Ammeter state monitoring method, system and equipment based on behavior baseline, and medium

Abstract

The invention discloses a behavior baseline-based ammeter state monitoring method, system, equipment and medium, and relates to the technical field of equipment behavior anomaly detection, and the method comprises the steps: collecting operation parameter time sequence data of an intelligent ammeter in a preset monitoring region during operation, carrying out the combined processing of the operation parameter time sequence data, extracting the multi-dimensional features of the region, and carrying out the detection of the behavior baseline; based on regional multi-dimensional features in a historical normal state, a regional behavior baseline is established through unsupervised learning, regional multi-dimensional features at a current monitoring moment are acquired, a regional consistency index is generated through matching degree calculation with the regional behavior baseline, and an abnormal ammeter is identified according to an individual deviation degree. And based on the region consistency index and a preset period, triggering update maintenance of the region behavior baseline. According to the method, visual visualization and a model self-evolution mechanism are combined, and the state of the industrial electric meter is changed from passive warning to active, accurate and large-scale predictive intelligent operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of equipment behavior anomaly detection technology, specifically to a method, system, device, and medium for monitoring the status of electricity meters based on behavioral baselines. Background Technology

[0002] With the deepening of the digital transformation of the power industry, smart meters, as key terminals for realizing industrial energy efficiency management and intelligent power distribution, have been widely deployed on the industrial user side, undertaking multiple tasks such as energy consumption collection and power monitoring. However, the current meter systems in industrial scenarios generally suffer from fragmented monitoring modes. Most systems still use a single meter as the monitoring object and judge anomalies by setting fixed thresholds. This isolated perspective ignores the regional correlation of industrial loads. Fluctuations in a single meter may not represent a real fault, while synchronous changes within a region are more likely to reflect systemic problems, resulting in a significant reduction in the effectiveness of monitoring.

[0003] In addition, existing systems face multiple challenges, including large environmental differences, lack of cross-meter comparison capabilities, and difficulty in updating models. Differences in meter models and load types within the same area make it difficult to adapt to a unified threshold, which can easily lead to false alarms or missed alarms. Due to the lack of a group comparison mechanism, the system cannot identify "behavioral islands" where a meter deviates from the group's collaborative behavior, thus missing early warnings such as hardware degradation or communication failures. Even with the introduction of machine learning models, the high cost of training and maintaining each meter independently, as well as the inherent defect that the models cannot be automatically updated with environmental changes, severely restrict their practical application in complex industrial environments. Summary of the Invention

[0004] In view of the above-mentioned existing problems, the present invention provides a method, system, device and medium for monitoring the status of electricity meters based on behavioral baselines, in order to solve the problems of fragmented monitoring modes, inconsistent thresholds due to large environmental differences, lack of cross-meter comparison capability and difficulty in model updating in the prior art.

[0005] To address the aforementioned technical issues, a meter status monitoring method based on behavioral baselines is proposed, including: The system collects time-series data of operating parameters of smart meters within a preset monitoring area, performs joint processing on the time-series data, and extracts multi-dimensional features of the area. Based on the multi-dimensional features of the area under historical normal conditions, it establishes a regional behavior baseline through unsupervised learning, obtains the multi-dimensional features of the area at the current monitoring time, and generates a regional consistency index by calculating the matching degree with the regional behavior baseline. It identifies abnormal meters based on individual deviations and triggers the update and maintenance of the regional behavior baseline based on the regional consistency index and a preset period.

[0006] As a preferred embodiment of the meter status monitoring method based on behavioral baseline described in this invention, the joint processing includes: cleaning the time-series data of operating parameters, removing outliers and filling in missing values, aligning the cleaned data according to a unified time base, and dividing it into fixed-length time windows. For the data within each time window, feature extraction is performed in parallel, and the features of all meters within the same time window are combined to obtain the regional feature matrix of the current window.

[0007] As a preferred embodiment of the meter status monitoring method based on behavioral baseline described in this invention, the parallel execution feature extraction includes: calculating statistical features reflecting the load level and volatility of the meter itself, analyzing the periodic features existing in the meter behavior, capturing the transient features in the meter behavior, statistically analyzing the event frequency features of preset type alarm events, and calculating the correlation strength features between the behavior sequences of two meters.

[0008] As a preferred embodiment of the meter status monitoring method based on behavioral baseline described in this invention, the establishment of the regional behavioral baseline includes: collecting regional feature matrices of historical time windows to form a baseline training dataset; performing unsupervised learning on the feature vectors in the baseline training dataset; and dividing different cluster centers, each cluster center representing a typical regional behavioral pattern. For each cluster center learned, a distance discrimination threshold is determined based on the distribution of feature vectors to define the data range belonging to the current pattern. All cluster centers and distance discrimination thresholds together constitute the regional behavior baseline.

