this solution can automatically generate targeted maintenance suggestions according to the type and severity of the abnormal state, and send alarm messages in a timely manner through the remote communication network, greatly improving the efficiency and accuracy of metering box fault handling. This technical solution completely solves the deficiencies of the traditional metering box monitoring system in terms of real-time performance, accuracy, self-adaptability, and intelligence, providing a strong guarantee for the safe and stable operation of the power system.

[0026] Embodiment 2

[0027] Refer to Figure 1 - Figure 2 , which is the second embodiment of the present invention.

[0028] In the embodiment of the present application, in step S100, the first monitoring data of the first target metering box is obtained. The first monitoring data includes metering box electrical parameter data and ambient environment data, and includes the following steps A1 - A2: A1: The first monitoring data includes: Current data, voltage data, power data, power factor data, temperature data, and vibration data; The first monitoring data is obtained by real - time acquisition through multiple sensors arranged inside and outside the metering box.

[0029] Specifically, in A1, the current data is the three - phase current value collected by a Hall current sensor, with a sampling frequency of 100Hz, a measurable range of 0 - 200A, and an accuracy of ±0.5%; the voltage data is the three - phase voltage value collected by a voltage transformer, with a sampling frequency of 100Hz, a measurable range of 0 - 400V, and an accuracy of ±0.2%; the power data includes active power and reactive power, which is calculated from the current and voltage data; the power factor data reflects the load characteristics and is calculated from the ratio of active power to apparent power; the temperature data measures the temperature of key parts of the metering box, such as contact points and busbars, through a thermocouple or an infrared sensor, with a sampling frequency of 0.2Hz, a measurement range of - 20°C to 120°C, and an accuracy of ±1°C; the vibration data measures the vibration condition of the metering box through an acceleration sensor, with a sampling frequency of 200Hz, a measurement range of 0 - 10g, and is used to detect problems such as mechanical looseness and poor contact.

[0030] In an alternative embodiment, the first monitoring data may further include humidity data, which is collected by a humidity sensor for the internal environment humidity of the metering box, with a sampling frequency of 0.1Hz, a measurement range of 0 - 100%RH, and an accuracy of ±3%RH, and is used to evaluate the impact of environmental conditions on the insulation performance of the metering box.

[0031] In another alternative embodiment, the first monitoring data may further include power quality parameters, such as harmonic content, voltage fluctuation, and flicker data, which are collected by a power quality analyzer, with a sampling frequency of 1kHz, and are used to evaluate the impact of grid quality on the operation of the metering box.

[0032] A2: Sensor layout scheme and data acquisition system The internal sensor layout of the metering box adopts a hierarchical and zonal strategy, and a variety of sensors are installed at key positions such as incoming terminals, outgoing terminals, circuit breakers, current transformers, and energy meters. External sensors mainly include environmental temperature and humidity sensors and vibration sensors, which are installed at appropriate positions on the metering box shell. All sensors are connected through a data acquisition unit (DAU). The DAU is equipped with a processor and a communication module, which can perform preliminary filtering and compression on the collected raw data and transmit the data to the cloud platform or edge computing node in real time through wireless communication methods such as 4G / 5G, NB-IoT, or LoRa.

[0033] Exemplarily, the above-mentioned sensor systems are installed in 10 metering boxes in a substation area. Each metering box is equipped with 12 - 15 sensors to form a sensor network. Each sensor node has a unique identifier to facilitate the system to identify the data source. The data acquisition frequency is dynamically adjusted according to the importance and change characteristics of parameters. For example, key parameters such as current and voltage adopt a higher sampling rate, and slow-changing parameters such as temperature adopt a lower sampling rate to balance the monitoring accuracy and system resource occupancy.

[0034] In the embodiment of the present application, in step S200, feature extraction is performed on the first monitoring data to obtain the second monitoring data, including the following steps B1: B1: Feature extraction includes: Extract the features of the first monitoring data from two dimensions of time domain and frequency domain; Based on the importance scoring of the features, select key features to form the second monitoring data.

[0035] Specifically, in B1, the time domain feature extraction mainly focuses on the statistical characteristics and change patterns of electrical parameter data and environmental data in the time dimension. For current, voltage, and power data, the extracted time domain features include statistics such as mean, standard deviation, peak value, valley value, peak value, skewness, kurtosis, zero crossing rate, quartile, and maximum change rate; for temperature and vibration data, the extracted time domain features include indicators such as average temperature, temperature gradient, temperature rise rate, vibration amplitude, and vibration energy. In addition, the correlation coefficient matrix between each parameter is also calculated to capture the coupling relationship between multiple parameters, such as the correlation between power and temperature rise, and the association between load change and vibration characteristics.

