Current and voltage monitoring method and system based on metering device

Abstract

The invention belongs to the technical field of current and voltage monitoring, and discloses a current and voltage monitoring method and system based on a metering device. The method comprises the following steps: acquiring real-time multi-modal monitoring data of target monitoring equipment by using a corresponding metering device according to a preset multi-modal data feature engineering space, and transmitting the real-time multi-modal monitoring data to an edge computing gateway; an edge computing gateway is used to preprocess the real-time multi-modal monitoring data, and the obtained preprocessed real-time multi-modal monitoring data is uploaded to a cloud data center; and in the cloud data center, current and voltage monitoring is carried out on the preprocessed real-time multi-modal monitoring data of the plurality of target monitoring devices, and a generated real-time abnormity maintenance strategy is sent to the corresponding target monitoring device. According to the invention, the problems of single monitoring dimension, insufficient real-time performance, low intelligent degree and poor adaptive ability in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of current and voltage monitoring, and particularly relates to a current and voltage monitoring method and system based on a measuring device. BACKGROUND

[0002] On-line monitoring of current and voltage data of electrical equipment used by enterprises, factories and workshops, timely discovery of abnormal conditions of target monitoring equipment and maintenance and early warning are important measures to ensure normal and safe production and work. With the continuous expansion and increasing complexity of the scale of enterprises, factories and workshops, higher requirements are put forward for real-time monitoring and abnormal processing of current and voltage.

[0003] The prior art has many defects, including: 1) Single monitoring dimension: the prior art mainly focuses on single parameters such as current and voltage, lacks comprehensive consideration of device state parameters (such as temperature, humidity, vibration, etc.) and environmental parameters (such as weather, load, etc.), and is difficult to comprehensively evaluate the running state and health level of the equipment; 2) Insufficient real-time performance: traditional periodic inspection methods are inefficient and difficult to discover and handle abnormalities in real time. Even if real-time monitoring is used, it often only focuses on the overall running state of the electrical system, and the abnormal detection and diagnosis capability of a single device is limited; 3) Low degree of intelligence: rule-based expert systems and traditional machine learning algorithms are difficult to cope with complex and unknown fault conditions, and the updating and maintenance of the rule base require a lot of manual intervention. These methods have a low degree of intelligence and are difficult to achieve adaptive abnormal detection and diagnosis; 4) Poor adaptability: existing abnormal maintenance strategy generation methods often rely on human experience and lack intelligence and adaptability, making it difficult to cope with diverse operating scenarios and device types. SUMMARY

[0004] In order to solve the problems of single monitoring dimension, insufficient real-time performance, low degree of intelligence and poor adaptability of the prior art, the application aims to provide a current and voltage monitoring method and system based on a measuring device.

[0005] The technical solution adopted by the application is: A current and voltage monitoring method based on a measuring device, comprising the following steps: According to the pre-set multi-modal data feature engineering space, using the corresponding measuring device, collecting real-time multi-modal monitoring data of the target monitoring equipment, and transmitting the data to the edge computing gateway; Using the edge computing gateway, pre-processing the real-time multi-modal monitoring data, and uploading the obtained pre-processed real-time multi-modal monitoring data to the cloud data center; The current-voltage monitoring of the pre-processed real-time multi-modal monitoring data of a plurality of target monitoring devices is performed in a cloud data center, and the generated real-time abnormal maintenance strategy is sent to the corresponding target monitoring device.

[0006] Further, the multi-modal data feature engineering space includes current-voltage monitoring input features of the target monitoring device, device state monitoring input features of the target monitoring device, and environment monitoring input features of the working environment. The real-time multi-modal monitoring data includes real-time current-voltage monitoring data of the target monitoring device, real-time device state monitoring data of the target monitoring device, and real-time environment monitoring data of the working environment.

[0007] Further, the edge computing gateway is used to pre-process the real-time multi-modal monitoring data, and the pre-processed real-time multi-modal monitoring data is uploaded to the cloud data center, including the following steps: According to the real-time multi-modal monitoring data, the edge computing gateway is used to adjust the dynamic sampling rate of the metering device, the adjusted dynamic sampling rate is returned to the corresponding metering device, and the real-time multi-modal monitoring data is updated. The updated real-time multi-modal monitoring data is processed by spatio-temporal alignment to obtain real-time multi-modal monitoring data after spatio-temporal alignment. The real-time multi-modal monitoring data after spatio-temporal alignment is enhanced to obtain pre-processed real-time multi-modal monitoring data, which is uploaded to the cloud data center.

[0008] Further, according to the real-time multi-modal monitoring data, the edge computing gateway is used to adjust the dynamic sampling rate of the metering device, the adjusted dynamic sampling rate is returned to the corresponding metering device, and the real-time multi-modal monitoring data is updated, including the following steps: According to the real-time multi-modal monitoring data, the device load prediction model pre-trained in the edge computing gateway is used for device load prediction to obtain real-time device load prediction results. According to the real-time device load prediction results, the dynamic sampling rate of the metering device is adjusted using the three-level sampling adjustment rules to obtain the corresponding adjusted dynamic sampling rate. The adjusted dynamic sampling rate of the metering device is returned to the corresponding metering device, and real-time data acquisition is performed again to obtain updated real-time multi-modal monitoring data.

