A method and system for monitoring the operating status of an electric meter box

Abstract

This invention relates to the field of power equipment condition monitoring, specifically to a method and system for monitoring the operating status of electricity meter boxes. An electricity meter box operating status monitoring system includes: a status perception module, an intelligent analysis module, a diagnosis and evaluation module, and an operation and maintenance decision-making module. This invention achieves accurate modeling and dynamic adaptation of the operating status of electricity meter boxes by constructing a personalized dynamic health benchmark model and combining it with real-time acquisition of multi-dimensional sensor data. This effectively eliminates the misjudgment problem caused by fixed threshold monitoring under different seasons, time periods, and load conditions, thereby improving the accuracy of anomaly identification and the system's environmental adaptability.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring, and specifically to a method and system for monitoring the operating status of an electric meter box. Background Technology

[0002] As a key node device connecting the power distribution network and end users, the meter box plays an important role in power metering, power safety assurance, and power distribution management. With the continuous improvement of the informatization level of the power distribution system, some meter boxes have basic status acquisition and remote meter reading capabilities, and can realize the timed reporting of parameters such as voltage and current and basic alarms for some abnormal situations. However, in the currently widely used meter box operation monitoring methods, the assessment of equipment operation status often relies on single-point sensors and fixed threshold judgments, which is difficult to adapt to the natural fluctuations of operating parameters of meter boxes under different seasons, time periods, and load conditions. Especially when facing complex environmental disturbances and abnormal behavior identification scenarios, traditional methods still have considerable room for improvement in terms of anomaly detection accuracy, alarm timeliness, and response strategy generation.

[0003] Existing systems generally lack the ability to integrate and model multi-dimensional operational data from meter boxes, making it difficult to identify complex anomalies caused by a combination of factors such as electricity theft, strong magnetic interference, or external damage. In addition, most monitoring platforms have not yet established dynamic behavior models at the individual meter box level, making it impossible to combine information such as equipment attributes, geographical location, and load characteristics to achieve targeted assessments. At the same time, at the operation and maintenance management level, current strategies mainly rely on manual experience or fixed rule pushes, making it difficult to formulate the optimal operation and maintenance response plan based on the health status of equipment, risk level, and asset value. Overall, it is still mainly based on "reactive processing" and has not yet formed a closed-loop predictive maintenance mechanism.

[0004] Based on this need, this invention proposes a method and system for monitoring the operating status of electric meter boxes, constructing an intelligent monitoring and operation and maintenance closed loop from perception, diagnosis, evaluation to decision-making. Summary of the Invention

[0005] This invention constructs a personalized dynamic health benchmark model for the meter box and combines it with real-time acquisition of multi-dimensional sensor data to achieve accurate modeling and dynamic adaptation of the meter box's operating status. This effectively eliminates the misjudgment problem of fixed threshold monitoring under different seasons, time periods, and load conditions, thereby improving the accuracy of anomaly identification and the system's environmental adaptability.

[0006] A method for monitoring the operating status of an electricity meter box, comprising:

[0007] Step S1: Deploy a multi-dimensional sensor array inside the meter box, including temperature, humidity, door magnetic, current, voltage, smoke, and tilt sensors; among them, the door magnetic sensor is used to detect the opening and closing status of the meter box door and generate access control event data; continuously collect data from each sensor to establish a dynamic health benchmark model for each meter box, which includes normal range curves of parameters that vary with season, date, and time.

[0008] Step S2: Perform multimodal fusion analysis based on multidimensional sensor data to identify composite abnormal events;

[0009] Step S3: Construct a digital twin for each meter box. The digital twin integrates the dynamic health benchmark model of the meter box and receives multi-dimensional sensor data. Perform deviation-driven anomaly detection based on the digital twin, and trigger an alarm by comparing the statistically significant deviation between real-time monitoring data and the predicted value of the benchmark model. At the same time, construct an electricity theft behavior identification model to analyze the spatiotemporal correlation between electrical characteristics and access control events to identify electricity theft behavior. Calculate the equipment health index by comprehensively considering parameter deviation, trend deterioration, and event frequency.

[0010] Step S4: Construct a three-dimensional assessment model based on equipment health index, geographical location risk level, and asset value. Dynamically generate differentiated operation and maintenance strategies based on the three-dimensional assessment model, including immediate response, planned maintenance, economical replacement, or continuous observation. Output inspection route planning, spare parts demand forecast, and operation and maintenance budget plan based on the assessment results of all meter boxes in the area.

[0011] Preferably, a dynamic health benchmark model is established for each meter box, and the specific operation is as follows:

[0012] Step S1.1: Continuously collect environmental parameters, electrical parameters, and safety status parameters of the meter box at a preset sampling frequency; the environmental parameters include temperature and humidity; the electrical parameters include current and voltage; the safety status parameters include door magnetic status, smoke alarm signal, and tilt angle; for the collected raw data, firstly, outliers are identified based on statistical distribution characteristics, specifically by calculating the difference between each data point and the mean of the data within the sliding window. When the difference exceeds three times the standard deviation of the window, the point is determined to be an outlier; secondly, for the identified outliers, linear interpolation of the adjacent normal data points is used for replacement; finally, timestamp alignment and unit unification are performed on the processed data to generate a quality-optimized standard dataset;

[0013] Step S1.2: Based on the standard dataset, construct a baseline curve for each monitoring parameter in an independent dimension:

[0014] Daily cycle benchmark curve: The data is grouped by hour, and the average value and standard deviation of the parameter corresponding to each hour are calculated. The average value is used as the benchmark value for that hour, and the normal fluctuation range of the parameter for that hour is defined by plus or minus three times the standard deviation, thus forming a benchmark curve with a 24-hour cycle.

[0015] Seasonal baseline curve: Group the data by month, calculate the average value of the parameter for each month, and use cubic spline interpolation to fit these monthly average values ​​to generate a smooth seasonal baseline curve that reflects the parameter's variation pattern throughout the year.

[0016] Load-related baseline curves: For current and temperature parameters, linear regression is performed with current as the independent variable and temperature as the dependent variable to obtain the temperature-current relationship; for voltage parameters, linear regression is performed with load rate as the independent variable and voltage deviation as the dependent variable to obtain the voltage-load rate relationship.

[0017] Step S1.3: Based on the standard dataset, construct a benchmark model for the correlation between different monitoring parameters for cross-validation:

[0018] Temperature-current coupling model: Based on historical data collected by temperature and current sensors, temperature and current sequences are generated respectively. The Pearson correlation coefficient between the temperature and current sequences is calculated, and a univariate linear regression equation is established. The correlation coefficient and the coefficient of the regression equation are used as the benchmark for normal coupling relationship.

[0019] Humidity-Seasonal Correlation Model: Calculate the average humidity value for each month and create a lookup table of the correspondence between humidity and month, which serves as a benchmark for determining whether the humidity is normal.