[0009] As a preferred embodiment of the meter status monitoring method based on behavioral baseline described in this invention, the generation of the regional consistency index includes: calculating the Mahalanobis distance to all pattern centers in the RBB for the feature vector of each meter at the current moment, and taking the minimum value as the individual deviation of the current meter; when the current deviation exceeds the preset comprehensive threshold, i.e. the maximum value of the threshold of all pattern centers, the current meter behavior is determined to be abnormal. The formula for calculating individual deviation is expressed as: in, Let be the individual deviation of meter i at the current time t. For the minimum value operator, For each pattern in the regional behavior baseline, For electricity meter index; The regional consistency index is calculated by quantifying the deviation of all individual meters within the region as a ratio of the mean to the maximum deviation. The formula is as follows: in, Let be the regional consistency index at time t. This represents the total number of smart meters in the area.

[0010] As a preferred embodiment of the meter status monitoring method based on behavioral baseline described in this invention, the method for identifying abnormal meters includes: calculating the cosine similarity of the feature vectors of two meters, generating a similarity matrix and visualizing it through a heatmap; meters in rows or columns with significantly different colors in the heatmap are considered behavioral islands and require key investigation; and converting individual deviations into standardized health scores and mapping them to the 0 to 1 range using a formula. The formula for calculating cosine similarity is expressed as: in, Let be the cosine similarity between meters i and j. Let i be the feature vector of meter i at the current moment. Let j be the feature vector of meter j at the current moment. For vectors norm, For vectors The norm; Set a tiered threshold: a health score of 0.8 or higher is considered healthy, a score between 0.6 and 0.8 is considered a warning, and a score less than 0.6 is considered abnormal. The health score is displayed on the cloud platform as a health score matrix and historical trend curve, providing an operational view. The formula for calculating health is expressed as: in, Let i be the health score of meter i at time t. The threshold is used for comprehensive judgment.

[0011] As a preferred embodiment of the meter status monitoring method based on behavioral baseline described in this invention, the updating and maintenance of the regional behavioral baseline includes designing two adaptive update mechanisms: a parameter progressive update mechanism and a model reconstruction triggering mechanism. The parameter incremental update mechanism includes tracking the long-term gradual trend of the load, automatically triggering the update cycle at the preset time. At the end of each cycle, the system calculates the average vector of regional features within the latest time window and uses the exponentially weighted moving average method, combined with the learning rate, to adjust the mean and covariance parameters of the baseline. The model reconstruction triggering mechanism is triggered when the regional consistency index of five consecutive time windows is lower than the preset triggering threshold. The system will determine that the current baseline has failed and automatically start the retraining process, collect one week of valid data, re-execute cluster analysis, and learn the latest number of patterns, parameter set and discrimination boundary. The update formula is expressed as: in, For the first The mean vector of the current pattern updated in the regional behavior baseline after each update cycle. This is the original mean vector of the current pattern at the end of the previous update cycle. For learning rate, In the first The average vector of all valid feature vectors in the latest time window within each update period. For the first The covariance matrix of the current mode after each update cycle. This is the original covariance matrix of the current mode at the end of the previous update cycle. The current observed eigenvectors used to update the covariance. The current observed feature vector Compared with the current updated mean The outer product of the deviation vectors, To update the periodic index; Once the baseline model of the source region's regional behavior is verified and determined to be stable and accurate, the cloud platform will encapsulate the GMM parameter set. When a target region with similar production characteristics needs to be deployed, the cloud directly sends the encapsulated parameter set to the edge nodes of the target region. The edge nodes use the received parameter set to initialize the local RBB model. After the model is put into operation, it will be adjusted locally by combining local data through an adaptive update mechanism.

[0012] The beneficial effects of this preferred technical solution are as follows: by calculating the individual deviation degree, the system can accurately identify the electricity meter that deviates from the normal mode, realize the location of the abnormal point, and evaluate the degree of coordination of the behavior of all electricity meters in the entire area through the regional consistency index. When the index drops significantly, even if the deviation degree of a single electricity meter does not exceed the threshold, it can still give an early warning of a large-scale coordination anomaly, so that maintenance personnel can not only quickly locate the source of the fault, but also grasp the overall health of the system, avoiding one-sided judgment.

[0013] As a preferred embodiment of the meter status monitoring system based on behavioral baselines described in this invention, it is characterized by including a data acquisition and preprocessing module, a multi-dimensional feature extraction module, a regional behavioral baseline modeling module, an anomaly detection and status assessment module, and a baseline adaptive update and visualization module.

[0014] The data acquisition and preprocessing module is used to collect multi-dimensional operating parameters in real time, remove outliers identified by statistical methods, imput missing values, synchronize all meter data to a unified time grid, and divide it into sliding time windows of fixed length.

[0015] The multi-dimensional feature extraction module is used to extract comprehensive features from the data within each time window. It extracts five types of features through parallel computing, combines the features of all electricity meters into a regional feature matrix, and transforms the original electrical quantity data into a feature representation containing behavioral semantics.