[0009] Further, the updated real-time multi-modal monitoring data is processed by spatio-temporal alignment to obtain real-time multi-modal monitoring data after spatio-temporal alignment, including the following steps: The timestamps of the real-time current-voltage monitoring data, real-time device state monitoring data, and real-time environment monitoring data of the updated real-time multi-modal monitoring data are standardized to obtain corresponding standardized timestamps. Synchronize the standardized time stamps of the real-time current and voltage monitoring data, the real-time equipment state monitoring data, and the real-time environmental monitoring data to obtain corresponding synchronized time stamps; Collect the recording position data corresponding to each modality in the real-time multi-modal monitoring data, and convert all the recording position data to a unified coordinate system; According to the synchronized time stamps of the real-time multi-modal monitoring data and the adjusted dynamic sampling rates of the corresponding measurement devices of each modality, use the linear interpolation method to perform spatio-temporal alignment processing on the real-time current and voltage monitoring data, the real-time equipment state monitoring data, and the real-time environmental monitoring data to obtain initial spatio-temporally aligned real-time multi-modal monitoring data; Use an error compensation algorithm to fine-tune the initial spatio-temporally aligned real-time multi-modal monitoring data to obtain final spatio-temporally aligned real-time multi-modal monitoring data.

[0010] Further, perform feature enhancement on the spatio-temporally aligned real-time multi-modal monitoring data to obtain preprocessed real-time multi-modal monitoring data, and upload it to the cloud data center, including the following steps: Perform normalization processing on the spatio-temporally aligned real-time multi-modal monitoring data to obtain normalized real-time multi-modal monitoring data; Use a wavelet transform algorithm to transform the normalized real-time multi-modal monitoring data to obtain corresponding real-time multi-scale features; Use a Fourier transform algorithm to transform the normalized real-time multi-modal monitoring data to obtain corresponding real-time frequency domain features; Use a principal component analysis algorithm to reduce the dimensionality of the normalized real-time multi-modal monitoring data to obtain corresponding real-time reduced dimension features; Use an autoencoder to perform feature learning and dimensionality reduction on the normalized real-time multi-modal monitoring data to obtain corresponding real-time reconstructed features; Use a short-time Fourier transform algorithm to transform the normalized real-time multi-modal monitoring data to obtain corresponding real-time time-frequency features; Perform feature fusion on the real-time multi-scale features, the real-time frequency domain features, the real-time reduced dimension features, the real-time reconstructed features, and the real-time time-frequency features to obtain corresponding preprocessed real-time multi-modal monitoring data; Through a pre-set secure encryption channel, upload the preprocessed real-time multi-modal monitoring data to the cloud data center after encryption.

[0011] Further, in the cloud data center, perform current and voltage monitoring on the preprocessed real-time multi-modal monitoring data of a plurality of target monitoring devices, and send the generated real-time abnormal maintenance strategy to the corresponding target monitoring device, including the following steps: In the cloud data center, the pre-processed real-time multi-modal monitoring data of the target monitoring devices are clustered using the ST-DBSCAN algorithm to obtain a plurality of real-time multi-modal monitoring data clusters of the same type of monitoring target; The pre-trained anomaly analysis model is used to perform anomaly analysis on each real-time multi-modal monitoring data cluster to obtain a corresponding real-time anomaly analysis result. According to the real-time anomaly analysis result, the pre-trained anomaly maintenance strategy generation model is used to generate an anomaly maintenance strategy to obtain a real-time anomaly maintenance strategy and send it to the corresponding target monitoring device.

[0012] Further, in the cloud data center, the pre-processed real-time multi-modal monitoring data of the target monitoring devices are clustered using the ST-DBSCAN algorithm to obtain a plurality of real-time multi-modal monitoring data clusters of the same type of monitoring target, including the following steps: In the cloud data center, the pre-processed real-time multi-modal monitoring data of each target monitoring device is converted into corresponding real-time data points. A dynamic computing mechanism is introduced to obtain the dynamic neighborhood radius of a plurality of real-time data points, and the spatiotemporal distance between different data points is defined. All real-time data points are traversed, and the ST-DBSCAN algorithm is used for clustering to obtain a plurality of real-time multi-modal monitoring data clusters of the same type of monitoring target.

[0013] Further, the anomaly analysis model is constructed based on the LSTM-GAT-Transformer-MLP algorithm. The anomaly maintenance strategy generation model is constructed based on the MA-DDPG-PPO-DAN algorithm.

[0014] A current and voltage monitoring system based on a metering device is used to implement a current and voltage monitoring method, which includes a cloud data center, a plurality of edge computing gateways, and a plurality of metering devices.

