Real-time metering method and system for intelligent modular power metering boxes

Abstract

This invention relates to the field of electricity metering and discloses a real-time metering method for an intelligent modular electricity metering box, comprising: deploying a sensor array inside the electricity metering box; collecting raw electricity data and environmental parameters through the sensor array to obtain a raw electricity metering data stream; performing dynamic threshold filtering on the raw electricity metering data stream to obtain clean electricity metering data; constructing an electricity topology model based on a graph neural network, and inputting the key electricity parameters into the electricity topology model to obtain initial electricity topology network data; and detecting abnormal nodes in the electricity topology model using an isolated forest algorithm to obtain optimized electricity topology data. This invention, by employing dynamic threshold filtering and Kalman filtering algorithms, can effectively remove noise and outliers from the raw electricity data, ensuring the reliability of the electricity metering data, especially in complex environments where errors may occur due to factors such as electromagnetic interference.

Description

Technical Field

[0001] This invention relates to the field of electricity metering technology, specifically to a real-time metering method and system for intelligent modular electricity metering boxes. Background Technology

[0002] A metering box is a collection of metering instruments and auxiliary equipment necessary for measuring electrical energy, including electricity meters, voltage and current transformers and their secondary circuits, electricity metering panels, cabinets, boxes, etc.

[0003] Existing equipment suffers from the following drawbacks: Traditional methods for processing electrical energy data often employ simple filters or fixed thresholds to remove noise. This may not be effective in handling noise and outliers in complex environments, such as electromagnetic interference, leading to inaccurate and unreliable electrical energy metering data. Traditional methods cannot dynamically adjust thresholds, making it difficult to adapt to environmental changes and potentially ignoring or incorrectly processing some outliers. Furthermore, traditional methods typically use fixed algorithms to analyze the power grid topology, lacking the ability to accurately identify potential problems within the grid. Fault points or unreasonable power distribution in the power grid may be missed, failing to detect potential problems in a timely manner.

[0004] Therefore, this application proposes a real-time metering method and system for intelligent modular power metering boxes to solve the aforementioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time metering method and system for intelligent modular power metering boxes, in order to solve the problem that the traditional methods often use simple filters or fixed thresholds to remove noise when processing power data. This may not be able to effectively deal with noise and outliers in complex environments, such as electromagnetic interference, resulting in inaccurate and unreliable power metering data.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time metering method for an intelligent modular energy metering box, comprising:

[0007] A sensor array is deployed inside the power metering box; raw power data and environmental parameters are collected through the sensor array to obtain raw power metering data stream; dynamic threshold filtering is performed on the raw power metering data stream to obtain clean power metering data;

[0008] Principal component analysis is performed on the clean energy metering data to obtain key energy parameters; an energy topology model is constructed based on a graph neural network, and the key energy parameters are input into the energy topology model to obtain initial energy topology network data; anomaly nodes in the energy topology model are detected by the isolated forest algorithm to obtain optimized energy topology data.

[0009] Environmental data is acquired, and environmental effect compensation is performed on the optimized power topology data based on the environmental data to obtain compensated optimized power topology data; the clean energy metering data is dynamically corrected based on the compensated optimized power topology data using a Kalman filter algorithm to obtain actual energy metering data; the actual energy metering data is mapped to standard energy units using edge computing nodes to obtain standardized energy metering data.

[0010] The standardized electricity metering data is input into a pre-trained fusion model, which uses LSTM and CNN to extract temporal features and fuse spatial features to obtain electricity usage pattern labels. The electricity usage pattern labels are compared with preset electricity usage standard data to calculate the electricity usage compliance score and obtain a smart metering report. The smart metering report includes real-time metering values, electricity usage patterns, and compliance scores.

[0011] Specifically, a sensor array is deployed inside the electricity metering box; the sensor array collects raw electricity data and environmental parameters to obtain a raw electricity metering data stream, including:

[0012] A modular sensor array architecture is designed inside the power metering box. The sensor array includes a current transformer, a voltage sensor, and a temperature and humidity sensor.

[0013] A hypercommunication link is established through the network to periodically collect raw power data and environmental parameters, and add timestamps and device ID tags to obtain the raw power metering data stream.

[0014] The process involves dynamic threshold filtering of the raw data to obtain clean energy metering data, including:

[0015] A dynamic threshold benchmark library is constructed based on historical power data. The mean and standard deviation of the data within the window are calculated using a sliding window algorithm to obtain the adaptive threshold range. An environmental interference model is established by combining temperature and humidity sensor data to obtain environmental adaptive threshold parameters.

[0016] Real-time deviation detection is performed on the raw power data stream, and the absolute error between the current data point and the mean of the sliding window is calculated; the soft threshold shrinkage method is used to denoise the high-frequency noise band to obtain preliminary denoised data;

[0017] Cyclic redundancy check is used to check the integrity of data packets in order to obtain intermediate data after verification;

[0018] The verified intermediate data is input into a Kalman filter, and state estimation is performed by combining dynamic threshold parameters and wavelet denoising results. Data deviation is corrected through prediction update cycles to obtain clean energy metering data.