[0016] The regional behavior baseline modeling module is used to perform probabilistic clustering of the historical regional feature matrix using a Gaussian mixture model, identify behavior patterns that represent different typical working conditions, estimate the probability weight, mean vector and covariance matrix for each pattern, and set a discrimination threshold for each pattern based on Mahalanobis distance and chi-square distribution to obtain the regional behavior baseline.

[0017] The anomaly detection and status assessment module is used to calculate the Mahalanobis distance from the feature vector of each meter to all baseline patterns, take the minimum value as the individual deviation of the meter, compare it with the threshold to determine single-point anomalies, and calculate the regional consistency index by combining the deviations of all meters in the region. By calculating and visualizing the cosine similarity matrix of features between meters, the deviation is linearly mapped to a health score, and the levels of health, warning, and anomaly are divided.

[0018] The baseline adaptive update and visualization module is used to perform baseline adaptive updates through a dual-mode adaptive mechanism, generate a brand-new baseline model, provide a health matrix diagram and a historical trend curve of health on the cloud platform, transform the analysis results into an operation and maintenance view, support predictive maintenance decisions, and migrate the baseline model between the same areas.

[0019] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a method for monitoring the status of an electricity meter based on a behavioral baseline.

[0020] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for monitoring the state of an electricity meter based on a behavioral baseline.

[0021] The beneficial effects of this invention are as follows: By collecting and preprocessing the operating sequence data of multiple meters within a region, this invention lays the data foundation for cross-device comparison. It extracts multi-dimensional features covering the time domain, frequency domain, and the strength of correlation between meters in parallel, constructing a behavioral fingerprint that comprehensively characterizes individual behavior and group relationships. Furthermore, it utilizes unsupervised learning to automatically summarize multimodal regional behavioral baselines, accurately describing complex and ever-changing normal operating conditions in a probabilistic manner, replacing rigid fixed thresholds. In the detection phase, it calculates the correlation-sensitive Mahalanobis distance to obtain individual deviations and aggregates them to generate a regional consistency index, simultaneously achieving accurate single-point anomaly localization and systematic regional risk warning. Similarity heatmaps and standardized health visualization improve the efficiency of operation and maintenance diagnosis. By combining a progressive update and a mutation-triggered model reconstruction mechanism, it ensures the long-term adaptability of the baseline model to slow drift and drastic changes. The cloud-supported model migration strategy enables rapid knowledge reuse and large-scale, economical deployment of the system. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 The above is a flowchart of a method for monitoring the status of electricity meters based on behavioral baselines, provided as an embodiment of the present invention.

[0024] Figure 2 A system flowchart of a meter status monitoring system based on behavioral baselines provided in one embodiment of the present invention. Detailed Implementation

[0025] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0026] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for monitoring the status of electricity meters based on behavioral baselines is provided, comprising: S100: Collects time-series data of operating parameters of smart meters within a preset monitoring area, and performs joint processing on the time-series data of operating parameters to extract multi-dimensional features of the area.

[0027] S200: Based on the multidimensional features of the region under historical normal conditions, a regional behavior baseline is established through unsupervised learning, and the multidimensional features of the region at the current monitoring time are obtained. By calculating the matching degree with the regional behavior baseline, a regional consistency index is generated.

[0028] S300: Identifies abnormal meters based on individual deviation and triggers updates and maintenance of the regional behavior baseline based on the regional consistency index and a preset cycle.

[0029] It should be noted that this invention, by constructing a closed-loop system from data preprocessing, behavioral baseline modeling, multi-dimensional anomaly assessment to visualized operation and maintenance and adaptive evolution, has achieved a leap from isolated single-point monitoring to regional collaborative intelligent analysis. This solves the problems of rigid thresholds and lack of a group perspective in traditional methods, and significantly improves the accuracy of anomaly identification, the depth of diagnosis, and the robustness and economy of long-term model application.

[0030] Example 2, refer to Figure 1 This is a second embodiment of the present invention, which provides a method for monitoring the status of electricity meters based on behavioral baselines, including: In this embodiment of the application, step S100, collecting the timing data of the operating parameters, includes steps S101 to S104: S101: Each meter collects data synchronously according to a preset sampling period (e.g., every 15 minutes).

[0031] S102: The collected content covers multiple dimensions, including: basic electrical quantities represented by three-phase voltage and three-phase current; power and energy efficiency quantities with total active power, total reactive power and power factor as the core; power quality characterized by total harmonic distortion rate or harmonic content; and events and status quantities that record the number of occurrences and corresponding timestamps of preset discrete events such as voltage loss, current loss and communication interruption.

[0032] S103: All collected data are accompanied by precise timestamps and are uploaded in real time to the edge concentrator deployed in the monitoring area via communication protocols such as industrial fieldbus (RS-485), carrier wave (HPLC), or wireless network (LoRa).