[0015] The beneficial effects of the present application are: This invention provides a current and voltage monitoring method and system based on a metering device. By integrating multimodal data such as current and voltage parameters, equipment status parameters, and environmental parameters, it overcomes the limitations of existing technologies that only focus on a single parameter, achieving more comprehensive and detailed monitoring of the monitoring target and improving the accuracy and reliability of current and voltage anomaly detection. An edge computing gateway is used for real-time multimodal monitoring data preprocessing, including dynamic sampling adjustment, spatiotemporal alignment, and feature enhancement, effectively reducing data transmission volume, alleviating cloud computing pressure, and improving data processing efficiency and real-time performance. The ST-DBSCAN algorithm is used for spatiotemporal clustering analysis, grouping monitoring targets with similar spatiotemporal behavior and electrical characteristics into the same data cluster, providing more refined data grouping for subsequent anomaly detection and maintenance strategy generation, improving the model's specificity and generalization ability. An anomaly analysis model is also used. Features in the time domain, frequency domain, and time-frequency domain are extracted separately, and current and voltage anomalies, equipment status anomalies, and environmental anomalies are simultaneously identified through a multi-task learning framework, significantly improving the intelligence, accuracy, and robustness of anomaly detection. An anomaly maintenance strategy generation model is used to construct regional and equipment intelligent agent layers, enabling hierarchical decision-making for regional resource scheduling and single-equipment maintenance strategy optimization. A transfer learning framework is used to quickly adapt to new equipment types and operating scenarios, generating intelligent and adaptive anomaly maintenance strategies. Data-driven anomaly detection and intelligent maintenance strategy generation reduce reliance on human experience in current and voltage monitoring and maintenance, improving the automation level and efficiency of power system operation and maintenance. Real-time monitoring, accurate anomaly detection, and intelligent maintenance strategy generation enable timely detection and handling of equipment anomalies, improving the operational reliability and maintenance efficiency of the power system and reducing operation and maintenance costs.

[0016] Other beneficial effects of the present invention will be further explained in the specific embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart of the current and voltage monitoring method based on a metering device in this invention.

[0018] Figure 2 This is a structural block diagram of the current and voltage monitoring system based on a metering device in this invention. Detailed Implementation

[0019] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1: like Figure 1 As shown, this embodiment provides a current and voltage monitoring method based on a metering device, including the following steps: S1: According to the pre-set multi-modal data feature engineering space, use the corresponding measuring device to collect real-time multi-modal monitoring data of the target monitoring equipment, and transmit it to the edge computing gateway; The multi-modal data feature engineering space includes current-voltage monitoring input features of the target monitoring equipment, device state monitoring input features of the target monitoring equipment, and environmental monitoring input features of the working environment; The real-time multi-modal monitoring data includes real-time current-voltage monitoring data of the target monitoring equipment, real-time device state monitoring data of the target monitoring equipment, and real-time environmental monitoring data of the working environment; The measuring device for real-time current-voltage monitoring data is a current-voltage monitoring measuring device, which includes a wide-range Hall sensor and a digital Rogowski coil; Wide-range Hall sensor: Hall sensors can be used to measure current, with the advantage of non-contact measurement, suitable for high-current measurement occasions; wide-range Hall sensors can adapt to different sizes of current changes, and are very useful for synchronous acquisition of fundamental and harmonic components; Digital Rogowski coil: Rogowski coil is a sensor for measuring high-voltage pulse or transient current, suitable for transient current measurement in power systems; digital Rogowski coil can provide high-precision current measurement and can capture rapid changes in current, which is particularly important for harmonic analysis; The measuring device for real-time device state monitoring data is a device state monitoring measuring device, which includes a micro-vibration sensor, a partial discharge detector, and an infrared thermal imager; Micro-vibration sensor: used to detect the micro-vibration of the equipment during operation, which can help diagnose mechanical failure or abnormal wear; Partial discharge detector: a partial discharge detector is used to detect partial discharge phenomena in insulating materials, and is an important tool for evaluating the insulation state of power equipment; Infrared thermal imager: by detecting the temperature distribution data of the equipment surface, it can find potential overheating problems, which is an important means to prevent equipment failure; The measuring device for real-time environmental monitoring data is an environmental monitoring measuring device, which includes a temperature and humidity sensor, a gas concentration sensor, and a mechanical vibration sensor; Temperature and humidity sensor: monitors environmental temperature and humidity, which is very important for evaluating whether the equipment operating environment is suitable; Gas concentration sensor: monitoring gas concentration is crucial for ensuring safe operation of equipment; Mechanical vibration sensor: similar to the micro-vibration sensor, but may focus more on larger mechanical vibration monitoring, used to evaluate the mechanical stability of the equipment operating environment; S2: using the edge computing gateway, pre-processing the real-time multi-modal monitoring data, and uploading the obtained pre-processed real-time multi-modal monitoring data to the cloud data center, including the following steps: S2-1: using the edge computing gateway, adjusting the dynamic sampling rate of the metering device according to the real-time multi-modal monitoring data, returning the adjusted dynamic sampling rate to the corresponding metering device, and updating the real-time multi-modal monitoring data, including the following steps: S2-1-1: using the pre-trained device load prediction model in the edge computing gateway, performing device load prediction according to the real-time multi-modal monitoring data, and obtaining real-time device load prediction results; The device load prediction model is constructed based on the Long Short-Term Memory (LSTM) algorithm, which can predict the load of the target monitoring device according to the data characteristics of the real-time current and voltage monitoring data in the real-time multi-modal monitoring data; S2-1-2: using a three-level sampling adjustment rule to adjust the dynamic sampling rate of the metering device according to the real-time device load prediction results, and obtaining the corresponding adjusted dynamic sampling rate; Three-level sampling adjustment rule: Light load mode (<30% load): 10Hz sampling + 1s sliding window; Normal mode: 50Hz sampling + sampling rate compensation value; Fault mode (>70% load): 100Hz sampling + sampling rate compensation value; S2-1-3: returning the adjusted dynamic sampling rate of the metering device to the corresponding metering device, and re-performing real-time data acquisition to obtain updated real-time multi-modal monitoring data; S2-2: performing spatio-temporal alignment processing on the updated real-time multi-modal monitoring data to obtain spatio-temporally aligned real-time multi-modal monitoring data, including the following steps: S2-2-1: standardizing the timestamps of the real-time current and voltage monitoring data, real-time device state monitoring data, and real-time environmental monitoring data of the updated real-time multi-modal monitoring data to obtain corresponding standardized timestamps; For example, the timestamp of real-time current and voltage monitoring data A is 2023-10-01 12:00:00.000, coordinated universal time +8, summer time adjustment amount is 0, the timestamp of real-time current and voltage monitoring data B is 2023-10-01 12:00:00.500, time zone is coordinated universal time +8, summer time adjustment amount is 0, after conversion to coordinated universal time, the timestamp of real-time current and voltage monitoring data A is 2023-10-01 04:00:00.000, and the timestamp of real-time current and voltage monitoring data B is 2023-10-01 04:00:00.500; S2-2-2: Synchronize the standardized timestamps of real-time current-voltage monitoring data, real-time device state monitoring data, and real-time environmental monitoring data to obtain corresponding synchronized timestamps; For example, through network time protocol synchronization, the timestamp offset of real-time current-voltage monitoring data A is +0.002s, and the timestamp offset of real-time current-voltage monitoring data B is -0.001s; the synchronized timestamps are 2023-10-01 04:00:00.002 and 2023-10-01 04:00:00.499, respectively; S2-2-3: Collect the recording position data corresponding to each modality in real-time multi-modal monitoring data, and convert all recording position data to a unified coordinate system; The coordinates of real-time current-voltage monitoring data A are (120.1, 30.2, 50), and the coordinates of real-time current-voltage monitoring data B are (121.2, 31.3, 60); after conversion to the unified coordinate system, the coordinates are (120.1, 30.2, 50) and (121.2, 31.3, 60), respectively; S2-2-4: According to the synchronized timestamps of real-time multi-modal monitoring data and the adjusted dynamic sampling rates of each modality's measurement device, use linear interpolation method to perform spatio-temporal alignment processing on real-time current-voltage monitoring data, real-time device state monitoring data, and real-time environmental monitoring data to obtain initial spatio-temporally aligned real-time multi-modal monitoring data; S2-2-5: Use error compensation algorithm to fine-tune the initial spatio-temporally aligned real-time multi-modal monitoring data to obtain the final spatio-temporally aligned real-time multi-modal monitoring data, ensuring that the error is within ±1ms; S2-3: Perform feature enhancement on the spatio-temporally aligned real-time multi-modal monitoring data to obtain pre-processed real-time multi-modal monitoring data, and upload it to the cloud data center, including the following steps: S2-3-1: Perform normalization processing on the spatio-temporally aligned real-time multi-modal monitoring data to obtain normalized real-time multi-modal monitoring data; The formula is:

[0021] In the formula, is the normalized real-time multi-modal monitoring data; is the spatio-temporally aligned real-time multi-modal monitoring data; is the mean of the spatio-temporally aligned real-time multi-modal monitoring data; is the standard deviation of the spatio-temporally aligned real-time multi-modal monitoring data; S2-3-2: Use wavelet transform algorithm to transform the normalized real-time multimodal monitoring data to obtain the corresponding real-time multi-scale features; The formula is:

[0022] In the formula, For real-time multi-scale features; For scale parameters; These are translation parameters; This is normalized real-time multimodal monitoring data; These are wavelet basis functions; For time indication; S2-3-3: The Fourier transform algorithm is used to transform the normalized real-time multimodal monitoring data to obtain the corresponding real-time frequency domain features; The formula is:

[0023] In the formula, Real-time frequency domain characteristics; For frequency; For complex units; S2-3-4: Using principal component analysis algorithm, the normalized real-time multimodal monitoring data is dimensionality reduced to obtain the corresponding real-time dimensionality-reduced features; The formula is:

[0024] In the formula, Features after real-time dimensionality reduction; Principal component matrix; It is the transpose symbol; S2-3-5: Using an autoencoder, feature learning and dimensionality reduction are performed on the normalized real-time multimodal monitoring data to obtain the corresponding real-time reconstructed features; The formula is:

[0025] In the formula, Real-time frequency domain characteristics; For encoder functions; For decoder functions; S2-3-6: Use the short-time Fourier transform algorithm to transform the normalized real-time multimodal monitoring data to obtain the corresponding real-time time-frequency features; The formula is:

[0026] In the formula, Real-time time-frequency characteristics; for the normalized real-time multi-modal monitoring data; for the time indication quantity; for the window function; S2-3-7: feature fusion is performed on the real-time multi-scale features, the real-time frequency domain features, the real-time dimensionality-reduced features, the real-time reconstructed features and the real-time time-frequency features to obtain corresponding preprocessed real-time multi-modal monitoring data; The formula is:

[0027] In the formula, for the real-time fusion feature; for the attention weighting function; S2-3-8: the preprocessed real-time multi-modal monitoring data is uploaded to the cloud data center through a pre-set secure encryption channel and is encrypted; S3: in the cloud data center, current and voltage monitoring is performed on the preprocessed real-time multi-modal monitoring data of the target monitoring devices, and a generated real-time abnormal maintenance strategy is sent to the corresponding target monitoring device, including the following steps: S3-1: in the cloud data center, a Spatial Temporal-Density-Based Spatial Clustering of Applications with Noise (ST-DBSCAN) algorithm is used to cluster the preprocessed real-time multi-modal monitoring data of the target monitoring devices to obtain a real-time multi-modal monitoring data cluster of the same type of monitoring target, including the following steps: S3-1-1: in the cloud data center, the preprocessed real-time multi-modal monitoring data of each target monitoring device is converted into corresponding real-time data points; The real-time data points are used as a multi-modal data set Each data point is represented as a vector in the set , including the position coordinates (x, y, z) of the unified coordinate system, the synchronized time stamp and the real-time fusion feature , and the data point can be represented as ; S3-1-2: a dynamic computing mechanism is introduced to obtain the dynamic neighborhood radius of the real-time data points, and the spatial distance between different data points is defined; The dynamic neighborhood radius includes a spatial dynamic neighborhood radius , a time dynamic neighborhood radius and a feature space dynamic neighborhood radius ; preset a base range based on the monitored target type and regional characteristics , ], , minimum and maximum values for the spatial dynamic neighborhood radius; for the current multi-modal data subset to be processed (e.g., the monitored target devices within the same voltage level or geographical region), calculate the statistical characteristics of their spatial coordinates, such as the average spatial distance ; get the spatial dynamic neighborhood radius ; a scaling factor determined empirically or through cross-validation; preset a base range based on the monitored target type and operational characteristics , ], , minimum and maximum values for the temporal dynamic neighborhood radius; for the current multi-modal data subset to be processed , calculate the statistical characteristics of their timestamps, such as the average time interval ; dynamic can be set as: , where is a scaling factor determined empirically or through cross-validation; include the feature vectors in the neighborhood calculation, defining the distance metric (e.g., Euclidean distance) of the feature space; based on the distribution of the feature vectors, calculate their statistical characteristics, such as the average feature distance ; dynamic can be set as: , where the formula for the spatio-temporal distance is:

[0028] In the formula, is the spatio-temporal distance between data points and ; is the spatial Euclidean distance between data points and ; is the absolute value of the time difference between data points and ; is the Euclidean distance between feature vectors; are weight coefficients used to balance the importance of space, time and feature space, which can be adjusted according to specific application scenarios; Determine the density threshold, core point, boundary point and noise point: The formula of the density threshold is:

[0029] In the formula, is the density threshold of data point ; is the global density threshold; is the local density estimate of data point ; If , data point is a core point, is the field of data point ; If and data point is in the neighborhood of a core point, data point is a boundary point; If and data point is not in the neighborhood of any core point, data point is a noise point; S3-1-3: Traverse all real-time data points, use the ST-DBSCAN algorithm to cluster, and obtain a plurality of real-time multi-modal monitoring data clusters of the same type of monitoring target; The formula of the ST-DBSCAN algorithm is:

[0030] In the formula, is the spatio-temporal density of data point ; is the field of data point ; is the spatio-temporal distance between data points and ; and are data point indicators; is a dynamic neighborhood radius; is a natural base; The clustering process is: Initialization: all points are marked as unvisited; Traverse each data point in the multi-modal data set ; If data point has been visited, skip; Otherwise, mark data point as visited; Compute the neighborhood of data point If data point is a core point; create a new cluster ; add data point to cluster ; for each point data point in : if data point is not visited, mark data point as visited; if data point is a core point, recursively add the neighborhood points (points satisfying density reachable) of data point to cluster ; add data point to cluster ; If data point is not a core point: mark data point as a noise point (temporarily mark); Repeat the above process until all points are visited; Post-processing: check all points marked as noise, if they fall within the neighborhood of the core points of any cluster, reclassify them as boundary points of that cluster; S3-2: using a pre-trained anomaly analysis model, performing anomaly analysis on each real-time multi-modal monitoring data cluster to obtain the corresponding real-time anomaly analysis result; The anomaly analysis model is constructed based on an LSTM-Graph Attention Network (GAT)-Transformer-Multi-Layer Perceptron (MLP) algorithm, and the anomaly analysis model includes a time series feature extraction module constructed based on an LSTM algorithm, a graph structure feature extraction module constructed based on a GAT algorithm, a dependency relationship extraction module constructed based on a Transformer algorithm, and an anomaly analysis module constructed based on an MLP algorithm. The anomaly analysis module is connected with the time series feature extraction module, the graph structure feature extraction module, and the dependency relationship extraction module respectively. For each data cluster, the long short-term memory network (LSTM) of the time series feature extraction module is used to extract the time series features of the real-time multi-modal monitoring data, the graph attention network (GAT) of the graph structure feature extraction module is used to mine the correlation between data points, the Transformer network of the dependency relationship extraction module is used to capture the long-distance dependency relationship between environmental parameters and current and voltage, and the multi-layer perceptron of the anomaly analysis module is used to fuse and predict the features.​ using a pre-trained anomaly analysis model, performing anomaly analysis on each real-time multi-modal monitoring data cluster to obtain corresponding real-time anomaly analysis results, including the following steps: S3-2-1: using a time sequence feature extraction module of the pre-trained anomaly analysis model, extracting real-time time sequence features of the real-time multi-modal monitoring data; S3-2-2: using a graph structure feature extraction module of the anomaly analysis model, extracting real-time correlation graph structure features of the real-time multi-modal monitoring data; S3-2-3: using a dependency relationship extraction module of the anomaly analysis model, extracting real-time dependency relationship features of the real-time multi-modal monitoring data; S3-2-4: using an anomaly analysis module of the anomaly analysis model, performing anomaly analysis according to the real-time time sequence features, the real-time correlation graph structure features, and the real-time dependency relationship features to obtain corresponding real-time anomaly analysis results; The real-time anomaly analysis results include types, locations, and severity of real-time current and voltage anomalies, real-time device state anomalies, and real-time environmental anomalies; S3-3: according to the real-time anomaly analysis results, using a pre-trained anomaly repair strategy generation model to perform anomaly repair strategy generation to obtain real-time anomaly repair strategies and send them to corresponding target monitoring devices; The anomaly repair strategy generation model is constructed based on a Multi-Agent Deep Deterministic Policy Gradient (MA-DDPG)-Proximal Policy Optimization (PPO)-Domain Adaptation Network (DAN) algorithm, and the anomaly repair strategy generation model includes a regional agent layer constructed based on the MA-DDPG algorithm, a device agent layer constructed based on the PPO algorithm, and a transfer learning module constructed based on the DAN algorithm connected in sequence; The regional agent layer is responsible for regional-level resource scheduling, including maintenance personnel allocation, spare parts allocation, etc., including detection agents, maintenance agents, and coordination agents, the state space of the detection agent is a spatiotemporal feature vector + device state, the action space is an anomaly level (0-4 levels), the state space of the maintenance agent is a fault feature + historical maintenance record, and the action space is a maintenance scheme (12 types), the state space of the coordination agent is a system load + maintenance resource, and the action space is a resource scheduling strategy; The device agent layer is responsible for fine maintenance strategy optimization of a single device, the state space is the detailed running state, historical fault record, and current maintenance resource allocation of a single device, and the action space is the execution of routine inspection component maintenance decisions; The transfer learning module is used to accelerate policy learning in new scenarios, realize cross-scenario policy transfer, learn the feature correspondence between the source domain (learned scenario) and the target domain (new scenario), and realize domain adaptation at the feature level; According to the real-time anomaly analysis result, an abnormal maintenance strategy generation model is used to generate an abnormal maintenance strategy, and a real-time abnormal maintenance strategy is obtained and sent to the corresponding target monitoring device, including the following steps: S3-3-1: Using the pre-trained abnormal maintenance strategy generation model, the real-time anomaly analysis result and the corresponding device cluster health index, maintenance resource state (available personnel, spare parts inventory, etc.) from the abnormal maintenance strategy generation model output are used as the first real-time state input of the regional agent layer; S3-3-2: Using the regional agent layer of the abnormal maintenance strategy generation model, a regional-level resource scheduling strategy is generated according to the first real-time state, and a real-time regional-level resource scheduling strategy is obtained, such as the number of maintenance personnel and the number of spare parts allocated to each device cluster; Actor network: generate the optimal action (resource scheduling strategy) for each regional agent under a given state; Critic network: evaluate the value (Q value) of a given state-action pair; Experience replay: store the experience (state, action, reward, next state, end or not) of the agent and randomly sample from it for training; Target network: update the target network of Actor and Critic using a soft update policy to maintain the stability of network parameters; Multi-agent collaboration: consider the cooperation and competition between different regional agents, and realize global optimization through a central critic mechanism or other multi-agent coordination algorithms; S3-3-3: The real-time regional-level resource scheduling strategy output by the regional agent layer, the real-time anomaly analysis result output by the abnormal maintenance strategy generation model, and the corresponding real-time device state monitoring data are used as the second real-time state input of the device agent layer; S3-3-4: Using the device agent layer of the abnormal maintenance strategy generation model, a real-time single-device-level maintenance strategy is generated, which is a maintenance operation suggestion for a specific device, such as specific maintenance steps, spare parts replacement schemes, and operation parameter adjustments; Policy network: generate the action probability distribution under a given state; Value network: evaluate the value function of a given state; Advantage function: calculate the advantage function of the current policy relative to the old policy; Proximal objective function: use the objective function to limit the policy update range to ensure the stability of the policy. Constraint optimization module: introduce safety constraints (such as operation procedures, safety limits) and cost constraints (such as maintenance cost, downtime loss), and incorporate them into the optimization objective through penalty terms or constraint terms; S3-3-5: integrate real-time regional resource scheduling strategy and real-time single device maintenance strategy to obtain real-time abnormal maintenance strategy, and send the real-time abnormal maintenance strategy to the corresponding target monitoring device through a secure encryption channel.