[0020] Voltage stability model: Based on the voltage-load rate relationship, calculate the residual sequence between the actual voltage value and the predicted value of the relationship, and calculate the standard deviation of the residual sequence as the benchmark for voltage stability;

[0021] Step S1.4: Based on the newly collected normal operation data, recalculate the statistical characteristics to update the baseline curve and correlation of the dynamic health baseline model.

[0022] Preferably, multimodal fusion analysis is performed based on multidimensional sensor data to identify composite abnormal events. The specific operation is as follows:

[0023] Step S2.1: Pre-set an event discrimination rule base containing a basic rule layer and a composite rule layer; the basic rule layer is used to define the primary anomaly conditions of each single sensor data, and the composite rule layer is used to define the composite anomaly mode triggered by the combination of multiple different types of sensor data;

[0024] Step S2.2: Monitor the data stream of each sensor in real time. When the data of any sensor meets the primary anomaly condition of the basic rule layer, immediately start a correlation analysis time window. Within this time window, synchronously monitor the data status of a set of preset correlation sensors.

[0025] Step S2.3: Based on the composite rule layer, perform weighted fusion calculation on the multi-sensor anomaly evidence captured within the correlation analysis time window. The multi-sensor anomaly evidence refers to the anomaly state record formed when the data from each sensor meets the primary anomaly conditions in the basic rule layer within the correlation analysis event window. Different types of sensor anomaly evidence are assigned different basic weights to characterize their risk severity. At the same time, the basic weights are dynamically adjusted according to the dynamic characteristics of the anomaly data, and finally, a comprehensive risk score is obtained.

[0026] Step S2.4: Compare the comprehensive risk score with the preset multi-level risk thresholds to generate fusion event alarms with different priorities; and determine the corresponding data encapsulation and uploading strategy according to the priority of the alarm.

[0027] Preferably, the specific implementation methods of digital twin construction and anomaly detection, electricity theft identification, and equipment health assessment in step S3 include:

[0028] Step S3.1: Construct a digital twin of the meter box, which includes a static attribute model and a dynamic behavior model; the static attribute model records the equipment model, rated parameters, installation location, and topological connection relationship in the power distribution network; the dynamic behavior model integrates the dynamic health benchmark model and establishes a state evolution equation based on physical laws to simulate the thermal dynamic process and electrical characteristic changes of the meter box under different load conditions and ambient temperatures;

[0029] Step S3.2: Perform deviation-driven anomaly detection, specifically: acquire real-time monitoring data sequences using a sliding window method, align them with the normal parameter sequences for the corresponding time period in the dynamic health benchmark model; calculate the relative deviation of each corresponding data point between the two sequences to form a deviation sequence; standardize the deviation sequence and calculate its degree of deviation from the historical normal deviation distribution; when the degree of deviation exceeds the control limit set based on the standard deviation of the historical normal deviation for multiple consecutive monitoring cycles, trigger a progressive early warning.

[0030] Step S3.3: Establish an electricity theft identification model. The inputs to this model include: current and voltage waveform data, timestamp sequences and durations of gate state events, and load curve morphology features. The timestamp sequences and durations of gate state events are derived from gate magnetic sensor data; the load curve morphology features are composed of the load rate change trend calculated from current and voltage data; local abnormal pattern features are extracted from the current and voltage waveforms using a one-dimensional convolutional neural network; gate state events are encoded as time series features; an attention mechanism is used to weight and fuse the features extracted from the current and voltage waveform data, gate state events, and load curves based on their importance; finally, an electricity theft probability score between 0 and 1 is output, with a higher score indicating a greater suspicion of electricity theft.

[0031] Step S3.4: Calculate the equipment health index. The specific process is as follows: First, calculate the instantaneous anomaly index by statistically analyzing the degree to which each monitoring parameter deviates from its dynamic health benchmark, and then summing the weighted values ​​to obtain the current anomaly level. Second, calculate the trend deterioration index by using linear regression analysis to assess the rate of change of temperature, current, and voltage over a recent period. Then, calculate the event frequency index by statistically analyzing the frequency and severity of various abnormal events that have occurred recently. Finally, use an objective weighting method to determine the weight of each index, normalize the three indicators, and then weight and fuse them to map them to an equipment health index of 0-100 points.

[0032] Preferably, the electricity theft identification model is based on a convolutional neural network, which includes an input layer, a feature extraction layer, a feature fusion layer, and an output layer.

[0033] Preferably, the specific implementation method of generating the three-dimensional evaluation and dynamic operation and maintenance strategy in step S4 includes:

[0034] Step S4.1: Map the equipment health index to a status dimension score; in the risk dimension, obtain the risk score by weightedly integrating the calculation results of three factors: geographical location importance, power supply user type, and power supply range impact; in the value dimension, obtain the value score by weightedly integrating the calculation results of three factors: equipment remaining lifespan, maintenance history cost, and asset residual value; finally, combine them to form a three-dimensional feature vector;

[0035] Step S4.2: Collect historical data containing three-dimensional feature vectors, operation and maintenance strategies, and execution results; using operation and maintenance costs, power supply reliability, and security risks as optimization objectives, use a multi-objective optimization algorithm to obtain the Pareto optimal solution set for each three-dimensional region; associate the optimal solution with the operation and maintenance strategy to construct a mapping relationship from the three-dimensional region to the strategy, forming a dynamic decision matrix;

[0036] Step S4.3: When the three-dimensional feature vector is updated, the k-nearest neighbor algorithm is used to find the nearest historical best decision point in the decision matrix, and the operation and maintenance strategy is determined by weighted voting; a structured work order containing maintenance content, completion deadline and resource requirements is automatically generated according to the selected strategy.

[0037] Step S4.4: Regularly collect three-dimensional status data of all meter boxes, identify low-health clustering areas based on density clustering algorithm; combine geographical distribution information, use a path planning algorithm with capacity constraints to generate the optimal inspection route; use Monte Carlo method to simulate equipment status development path, statistically analyze spare parts demand probability distribution, and generate demand forecast and budget estimate.

[0038] Step S4.5: Continuously collect and quantitatively evaluate the effectiveness data of strategy execution; dynamically adjust the strategy recommendation priority in the decision matrix based on the effectiveness score; periodically recalculate the strategy mapping relationship between the three-dimensional feature vector and the optimal operation and maintenance strategy based on the accumulated feedback data, that is, update the recommendation rules in the decision matrix.

[0039] A meter box operation status monitoring system, comprising:

[0040] The state perception module includes a multi-dimensional sensor array unit and a dynamic health benchmark modeling unit. The multi-dimensional sensor array unit contains temperature, humidity, door magnetic, current, voltage, smoke, and tilt sensors, which are used to continuously collect data from each sensor. The door magnetic sensor is used to detect the open and closed status of the meter box door and generate access control event data. The dynamic health benchmark modeling unit is used to establish a dynamic health benchmark model for each meter box.