[0019] Specifically, principal component analysis is performed on the clean energy metering data to identify key energy parameters; an energy topology model is constructed based on a graph neural network to correlate the energy data of each circuit, thereby obtaining initial energy topology network data, including:

[0020] The clean energy metering data is standardized, and principal components are screened by cumulative variance contribution rate to obtain a dimension-reduced feature matrix; the dimension-reduced feature matrix includes harmonic content, three-phase imbalance degree and power factor;

[0021] A multimodal graph structure is constructed, and each circuit in the power metering box is defined as a graph node; the correlation of power parameters between the graph nodes is calculated using the Pearson correlation coefficient to obtain an initial power topology graph;

[0022] A graph attention network is deployed to perform multi-layer graph convolution operations on the initial topology graph; edge weights are dynamically adjusted through an attention mechanism to obtain optimized power topology network data.

[0023] Specifically, the isolated forest algorithm is used to detect abnormal nodes in the topology network and optimize the power metering path to obtain optimized power topology data, including:

[0024] Extract node feature vectors from the initial power topology network data; construct a node feature matrix and preserve edge connectivity to obtain topological feature data suitable for isolated forest input;

[0025] An isolated forest model is constructed by randomly sampling a subspace on the node feature matrix to build multiple isolated trees; anomaly scores are calculated based on path lengths to obtain anomaly detection results.

[0026] The topology network is optimized based on the anomaly detection results, and abnormal nodes are removed. The shortest path of the optimized topology network is calculated based on the energy metering path efficiency requirements to obtain the optimized edge weight matrix and node list.

[0027] The optimization effect was verified by simulating the injection of abnormal data, and the anomaly detection accuracy and path delay were calculated to obtain optimized power topology network data.

[0028] Specifically, environmental effect compensation is applied to the optimized power topology data, and an electromagnetic interference compensation model is established by combining temperature and humidity sensor data. The metering data is then dynamically corrected using a Kalman filter algorithm to obtain the compensated actual power metering value.

[0029] Node features and topological attributes are extracted from the optimized power topology data; temperature and humidity sensor data are collected to construct a multimodal dataset containing power parameters and environmental parameters to obtain an environmental power correlation feature matrix;

[0030] An electromagnetic interference compensation model is constructed based on a graph attention network. The nonlinear effects of the multi-modal dataset on the model are learned through a multi-layer graph convolutional learning environment to obtain node-level electromagnetic interference compensation coefficients.

[0031] The node-level electromagnetic interference compensation coefficient is applied to the optimized power topology data to reverse the voltage and current parameters to obtain the preliminary corrected data after compensation. Edge computing nodes are deployed to perform Kalman filtering, using the compensated data as the observation value and the model prediction value as the state estimate. The noise is dynamically corrected through prediction updates to obtain the actual power metering value with the noise level below the threshold.

[0032] The compensation effect was verified by simulating environmental changes, and the error of the actual electricity metering value before and after correction was calculated to obtain the compensated actual electricity metering value.

[0033] This involves using edge computing nodes to map actual metering values ​​to standard energy units to obtain standardized energy metering data, including:

[0034] Deploy a unit conversion engine on the edge computing node to receive the compensated actual power metering value; identify the power parameter type through a pre-trained neural network model to obtain initial standardized data;

[0035] A dynamic calibration strategy is deployed, and the deviation between the initial standardized data and the standard value is calculated by combining the historical data of the standard energy meter stored at the edge node to obtain intermediate standardized data.

[0036] A time-series model within an edge computing node is used to perform nonlinear error correction on the intermediate standardized data; by capturing the time correlation of power parameters, the cumulative error in the unit conversion process is predicted and corrected to obtain standardized power metering data.

[0037] The standardized electricity metering data is synchronized to the cloud blockchain platform, and a data fingerprint is generated through a hash algorithm and stored on the blockchain to obtain a standardized electricity metering data stream.

[0038] Specifically, the standardized electricity metering data is input into a pre-trained fusion model, and electricity usage pattern labels are obtained through temporal feature extraction and spatial feature fusion, including:

[0039] The standardized electricity metering data is divided into multi-scale time windows to obtain short-time, medium-time, and long-time time series segments; frequency domain features of each segment are extracted by wavelet transform, and a multimodal feature matrix is ​​constructed by combining time domain statistics to obtain enhanced data;

[0040] A parallel LSTM and CNN architecture is deployed, with the LSTM branch capturing the dynamic changes of the enhanced data and the CNN branch extracting the spatial distribution patterns of frequency domain features. The outputs of the two branches are fused through a cross-attention mechanism to obtain a fused feature vector.

[0041] The process includes inputting the standardized electricity metering data into a pre-trained fusion model, extracting time-series features and fusing spatial features to obtain electricity usage pattern labels, and further includes:

[0042] The fused feature vector is input into a pre-trained classification layer and combined with an energy usage scenario label library for pattern matching; a reinforcement learning algorithm is used to dynamically adjust the classification threshold to obtain energy usage pattern labels.

[0043] The accuracy of the labels is verified by backtracking historical electricity data, and the classification F1 value is calculated to obtain the electricity usage pattern label stream.

[0044] A real-time metering system for intelligent modular energy metering boxes, applicable to the aforementioned real-time metering method for intelligent modular energy metering boxes, includes:

[0045] The data filtering unit deploys a sensor array inside the power metering box; it collects raw power data and environmental parameters through the sensor array to obtain a raw power metering data stream; and it performs dynamic threshold filtering on the raw power metering data stream to obtain clean power metering data.