[0036] The frequency domain feature extraction mainly uses methods such as fast Fourier transform (FFT), wavelet transform, and Hilbert-Huang transform to convert the time domain signal to the frequency domain or time-frequency domain for analysis. Perform harmonic analysis on current and voltage signals to extract fundamental wave components, the content of each harmonic, and the total harmonic distortion rate (THD); perform spectrum analysis on vibration signals to identify characteristic frequency components and their amplitudes for detecting abnormal vibration modes; perform time-frequency analysis on temperature signals to extract their periodic change characteristics and non-stationary characteristics.

[0037] It should be noted that the feature extraction process adopts an adaptive window technology, which dynamically adjusts the length of the analysis window according to the changing characteristics of the signal. A shorter window is used for rapidly changing signals, and a longer window is used for slowly changing signals to balance the accuracy and real-time performance of feature extraction. At the same time, different preprocessing methods are applied to different types of data. For example, band-pass filtering is performed on current and voltage data to eliminate high-frequency noise, and low-pass filtering is performed on temperature data to eliminate sudden interference.

[0038] In an alternative embodiment, the feature extraction can also adopt nonlinear dynamics methods, such as phase space reconstruction, Lyapunov exponent calculation, etc., to extract the chaotic characteristics and complexity indicators of the signal, which are used to capture the nonlinear dynamic behavior of the system and have unique advantages for identifying certain complex fault modes.

[0039] In another alternative embodiment, the feature extraction can also be combined with knowledge graph technology to encode the experience of domain experts into a rule base to guide the feature extraction process. For example, the feature extraction parameters are automatically adjusted according to the specific device model and operating conditions, or specific types of feature patterns are targeted according to historical fault cases.

[0040] In the embodiment of the present application, in step S300, inputting the second monitoring data into a preset abnormal state detection model includes the following steps C1: C1: The construction method of the abnormal state detection model includes: Collect the normal operation data and various types of abnormal operation data of the historical metering box to construct a training data set; Use the training data set to train the abnormal state detection model; Use the validation data set to evaluate and optimize the performance of the trained abnormal state detection model.

[0041] Specifically, in C1, the historical data collection includes two types of data: one is the actual operation data, which is the real operation records obtained from the on-site metering box through the monitoring system, covering the normal operation data under different seasons and different load conditions, as well as various types of fault data that actually occurred; the other is the simulation data, which is various types of abnormal condition data generated by electrical simulation software and physical test platforms, making up for the deficiency of actual fault data, especially for high-risk or low-probability fault types.

[0042] The data annotation adopts a multi-level annotation strategy: the first level is the normal / abnormal binary classification annotation; the second level is the specific abnormal type annotation, such as overload, overvoltage, undervoltage, etc.; the third level is the fault severity annotation, which is divided into three levels: minor, medium, and severe. The annotation process combines automated tools and expert review to ensure the annotation quality. The training data set is divided into a training set and a validation set according to a ratio of 7:3, and at the same time, ensure that the distribution ratios of various abnormalities in the two data sets are similar.

[0043] The abnormal state detection model adopts a hierarchical design, including an anomaly detection layer and an anomaly classification layer. The anomaly detection layer mainly solves the binary classification problem of normal / abnormal, and uses unsupervised or semi-supervised learning methods such as Isolation Forest, One-Class SVM, and Autoencoder, which can effectively handle the problem of data imbalance; the anomaly classification layer is responsible for further identifying specific anomaly types, and uses supervised learning methods such as Random Forest, Gradient Boosting Decision Tree (GBDT), and Deep Neural Network (DNN), and is trained using labeled abnormal data.

[0044] The model training adopts a cross-validation method, and selects appropriate hyperparameter optimization strategies for different algorithms, such as grid search, random search, or Bayesian optimization. To improve the robustness of the model, data augmentation techniques are also adopted, and training samples are expanded by adding Gaussian noise, amplitude modulation, and time shift.

[0045] It should be noted that the model performance evaluation adopts a multi-index comprehensive evaluation system, including conventional indexes such as Accuracy, Precision, Recall, F1 score, and AUC value, as well as specific indexes such as fault detection delay time, false alarm rate, and missed alarm rate. For different types of anomalies, differentiated evaluation index weights are set. For example, for high-risk fault types (such as severe overload), more attention is paid to Recall; for frequently occurring minor anomalies, more attention is paid to Precision to reduce false alarms.

[0046] In an optional implementation, the abnormal state detection model can also adopt an ensemble learning method, such as Stacking or Blending, to combine the prediction results of multiple base models to form a more robust and accurate ensemble model. The base models can include statistical models, traditional machine learning models, and deep learning models, making full use of the advantages of different methods to improve the overall detection performance.