[0033] S104: The multi-source, multi-dimensional data streams from different meters are aggregated at the edge concentrator to form a unified raw dataset with time-series labels, laying the foundation for subsequent analysis.

[0034] In an optional implementation, in step S100, the acquisition of operating parameter timing data also includes data acquisition based on a hybrid mode of event triggering and periodic acquisition. While maintaining periodic acquisition, the meter introduces an event triggering mechanism. When the instantaneous change rate of any parameter is detected to exceed a preset sensitive threshold, high-frequency acquisition is immediately started and continues for a period of time to record detailed waveform data of the current event (the entire process of voltage sag) and report it together with the event identifier.

[0035] In another optional implementation, in step S100, the collection of time-series data of operating parameters also includes the collaborative collection of non-electrical parameters. While collecting electrical parameters, the smart meter or its associated sensor module synchronously collects time-series data of environmental parameters of the cabinet, such as temperature, humidity, and cabinet door opening and closing status. The data and electrical data share the same timestamp and are uploaded together. The environmental data is also used as a set of additional operating parameters to participate in subsequent feature extraction.

[0036] Furthermore, in this embodiment of the application, in step S100, the joint processing includes steps S111~S113: S111: Data cleaning includes using statistical methods (3σ principle) to identify and remove outliers caused by transient interference; for data gaps caused by communication delays or packet loss, linear interpolation is used to fill in the gaps. The formula for linear interpolation is as follows: in, For the missing values ​​that need to be imputed at time point z, For the timestamps of missing data points that need to be interpolated and padded, Let z be the value of the nearest valid data point before the missing point. for The corresponding timestamp, Let z be the value of the nearest valid data point after the missing point z. This is the corresponding timestamp.

[0037] S112: To ensure the accuracy of cross-device analysis, the data streams of all meters are timestamped and resampled based on the clock of the edge concentrator to achieve synchronous time alignment and unify the data onto an equally spaced time grid.

[0038] S113: Divide the continuous and aligned time series into sliding time windows of fixed length (30 minutes or 1 hour) with partial overlap; each window contains all parameter data of all meters in the current time period, forming a three-dimensional data block of "meter × parameter × time point".

[0039] In an optional implementation, in step S100, the joint processing further includes robust cleaning and spline interpolation based on quantiles (IQR), identifying outliers using the interquartile range method, calculating the upper quartile (Q3) and lower quartile (Q1) of the data sequence, defining data points outside the outlier threshold, and filling in missing values ​​using cubic spline interpolation to generate a smoother curve that better fits the overall trend of the data.

[0040] In another optional implementation, in step S100, the joint processing may further include cleaning and filling based on the collaborative data of neighboring meters. When a meter's data point is missing or abnormal, not only its own historical data is considered, but also the data of the physically or logically nearest and normally functioning remaining meters at the same time are introduced as a reference. By calculating the historical correlation model (such as linear regression relationship) between the meter and neighboring meters on the current parameters, the reasonable value at the current time is predicted or corrected for filling or replacement.

[0041] Furthermore, in this embodiment of the application, in step S100, the extraction of multidimensional features of the region includes steps S121 to S125: S121: The time-domain feature set consists of calculating the mean to characterize the average load for the active power sequence, calculating the variance to measure the fluctuation intensity, and using skewness and kurtosis to analyze the symmetry and steepness of the data distribution. The stability coefficient is calculated to quantify relative volatility and eliminate the influence of absolute dimensions. The communication stability of the meter is monitored by analyzing the mean and variance of the data message reporting time intervals, forming a time-domain feature set. The formula for calculating the stability coefficient is expressed as follows: in, Let i be the active power sequence of meter i. Let be the mean of the power sequence of meter i within the time window. Let be the stability coefficient of meter i. Let i be the standard deviation of the power series of meter i. This is the electricity meter index.

[0042] S122: Perform Discrete Fourier Transform (DFT) on the power sequence within the window to transform it from the time domain to the frequency domain, obtain the spectral representation, and extract the amplitude of specific periodic components that are strongly correlated with the production rhythm, such as the amplitude corresponding to the 24-hour cycle. The formula for obtaining the spectrum representation is as follows: in, Let F be the Fourier transform coefficients of the power sequence of meter i at frequency F. Let be the power value of meter i at time t. This represents the total number of sampling points within the time window. For frequency components, For time indexing, For rotation factor, The imaginary unit, For electricity meter index; To quantify the significance of the most dominant periodic pattern, the dominant frequency energy ratio is calculated, which is the proportion of the energy of the dominant frequency component to the total energy of the signal, effectively identifying the dominant periodic behavior in the data. The formula for calculating the power-to-frequency ratio is expressed as: in, The main frequency energy ratio of meter i. The Fourier coefficients are at the dominant frequency, i.e., the frequency with the highest energy. The total energy of the frequency components. This is the electricity meter index.