[0046] The power topology unit performs principal component analysis on the clean energy metering data to obtain key power parameters; constructs a power topology model based on a graph neural network, inputs the key power parameters into the power topology model to obtain initial power topology network data; and detects abnormal nodes in the power topology model using the isolated forest algorithm to obtain optimized power topology data.

[0047] The data correction unit acquires environmental data and performs environmental effect compensation on the optimized power topology data based on the environmental data to obtain compensated optimized power topology data; it then uses a Kalman filter algorithm to dynamically correct the clean energy metering data based on the compensated optimized power topology data to obtain actual energy metering data; finally, it uses edge computing nodes to map the actual energy metering data to standard energy units to obtain standardized energy metering data.

[0048] The electricity usage standardization unit inputs the standardized electricity metering data into a pre-trained fusion model, extracts time-series features and fuses spatial features to obtain electricity usage pattern labels; compares the electricity usage pattern labels with the preset electricity usage standardization data, calculates the electricity usage compliance score, and obtains a smart metering report; wherein, the smart metering report includes real-time metering values, electricity usage patterns, and compliance scores.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] This invention effectively removes noise and outliers from the original power data by employing dynamic threshold filtering and Kalman filtering algorithms, thereby improving the accuracy of power metering. This ensures the reliability of power metering data, especially in complex environments where factors such as electromagnetic interference may cause errors.

[0051] This invention constructs an electrical topology model using a graph neural network and employs the isolated forest algorithm to detect abnormal nodes in the topology network. This can accurately identify potential problems in the power grid, such as fault points or unreasonable power distribution. This provides strong support for power grid optimization and fault diagnosis, and helps to promptly discover and resolve problems in the power transmission process.

[0052] This invention effectively addresses the impact of environmental changes on electricity metering by combining temperature and humidity sensor data with an electromagnetic interference compensation model. For example, temperature and humidity changes may affect the accuracy of the equipment. By using a compensation model for correction, the measurement results are ensured to be unaffected by external environmental fluctuations, thereby improving the stability and reliability of metering.

[0053] This invention extracts key parameters from electrical energy data through principal component analysis and combines them with an LSTM-CNN fusion model. This allows for the extraction of temporal and spatial features while measuring electrical energy, helping to construct more accurate electrical energy usage patterns. The generation of electrical energy usage pattern labels helps to further analyze and predict the usage behavior of equipment.

[0054] This invention generates an electricity compliance score by comparing electricity usage pattern tags with preset electricity usage standard data and automatically generates a smart metering report. This provides real-time compliance detection for users or operators, helps to promptly identify non-compliant behaviors in electricity usage, and promotes energy conservation and emission reduction.

[0055] This invention utilizes edge computing nodes to process and transmit data, enabling rapid local energy metering and analysis. This reduces reliance on central servers and improves system response speed and processing efficiency. This is particularly important for energy management systems that require real-time feedback. The system can respond immediately and issue alarms when equipment or power grid problems occur, monitor energy metering data and power topology anomalies in real time, proactively issue fault warnings, and promptly repair potential faults or optimize maintenance strategies, thereby reducing power grid downtime and fault repair costs. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the overall method's steps in one embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0058] In the diagram: 1. Data filtering unit; 2. Power topology unit; 3. Data correction unit; 4. Power consumption specification unit. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1-2 This invention provides a technical solution: a real-time metering method for an intelligent modular energy metering box, comprising:

[0061] S1. Deploy a sensor array inside the power metering box; collect raw power data and environmental parameters through the sensor array to obtain raw power metering data stream; perform dynamic threshold filtering on the raw power metering data stream to obtain clean power metering data;

[0062] S2. Perform principal component analysis on the clean energy metering data to obtain key energy parameters; construct an energy topology model based on a graph neural network, input the key energy parameters into the energy topology model to obtain initial energy topology network data; detect abnormal nodes in the energy topology model using the isolated forest algorithm to obtain optimized energy topology data.

[0063] S3. Acquire environmental data, and perform environmental effect compensation on the optimized power topology data based on the environmental data to obtain compensated optimized power topology data; use a Kalman filter algorithm to dynamically correct the clean energy metering data based on the compensated optimized power topology data to obtain actual energy metering data; use edge computing nodes to map the actual energy metering data to standard energy units to obtain standardized energy metering data.

[0064] S4. Input the standardized electricity metering data into a pre-trained fusion model. The fusion model uses LSTM and CNN to extract temporal features and fuse spatial features to obtain electricity usage pattern labels. Compare the electricity usage pattern labels with the preset electricity usage standard data to calculate the electricity compliance score to obtain a smart metering report. The smart metering report includes real-time metering values, electricity usage patterns, and compliance scores.