[0047] In another optional implementation, the abnormal state detection model can also introduce an attention mechanism, enabling the model to automatically focus on key time periods and key features, and improving the ability to capture complex time series patterns. This method is particularly suitable for processing the fusion analysis of long sequence data and multi-source heterogeneous data.

[0048] Exemplarily, the abnormal detection system of the metering box in a regional substation adopts the above model architecture. After 3 months of data collection, about 50,000 normal operation records and 8,000 various abnormal records are accumulated. After the model training is completed, the overall detection accuracy rate on the independent test set reaches 95.8%, the average detection delay time is less than 5 minutes, the false alarm rate is controlled below 2%, and the missed alarm rate is less than 1%, meeting the actual operation and maintenance requirements.

[0049] In the embodiment of the present application, in step S400, based on the output result of the abnormal state detection model, identifying the abnormal state of the first target metering box includes the following steps D1 - D2: D1: The abnormal states include: Overload state, underload state, overvoltage state, undervoltage state, three - phase unbalance state, harmonic anomaly state, leakage anomaly state, and temperature anomaly state; The abnormal state detection model outputs the probability values of various abnormal states, and determines whether there is an abnormality according to a preset threshold.

[0050] Specifically, in D1, the abnormal state detection model outputs the probability value for each abnormal state, indicating the possibility that the current metering box is in this abnormal state. The overload state means that the current of the metering box exceeds the critical value of the rated capacity. The set thresholds are 1.1 times, 1.2 times, and 1.5 times the rated current, corresponding to mild, moderate, and severe overload respectively; the underload state means that the metering box operates at a low load for a long time, and the threshold is set to 0.1 times the rated current; the overvoltage state means that the voltage exceeds 1.1 times the rated voltage; the undervoltage state means that the voltage is lower than 0.9 times the rated voltage; the three - phase unbalance state means that the unbalance degree of the three - phase current or voltage exceeds 5%; the harmonic anomaly state means that the total harmonic distortion rate (THD) exceeds the limit value specified in the national standard GB / T 14549 - 93; the leakage anomaly state means that the leakage current exceeds 30 mA; the temperature anomaly state means that the temperature of the key part exceeds the safety threshold, such as the contact point temperature exceeds 65 °C.

[0051] The probability values output by the model are normalized by the softmax function to ensure that the sum of the probabilities of all abnormal states is 1. The system determines whether there is an abnormality according to a preset probability threshold (usually 0.6 - 0.8), and at the same time considers the possibility of multiple abnormalities occurring simultaneously, such as overload often accompanied by temperature anomaly. To reduce the false alarm rate, the system also implements a time - persistence check, requiring that the abnormal state persists within a certain time window (such as 30 seconds to 5 minutes, varying according to the abnormal type) to trigger an alarm.

[0052] It should be noted that the influence of environmental factors and equipment characteristics is considered in the process of calculating the probability of abnormal states. For example, for abnormal temperature judgment, the model will automatically adjust the threshold according to the environmental temperature, appropriately increasing the threshold in summer with high temperature and appropriately decreasing the threshold in winter with low temperature; for abnormal judgment related to load, the model will consider the daily load curve and seasonal variations, appropriately relaxing the overload judgment criteria during peak electricity consumption periods and increasing the sensitivity during off-peak electricity consumption periods. This adaptive threshold mechanism greatly improves the accuracy and practicality of abnormal detection.

[0053] In an alternative implementation, abnormal state recognition can also be based on time series pattern matching. By comparing the current time series features with a pre-defined abnormal pattern library, complex dynamic abnormal patterns can be identified. The pattern library contains time series feature templates for various typical faults, and algorithms such as dynamic time warping (DTW) are used to calculate the similarity to achieve precise matching of abnormal patterns.

[0054] In another alternative implementation, abnormal state recognition can also adopt a hybrid method that combines a rule engine and a machine learning model, encoding expert experience into a rule set to complement the data-driven model. The rule engine processes clearly defined abnormal situations, and the machine learning model is responsible for handling complex and new abnormal patterns. The two work together to improve the reliability and interpretability of detection.

[0055] D2: Abnormal state severity assessment and priority assignment The system conducts a severity assessment of the identified abnormal states, mainly considering the following factors: Degree of deviation from the normal value: such as the ratio of the overload current to the rated current; Duration: The duration of the abnormal state; Rate of change: The speed and acceleration of parameter changes; Scope of influence: The range of equipment and the number of users that the abnormality may affect; Safety risk: The level of potential safety hazards; Based on the severity assessment results, the system automatically assigns alarm priorities, usually divided into four levels: Critical, High, Medium, and Low. The priority determines the urgency of alarm handling and the resource allocation strategy. Critical-level alarms require immediate response, high-level alarms need to be processed within 2 hours, medium-level alarms need to be processed within 24 hours, and low-level alarms can be processed during routine maintenance.