[0043] S123: Employing wavelet packet decomposition technology, the power signal is finely decomposed into multiple different frequency bands. By calculating the energy of each major frequency band, it identifies transient impacts, voltage dips, or brief but critical dynamic events such as equipment start-up and shutdown that are difficult to detect with traditional Fourier transform.

[0044] S124: Extracting event statistical features includes directly counting the frequency of occurrence of various predefined events (such as undervoltage and overcurrent) within the time window, and calculating the entropy value of the event occurrence time distribution.

[0045] Entropy assesses randomness by measuring the probability distribution of events occurring in different sub-periods. High entropy values ​​indicate that events are randomly distributed over time, while low entropy values ​​suggest that events are concentrated in a specific period. The formula for calculating the entropy of the time distribution is expressed as: in, Let i be the entropy value of the event time distribution of meter i. This represents the probability distribution of an event occurring in different sub-time periods within a time window. For probability The natural logarithm, This is the electricity meter index.

[0046] S125: The linear correlation of load changes is measured by calculating the Pearson cross-correlation coefficient of the active power sequence between any two meters i and j in the region. The coefficient ranges from [-1, 1]. The closer the absolute value is to 1, the stronger the synergy of the load changes of the two meters. The formula for calculating the Pearson cross-correlation coefficient is as follows: in, Let Pearson cross-correlation coefficient be the number between meter i and meter j. Let be the covariance of the sequences of meters i and j. Let i be the standard deviation of the power series of meter i. Let j be the standard deviation of the power series of meter j. and For electricity meter index; After extracting all individual features and cross-correlation features, the multi-dimensional feature vectors of each meter are combined to form the regional feature matrix of the current time window, which reflects the overall electricity consumption behavior pattern of the entire region at the current moment. The region feature matrix is ​​represented as follows: in, The feature matrix of the region within time window t1, This is the multidimensional feature vector of the electricity meter within time window t1. This is the matrix transpose symbol.

[0047] In an optional implementation, in step S121, the construction of the time-domain feature set further includes, before calculating the global statistics, dividing the power sequence within the time window into "rising segments", "stable segments" and "falling segments" according to its local extreme points, and calculating the duration, average slope (rate of change) and variance within each segment respectively. The final features include the number of segments of each type, average duration and maximum slope.

[0048] In another optional implementation, in step S121, the construction of the time-domain feature set may further include: segmenting and aggregating the power time series, dividing the data points within the window into several sub-segments on an average basis, using the mean of the sub-segments to represent the current segment, symbolizing the mean sequence, that is, mapping it to a limited number of symbols according to the numerical range it falls into, and using the frequency of a specific pattern appearing in the symbol sequence as a feature.

[0049] In step S200, establishing the regional behavior baseline includes steps S201 to S204: S201: Collect the regional feature matrices corresponding to all time windows marked as operating normally within a month to form the baseline training dataset.

[0050] S202: Perform unsupervised learning on the feature vectors in the baseline training dataset to discover and divide multiple distinct cluster centers, each representing a typical regional behavior pattern.

[0051] S203: After the model training is completed, a discrimination boundary based on Mahalanobis distance is defined for each learned pattern k to determine the normal range. The Mahalanobis distance from the data point to the center of each pattern is calculated and compared with the threshold determined from the chi-square distribution based on the preset confidence level (95%) to determine whether the data conforms to the current pattern. The formula for calculating Mahalanobis distance is expressed as: in, Let be the multidimensional feature vector of meter i. Let be the mean vector of the k-th Gaussian component, i.e., a point in the feature space, representing the center of the current behavior pattern k. Let be the inverse of the covariance matrix of the k-th mode. Here is the matrix transpose symbol, and here is the eigenvector. The Mahalanobis distance to the center of the k-th pattern.

[0052] S204: The regional behavior baseline is defined as a parameterized set RBB containing all mode parameters and thresholds, where each component corresponds to an interpretable typical regional operating condition ("night shift standby mode" or "day shift production mode"). The parameterized set RBB is represented as follows: in, The mixing weights for the k-th Gaussian component. Let be the mean vector of the k-th Gaussian component. Let be the covariance matrix of the k-th Gaussian component. It is a parameterized set.

[0053] Furthermore, in this embodiment of the application, in step S202, the unsupervised learning includes modeling the feature vectors in the dataset using a Gaussian Mixture Model (GMM), specifically including steps A1 to A3: A1: GMM assumes that the data is a mixture of K multivariate Gaussian distributions, and each Gaussian component (defined by the mean, covariance matrix and mixture weights) represents a typical regional behavior pattern.

[0054] A2: Using the Expectation-Maximization (EM) algorithm, the optimal model parameters are automatically learned from the training data. The model probability density function is expressed as: in, For feature vectors The probability density of occurrence Let be the multidimensional feature vector of meter i. This represents the number of Gaussian components, i.e., the total number of regional behavioral patterns. The mixing weights for the k-th Gaussian component. Let be the probability density function of a multivariate Gaussian distribution. Let be the mean vector of the k-th Gaussian component, i.e., a point in the feature space, representing the center of the current behavior pattern k. Let be the covariance matrix of the k-th Gaussian component.