[0065] It's important to note that during operation, a sensor array refers to an integrated group of sensor devices used together to monitor and collect different types of data. In an electricity metering box, the sensor array might include temperature and humidity sensors, current sensors, power sensors, etc., for real-time acquisition of electricity and environmental data. 5G networks, the fifth generation of mobile communication networks, offer higher transmission speeds, lower latency, and greater network capacity compared to previous generations. In electricity metering systems, 5G's role is to rapidly transmit raw electricity data and environmental parameters, ensuring data real-time performance and accuracy. Dynamic threshold filtering is a data processing technique that dynamically adjusts thresholds based on real-time data changes to remove noise or outliers. This improves data clarity, allowing subsequent analysis to be based on more accurate and cleaner data. Principal Component Analysis (PCA) is a commonly used dimensionality reduction technique that projects data from a high-dimensional space to a low-dimensional space through linear transformation, thereby extracting the main features of the data. In this method, PCA is used to extract the most important electricity parameters from the raw electricity data, reducing data complexity. Graph neural networks are neural network algorithms based on graph structures, capable of processing and analyzing graph data. In electricity metering, graph neural networks are used to construct electricity topology models, helping to identify the distribution and transmission of electricity by analyzing the relationships between nodes in the power grid. Isolation forest is a tree-based unsupervised learning algorithm primarily used for anomaly detection. It constructs multiple random trees to "isolate" data points; if a point is quickly isolated across multiple trees, it is considered abnormal data. In electricity topology, the isolation forest algorithm is used to detect abnormal nodes in the system, such as faulty equipment or incorrect electricity transmission. Electromagnetic interference (EMI) is noise or unwanted signals generated by electrical equipment that can affect the measurement accuracy of sensors. EMI compensation models, combining data such as ambient temperature and humidity, use mathematical models to compensate for errors caused by interference, ensuring the accuracy of electricity metering. Kalman filtering is an algorithm used for data estimation and correction. It dynamically estimates the system state using prior information and current observation data. In electricity metering, Kalman filtering can be used to correct raw data and sensor noise, resulting in more accurate actual electricity metering values. Edge computing is a distributed computing approach where data processing is completed closer to the data source (such as a local server or device), reducing transmission latency and bandwidth requirements. Edge computing nodes in the electricity metering system are responsible for real-time processing and analysis of electricity data, thereby improving the system's response speed and real-time performance. LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN) suitable for processing time-series data. CNN (Convolutional Neural Network) excels at processing image and spatial data. In this method, LSTM is used to extract time-series features from the electricity data, while CNN is used to fuse spatial features. Combining the two enables efficient analysis of electricity usage patterns.Electricity usage pattern tags are generated by analyzing electricity usage data and reflect electricity usage behavior. For example, they can display device usage habits, load changes, or energy consumption trends. Electricity compliance scores assess whether electricity usage meets prescribed standards or energy-saving requirements by comparing electricity usage patterns with preset electricity usage regulations. If the behavior does not comply with regulations, the score is lower; conversely, the score is higher. Smart metering reports include real-time electricity metering data, electricity usage pattern analysis, and compliance scores. This report provides users or managers with detailed information on electricity usage, helping them make smarter decisions, optimize electricity usage, and reduce energy waste.

[0066] In one embodiment, a sensor array is deployed inside the power metering box; raw power data and environmental parameters are periodically collected via a 5G network to obtain a raw power metering data stream, including:

[0067] A modular sensor array architecture is designed inside the power metering box. The sensor array includes a current transformer, a voltage sensor, and a temperature and humidity sensor.

[0068] A super communication link is established through the 5G network to periodically collect raw power data and environmental parameters, and add timestamps and device ID tags to obtain raw power metering data streams;

[0069] The raw data is subjected to dynamic threshold filtering, and harmonic interference is filtered out by wavelet transform denoising algorithm; the data integrity is verified in real time through edge computing nodes to obtain clean energy metering data;

[0070] Clean data is input into a pre-trained graph neural network model to construct a loop power topology graph; features are extracted through a node embedding algorithm and combined with an LSTM time series prediction model to obtain electricity consumption pattern labels.

[0071] Dynamic wavelet Kalman filter formula:

[0072]

[0073] in, Indicates scale and time delay The wavelet coefficients of the signal reflect the characteristics of the signal at different time locations and scales. This indicates the raw electricity metering data over time. Changing signals The conjugate of the wavelet function is represented by the wavelet basis function at different scales. and time offset The transformation under, This represents the scale parameter, which controls the width of the wavelet basis functions. A larger scale results in lower frequency components, and vice versa. This represents the signal after denoising, the signal recovered after wavelet coefficient denoising processing. Indicates based on threshold The wavelet coefficients are processed to remove noise. Typically, smaller absolute values ​​of the wavelet coefficients are suppressed or discarded, while larger coefficients are retained to preserve the useful components of the signal. Wavelet functions representing scaling are used to reconstruct signals. This represents the set of all scales, encompassing components across different frequency ranges. This represents the final data obtained after cleaning the electrical energy data, through wavelet denoising and Kalman filtering. The state transition matrix of the Kalman filter represents the dynamic changes of the system from one time step to the next. The control input matrix of the Kalman filter represents the effect of the external control signal on the system state. It represents the control input signal, which usually indicates the impact of external factors (such as environmental parameters, equipment status, etc.) on the use of electrical energy.