[0056] Exemplarily, a certain metering box monitoring system detected 57 abnormal states in a week, including 15 overloads, 12 three-phase imbalances, 10 harmonic anomalies, 8 temperature anomalies, and 12 other types. After severity assessment, 3 were assigned as emergency level, 12 as high level, 25 as medium level, and 17 as low level. Maintenance personnel processed according to the priority order, effectively improving the resource utilization efficiency.

[0057] In the embodiment of the present application, when an abnormal state is identified in step S500, an alarm message is generated and sent to the monitoring terminal through a remote communication network, including the following steps E1 - E7: E1: It also includes: Based on historical abnormal state data and equipment maintenance records, online learning and updating are performed on the abnormal state detection model; Online learning and updating are achieved through incremental learning methods, enabling the model to adapt to changes in the operating environment and aging characteristics of the metering box.

[0058] Specifically, in E1, the system continuously updates the abnormal state detection model using online learning methods based on historical abnormal state data and equipment maintenance records. The online learning process mainly includes three links: data incremental processing, parameter update, and model evaluation.

[0059] In the data incremental processing stage, the system adds newly collected labeled data (alarms confirmed by humans) to the historical database, constructs a sliding window data set, and retains data for the most recent period (such as 3 - 6 months) for model update. For particularly valuable cases (such as rare faults or misreporting and misjudgment cases), the system will mark them as "key samples" and retain them for a long time.

[0060] In the parameter update stage, different incremental learning strategies are adopted according to the model type. For tree-based models (such as random forest, GBDT), an incremental tree strategy is adopted to retain the original tree structure and add new decision trees; for neural network models, a transfer learning method is adopted to freeze the parameters of some network layers and only update the parameters of the output layer or the last few layers; for linear models, an online gradient descent method is used to directly update the parameters. The update frequency is dynamically adjusted according to the data accumulation situation, usually once a week or once a month.

[0061] In the model evaluation stage, the system uses the validation data set to evaluate the performance of the updated model and compares it with the original model. Only when the new model has a significant improvement in key indicators and passes the stability test will it officially replace the model running online. At the same time, the system retains the model version history to support model rollback operations.

[0062] It should be noted that during the online learning process, special attention is paid to the ability of the model to adapt to the changes in the operating environment and aging characteristics of the metering box. Environmental changes include seasonal changes, load pattern changes, and power grid quality changes, etc.; aging characteristics include increased contact resistance, decreased insulation performance, and component parameter drift, etc. The system automatically adjusts the feature weights and model parameters by comparing the data distribution changes in different periods to ensure that the model maintains high accuracy throughout the entire life cycle of the equipment.

[0063] In an alternative embodiment, online learning can also adopt an Active Learning strategy. The system actively identifies the most informative uncertain samples and requests experts to annotate them to obtain the maximum model improvement effect at the minimum annotation cost. This method is particularly suitable for scenarios where abnormal samples are scarce.

[0064] In another alternative embodiment, online learning can also be combined with the Transfer Learning method to utilize the knowledge from similar types of devices to accelerate the model training process for newly deployed devices. Through domain adaptation technology, the system can transfer the model knowledge trained on a large number of general metering boxes to metering boxes of specific types or in specific environments, significantly shortening the model convergence time.

[0065] E2: Automatically generate maintenance suggestions based on the type and severity of the abnormal state; According to the set alarm level, select the corresponding notification method to send the alarm information and maintenance suggestions to the relevant maintenance personnel and management system.

[0066] Specifically, in E2, the system automatically generates targeted maintenance suggestions based on the identified abnormal state type, severity, and historical maintenance experience. The maintenance suggestion generation process includes three links: fault diagnosis, cause analysis, and treatment plan recommendation.

[0067] In the fault diagnosis link, the system matches the detected abnormal state with the fault patterns in the knowledge base to determine the possible fault types. The knowledge base contains the characteristic patterns, discrimination rules, and case bases of various typical faults, which are continuously enriched and optimized as the system runs. For complex faults, the system may give multiple possible diagnostic results and mark the probability values.

[0068] In the cause analysis link, the system analyzes the potential causes of the fault based on the causal relationship model. The causal relationship model describes the correlation between various parameters, environmental factors, and faults through a Bayesian network or decision tree structure. For example, for an overload fault, the possible causes include increased electrical equipment, short circuit faults, meter measurement errors, etc.; for abnormal temperature, the possible causes include poor contact, insufficient environmental heat dissipation, overload operation, etc.