[0055] A3: The mean serves as the center point of the behavioral pattern in the feature space, while the covariance matrix describes the shape of the current pattern and the correlation between the features.

[0056] In an optional implementation, in step S202, the unsupervised learning further includes calculating the local density of each feature vector, i.e., the number of remaining points within a certain distance range, and the minimum distance between each feature vector and a point with higher density, and selecting a point with high local density and a large distance from the high-density point as the cluster center, and assigning the remaining points to the category of the nearest point with a higher density than itself.

[0057] In another optional implementation, in step S202, the unsupervised learning may further include training a deep autoencoder to compress high-dimensional feature vectors into low-dimensional latent space vectors, a decoder to reconstruct the original features from the latent space vectors, training to minimize the reconstruction error, and after training, using the encoder to map all historical feature vectors to the latent space, and using the K-means algorithm to cluster on the low-dimensional, redundancy-free latent space representation to obtain cluster centers.

[0058] Furthermore, in step S200, the generation of the regional consistency index includes steps S211-S212: S211: For the feature vector of each meter at the current moment, calculate the Mahalanobis distance to all pattern centers in the RBB, and take the minimum value as the individual deviation of the current meter. When the current deviation exceeds the preset comprehensive threshold, i.e. the maximum value of the threshold of all pattern centers, the current meter behavior is determined to be abnormal. The formula for calculating individual deviation is expressed as: in, Let be the individual deviation of meter i at the current time t. For the minimum value operator, For each pattern in the regional behavior baseline, This is the electricity meter index.

[0059] S212: Calculate the regional consistency index, which is quantified by the ratio of the mean to the maximum deviation of all individual meters within the region. The formula is as follows: in, Let be the regional consistency index at time t. This represents the total number of smart meters in the area. The closer the value is to 1, the more consistent the behavior of all meters in the area is with the baseline. A significant decrease in the value indicates a widespread anomaly in synergy.

[0060] In step S300, the identification of abnormal meters includes steps S301 to S303: S301: Calculate the cosine similarity of the feature vectors of two meters, generate a similarity matrix and visualize it through a heatmap. The meters in rows or columns with significantly different colors in the heatmap are considered "behavioral islands" and need to be investigated in detail. The formula for calculating cosine similarity is expressed as: in, Let be the cosine similarity between meters i and j. Let i be the feature vector of meter i at the current moment. Let j be the feature vector of meter j at the current moment. For vectors norm, For vectors The norm of .

[0061] S302: Convert individual deviations into standardized health scores, mapping them to a 0-1 range using a formula; The formula for calculating health is expressed as: in, Let i be the health score of meter i at time t. The threshold is used for comprehensive judgment.

[0062] S303: Set the grading threshold, when For health, As an early warning, If an anomaly is detected, it will be displayed on the cloud platform as a health matrix chart and historical trend curve, providing an operational view.

[0063] Furthermore, in step S300, the updating and maintenance of the regional behavior baseline includes designing two adaptive update mechanisms: a parameter progressive update mechanism and a model reconstruction triggering mechanism, specifically including steps S311~S313: S311: The parameter gradual update mechanism includes tracking the long-term gradual trend of the load, automatically triggering the update cycle at the preset time. At the end of each cycle, the system calculates the average vector of regional features within the latest time window and uses the exponentially weighted moving average method, combined with the learning rate, to adjust the mean and covariance parameters of the baseline. The update formula is expressed as: in, For the first The mean vector of the current pattern updated in the regional behavior baseline after each update cycle. This is the original mean vector of the current pattern at the end of the previous update cycle. For learning rate, In the first The average vector of all valid feature vectors in the latest time window within each update period. For the first The covariance matrix of the current mode after each update cycle. This is the original covariance matrix of the current mode at the end of the previous update cycle. The current observed eigenvectors used to update the covariance. The current observed feature vector Compared with the current updated mean The outer product of the deviation vectors, To update the periodic index.

[0064] S312: Model Reconstruction Trigger Mechanism. When the regional consistency index of five consecutive time windows is lower than the preset trigger threshold, the system will determine that the current baseline has failed and automatically start the retraining process, collect one week of valid data, re-execute cluster analysis, and learn the latest number of patterns, parameter set and discrimination boundary.

[0065] S313: Once the regional behavior baseline model of the source region is verified and determined to be stable and accurate, the cloud platform will encapsulate the GMM parameter set. When a target region with similar production characteristics needs to be deployed, the cloud directly sends the encapsulated parameter set to the edge node of the target region. The edge node uses the received parameter set to initialize the local RBB model. After the model is put into operation, it makes localized adjustments by combining local data through an adaptive update mechanism.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0067] Example 3, referring to Figure 2The third embodiment of the present invention provides a meter status monitoring system based on behavioral baselines, including a data acquisition and preprocessing module, a multi-dimensional feature extraction module, a regional behavioral baseline modeling module, an anomaly detection and status assessment module, and a baseline adaptive update and visualization module.