[0041] The intelligent analysis module is used to perform multimodal fusion analysis based on multidimensional sensor data to identify complex abnormal events;

[0042] The diagnosis and assessment module includes a digital twin management unit, a deviation-driven anomaly detection unit, an electricity theft identification unit, and a health status assessment unit. The digital twin management unit integrates a dynamic health baseline model of the meter box and receives multi-dimensional sensor data. The deviation-driven anomaly detection unit performs deviation-driven anomaly detection based on the digital twin, triggering alarms by comparing the statistically significant deviation between real-time monitoring data and the baseline model's predicted values. The electricity theft identification unit constructs an electricity theft identification model to analyze the spatiotemporal correlation between electrical characteristics and access control events, identifying electricity theft behavior. The health status assessment unit calculates the equipment health index by comprehensively considering parameter deviations, trend deterioration, and event frequency.

[0043] The operation and maintenance decision module is used to build a three-dimensional assessment model based on equipment health index, geographical location risk level and asset value. Based on the three-dimensional assessment model, it dynamically generates differentiated operation and maintenance strategies, including immediate response, planned maintenance, economic replacement or continuous observation. Based on the assessment results of all meter boxes in the area, it outputs inspection route planning, spare parts demand forecast and operation and maintenance budget plan.

[0044] The present invention has the following advantages:

[0045] 1. This invention constructs a personalized dynamic health benchmark model for the meter box and combines it with real-time acquisition of multi-dimensional sensor data to achieve accurate modeling and dynamic adaptation of the meter box's operating status. This effectively eliminates the misjudgment problem of fixed threshold monitoring under different seasons, time periods, and load conditions, thereby improving the accuracy of anomaly identification and the system's environmental adaptability.

[0046] 2. This invention constructs a digital twin in the cloud, integrates deviation-driven anomaly detection, deep learning electricity theft identification model and health index quantitative assessment mechanism, and further combines equipment status, risk level and asset value to form a three-dimensional assessment and intelligent strategy generation system. It can dynamically formulate differentiated operation and maintenance decisions based on actual operational risks, and significantly improve the advance warning of faults and the efficiency of operation and maintenance resource utilization. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the meter box operation status monitoring system used in an embodiment of the present invention. Detailed Implementation

[0048] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0049] Example 1: A method for monitoring the operating status of an electricity meter box, comprising:

[0050] Step S1: Deploy a multi-dimensional sensor array inside the meter box, including sensors for temperature, humidity, door magnetic sensors, current, voltage, smoke, and tilt. The door magnetic sensor detects the opening and closing status of the meter box door, generating access control event data (including opening events, duration, etc.). Continuously collect data from each sensor to establish a dynamic health benchmark model for each meter box. This model includes normal range curves for parameters that vary with season, date, and time. The dynamic health benchmark model is not based on a neural network training model; it does not have a hierarchical structure or weight parameters. Instead, it is a set of benchmarks constructed based on historical data from the multi-dimensional sensors using statistical analysis methods (such as mean, standard deviation, correlation coefficient, linear regression, etc.) to reflect the normal fluctuation range and correlation of parameters in the meter box at different time scales (such as hours, months, load conditions). This set is used for quantitative comparison of real-time data deviation.

[0051] Step S2: Perform multimodal fusion analysis based on multidimensional sensor data to identify composite abnormal events;

[0052] Step S3: Construct a digital twin for each meter box. The digital twin integrates the dynamic health benchmark model of the meter box and receives multi-dimensional sensor data. Perform deviation-driven anomaly detection based on the digital twin. Trigger an alarm by comparing the statistical significance deviation between real-time monitoring data and the predicted value of the benchmark model. At the same time, construct an electricity theft behavior identification model to analyze the spatiotemporal correlation between electrical characteristics and access control events to identify electricity theft behavior. Calculate the equipment health index by comprehensively considering parameter deviation, trend deterioration, and event frequency to complete the quantitative assessment of the meter box status.

[0053] Step S4: Construct a three-dimensional assessment model based on equipment health index, geographical location risk level, and asset value. Dynamically generate differentiated operation and maintenance strategies based on the three-dimensional assessment model, including immediate response, planned maintenance, economical replacement, or continuous observation. Output inspection route planning, spare parts demand forecast, and operation and maintenance budget plan based on the assessment results of all meter boxes in the area.

[0054] Establish a dynamic health benchmark model for each meter box. The specific steps are as follows:

[0055] Step S1.1: Continuously collect environmental parameters, electrical parameters, and safety status parameters of the meter box at a preset sampling frequency; the environmental parameters include temperature and humidity; the electrical parameters include current and voltage; the safety status parameters include door magnetic status, smoke alarm signal, and tilt angle; for the collected raw data, firstly, outliers are identified based on statistical distribution characteristics, specifically by calculating the difference between each data point and the mean of the data within the sliding window. When the difference exceeds three times the standard deviation of the window, the point is determined to be an outlier; secondly, for the identified outliers, linear interpolation of the adjacent normal data points is used for replacement; finally, timestamp alignment and unit unification are performed on the processed data to generate a quality-optimized standard dataset;

[0056] Step S1.2: Based on the standard dataset, construct a baseline curve for each monitoring parameter in an independent dimension:

[0057] Daily cycle benchmark curve: The data is grouped by hour, and the average value and standard deviation of the parameter corresponding to each hour are calculated. The average value is used as the benchmark value for that hour, and the normal fluctuation range of the parameter for that hour is defined by plus or minus three times the standard deviation, thus forming a benchmark curve with a 24-hour cycle.

[0058] Seasonal baseline curve: Group the data by month, calculate the average value of the parameter for each month, and use cubic spline interpolation to fit these monthly average values ​​to generate a smooth seasonal baseline curve that reflects the parameter's variation pattern throughout the year.

[0059] Load-related baseline curves: For current and temperature parameters, linear regression is performed with current as the independent variable and temperature as the dependent variable to obtain the temperature-current relationship; for voltage parameters, linear regression is performed with load rate as the independent variable and voltage deviation as the dependent variable to obtain the voltage-load rate relationship.

[0060] Step S1.3: Based on the standard dataset, construct a benchmark model for the correlation between different monitoring parameters for cross-validation:

[0061] Temperature-current coupling model: Based on historical data collected by temperature and current sensors, temperature and current sequences are generated respectively. The Pearson correlation coefficient between the temperature and current sequences is calculated, and a univariate linear regression equation is established. The correlation coefficient and the coefficient of the regression equation are used as the benchmark for normal coupling relationship.

[0062] Humidity-Seasonal Correlation Model: Calculate the average humidity value for each month and create a lookup table of the correspondence between humidity and month, which serves as a benchmark for determining whether the humidity is normal.