[0069] In the processing solution recommendation stage, the system selects the most suitable processing solution from the solution library based on the fault type, cause analysis, and equipment conditions. The processing solutions include three levels: emergency measures, temporary measures, and fundamental solutions. The system will consider factors such as maintenance resource availability, maintenance difficulty, and cost-effectiveness to recommend the optimal processing path.

[0070] It should be noted that the generation process of maintenance suggestions integrates two methods: rule-based reasoning and case-based reasoning. Rule-based reasoning is based on expert knowledge and industry standards and is applicable to dealing with clearly defined fault scenarios; case-based reasoning retrieves similar cases from the historical maintenance case library and is applicable to dealing with new or complex fault scenarios. The system continuously optimizes the accuracy and practicality of maintenance suggestions by continuously learning historical maintenance records.

[0071] In an alternative implementation, the generation of maintenance suggestions can also be combined with predictive analysis, which not only provides a processing solution for the current fault but also predicts potential development trends and gives preventive suggestions. For example, by analyzing the current growth trend, it predicts the time point when overload may occur in the future and recommends capacity expansion or load adjustment in advance.

[0072] E3: According to the set alarm level, select the corresponding notification method to send the alarm information and maintenance suggestions to relevant maintenance personnel and management systems.

[0073] Specifically, in E3, the system selects different notification methods and recipients according to the alarm level. The alarm level is usually divided into four levels: Critical, High, Medium, and Low. Critical-level alarms are pushed simultaneously through multiple channels, including SMS, phone calls, mobile application pushes, and emails, to ensure that maintenance personnel can receive the notification in the first place; High-level alarms are pushed via SMS and mobile application pushes; Medium-level alarms are mainly pushed through mobile applications; Low-level alarms are sent via email or internal system messages.

[0074] The alarm recipients are determined according to the "Responsibility Assignment Matrix" (RACI Matrix), including four types of roles: Responsible, Accountable, Consulted, and Informed. Critical-level alarms will notify both the direct responsible person and the approver; High-level alarms notify the direct responsible person and carbon copy the approver; Medium-level alarms only notify the direct responsible person; Low-level alarms are summarized according to the equipment maintenance plan and notified regularly. The system also supports an escalation mechanism. When the direct responsible person fails to confirm the alarm within the specified time, the system will automatically escalate the alarm to a higher management level.

[0075] The content and format of the alarm information are optimized according to the receiving method and the recipient. SMS and phone notifications adopt a concise format, only containing the most critical information: device ID, fault type, severity level, and location; mobile application push and email adopt a detailed format, including complete alarm information, fault diagnosis, maintenance suggestions, historical records, and relevant resource links. The system also supports information localization, automatically converting languages, time zones, and measurement units according to user settings.

[0076] It should be noted that the alarm notification system has intelligent suppression and aggregation functions to avoid disturbing maintenance personnel caused by "alarm storms". Intelligent suppression means automatically suppressing certain types of alarms under specific conditions (such as known power outages, during planned maintenance); alarm aggregation means merging multiple alarms of the same device or related devices within a short period into a comprehensive alarm to reduce duplicate notifications. The system also implements alarm lifecycle management, tracking the whole process of each alarm from issuance to confirmation, handling, and closing to ensure that all alarms are properly handled.

[0077] In an alternative implementation, alarm notifications can also be intelligently allocated based on geographical location and workload. The system determines the current location of the maintenance personnel through GPS positioning or base station positioning, and preferentially allocates alarms to the personnel who are the closest and have a reasonable workload, reducing response time and traffic costs.

[0078] In another alternative implementation, alarm notifications can also be combined with augmented reality (AR) technology to directly display visual information such as device location, fault points, and repair guidance in the mobile application, helping maintenance personnel to locate and handle problems more quickly and accurately, especially suitable for fault handling in complex environments.

[0079] E4: Alarm Information Processing and Feedback Mechanism After the alarm information is sent out, the system will start processing process tracking, mainly including the following links: Alarm Confirmation: After receiving the alarm, the maintenance personnel need to confirm the received alarm through the system within the specified time (15 minutes for emergency level, 1 hour for high level, 4 hours for medium level, 24 hours for low level).

[0080] Status Update: During the on-site handling process, the maintenance personnel use the mobile terminal to update the handling status in real-time, including status such as arriving at the scene, fault confirmation, in handling, temporarily resolved, and finally resolved.

[0081] Result Feedback: After the fault handling is completed, the maintenance personnel need to fill in the handling result report, including information such as the actual fault cause, handling method, materials used, and consumption time.

[0082] Knowledge Deposition: The system automatically associates the handling result with the initial alarm information to form a complete case and incorporates it into the knowledge base for reference in future similar faults.