[0068] The data acquisition and preprocessing module is used to collect multi-dimensional operating parameters in real time, remove outliers identified by statistical methods, imput missing values, synchronize all meter data to a unified time grid, and divide it into sliding time windows of fixed length.

[0069] The multi-dimensional feature extraction module is used to extract comprehensive features from the data within each time window. It extracts five types of features through parallel computing, combines the features of all electricity meters into a regional feature matrix, and transforms the original electrical quantity data into a feature representation containing behavioral semantics.

[0070] The regional behavior baseline modeling module is used to perform probabilistic clustering of the historical regional feature matrix using a Gaussian mixture model, identify behavior patterns that represent different typical working conditions, estimate the probability weight, mean vector and covariance matrix for each pattern, and set a discrimination threshold for each pattern based on Mahalanobis distance and chi-square distribution to obtain the regional behavior baseline.

[0071] The anomaly detection and status assessment module is used to calculate the Mahalanobis distance from the feature vector of each meter to all baseline patterns, take the minimum value as the individual deviation of the meter, compare it with the threshold to determine single-point anomalies, and calculate the regional consistency index by combining the deviations of all meters in the region. By calculating and visualizing the cosine similarity matrix of features between meters, the deviation is linearly mapped to a health score, and the levels of health, warning, and anomaly are divided.

[0072] The baseline adaptive update and visualization module is used to perform baseline adaptive updates through a dual-mode adaptive mechanism, generate a brand-new baseline model, provide a health matrix diagram and a historical trend curve of health on the cloud platform, transform the analysis results into an operation and maintenance view, support predictive maintenance decisions, and migrate the baseline model between the same areas.

[0073] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0074] Example 4, the fourth embodiment of the present invention, differs from the previous three embodiments in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0075] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0076] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0077] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

Claims

1. A method for monitoring the status of electricity meters based on behavioral baselines, characterized in that: include, Collect time-series data of operating parameters of smart meters within a preset monitoring area, and perform joint processing on the time-series data of operating parameters to extract multi-dimensional features of the area; Based on the regional multidimensional features under historical normal conditions, a regional behavior baseline is established through unsupervised learning, and the regional multidimensional features at the current monitoring time are obtained. By calculating the matching degree with the regional behavior baseline, a regional consistency index is generated. Abnormal meters are identified based on individual deviations, and the regional behavior baseline is updated and maintained based on the regional consistency index and a preset cycle.

2. The meter status monitoring method based on behavioral baseline as described in claim 1, characterized in that: The joint processing includes cleaning the time-series data of the operating parameters, removing outliers and filling in missing values, aligning the cleaned data according to a unified time base, and dividing it into fixed-length time windows. For the data within each time window, feature extraction is performed in parallel, and the features of all meters within the same time window are combined to obtain the regional feature matrix of the current window.

3. The meter status monitoring method based on behavioral baseline as described in claim 2, characterized in that: The parallel execution feature extraction includes calculating statistical features reflecting the load level and volatility of the electricity meter itself, analyzing the periodic features in the electricity meter behavior, capturing the transient features in the electricity meter behavior, statistically analyzing the event frequency features of preset type alarm events, and calculating the correlation strength features between the behavior sequences of two electricity meters.

4. The meter status monitoring method based on behavioral baseline as described in claim 3, characterized in that: The establishment of the regional behavior baseline includes: collecting regional feature matrices from historical time windows to form a baseline training dataset; performing unsupervised learning on the feature vectors in the baseline training dataset; and dividing the data into different cluster centers, each cluster center representing a typical regional behavior pattern. For each cluster center learned, a distance discrimination threshold is determined based on the distribution of feature vectors to define the data range belonging to the current pattern. All cluster centers and distance discrimination thresholds together constitute the regional behavior baseline.

5. The meter status monitoring method based on behavioral baseline as described in claim 4, characterized in that: The generated regional consistency index includes calculating the Mahalanobis distance to all pattern centers in the RBB for the feature vector of each meter at the current time, and taking the minimum value as the individual deviation of the current meter. When the current deviation exceeds the preset comprehensive threshold, i.e. the maximum value of the threshold of all pattern centers, the current meter behavior is determined to be abnormal. The formula for calculating individual deviation is expressed as: in, Let be the individual deviation of meter i at the current time t. For the minimum value operator, For each pattern in the regional behavior baseline, For electricity meter index; The regional consistency index is calculated by quantifying the deviation of all individual meters within the region as a ratio of the mean to the maximum deviation. The formula is as follows: in, Let be the regional consistency index at time t. This represents the total number of smart meters in the area.