[0063] Voltage stability model: Based on the voltage-load rate relationship, calculate the residual sequence between the actual voltage value and the predicted value of the relationship, and calculate the standard deviation of the residual sequence as the benchmark for voltage stability;

[0064] Step S1.4: Based on the newly collected normal operation data, recalculate the statistical characteristics to update the baseline curve and correlation of the dynamic health baseline model;

[0065] After completing the initial self-learning, the system enters the normal operation phase and starts the model update process: For newly collected data that is determined to be normal, the existing baseline curve and model are integrated using an exponentially weighted moving average algorithm, where new data is assigned a weight of 0.05 and historical baselines are assigned a weight of 0.95, thus achieving a gradual update of the model; at the same time, every full month, a local recalculation based on the data of the past three months is automatically triggered to optimize the seasonal baseline curve.

[0066] In existing conventional technologies, uniform upper and lower limits or simple seasonal corrections are usually used, which cannot adapt to different enclosure materials, installation orientations, load characteristics, and micro-environmental differences. This leads to frequent false alarms and missed alarms in scenarios such as high temperature sunny days and sudden load increases. This model continuously absorbs multi-dimensional data from the normal operation of a single enclosure and automatically updates its "daily-seasonal-load" three-dimensional baseline curve and parameter coupling relationship. It achieves adaptive drift of thresholds with enclosure, season, and load, thereby reducing the false alarm rate and detecting deviation trends in the early stages of faults, improving the accuracy of anomaly detection and environmental robustness.

[0067] Multimodal fusion analysis is performed based on multidimensional sensor data to identify complex anomalous events. The specific operation is as follows:

[0068] Step S2.1: Pre-set an event discrimination rule base containing a basic rule layer and a composite rule layer; the basic rule layer is used to define the primary abnormal conditions of each individual sensor data. For example, the primary alarm condition of the temperature sensor is set to exceed 85℃. This threshold is set based on the long-term tolerance temperature critical point of common insulation materials in meter boxes; the primary alarm condition of the humidity sensor is set to exceed 90%RH for 3 consecutive minutes. This duration is used to distinguish between instantaneous interference and continuous condensation risk; the primary alarm condition of the door magnetic sensor is set to be continuously opened for more than 5 minutes within the non-maintenance time window. This duration is set based on distinguishing between normal passage and abnormal stagnation; the composite rule layer is used to define composite abnormal modes triggered by the combination of multiple different types of sensor data. For example, "temperature rises by more than 10℃ within 1 minute and current exceeds the rated value" is defined as an overload risk mode. This combination rule is set based on identifying heat accumulation as a precursor to failure; "vibration acceleration exceeds 5g and box tilt angle exceeds 15 degrees" is defined as a physical impact risk mode. This combination rule is set based on comprehensively judging the severity of external force damage.

[0069] Step S2.2: Monitor the data streams of each sensor in real time. When any sensor data meets the primary anomaly condition of the basic rule layer, immediately start a correlation analysis time window, which can be set to 30 seconds. The time setting is based on balancing the speed of fault development and the system processing capacity. Within this time window, the data status of a set of preset correlated sensors is monitored simultaneously. For example, for temperature anomalies, monitor the current sensor and the door magnetic sensor; for door magnetic anomalies, monitor the current sensor and the vibration sensor.

[0070] Step S2.3: Based on the composite rule layer, perform weighted fusion calculation on the multi-sensor anomaly evidence captured within the correlation analysis time window. The multi-sensor anomaly evidence refers to the abnormal state record formed when the data from each sensor meets the primary anomaly conditions in the basic rule layer within the correlation analysis event window. Different types of sensor anomaly evidence are assigned different basic weights to characterize their risk severity. Temperature anomalies and smoke detection can be assigned the highest weight of 10 points; door status anomalies and tilting vibration can be assigned a medium weight of 6 points; and environmental parameter anomalies can be assigned a basic weight of 3 points. This weight allocation is based on the contribution statistics of each evidence type in historical fault data. At the same time, the basic weights are dynamically adjusted according to the dynamic characteristics of the anomaly data. For example, when the detected temperature change rate exceeds 5℃ / minute, 3 points are added to the basic weight of temperature anomalies. This adjustment is based on the rapid development of the anomaly. Finally, a comprehensive risk score is obtained by summarizing the results.

[0071] Step S2.4: Compare the comprehensive risk score with the preset multi-level risk thresholds. A three-level threshold of 15, 25, and 35 points can be set for general alarms, important alarms, and emergency alarms. This threshold setting is based on the balance achieved between false alarm rate and false negative rate through historical data testing. Based on this, generate fusion event alarms with different priorities. And according to the priority of the alarm, determine the corresponding data encapsulation and uploading strategy (e.g., encapsulate the original data fragments 120 seconds before and after the alarm time for emergency alarms; this length is set to cover the entire typical fault development process).

[0072] The specific implementation methods for digital twin construction and anomaly detection, electricity theft identification, and equipment health assessment in step S3 include:

[0073] Step S3.1: Construct a digital twin of the meter box, which includes a static attribute model and a dynamic behavior model. The static attribute model records the equipment model, rated parameters, installation location, and topological connection relationship in the distribution network (recording the meter box's model specifications, rated current value, GPS coordinate location, and its connection relationship with transformers and branch lines in the distribution network; this information is set based on establishing a unique identification of the equipment and electrical association relationships). The dynamic behavior model integrates the dynamic health benchmark model and establishes a state evolution equation based on physical laws to simulate the thermal dynamic process and electrical characteristic changes of the meter box under different load conditions and ambient temperatures. For example, it establishes a box internal temperature evolution equation based on thermodynamic laws, considering the heat generation of the relationship between the square of the current and resistance, the heat dissipation coefficient of the box material, and the influence of ambient temperature. This modeling is based on accurately simulating the physical essence of the changes in the operating state of the equipment.

[0074] Step S3.2: Perform deviation-driven anomaly detection, specifically: acquire real-time monitoring data sequences using a sliding window method, for example, a 30-minute sliding window, the length of which is set to cover typical load change cycles; align these sequences with the normal parameter sequences for the corresponding time periods in the dynamic health benchmark model; calculate the relative deviation of each corresponding data point between the two sequences to form a deviation sequence; standardize the deviation sequence and calculate its degree of deviation from the historical normal deviation distribution; when this deviation exceeds the control limit set based on the historical normal deviation standard deviation for multiple consecutive monitoring periods, for example, triggering a Level 1 warning when the deviation exceeds twice the historical standard deviation for 3 consecutive periods, and upgrading to a Level 2 warning for 5 consecutive periods, this setting is based on achieving a balance between timely warning and avoiding false alarms; trigger a progressive warning (i.e., the warning level increases progressively with the increase of the continuous exceedance period, for example, Level 1 for three consecutive periods, and Level 2 for five consecutive periods);

[0075] Step S3.3: Establish an electricity theft identification model. The inputs to this model include: current and voltage waveform data, timestamp sequences and durations of gate state events, and load curve morphology features. The timestamp sequences and durations of gate state events are derived from gate magnetic sensor data; the load curve morphology features are composed of the load rate change trend calculated from the current and voltage data. Local abnormal pattern features are extracted from the current and voltage waveforms using a one-dimensional convolutional neural network. Gate state events are encoded as time-series features. An attention mechanism is used to weight and fuse the features extracted from the current and voltage waveform data, gate state events, and load curves based on their importance. For example, a weight of 0.6 is assigned to the current waveform features, 0.3 to the gate state features, and 0.1 to the load features. This weighting is based on the difference in contribution of each feature to electricity theft identification. Finally, an electricity theft probability score between 0 and 1 is output, with a higher score indicating a greater suspicion of electricity theft.