[0083] Quality assessment: Based on processing time, resolution effect and user satisfaction, the system automatically calculates the processing quality score as the basis for personnel performance appraisal.

[0084] This closed-loop alarm processing process ensures that each alarm is handled promptly and effectively, while accumulating a large amount of actual case data, providing a basis for model optimization and knowledge base enrichment.

[0085] E5: Alarm Visualization and Analysis The system provides multi-dimensional alarm visualization and analysis functions to help managers understand equipment health status and maintenance efficiency: Alarm distribution map: Displays the alarm distribution in different areas and time periods in the form of a heat map.

[0086] Alarm trend analysis: Displays the changing trends of various alarms over time, identifying periodic patterns and abnormal growth points.

[0087] Root cause analysis: Cluster related alarms to discover potential common root causes, such as power grid quality issues or environmental factors.

[0088] Maintenance efficiency analysis: Statistics on the response time, processing time and complete resolution rate of various alarms are collected to evaluate the efficiency of the maintenance team.

[0089] Predictive analysis: Predict future alarm trends based on historical data to provide a basis for resource planning and preventive maintenance.

[0090] For example, a power supply company uses the system to monitor 2,000 meter boxes in its jurisdiction, generating an average of about 3,000 alarms per month. Through multi-level screening and aggregation, about 500 alarms ultimately require on-site processing. The average response time of the maintenance team is 45 minutes, the processing completion time is 3.5 hours, and the alarm resolution rate reaches 98.5%. The accuracy of the maintenance suggestions generated by the system reaches 92%, effectively reducing the diagnosis time and repair rate of maintenance personnel.

[0091] In the implementation manner of the present application, step S500 also includes: E6: Based on historical abnormal state data and equipment maintenance records, the abnormal state detection model is learned and updated online; Online learning and updating are achieved through incremental learning, which enables the model to adapt to the changes and aging characteristics of the meter box operating environment.

[0092] Specifically, in E6, the system uses incremental learning to update the anomaly detection model online to adapt to the changes in the operating environment and aging characteristics of the metering box. Incremental learning is a learning method that continuously integrates new data information while retaining the learned knowledge, avoiding the high computational cost of traditional batch learning that requires retraining the entire model.

[0093] The incremental learning process first conducts quality assessment and preprocessing on the newly collected data to ensure the accuracy and representativeness of the data. Then, corresponding incremental update strategies are selected according to the model type: for statistical-based models such as Gaussian mixture models, directly update the statistical parameters; for instance-based models such as k-nearest neighbor algorithms, expand the instance library; for neural network-based models, use Progressive Learning or Knowledge Distillation methods to update the parameters.

[0094] The model update frequency is dynamically adjusted according to the data accumulation speed and the environmental change rate. For newly installed metering boxes, a higher update frequency (such as once a week) is adopted in the early stage; for stable-running devices, the update frequency is reduced (such as once a month or once a quarter). The system will also trigger special updates when a significant performance decline or a significant environmental change is detected.

[0095] It should be noted that the online learning process pays special attention to two types of characteristic changes in the metering box: environmental changes and aging characteristics. Environmental changes mainly include seasonal changes, changes in electricity consumption patterns, and changes in surrounding environmental conditions, etc.; aging characteristics are mainly manifested as the slow drift of device parameters over time, such as an increase in contact resistance and a decrease in insulation performance. The system establishes an evolution model of device parameters over time by tracking long-term data trends, predicts future change trends, and adjusts the reference benchmark for anomaly detection accordingly.

[0096] In an alternative implementation, online learning can also adopt the Federated Learning framework to achieve collaborative model updates among multiple devices while protecting data privacy. After the local models of different metering boxes are trained on edge devices, only the model parameters rather than the original data are uploaded to the central server. The server integrates the parameter updates of each node and distributes the global model to achieve a balance between knowledge sharing and privacy protection.

[0097] In another alternative implementation, online learning can also combine the Meta-learning method to enable the model to learn how to learn new knowledge more efficiently. The system extracts more abstract pattern representations and learning strategies through multi-task learning and cross-scenario training, enabling the model to quickly adapt to new environments and new device characteristics with the least number of samples, which is especially suitable for large-scale monitoring systems with diverse device types and complex deployment environments.

[0098] Exemplarily, a certain power grid company deployed 50 metering box monitoring devices in a total of 5 substation areas, and the initial model was trained with simulated data in a laboratory environment. After deployment, through three months of incremental learning, the average detection accuracy of the model for various anomalies increased from the initial 86% to 94.5%, and the false alarm rate decreased from 8% to 3.2%, effectively adapting to the on-site environmental conditions. Especially during the seasonal transition period (such as from summer to autumn), the system quickly adjusted the determination threshold for temperature anomalies through incremental learning, avoiding a large number of seasonal false alarms.