[0074] In this electricity metering system, the sensor array, designed with a modular architecture, includes current transformers, voltage sensors, and temperature and humidity sensors, enabling precise acquisition of electrical energy and environmental parameters. A communication link is established via a 5G network to periodically collect raw electricity data streams, adding timestamps and device IDs to each data packet to ensure accuracy and traceability. The raw data undergoes dynamic threshold filtering and wavelet transform denoising algorithms to remove grid harmonic interference and improve data quality. Edge computing nodes verify data integrity in real time, ensuring data is unaffected by network latency. Then, the cleaned data is input into a pre-trained graph neural network model to analyze electricity flow patterns by constructing an electricity topology graph. A node embedding algorithm extracts grid topology features, which, combined with an LSTM time-series prediction model, predicts future electricity consumption patterns and generates consumption pattern labels to guide users in optimizing their electricity usage and improving energy efficiency.

[0075] In one embodiment, dynamic threshold filtering is performed on the raw data to obtain clean energy metering data, including:

[0076] A dynamic threshold benchmark library is constructed based on historical power data. The mean and standard deviation of the data within the window are calculated using a sliding window algorithm to obtain the adaptive threshold range. An environmental interference model is established by combining temperature and humidity sensor data to obtain environmental adaptive threshold parameters.

[0077] Real-time deviation detection is performed on the raw power data stream, and the absolute error between the current data point and the mean of the sliding window is calculated; the soft threshold shrinkage method is used to denoise the high-frequency noise band to obtain preliminary denoised data;

[0078] Cyclic redundancy check is used to check the integrity of data packets in order to obtain intermediate data after verification;

[0079] Intermediate data is input into a Kalman filter, and state estimation is performed by combining dynamic threshold parameters with wavelet denoising results. Data bias is corrected by prediction update cycle to obtain clean energy metering data with noise levels below the threshold.

[0080] In this design, dynamic threshold filtering technology is used for real-time deviation detection in the electricity metering system. By establishing a dynamic threshold benchmark library and an environmental interference model, the error threshold of the data is adjusted in real time. A sliding window algorithm calculates the mean and standard deviation of the data, providing an adaptive threshold range for the raw electricity data to adapt to the impact of environmental changes on electricity metering. Environmental data from temperature and humidity sensors further optimize the threshold parameters, making the system more adaptable to environmental interference. The raw data undergoes real-time deviation detection, combined with a wavelet denoising algorithm to remove high-frequency noise. Data integrity is confirmed through cyclic redundancy check to ensure data accuracy. Based on this, a Kalman filter further estimates the state of the data, correcting data deviations by dynamically adjusting the threshold and denoising results, ultimately outputting clean electricity metering data, ensuring data accuracy and stability. This processing effectively reduces environmental interference and noise, ensuring high-quality electricity metering data and providing accurate and reliable input data for subsequent electricity analysis and prediction.

[0081] In one embodiment, principal component analysis is performed on clean energy metering data to identify key energy parameters; an energy topology model is constructed based on a graph neural network to correlate the energy data of each circuit, thereby obtaining initial energy topology network data, including:

[0082] Z-score standardization was performed on the clean energy metering data, and principal components were screened by cumulative variance contribution rate to obtain a dimension-reduced feature matrix that includes harmonic content, three-phase imbalance, and power factor.

[0083] A multimodal graph structure is constructed, and each circuit in the power metering box is defined as a graph node; the correlation of power parameters between nodes is calculated using the Pearson correlation coefficient to obtain the initial power topology graph.

[0084] A graph attention network is deployed to perform multi-layer graph convolution operations on the initial topology graph; edge weights are dynamically adjusted through an attention mechanism to obtain optimized power topology network data that includes node embeddings and edge weights.

[0085] Soft threshold shrinkage denoising formula:

[0086]

[0087]

[0088] in, The denoised signal represents the original signal. The result after applying soft thresholding, Represents a symbolic function, indicating symbols, This represents the original signal, which may contain noise. A function representing the soft threshold, based on time. Dynamically changing threshold The standard deviation of noise describes how the noise intensity changes over time. It represents the noise component in a signal. Indicates the length or size of the data. This represents a parameter that controls threshold sensitivity. It determines... Flexibility that changes over time These are typically local statistics of the signal, which may represent the signal’s variation or smoothness, and are used to dynamically adjust the threshold.

[0089] This design first standardizes the clean energy metering data using Z-scores to ensure data uniformity and comparability. Principal components are selected using the cumulative variance contribution rate to extract key information from the energy data, such as harmonic content, three-phase imbalance, and power factor, thereby reducing the dimensionality of the high-dimensional data into a feature matrix and minimizing noise and redundant data. Next, a multimodal graph structure is constructed, with each circuit within the energy metering box as a node in the graph. The correlation of energy parameters between nodes is calculated using the Pearson correlation coefficient, forming an initial energy topology graph. This topology graph illustrates the energy flow relationships between circuits, providing a foundation for subsequent analysis. Finally, a graph attention network (GAT) is deployed, performing multi-layer graph convolution operations on the initial energy topology graph and dynamically adjusting edge weights using the attention mechanism. Through this optimization method, the energy topology network can more accurately reflect the energy relationships between nodes (circuits), improving the correlation and predictive accuracy of the energy data. This process helps to deepen the understanding of energy flow patterns, providing data support for power optimization and fault detection.