6. The meter status monitoring method based on behavioral baseline as described in claim 5, characterized in that: The method for identifying abnormal electricity meters includes calculating the cosine similarity of the feature vectors of two electricity meters, generating a similarity matrix and visualizing it through a heatmap. Electricity meters in rows or columns with significantly different colors in the heatmap are considered behavioral islands and need to be investigated in detail. Individual deviations are converted into standardized health scores and mapped to the 0 to 1 range using a formula. The formula for calculating cosine similarity is expressed as: in, Let be the cosine similarity between meters i and j. Let i be the feature vector of meter i at the current moment. Let j be the feature vector of meter j at the current moment. For vectors norm, For vectors The norm; Set a tiered threshold: a health score of 0.8 or higher is considered healthy, a score between 0.6 and 0.8 is considered a warning, and a score less than 0.6 is considered abnormal. The health score is displayed on the cloud platform as a health score matrix and historical trend curve, providing an operational view. The formula for calculating health is expressed as: in, Let i be the health score of meter i at time t. The threshold is used for comprehensive judgment.

7. The meter status monitoring method based on behavioral baseline as described in claim 6, characterized in that: The updating and maintenance of the regional behavior baseline includes designing two adaptive update mechanisms: a parameter progressive update mechanism and a model reconstruction triggering mechanism. The parameter incremental update mechanism includes tracking the long-term gradual trend of the load, automatically triggering the update cycle at the preset time. At the end of each cycle, the system calculates the average vector of regional features within the latest time window and uses the exponentially weighted moving average method, combined with the learning rate, to adjust the mean and covariance parameters of the baseline. The model reconstruction triggering mechanism is triggered when the regional consistency index of five consecutive time windows is lower than the preset triggering threshold. The system will determine that the current baseline has failed and automatically start the retraining process, collect one week of valid data, re-execute cluster analysis, and learn the latest number of patterns, parameter set and discrimination boundary. The update formula is expressed as: in, For the first The mean vector of the current pattern updated in the regional behavior baseline after each update cycle. This is the original mean vector of the current pattern at the end of the previous update cycle. For learning rate, In the first The average vector of all valid feature vectors in the latest time window within each update period. For the first The covariance matrix of the current mode after each update cycle. This is the original covariance matrix of the current mode at the end of the previous update cycle. The current observed eigenvectors used to update the covariance, The current observed feature vector Compared with the current updated mean The outer product of the deviation vectors, To update the periodic index; Once the baseline model of the source region's regional behavior is verified and determined to be stable and accurate, the cloud platform will encapsulate the GMM parameter set. When a target region with similar production characteristics needs to be deployed, the cloud directly sends the encapsulated parameter set to the edge nodes of the target region. The edge nodes use the received parameter set to initialize the local RBB model. After the model is put into operation, it will be adjusted locally by combining local data through an adaptive update mechanism.

8. A meter status monitoring system based on behavioral baselines, employing the meter status monitoring method based on behavioral baselines as described in any one of claims 1 to 7, characterized in that, It includes a data acquisition and preprocessing module, a multi-dimensional feature extraction module, a regional behavior baseline modeling module, an anomaly detection and state assessment module, and a baseline adaptive update and visualization module; The data acquisition and preprocessing module is used to collect multi-dimensional operating parameters in real time, remove outliers identified by statistical methods, imput missing values, synchronize all meter data to a unified time grid, and divide it into sliding time windows of fixed length. The multi-dimensional feature extraction module is used to extract comprehensive features from the data within each time window, extract five types of features through parallel computing, combine the features of all electricity meters into a regional feature matrix, and transform the original electrical quantity data into a feature representation containing behavioral semantics. The regional behavior baseline modeling module is used to perform probabilistic clustering of the historical regional feature matrix using a Gaussian mixture model, identify behavior patterns that represent different typical working conditions, estimate the probability weight, mean vector and covariance matrix for each pattern, and set a discrimination threshold for each pattern based on Mahalanobis distance and chi-square distribution to obtain the regional behavior baseline. The anomaly detection and status assessment module is used to calculate the Mahalanobis distance from each meter's feature vector to all baseline patterns, take the minimum value as the individual deviation of the meter, compare it with the threshold to determine single-point anomalies, and integrate the deviations of all meters in the region to calculate the regional consistency index. By calculating and visualizing the cosine similarity matrix of features between meters, the deviation is linearly mapped to a health score, and the levels of health, warning, and anomaly are divided. The baseline adaptive update and visualization module is used to perform baseline adaptive updates through a dual-mode adaptive mechanism, generate a brand-new baseline model, provide a health matrix diagram and a historical trend curve of health on the cloud platform, transform the analysis results into an operation and maintenance view, support predictive maintenance decisions, and migrate the baseline model between the same areas.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the meter status monitoring method based on behavioral baseline as described in 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 meter status monitoring method based on behavioral baseline as described in any one of claims 1 to 7.

Images