[0099] E7: Automatically generate maintenance suggestions based on the type and severity of the abnormal state; According to the set alarm level, select the corresponding notification method to send the alarm information and maintenance suggestions to the relevant maintenance personnel and management system.

[0100] Specifically, in E7, the system automatically generates targeted maintenance suggestions based on the identified abnormal state type and severity, combined with equipment history, environmental conditions, and maintenance experience. The maintenance suggestion generation process includes three main links: fault analysis, solution selection, and task planning.

[0101] In the fault analysis stage, the system comprehensively considers abnormal parameters, change trends, related parameter changes, and historical fault cases to infer the most likely fault causes. For a clear single fault, a deterministic diagnosis is directly given; for complex or ambiguous situations, multiple possible fault hypotheses and their probabilities are given, and further detection items are recommended for confirmation.

[0102] In the solution selection stage, the system screens applicable processing methods from the solution library and sorts them according to the following factors: solution efficiency (the possibility of completely solving the problem), implementation difficulty (required skills and tools), required time, cost, and potential risks. The solutions are divided into two categories: temporary measures and fundamental solutions. For emergencies, the system will recommend both rapid temporary measures and subsequent complete solutions.

[0103] In the task planning stage, the system converts the selected solution into specific work tasks, including a list of required tools and materials, operation steps, safety precautions, and quality inspection standards. For complex tasks, the system will generate a structured work flow chart, clearly showing the dependency relationships and critical paths between the steps.

[0104] It should be noted that the maintenance recommendation generation mechanism adopts a hybrid method that combines a knowledge base and data-driven approach. The knowledge base part contains expert experience and standard operating procedures, which are applicable to handling typical faults; the data-driven part is based on historical maintenance records and successful cases, and extracts best practices through machine learning methods, which are applicable to handling new or atypical faults. The system continuously optimizes the accuracy and practicality of maintenance recommendations by learning the maintenance feedback results, forming a closed-loop improvement mechanism.

[0105] In an alternative embodiment, the maintenance recommendation can also be combined with equipment condition assessment and remaining life prediction to provide long-term planning recommendations for equipment renewal and preventive maintenance. The system analyzes the historical data and current state of the equipment, evaluates the deterioration degree of key components, predicts the occurrence time and probability of potential faults, and reasonably arranges the preventive maintenance plan to achieve the transformation from passive response to active prevention.

[0106] In another alternative embodiment, the maintenance recommendation can also adopt a collaborative optimization method, comprehensively considering the maintenance requirements of multiple devices, resource constraints, and cost-benefit, to generate an optimal overall maintenance strategy. The system solves the multi-objective optimization problem through operations research methods (such as genetic algorithms, ant colony algorithms), balances factors such as maintenance urgency, distance, personnel skill matching degree, and equipment downtime cost, and formulates the best maintenance scheduling plan under limited resources.

[0107] Exemplarily, an abnormal three-phase imbalance (the current of phase C is significantly higher than that of phases A and B) accompanied by an increase in the temperature of phase C is detected in a metering box within a substation area. The maintenance recommendations generated by the system include: ① Preliminary diagnosis: It may be that the load of phase C is too large or the contact point is loose; ② On-site verification: Check the load distribution of each phase, measure the temperature and contact resistance of the contact point; ③ If the load is uneven, it is recommended to re-distribute the three-phase load; ④ If it is a contact point problem, it is recommended to immediately clean the contact surface and re-tighten it, and at the same time check whether the contact parts are burned and need to be replaced; ⑤ Verification after maintenance: Measure the three-phase current and temperature to confirm whether the problem is solved. After the maintenance personnel processed according to the recommendations, it was found that the problem was indeed caused by the loose incoming line terminal of phase C, and the fault was completely eliminated after processing.

[0108] In summary, the embodiment of the present invention realizes the intelligence and self-adaptability of remote monitoring of the metering box by obtaining the electrical parameters and environmental data of the metering box, extracting multi-dimensional features, using an artificial intelligence model to identify abnormal states, and combining online learning and expert knowledge to generate maintenance recommendations. This method can effectively improve the safety and reliability of the power system, reduce equipment failures and unplanned power outages, and at the same time optimize the allocation of maintenance resources and reduce the operation and maintenance costs.

[0109] Embodiment 3

[0110] The above is a schematic solution of a remote monitoring method for metering boxes based on artificial intelligence. It should be noted that the technical solution of the system for remote monitoring of metering boxes based on artificial intelligence and the technical solution of the above-mentioned remote monitoring method for metering boxes based on artificial intelligence belong to the same concept. For the details not described in detail in the technical solution of the system for remote monitoring of metering boxes based on artificial intelligence in this embodiment, reference can be made to the description of the technical solution of the above-mentioned remote monitoring method for metering boxes based on artificial intelligence.