[0031] Embodiment 2: As shown in Figure 2 The embodiment provides a current and voltage monitoring system based on a metering device, which is used to implement a current and voltage monitoring method, and includes a cloud data center, a plurality of edge computing gateways, and a plurality of metering devices. The cloud data center is in communication connection with the edge computing gateways of a plurality of monitoring areas, the plurality of metering devices are in communication connection with the edge computing gateways in the corresponding monitoring areas, and each metering device is in electrical connection with a corresponding target monitoring device. The metering device is used to collect real-time multi-modal monitoring data of the target monitoring device according to a preset multi-modal data feature engineering space, and transmit the real-time multi-modal monitoring data to the edge computing gateway. The edge computing gateway is used to pre-process the real-time multi-modal monitoring data, and upload the pre-processed real-time multi-modal monitoring data to the cloud data center. The cloud data center is used to perform current and voltage monitoring on the pre-processed real-time multi-modal monitoring data of a plurality of target monitoring devices, and send a generated real-time abnormal maintenance strategy to the corresponding target monitoring device.

[0032] The application provides a current-voltage monitoring method and system based on a metering device, which integrates current-voltage parameters, equipment state parameters and environmental parameters and other multi-modal data, overcomes the limitation of only focusing on a single parameter in the prior art, realizes more comprehensive and meticulous monitoring of the monitoring target, and improves the accuracy and reliability of current-voltage abnormality detection; an edge computing gateway is used for real-time preprocessing of multi-modal monitoring data, including dynamic sampling adjustment, time-space alignment and feature enhancement, which effectively reduces the data transmission amount, reduces the cloud computing pressure, improves the data processing efficiency and real-time performance; an ST-DBSCAN algorithm is used for time-space clustering analysis, the monitoring targets with similar time-space behaviors and electrical characteristics are divided into the same data cluster, which provides more refined data grouping for subsequent abnormality detection and maintenance strategy generation, improves the pertinence and generalization ability of the model; an abnormality analysis model is used to extract time domain, frequency domain and time-frequency domain features, and simultaneously identify current-voltage abnormality, equipment state abnormality and environmental abnormality through a multi-task learning framework, which significantly improves the intelligent degree, accuracy and robustness of abnormality detection; an abnormality maintenance strategy generation model is used to construct a regional agent layer and a device agent layer, realize hierarchical decision of regional resource scheduling and single device maintenance strategy optimization, and quickly adapt to new device types and operation scenarios through a transfer learning framework to generate intelligent and adaptive abnormality maintenance strategies; through data-driven abnormality detection and intelligent maintenance strategy generation, the dependence on artificial experience in the current-voltage monitoring and maintenance process is reduced, and the automation level and efficiency of the power system operation and maintenance are improved; through real-time monitoring, accurate abnormality detection and intelligent maintenance strategy generation, equipment abnormalities can be discovered and handled in time, the operation reliability and maintenance efficiency of the power system are improved, and the operation and maintenance cost is reduced.

[0033] The application is not limited to the above-mentioned optional embodiments, and anyone can derive other various forms of products under the inspiration of the application. The above-mentioned specific embodiments should not be understood as limiting the protection scope of the application, and the protection scope of the application should be defined by the claims, and the specification can be used to explain the claims.

Claims

1. A current-voltage monitoring method based on a metering device, characterized by: The method comprises the following steps: According to the preset multi-modal data feature engineering space, using the corresponding measuring device, the real-time multi-modal monitoring data of the target monitoring equipment is collected and transmitted to the edge computing gateway; Using the edge computing gateway, the real-time multi-modal monitoring data is preprocessed, and the preprocessed real-time multi-modal monitoring data is uploaded to the cloud data center; In the cloud data center, the preprocessed real-time multi-modal monitoring data of several target monitoring equipment is monitored, and the generated real-time abnormal maintenance strategy is sent to the corresponding target monitoring equipment.

2. A current-voltage monitoring method based on a metering device according to claim 1, characterized in that: The multi-modal data feature engineering space includes current-voltage monitoring input features of the target monitoring equipment, device state monitoring input features of the target monitoring equipment, and environment monitoring input features of the working environment; The real-time multi-modal monitoring data includes real-time current-voltage monitoring data of the target monitoring equipment, real-time device state monitoring data of the target monitoring equipment, and real-time environment monitoring data of the working environment.

3. A current-voltage monitoring method based on a metering device according to claim 2, characterized in that: Using the edge computing gateway, the real-time multi-modal monitoring data is preprocessed, and the preprocessed real-time multi-modal monitoring data is uploaded to the cloud data center, comprising the following steps: According to the real-time multi-modal monitoring data, using the edge computing gateway, the dynamic sampling rate of the measuring device is adjusted, the adjusted dynamic sampling rate is returned to the corresponding measuring device, and the real-time multi-modal monitoring data is updated; The updated real-time multi-modal monitoring data is processed by time and space alignment to obtain real-time multi-modal monitoring data after time and space alignment processing; According to the real-time multi-modal monitoring data, using the edge computing gateway, the dynamic sampling rate of the measuring device is adjusted, the adjusted dynamic sampling rate is returned to the corresponding measuring device, and the real-time multi-modal monitoring data is updated, comprising the following steps:

4. The current-voltage monitoring method based on a metering device according to claim 3, characterized in that: According to the real-time multi-modal monitoring data, using the pre-trained device load prediction model in the edge computing gateway, device load prediction is performed to obtain real-time device load prediction results; According to the real-time device load prediction results, using the three-level sampling adjustment rules, the dynamic sampling rate of the measuring device is adjusted to obtain the corresponding adjusted dynamic sampling rate; The adjusted dynamic sampling rate of the measuring device is returned to the corresponding measuring device, and real-time data collection is performed again to obtain updated real-time multi-modal monitoring data. The updated real-time multi-modal monitoring data is processed by time and space alignment to obtain real-time multi-modal monitoring data after time and space alignment processing, comprising the following steps:

5. A current-voltage monitoring method based on a metering device according to claim 4, characterized in that: The timestamps of the real-time current-voltage monitoring data, real-time device state monitoring data and real-time environment monitoring data of the updated real-time multi-modal monitoring data are standardized to obtain corresponding standardized timestamps; The standardized timestamps of the real-time current-voltage monitoring data, real-time device state monitoring data and real-time environment monitoring data are synchronized to obtain corresponding synchronized timestamps; The record position data corresponding to each modality in the real-time multi-modal monitoring data is collected, and all record position data is converted to a unified coordinate system; ​ According to the synchronized time stamp of the real-time multi-modal monitoring data and the adjusted dynamic sampling rate of the metering device corresponding to each mode, the real-time current and voltage monitoring data, the real-time equipment state monitoring data and the real-time environmental monitoring data are processed by time and space alignment using a linear interpolation method to obtain initial real-time multi-modal monitoring data after time and space alignment. The initial real-time multi-modal monitoring data after time and space alignment is fine-tuned using an error compensation algorithm to obtain final real-time multi-modal monitoring data after time and space alignment.

6. A current-voltage monitoring method based on a metering device according to claim 5, characterized in that: The real-time multi-modal monitoring data after time and space alignment is processed by feature enhancement to obtain preprocessed real-time multi-modal monitoring data, which is uploaded to a cloud data center, including the following steps: The real-time multi-modal monitoring data after time and space alignment is processed by normalization to obtain normalized real-time multi-modal monitoring data. The normalized real-time multi-modal monitoring data is transformed using a wavelet transform algorithm to obtain corresponding real-time multi-scale features. The normalized real-time multi-modal monitoring data is transformed using a Fourier transform algorithm to obtain corresponding real-time frequency domain features. The normalized real-time multi-modal monitoring data is reduced in dimension using a principal component analysis algorithm to obtain corresponding real-time reduced features. The normalized real-time multi-modal monitoring data is processed by feature learning and dimension reduction using an autoencoder to obtain corresponding real-time reconstructed features. The normalized real-time multi-modal monitoring data is transformed using a short-time Fourier transform algorithm to obtain corresponding real-time time-frequency features. The real-time multi-scale features, real-time frequency domain features, real-time reduced features, real-time reconstructed features and real-time time-frequency features are fused to obtain corresponding preprocessed real-time multi-modal monitoring data. The preprocessed real-time multi-modal monitoring data is encrypted and uploaded to the cloud data center through a pre-set secure encryption channel.

7. A current-voltage monitoring method based on a metering device according to claim 6, characterized in that: In the cloud data center, the preprocessed real-time multi-modal monitoring data of the target monitoring equipment is monitored by current and voltage, and the generated real-time abnormal maintenance strategy is sent to the corresponding target monitoring equipment, including the following steps: In the cloud data center, the preprocessed real-time multi-modal monitoring data of the target monitoring equipment is clustered using an ST-DBSCAN algorithm to obtain a real-time multi-modal monitoring data cluster of the same type of monitoring target. An abnormal analysis model trained in advance is used to analyze each real-time multi-modal monitoring data cluster to obtain a corresponding real-time abnormal analysis result. According to the real-time abnormal analysis result, an abnormal maintenance strategy generation model trained in advance is used to generate an abnormal maintenance strategy to obtain a real-time abnormal maintenance strategy and send it to the corresponding target monitoring equipment.

8. A current-voltage monitoring method based on a metering device according to claim 7, characterized in that: In the cloud data center, the preprocessed real-time multi-modal monitoring data of the target monitoring equipment is clustered using an ST-DBSCAN algorithm to obtain a real-time multi-modal monitoring data cluster of the same type of monitoring target, including the following steps: In the cloud data center, the preprocessed real-time multi-modal monitoring data of each target monitoring equipment is converted into corresponding real-time data points. A dynamic computer mechanism is introduced to obtain the dynamic neighborhood radius of a plurality of real-time data points and define the time and space distance between different data points. All real-time data points are traversed, clustering is performed using the ST-DBSCAN algorithm, and a plurality of real-time multi-modal monitoring data clusters of the same type of monitoring target are obtained.

9. A current-voltage monitoring method based on a metering device according to claim 8, characterized in that: The abnormality analysis model is constructed based on an LSTM-GAT-Transformer-MLP algorithm. The abnormality maintenance strategy generation model is constructed based on an MA-DDPG-PPO-DAN algorithm.

10. A current-voltage monitoring system based on a metering device for implementing the current-voltage monitoring method according to any one of claims 1 to 9, characterized in that: The system comprises a cloud data center, a plurality of edge computing gateways and a plurality of metering devices, the cloud data center is in communication connection with the edge computing gateways of a plurality of monitoring areas, the plurality of metering devices are in communication connection with the edge computing gateways in the corresponding monitoring areas, and each metering device is in electrical connection with the corresponding target monitoring equipment.

Images