[0076] Step S3.4: Calculate the equipment health index. The specific process is as follows: First, calculate the instantaneous anomaly index. By statistically analyzing the degree of deviation of each monitoring parameter from its dynamic health benchmark, the current anomaly level is obtained by weighted summation. Mahalanobis distance can be used to calculate the comprehensive deviation. This method is set based on considering the correlation between parameters. Second, calculate the trend deterioration index. By analyzing the slope of the changes in temperature, current, and voltage over a recent period through linear regression, the rate of condition deterioration is assessed (e.g., analyzing the trend of temperature parameter changes over the past 24 hours; the duration is set based on reflecting the daily cycle of equipment health status changes). Then, calculate the event frequency index. Statistically analyze the frequency and severity of various abnormal events that have occurred recently (e.g., statistically analyze all abnormal events that have occurred in the past 7 days and assign different weights according to severity; the duration is set based on covering typical event occurrence cycles). Finally, use an objective weighting method to determine the weight of each indicator (e.g., use the entropy weight method to automatically allocate weights based on the information content of each indicator; this method is set based on avoiding the influence of subjective factors). After normalizing the three indicators, they are weighted and merged to map to an equipment health index of 0-100 points.

[0077] The electricity theft detection model is based on a convolutional neural network, which includes an input layer, a feature extraction layer, a feature fusion layer, and an output layer.

[0078] The input layer receives multi-source input data, including current and voltage waveform data, timestamp sequences and durations of gate state events, and load curve morphology features. The feature extraction layer contains parallel one-dimensional convolutional branches that extract local anomaly features from the current and voltage waveforms and encode the timestamp sequences and durations of gate state events into time-series features. The feature fusion layer uses an attention mechanism to weightedly integrate the extracted multi-source features. The output layer outputs a probability score for electricity theft.

[0079] The specific steps for constructing the electricity theft detection model are as follows:

[0080] Historical operating data is used as training samples. Each training sample contains a set of time series data and its corresponding real labels. The time series data includes current and voltage waveforms, timestamp sequences and durations of gate state events, and load curve morphology characteristics. The real labels are labeled based on historically confirmed cases of electricity theft and normal operation records.

[0081] The training samples are divided into a training set and a validation set in chronological order to ensure that the time periods of the training set and the validation set do not overlap.

[0082] During training, the network parameters are optimized using the training set data, while the model performance is monitored using the validation set to prevent overfitting. The difference between the predicted output and the true label is minimized through the backpropagation algorithm, and the network parameters are continuously adjusted until the model's performance on the validation set reaches a stable state.

[0083] The specific implementation methods for generating the three-dimensional evaluation and dynamic operation and maintenance strategy in step S4 include:

[0084] Step S4.1: Linearly map the device health index to the status dimension, and use it as the score for the status dimension;

[0085] In the risk dimension, the risk score is calculated by comprehensively assessing three factors: the geographical importance of the meter box, the type of power supply user, and the degree of impact on the power supply range. The geographical importance is assessed based on the population density and distribution of important facilities in the area, the type of power supply user is classified according to the degree of social operational impact of the supplied load, and the degree of impact on the power supply range is quantified by the number of users who may experience power outages due to a fault in the meter box. The assessment results of the three factors are weighted and integrated according to preset weights to obtain the final risk dimension score.

[0086] The specific scoring calculation process is as follows: First, a hierarchical structure is constructed, which includes three criteria: geographical location importance, power supply user type, and power supply range influence. Then, the weight of each criterion is determined through pairwise comparison matrices, with geographical location importance having the highest weight, followed by power supply user type, and power supply range influence having the lowest weight.

[0087] In the assessment of geographical importance, the area where the meter box is located is divided into three levels according to population density: high, medium and low, corresponding to 3, 2 and 1 points respectively. At the same time, the distribution of important facilities is also considered. For example, if it is located within the power supply range of important facilities such as hospitals and schools, an additional 2 points are added; if it is located in a commercial area, an additional 1 point is added; and no points are awarded for ordinary residential areas.

[0088] In the assessment of power supply user types, the loads are classified and assigned values ​​according to their impact on social operation: loads that play a key supporting role in social operation, such as emergency command centers and large transportation hubs, are assigned 3 points; economic loads such as hospitals and large industrial enterprises are assigned 2 points; ordinary commercial loads are assigned 1 point; and residential loads are assigned 0 points.

[0089] In the assessment of the impact on the power supply range, the number of users who may experience a power outage due to the box failure is quantified: more than 100 households are assigned 3 points, 50-100 households are assigned 2 points, 10-50 households are assigned 1 point, and less than 10 households are assigned 0 points.

[0090] Finally, the scores of the three criteria are multiplied by their corresponding weights and then summed to obtain the final risk dimension score. The calculation formula is: Risk dimension score = Geographic location score × W1 + Power supply user type score × W2 + Power supply range impact score × W3, where W1, W2, and W3 are weight coefficients determined by the analytic hierarchy process, and W1 + W2 + W3 = 1.

[0091] In terms of value, the value score is calculated by comprehensively evaluating three factors: the remaining lifespan of the meter box equipment, the historical maintenance costs, and the residual value of the assets. The remaining lifespan of the equipment is assessed based on the comparison between the equipment's operating time and its designed lifespan. The historical maintenance costs are calculated by statistically analyzing the spare parts costs and labor costs in historical maintenance. The residual value of the assets is assessed based on the equipment depreciation model. The evaluation results of the three factors are weighted and integrated according to preset weights to obtain the final value score.

[0092] The calculation process for the value dimension score can be further explained in detail below:

[0093] The remaining life assessment of equipment is obtained by calculating the ratio of the equipment's operating time to its designed life. For example, if a certain model of meter box has a designed life of 15 years and has been in operation for 5 years, then the remaining life score is (15-5) / 15×100=67 points.

[0094] The historical maintenance cost assessment is calculated by statistically analyzing maintenance expenditures over the past three years. First, the average maintenance cost per unit time for this model of equipment under the same operating conditions during the same period is determined as a benchmark. Then, the ratio of the actual maintenance cost per unit time of the equipment to be assessed to the benchmark value is calculated. Finally, a piecewise function is used to map this ratio to a percentage score. Specifically, when the ratio is less than or equal to 0.8 (indicating excellent cost control), the score is 100 points; when the ratio is between 0.8 and 1.2 (indicating normal cost), the score linearly decreases from 100 points to 60 points; and when the ratio is greater than 1.2 (indicating excessive cost), the score linearly decreases from 60 points to 0 points, thus ensuring that the score is always within a reasonable range.