[0111] This embodiment also provides a remote monitoring system for metering boxes based on artificial intelligence, including: A data acquisition module, configured to obtain first monitoring data of a first target metering box, where the first monitoring data includes metering box electrical parameter data and surrounding environment data; A feature processing module, configured to extract features from the first monitoring data to obtain second monitoring data; An anomaly detection module, configured to input the second monitoring data into a preset anomaly status detection model, and identify the anomaly status of the first target metering box based on the output result of the anomaly status detection model; An alarm processing module, configured to generate an alarm message and send it to a monitoring terminal through a remote communication network when the anomaly status is identified.

[0112] This embodiment also provides an electronic device applicable to the situation of remote monitoring of metering boxes based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the remote monitoring method for metering boxes based on artificial intelligence as proposed in the above embodiment.

[0113] This embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the remote monitoring method for metering boxes based on artificial intelligence as proposed in the above embodiment.

[0114] The storage medium proposed in this embodiment and the remote monitoring method for metering boxes based on artificial intelligence proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0115] From the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software and necessary general-purpose hardware, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk, or optical disc of a computer, and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solution of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solution of the present invention, and all of them should be covered by the scope of the claims of the present invention.

Claims

1. A remote monitoring method for a metering box based on artificial intelligence, characterized in that: Including, obtaining the first monitoring data of the first target metering box, where the first monitoring data includes metering box electrical parameter data and surrounding environment data; Performing feature extraction on the first monitoring data to obtain second monitoring data; Inputting the second monitoring data into a preset abnormal state detection model; Based on the output result of the abnormal state detection model, identifying the abnormal state of the first target metering box; When an abnormal state is identified, generating an alarm message and sending it to the monitoring terminal through a remote communication network.

2. The remote monitoring method for a metering box based on artificial intelligence according to claim 1, characterized in that: The first monitoring data includes: Current data, voltage data, power data, power factor data, temperature data, and vibration data; The first monitoring data is obtained by real-time collection through multiple sensors arranged inside and outside the metering box.

3. The remote monitoring method for metering boxes based on artificial intelligence according to claim 2, characterized in that: The feature extraction includes: Extracting the features of the first monitoring data from two dimensions: time domain and frequency domain; Based on the importance scores of the features, selecting key features to form the second monitoring data.

4. The remote monitoring method for a metering box based on artificial intelligence according to claim 3, characterized in that: The construction method of the abnormal state detection model includes: Collecting normal operation data and various types of abnormal operation data of historical metering boxes to construct a training data set; Using the training data set to train the abnormal state detection model; Using a validation data set to perform performance evaluation and optimization on the trained abnormal state detection model.

5. The remote monitoring method for a metering box based on artificial intelligence according to claim 4, wherein: The abnormal states include: Overload state, underload state, overvoltage state, undervoltage state, three-phase imbalance state, harmonic anomaly state, leakage anomaly state, and temperature anomaly state; The abnormal state detection model outputs the probability values of various abnormal states and determines whether there is an abnormality according to a preset threshold.

6. The remote monitoring method for a metering box based on artificial intelligence according to claim 5, characterized in that: It also includes: Based on historical abnormal state data and equipment maintenance records, performing online learning and updating on the abnormal state detection model; The online learning and updating are achieved through an incremental learning method, enabling the model to adapt to changes in the operation environment and aging characteristics of the metering box.

7. The remote monitoring method of the metering box based on artificial intelligence according to claim 6, characterized in that: Automatically generating maintenance suggestions based on the type and severity of the abnormal state; According to the set alarm level, selecting the corresponding notification method to send the alarm message and maintenance suggestions to relevant maintenance personnel and management systems.

8. A remote monitoring system for a metering box based on artificial intelligence, based on the remote monitoring method for a metering box based on artificial intelligence according to any one of claims 1 to 7, characterized in that: It also includes a data acquisition module for obtaining the first monitoring data of the first target metering box, where the first monitoring data includes metering box electrical parameter data and surrounding environment data; A feature processing module for performing feature extraction on the first monitoring data to obtain second monitoring data; An abnormal detection module for inputting the second monitoring data into a preset abnormal state detection model and, based on the output result of the abnormal state detection model, identifying the abnormal state of the first target metering box; An alarm processing module for generating an alarm message and sending it to the monitoring terminal through a remote communication network when an abnormal state is identified.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for remote monitoring of a metering box based on artificial intelligence according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for remote monitoring of a metering box based on artificial intelligence according to any one of claims 1 to 7.

Images