[0090] In one embodiment, anomaly nodes in the topology network are detected using the isolated forest algorithm, and the power metering path is optimized to obtain optimized power topology data, including:

[0091] Extract node feature vectors from the initial power topology network data; construct a node feature matrix and preserve edge connectivity to obtain topological feature data suitable for isolated forest input;

[0092] Deploy a multi-scale isolated forest model, randomly sample subspaces on the feature matrix, and construct multiple isolated trees; calculate anomaly scores based on path lengths to obtain detection results containing anomaly node IDs and confidence levels;

[0093] The topology network is optimized based on the anomaly detection results, and abnormal nodes are removed or their edge connection weights are adjusted. In combination with the efficiency requirements of the power metering path, the shortest path is recalculated using the Dijkstra algorithm to obtain the optimized edge weight matrix and node list.

[0094] The optimization effect is verified by simulating the injection of abnormal data, and the anomaly detection accuracy and path delay are calculated; when the performance does not meet the standard, power topology network data with self-optimization capability is obtained.

[0095] In this design, the Isolation Forest algorithm is used in the electricity metering system to detect anomalous nodes in the electricity topology network and optimize the electricity metering path. First, node feature vectors are extracted from the initial topology network data to construct a node feature matrix, preserving edge connections to adapt to the input requirements of the Isolation Forest. Then, a multi-scale Isolation Forest model is deployed, constructing multiple isolated trees through random sampling of subspaces, calculating path lengths to obtain anomaly scores for each node, and determining whether a node is anomalous based on its score. After anomaly detection, the topology network is optimized by removing or adjusting the edge connection weights of anomalous nodes. Considering the efficiency of the electricity metering path, Dijkstra's algorithm is used to recalculate the shortest path, optimizing the edge weight matrix and node list in the topology network to improve the efficiency of electricity flow. Finally, the performance of the optimized topology network, including the accuracy of anomaly detection and path latency, is verified by simulating the injection of anomalous data. When performance does not meet the standard, the system will self-optimize, generating more efficient and adaptive electricity topology network data. This optimization process helps ensure the reliability and accuracy of the electricity metering system and improves the overall efficiency of electricity management.

[0096] In one embodiment, environmental effect compensation is performed on the optimized power topology data, and an electromagnetic interference compensation model is established by combining temperature and humidity sensor data; the metering data is dynamically corrected using a Kalman filter algorithm to obtain the compensated actual power metering value.

[0097] Node features and topological attributes are extracted from optimized power topology data; temperature and humidity sensor data are collected simultaneously to construct a multimodal dataset containing power parameters and environmental parameters, so as to obtain an environmental power correlation feature matrix;

[0098] An electromagnetic interference compensation model is constructed based on a graph attention network. The nonlinear influence of the environment on the power parameters is learned through multi-layer graph convolution to obtain the node-level electromagnetic interference compensation coefficients.

[0099] The compensation coefficient is applied to optimize the power topology data, and the voltage and current parameters are reversed to obtain the preliminary corrected data after compensation. Edge computing nodes are deployed to perform Kalman filtering. The compensated data is used as the observation value and the model prediction value is used as the state estimate. The noise is dynamically corrected through prediction updates to obtain the actual power metering value with the noise level below the threshold.

[0100] The compensation effect was verified by simulating environmental changes, and the data error before and after correction was calculated to obtain the actual power metering value with environmental adaptability.

[0101] This design first extracts node features and topological attributes from the optimized power topology data, while simultaneously collecting temperature and humidity sensor data to construct a multimodal dataset containing power and environmental parameters, thus obtaining an environmental-power correlation feature matrix. Next, an electromagnetic interference (EMI) compensation model is built using a Graph Attention Network (GAT). Through multi-layer graph convolution, the nonlinear effects of the environment on power parameters are learned, yielding EMI compensation coefficients for each node. These compensation coefficients are applied to optimize the power topology data, correcting parameters such as voltage and current to obtain preliminary compensation data. Then, edge computing nodes are deployed, and a Kalman filter algorithm is used to dynamically correct the compensated data. By using the compensated data as observations, the Kalman filter updates the state estimate using predicted values, dynamically correcting noise and ensuring that the noise level of the final power metering value is below a set threshold. Finally, simulation experiments of environmental changes are conducted to verify the compensation effect, calculating the data error before and after correction to ensure the system has good environmental adaptability and high-precision actual power metering values.

[0102] In one embodiment, edge computing nodes are used to map actual metering values ​​to standard energy units to obtain standardized energy metering data, including:

[0103] Deploy a unit conversion engine on the edge computing node to receive the compensated actual power metering value; identify the power parameter type through a pre-trained neural network model to obtain initial standardized data;

[0104] Deploy a dynamic calibration strategy, combine historical data of standard energy meters stored at the edge nodes, calculate the deviation between the initial standardized data and the standard value, and obtain the intermediate standardized data after calibration.

[0105] An LSTM time series model within an edge computing node is used to correct nonlinear errors in intermediate standardized data. By capturing the time correlation of power parameters, the accumulated error in the unit conversion process is predicted and corrected to obtain high-precision standardized power metering data.

[0106] Standardized data is synchronized to a cloud-based blockchain platform, and a hash algorithm is used to obtain data fingerprints, which are then stored on the blockchain to obtain a standardized electricity metering data stream with tamper-proof characteristics.