[0095] The current residual value of the asset is calculated using the straight-line depreciation method, and then a percentage score is calculated based on the residual value rate.

[0096] Residual value of assets = Original value of equipment × (1 - Number of years of use × Annual depreciation rate);

[0097] Asset residual value score = (asset residual value / original equipment value) × 100;

[0098] For example: If the original value of the equipment is 2000 yuan, it has been used for 5 years, and the annual depreciation rate is 6.7%, then:

[0099] The residual value of the asset is 2000 × (1 - 5 × 6.7%) = 1330 yuan;

[0100] Asset residual value score = (1330 / 2000) × 100 = 66.5 points;

[0101] Finally, the three scores are combined according to preset weights, such as 40% for remaining lifespan, 35% for maintenance costs, and 25% for asset residual value. The final value dimension score is 67 × 0.4 + maintenance cost score × 0.35 + asset residual value score × 0.25.

[0102] The weighting is based on the following criteria: remaining useful life has the highest weight, as it directly reflects the technical life of the equipment; maintenance cost has the second highest weight, reflecting the economic efficiency of equipment operation and maintenance; asset residual value has the lowest weight, mainly reflecting the book value of the equipment; the weighting allocation is determined based on the principle of equipment life cycle cost management.

[0103] Finally, the scores from the three dimensions are combined to form a three-dimensional feature vector;

[0104] Step S4.2: Establishing the dynamic decision matrix

[0105] Historical operation and maintenance data is collected, including the three-dimensional feature vectors of each meter box, the operation and maintenance strategies adopted, and the effect data after the strategy implementation. When implementing the specific data collection, the data collection time window is preferably set to 24 months to fully cover the typical operating cycle and seasonal variation of the equipment. Using operation and maintenance cost, power supply reliability index, and safety risk index as three optimization objectives, a multi-objective optimization algorithm is used to train the historical data (for example, using a genetic algorithm with an elite strategy to find the Pareto optimal solution set, the algorithm is selected based on its effectiveness and convergence in handling multi-objective optimization problems), resulting in Pareto optimal solution sets in different three-dimensional coordinate regions. Each Pareto optimal solution is associated with a specific operation and maintenance strategy. For example, for a region where the equipment health index is below 30, the risk level is above 80, and the value score is greater than 60, the optimal solution is identified as the "immediate replacement" strategy, forming a mapping relationship from three-dimensional spatial regions to operation and maintenance strategies, and constructing a dynamic decision matrix.

[0106] Step S4.3: Real-time strategy matching and work order generation

[0107] When the three-dimensional feature vector of the meter box is updated due to a change in state, the system calculates the coordinates of the vector in the three-dimensional evaluation space. In the dynamic decision matrix, the k-nearest neighbor algorithm is used to find several historical optimal decision points that are closest to the coordinates (for example, setting k=5 to find the 5 closest decision points; this parameter setting is based on achieving a balance between decision accuracy and computational efficiency). The most suitable operation and maintenance strategy for the current situation is determined through a weighted voting mechanism. Based on the selected operation and maintenance strategy, a structured work order is automatically generated, specifying the maintenance content, completion deadline, required skill level, and estimated resource consumption.

[0108] Step S4.4: Global Resource Optimization

[0109] Regularly collect three-dimensional status data of all meter boxes within the jurisdiction to form a status distribution map; identify low-health status areas with high-density clustering in the status space based on density clustering algorithm and mark them as priority areas of concern; combine the geographical distribution information of meter boxes and use a vehicle route planning algorithm with capacity constraints (e.g., use the cost-saving CW algorithm to solve the optimal route, with constraints including a maximum number of meter boxes inspected per day ≤ 20, based on the actual working capacity of the maintenance team) to generate the optimal inspection route considering timeliness and cost; use the Monte Carlo method to simulate the possible development path of equipment status within a certain period in the future, statistically analyze the expected demand probability distribution of various types of spare parts, and generate spare parts demand forecasts and budget range estimates.

[0110] Step S4.5: Strategy Feedback and Optimization

[0111] Establish a closed-loop feedback mechanism for strategy effectiveness, specifically including: continuously collecting actual effect data of implemented operation and maintenance strategies, including fault resolution time, cost deviation, and recurrence; quantitatively evaluating various strategies based on the effect data and generating effect scores. Specifically, a percentage-based scoring system can be constructed, including three dimensions: cost efficiency, time efficiency, and durability. The Q-learning algorithm is used to dynamically adjust the strategy weights, with the learning rate parameter set to 0.1 and the discount factor set to 0.9 to ensure the stability of the learning process; periodically recalculating the strategy mapping relationship in the decision matrix between the three-dimensional feature vector and the optimal operation and maintenance strategy based on accumulated feedback data, i.e., updating the recommendation rules in the decision matrix. For example, if the three-dimensional feature vector of a meter box is [health index = 25, risk score = 85, value score = 60], and the optimal strategy corresponding to this vector area in historical data is "immediate replacement", then this correspondence constitutes a strategy mapping. Subsequent similar vectors will directly recommend this strategy until feedback data shows that its effectiveness has declined and it is recalculated and replaced, ensuring that the decision rules are continuously optimized as the operating environment changes.

[0112] To objectively verify the actual effect of the technical solution described in this invention, a six-month comparative test was conducted on the operating performance of the existing system using the traditional fixed threshold monitoring method and the system deployed with this invention in a typical group of electricity meter boxes of the same area and scale. The test results are presented through quantitative indicators of the following three core dimensions, as shown in Tables 1-3, which fully demonstrate the significant advantages of this invention in terms of anomaly diagnosis accuracy, early warning timeliness, and operation and maintenance efficiency.

[0113] Table 1: Comparison of Accuracy Rates for Abnormal Diagnosis

[0114]

[0115] Table 2: Comparison of Fault Early Warning Timeliness

[0116]

[0117] Table 3: Comparison of Operation and Maintenance Resource Utilization Efficiency

[0118]

[0119] Tables 1–3 show the “existing conventional monitoring methods” which refer to the “single-point sensor + fixed threshold” scheme commonly used in the industry: independent sensors such as temperature, humidity, and smoke detectors are installed in the meter box, and the static alarm thresholds are uniformly set by the back-end main station (e.g., temperature ≥85 ℃, humidity ≥90 %RH, current ≥1.2×rated value). Once the real-time sampled value exceeds the threshold, an alarm is triggered. The threshold is not automatically adjusted according to the season, load, or individual differences in the meter box, and no multi-sensor fusion judgment is performed.

[0120] Example 2: A meter box operation status monitoring system, such as... Figure 1 As shown, it includes:

[0121] The status perception module includes a multi-dimensional sensor array unit and a dynamic health benchmark modeling unit. The multi-dimensional sensor array unit contains temperature, humidity, door magnetic, current, voltage, smoke, and tilt sensors, which are used to continuously collect data from each sensor. The door magnetic sensor is used to detect the open and closed status of the meter box door and generate access control event data. The dynamic health benchmark modeling unit is used to establish a dynamic health benchmark model for each meter box.

[0122] The intelligent analysis module is used to perform multimodal fusion analysis based on multidimensional sensor data to identify complex abnormal events;

[0123] The diagnosis and assessment module includes a digital twin management unit, a deviation-driven anomaly detection unit, an electricity theft identification unit, and a health status assessment unit. The digital twin management unit integrates a dynamic health baseline model of the meter box and receives multi-dimensional sensor data. The deviation-driven anomaly detection unit performs deviation-driven anomaly detection based on the digital twin, triggering alarms by comparing the statistically significant deviation between real-time monitoring data and the baseline model's predicted values. The electricity theft identification unit constructs an electricity theft identification model to analyze the spatiotemporal correlation between electrical characteristics and access control events, identifying electricity theft behavior. The health status assessment unit calculates the equipment health index by comprehensively considering parameter deviations, trend deterioration, and event frequency.

[0124] The operation and maintenance decision module is used to build a three-dimensional assessment model based on equipment health index, geographical location risk level and asset value. Based on the three-dimensional assessment model, it dynamically generates differentiated operation and maintenance strategies, including immediate response, planned maintenance, economic replacement or continuous observation. Based on the assessment results of all meter boxes in the area, it outputs inspection route planning, spare parts demand forecast and operation and maintenance budget plan.

[0125] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method of monitoring the operating condition of an electricity meter box, characterized by, include: Step S1: Deploy a multi-dimensional sensor array inside the meter box, including sensors for temperature, humidity, door magnetic sensors, current, voltage, smoke, and tilt. The door magnetic sensor detects the opening and closing status of the meter box door, generating access control event data. Continuously collect data from each sensor to establish a dynamic health benchmark model for each meter box. This model includes benchmark curves for each monitoring parameter in an independent dimension and correlation benchmark models between different monitoring parameters. The benchmark curves include daily cycle benchmark curves, seasonal cycle benchmark curves, and load-related benchmark curves. The correlation benchmark models include a temperature-current coupling model, a humidity-seasonal correlation model, and a voltage stability model. Step S2: Perform multimodal fusion analysis based on multidimensional sensor data to identify composite abnormal events; Step S3: Construct a digital twin for each meter box. The digital twin integrates the dynamic health benchmark model of the meter box and receives multi-dimensional sensor data. Perform deviation-driven anomaly detection based on the digital twin, and trigger an alarm by comparing the statistically significant deviation between real-time monitoring data and the predicted value of the benchmark model. At the same time, construct an electricity theft behavior identification model to analyze the spatiotemporal correlation between electrical characteristics and access control events to identify electricity theft behavior. Calculate the equipment health index by comprehensively considering parameter deviation, trend deterioration, and event frequency. Step S4: Construct a three-dimensional assessment model based on equipment health index, geographical location risk level, and asset value. Dynamically generate differentiated operation and maintenance strategies based on the three-dimensional assessment model, including immediate response, planned maintenance, economical replacement, or continuous observation. Output inspection route planning, spare parts demand forecast, and operation and maintenance budget plan based on the assessment results of all meter boxes in the area. The specific steps for step S2 are as follows: Step S2.1: Pre-set an event discrimination rule base containing a basic rule layer and a composite rule layer; the basic rule layer is used to define the primary anomaly conditions of each single sensor data, and the composite rule layer is used to define the composite anomaly mode triggered by the combination of multiple different types of sensor data; Step S2.2: Monitor the data stream of each sensor in real time. When the data of any sensor meets the primary anomaly condition of the basic rule layer, immediately start a correlation analysis time window. Within this time window, synchronously monitor the data status of a set of preset correlation sensors. Step S2.3: Based on the composite rule layer, perform weighted fusion calculation on the multi-sensor anomaly evidence captured within the correlation analysis time window. The multi-sensor anomaly evidence refers to the anomaly state record formed when the data from each sensor meets the primary anomaly conditions in the basic rule layer within the correlation analysis event window. Different types of sensor anomaly evidence are assigned different basic weights to characterize their risk severity. At the same time, the basic weights are dynamically adjusted according to the dynamic characteristics of the anomaly data, and finally, a comprehensive risk score is obtained. Step S2.4: Compare the comprehensive risk score with the preset multi-level risk thresholds to generate fusion event alarms with different priorities; and determine the corresponding data encapsulation and uploading strategy according to the priority of the alarm.

2. The method of claim 1, wherein, Establish a dynamic health benchmark model for each meter box. The specific steps are as follows: Step S1.1: Continuously collect environmental parameters, electrical parameters, and safety status parameters of the meter box at a preset sampling frequency; the environmental parameters include temperature and humidity; the electrical parameters include current and voltage; the safety status parameters include door magnetic status, smoke alarm signal, and tilt angle; for the collected raw data, firstly, outliers are identified based on statistical distribution characteristics, specifically by calculating the difference between each data point and the mean of the data within the sliding window. When the difference exceeds three times the standard deviation of the window, the point is determined to be an outlier; secondly, for the identified outliers, linear interpolation of the adjacent normal data points is used for replacement; finally, timestamp alignment and unit unification are performed on the processed data to generate a quality-optimized standard dataset; Step S1.2: Based on the standard dataset, construct a baseline curve for each monitoring parameter in an independent dimension: Daily cycle benchmark curve: The data is grouped by hour, and the average value and standard deviation of the parameter corresponding to each hour are calculated. The average value is used as the benchmark value for that hour, and the normal fluctuation range of the parameter for that hour is defined by plus or minus three times the standard deviation, thus forming a benchmark curve with a 24-hour cycle. Seasonal baseline curve: Group the data by month, calculate the average value of the parameter for each month, and use cubic spline interpolation to fit these monthly average values ​​to generate a smooth seasonal baseline curve that reflects the parameter's variation pattern throughout the year. Load-related baseline curves: For current and temperature parameters, linear regression is performed with current as the independent variable and temperature as the dependent variable to obtain the temperature-current relationship; for voltage parameters, linear regression is performed with load rate as the independent variable and voltage deviation as the dependent variable to obtain the voltage-load rate relationship. Step S1.3: Based on the standard dataset, construct a benchmark model for the correlation between different monitoring parameters for cross-validation: Temperature-current coupling model: Based on historical data collected by temperature and current sensors, temperature and current sequences are generated respectively. The Pearson correlation coefficient between the temperature and current sequences is calculated, and a univariate linear regression equation is established. The correlation coefficient and the coefficient of the regression equation are used as the benchmark for normal coupling relationship. Humidity-Seasonal Correlation Model: Calculate the average humidity value for each month and create a lookup table of the correspondence between humidity and month, which serves as a benchmark for determining whether the humidity is normal. Voltage stability model: Based on the voltage-load rate relationship, calculate the residual sequence between the actual voltage value and the predicted value of the relationship, and calculate the standard deviation of the residual sequence as the benchmark for voltage stability; Step S1.4: Based on the newly collected normal operation data, recalculate the statistical characteristics to update the baseline curve and correlation of the dynamic health baseline model.