[0107] This design involves several steps. First, the node receives the compensated electricity metering data and identifies the types of electricity parameters using a pre-trained neural network model, generating initial standardized data. Then, a dynamic calibration strategy is deployed, combining historical standard electricity meter data to calculate the deviation between the initial standardized data and the standard values, thus obtaining intermediate standardized data. Next, an LSTM time-series model is used to correct nonlinear errors in the intermediate data, capturing the temporal correlation of electricity parameters and predicting and correcting accumulated errors during unit conversion, ultimately yielding high-precision standardized electricity metering data. This standardized data is synchronized to a cloud-based blockchain platform, where a data fingerprint is generated using a hash algorithm and uploaded to the chain, ensuring data immutability and enhancing data credibility and security. This process not only improves the accuracy of electricity metering but also guarantees the immutability of the data stream, providing reliable support for power management and data monitoring.

[0108] In one embodiment, standardized electricity metering data is input into a pre-trained LSTM-CNN fusion model, and electricity usage pattern labels are obtained through temporal feature extraction and spatial feature fusion, including:

[0109] Standardized electricity metering data is divided into multi-scale time windows to obtain short-time, medium-time, and long-time time series segments; frequency domain features of each segment are extracted by wavelet transform, and multi-modal feature matrix is ​​constructed by combining time domain statistics to obtain enhanced data suitable for LSTM-CNN input;

[0110] A parallel LSTM-CNN architecture is deployed, with the LSTM branch capturing the dynamic changes in time-series segments and the CNN branch extracting the spatial distribution patterns of frequency domain features. A cross-attention mechanism fuses the outputs of both branches to obtain a fused feature vector containing temporal and spatial correlations. This design first divides standardized data into short-, medium-, and long-term time-series segments through multi-scale time windows. Then, frequency domain features are extracted using wavelet transform and combined with temporal statistics to construct a multimodal feature matrix, enhancing the data's expressiveness and adapting it to LSTM-CNN input. In terms of model architecture, parallel LSTM and CNN branches are employed. The LSTM branch is responsible for capturing the dynamic changes in the time-series segments, while the CNN branch focuses on extracting the spatial distribution patterns of frequency domain features. Through the cross-attention mechanism, the outputs of the LSTM and CNN are effectively fused to generate a fused feature vector containing both temporal and spatial correlation information. This vector can more accurately describe energy usage patterns and support subsequent tasks such as intelligent analysis, anomaly detection, and prediction.

[0111] In one embodiment, standardized electricity metering data is input into a pre-trained LSTM-CNN fusion model, and electricity usage pattern labels are obtained through temporal feature extraction and spatial feature fusion. The model further includes:

[0112] The fused feature vectors are input into the pre-trained classification layer and combined with the electricity usage scenario label library for pattern matching; a reinforcement learning algorithm is used to dynamically adjust the classification threshold to obtain electricity usage pattern labels.

[0113] The accuracy of the tags is verified by backtracking through historical electricity data, and the classification F1 value is calculated to obtain an adaptive electricity usage pattern tag stream.

[0114] This design inputs standardized electricity metering data into an LSTM-CNN model to extract spatial distribution patterns of temporal dynamic changes and frequency domain features, generating a fused feature vector. Then, combined with an electricity usage scenario label library, the fused feature vector is input into a pre-trained classification layer for pattern matching to identify electricity usage patterns. To improve classification accuracy, the system employs a reinforcement learning algorithm to dynamically adjust the classification threshold and optimize the label matching process. This strategy adaptively adjusts the model's response to different usage scenarios, enhancing the system's adaptability to complex patterns. Finally, the accuracy of the pattern labels is verified by backtracking historical electricity data, and the classification F1 score is calculated to evaluate the model's performance. This process ensures that the generated electricity usage pattern labels are adaptive, accurately reflecting different electricity usage scenarios and providing precise data support for power management and forecasting.

[0115] A real-time metering system for intelligent modular energy metering boxes, applicable to the aforementioned real-time metering method for intelligent modular energy metering boxes, includes:

[0116] Data filtering unit 1 is used to deploy a sensor array inside the power metering box; to collect raw power data and environmental parameters through the sensor array to obtain raw power metering data stream; and to perform dynamic threshold filtering on the raw power metering data stream to obtain clean power metering data.

[0117] Power topology unit 2 is used to perform principal component analysis on clean energy metering data to obtain key energy parameters; a power topology model is constructed based on a graph neural network, and the key energy parameters are input into the power topology model to obtain initial power topology network data; anomaly nodes in the power topology model are detected by the isolated forest algorithm to obtain optimized power topology data.

[0118] Data correction unit 3 is used to acquire environmental data, perform environmental effect compensation on the optimized power topology data based on the environmental data to obtain compensated optimized power topology data; dynamically correct the clean energy metering data based on the compensated optimized power topology data using the Kalman filter algorithm to obtain actual energy metering data; and map the actual energy metering data to standard energy units using edge computing nodes to obtain standardized energy metering data.

[0119] Electricity usage standardization unit 4 is used to input standardized electricity metering data into a pre-trained LSTM-CNN fusion model, extract temporal features and fuse spatial features to obtain electricity usage pattern labels; compare the electricity usage pattern labels with the preset electricity usage standardization data to calculate the electricity compliance score and obtain a smart metering report.