3. The method of claim 2, wherein, The specific implementation methods for digital twin construction and anomaly detection, electricity theft identification, and equipment health assessment in step S3 include: Step S3.1: Construct a digital twin of the meter box, which includes a static attribute model and a dynamic behavior model; the static attribute model records the equipment model, rated parameters, installation location, and topological connection relationship in the power distribution network; the dynamic behavior model integrates the dynamic health benchmark model and establishes a state evolution equation based on physical laws to simulate the thermal dynamic process and electrical characteristic changes of the meter box under different load conditions and ambient temperatures; Step S3.2: Perform deviation-driven anomaly detection, specifically: acquire real-time monitoring data sequences using a sliding window method, align them with the normal parameter sequences for the corresponding time period in the dynamic health benchmark model; calculate the relative deviation of each corresponding data point between the two sequences to form a deviation sequence; standardize the deviation sequence and calculate its degree of deviation from the historical normal deviation distribution; when the degree of deviation exceeds the control limit set based on the standard deviation of the historical normal deviation for multiple consecutive monitoring cycles, trigger a progressive early warning. Step S3.3: Establish an electricity theft identification model. The inputs to this model include: current and voltage waveform data, timestamp sequences and durations of gate state events, and load curve morphology features. The timestamp sequences and durations of gate state events are derived from gate magnetic sensor data; the load curve morphology features are composed of the load rate change trend calculated from current and voltage data; local abnormal pattern features are extracted from the current and voltage waveforms using a one-dimensional convolutional neural network; gate state events are encoded as time series features; an attention mechanism is used to weight and fuse the features extracted from the current and voltage waveform data, gate state events, and load curves based on their importance; finally, an electricity theft probability score between 0 and 1 is output, with a higher score indicating a greater suspicion of electricity theft. Step S3.4: Calculate the equipment health index. The specific process is as follows: First, calculate the instantaneous anomaly index by statistically analyzing the degree to which each monitoring parameter deviates from its dynamic health benchmark, and then summing the weighted values ​​to obtain the current anomaly level. Second, calculate the trend deterioration index by using linear regression analysis to assess the rate of change of temperature, current, and voltage over a recent period. Then, calculate the event frequency index by statistically analyzing the frequency and severity of various abnormal events that have occurred recently. Finally, use an objective weighting method to determine the weight of each index, normalize the three indicators, and then weight and fuse them to map them to an equipment health index of 0-100 points.

4. The method of claim 3, wherein, The electricity theft detection model is based on a convolutional neural network, which consists of an input layer, a feature extraction layer, a feature fusion layer, and an output layer.

5. The method for monitoring the operating status of an electricity meter box according to claim 4, characterized in that, The specific implementation methods for generating the three-dimensional evaluation and dynamic operation and maintenance strategy in step S4 include: Step S4.1: Map the equipment health index to a status dimension score; in the risk dimension, obtain the risk score by weightedly integrating the calculation results of three factors: geographical location importance, power supply user type, and power supply range impact; in the value dimension, obtain the value score by weightedly integrating the calculation results of three factors: equipment remaining lifespan, maintenance history cost, and asset residual value; finally, combine them to form a three-dimensional feature vector; Step S4.2: Collect historical data containing three-dimensional feature vectors, operation and maintenance strategies, and execution results; using operation and maintenance costs, power supply reliability, and security risks as optimization objectives, use a multi-objective optimization algorithm to obtain the Pareto optimal solution set for each three-dimensional region; associate the optimal solution with the operation and maintenance strategy to construct a mapping relationship from the three-dimensional region to the strategy, forming a dynamic decision matrix; Step S4.3: When the three-dimensional feature vector is updated, the k-nearest neighbor algorithm is used to find the nearest historical best decision point in the decision matrix, and the operation and maintenance strategy is determined by weighted voting; a structured work order containing maintenance content, completion deadline and resource requirements is automatically generated according to the selected strategy. Step S4.4: Regularly collect three-dimensional status data of all meter boxes, identify low-health clustering areas based on density clustering algorithm; combine geographical distribution information, use a path planning algorithm with capacity constraints to generate the optimal inspection route; use Monte Carlo method to simulate equipment status development path, statistically analyze spare parts demand probability distribution, and generate demand forecast and budget estimate. Step S4.5: Continuously collect and quantitatively evaluate the effectiveness data of strategy execution; dynamically adjust the strategy recommendation priority in the decision matrix based on the effectiveness score; periodically recalculate the strategy mapping relationship between the three-dimensional feature vector and the optimal operation and maintenance strategy based on the accumulated feedback data, that is, update the recommendation rules in the decision matrix.

6. A meter box operation status monitoring system, characterized in that, The system is applied to the method for monitoring the operating status of an electric meter box according to any one of claims 1-5, comprising: The state perception module includes a multi-dimensional sensor array unit and a dynamic health benchmark modeling unit. The multi-dimensional sensor array unit contains temperature, humidity, door magnetic, current, voltage, smoke, and tilt sensors, which are used to continuously collect data from each sensor. The door magnetic sensor is used to detect the open and closed status of the meter box door and generate access control event data. The dynamic health benchmark modeling unit is used to establish a dynamic health benchmark model for each meter box. The intelligent analysis module is used to perform multimodal fusion analysis based on multidimensional sensor data to identify complex abnormal events; The diagnosis and assessment module includes a digital twin management unit, a deviation-driven anomaly detection unit, an electricity theft identification unit, and a health status assessment unit. The digital twin management unit integrates a dynamic health baseline model of the meter box and receives multi-dimensional sensor data. The deviation-driven anomaly detection unit performs deviation-driven anomaly detection based on the digital twin, triggering alarms by comparing the statistically significant deviation between real-time monitoring data and the baseline model's predicted values. The electricity theft identification unit constructs an electricity theft identification model to analyze the spatiotemporal correlation between electrical characteristics and access control events, identifying electricity theft behavior. The health status assessment unit calculates the equipment health index by comprehensively considering parameter deviations, trend deterioration, and event frequency. The operation and maintenance decision module is used to build a three-dimensional assessment model based on equipment health index, geographical location risk level and asset value. Based on the three-dimensional assessment model, it dynamically generates differentiated operation and maintenance strategies, including immediate response, planned maintenance, economic replacement or continuous observation. Based on the assessment results of all meter boxes in the area, it outputs inspection route planning, spare parts demand forecast and operation and maintenance budget plan.

Images