[0120] The smart metering report includes real-time metering values, electricity usage patterns, and compliance scores.

[0121] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A real-time metering method for an intelligent modular energy metering box, characterized in that, include: Deploy a sensor array inside the power metering box; The sensor array is used to collect raw power data and environmental parameters to obtain raw power metering data stream; Dynamic threshold filtering is applied to the raw energy metering data stream to obtain clean energy metering data; Principal component analysis was performed on the clean energy metering data to obtain key energy parameters; A power topology model is constructed based on a graph neural network. The key power parameters are input into the power topology model to obtain initial power topology network data. Anomaly nodes in the power topology model are detected by the isolated forest algorithm to obtain optimized power topology data. Acquire environmental data, and perform environmental effect compensation on the optimized power topology data based on the environmental data to obtain compensated optimized power topology data; The clean energy metering data is dynamically corrected based on the compensated optimized power topology data using the Kalman filter algorithm to obtain the actual power metering data. The actual electricity metering data is mapped to standard electricity units using edge computing nodes to obtain standardized electricity metering data; Obtaining the compensated actual power consumption value includes the following steps: extracting node features and topological attributes from the optimized power topology data; collecting temperature and humidity sensor data to construct a multimodal dataset containing power parameters and environmental parameters to obtain an environmental power correlation feature matrix; constructing an electromagnetic interference compensation model based on a graph attention network, and learning the nonlinear influence of the environment on the multimodal dataset through multi-layer graph convolution to obtain node-level electromagnetic interference compensation coefficients; applying the node-level electromagnetic interference compensation coefficients to the optimized power topology data to perform reverse correction on voltage and current parameters to obtain preliminary corrected data after compensation; deploying edge computing nodes to perform Kalman filtering, using the compensated data as the observation value and the model prediction value as the state estimate, and dynamically correcting noise through prediction updates to obtain an actual power consumption value with a noise level below a threshold; verifying the compensation effect through simulated environmental change experiments, calculating the data error of the actual power consumption value before and after correction, to obtain the compensated actual power consumption value. The standardized electricity metering data is input into a pre-trained fusion model, which uses LSTM and CNN to extract temporal features and fuse spatial features to obtain electricity usage pattern labels. The electricity usage pattern labels are compared with preset electricity usage standard data to calculate the electricity usage compliance score and obtain a smart metering report. The smart metering report includes real-time metering values, electricity usage patterns, and compliance scores.

2. The real-time metering method for the intelligent modular power metering box according to claim 1, characterized in that, Deploy a sensor array inside the power metering box; The sensor array collects raw electrical energy data and environmental parameters to obtain a raw electrical energy metering data stream, including: A modular sensor array architecture is designed inside the power metering box. The sensor array includes a current transformer, a voltage sensor, and a temperature and humidity sensor. A communication link is established through the network to periodically collect raw power data and environmental parameters, and timestamps and device ID tags are added to obtain the raw power metering data stream.

3. The real-time metering method for the intelligent modular energy metering box according to claim 2, characterized in that, Dynamic threshold filtering is applied to the raw data to obtain clean energy metering data, including: A dynamic threshold benchmark library is constructed based on historical power data. The mean and standard deviation of the data within the window are calculated using a sliding window algorithm to obtain the adaptive threshold range. An environmental interference model is established by combining temperature and humidity sensor data to obtain environmental adaptive threshold parameters. Real-time deviation detection is performed on the raw power data stream, and the absolute error between the current data point and the mean of the sliding window is calculated; the soft threshold shrinkage method is used to denoise the high-frequency noise band to obtain preliminary denoised data; Cyclic redundancy check is used to check the integrity of data packets in order to obtain intermediate data after verification; The verified intermediate data is input into a Kalman filter, and state estimation is performed by combining dynamic threshold parameters and wavelet denoising results. Data deviation is corrected through prediction update cycles to obtain clean energy metering data.

4. The real-time metering method for the intelligent modular energy metering box according to claim 3, characterized in that, Principal component analysis was performed on the clean energy metering data to identify key energy parameters; A power topology model is constructed based on a graph neural network, and the power data of each circuit is correlated to obtain the initial power topology network data, including: The clean energy metering data is standardized, and principal components are screened by cumulative variance contribution rate to obtain a dimension-reduced feature matrix; the dimension-reduced feature matrix includes harmonic content, three-phase imbalance degree and power factor; A multimodal graph structure is constructed, and each circuit in the power metering box is defined as a graph node; the correlation of power parameters between the graph nodes is calculated using the Pearson correlation coefficient to obtain an initial power topology graph; A graph attention network is deployed to perform multi-layer graph convolution operations on the initial topology graph; edge weights are dynamically adjusted through an attention mechanism to obtain optimized power topology network data.

5. The real-time metering method for the intelligent modular energy metering box according to claim 4, characterized in that, The isolated forest algorithm is used to detect abnormal nodes in the topology network and optimize the power metering path to obtain optimized power topology data, including: Extract node feature vectors from the initial power topology network data; construct a node feature matrix and preserve edge connectivity to obtain topological feature data suitable for isolated forest input; An isolated forest model is constructed by randomly sampling a subspace on the node feature matrix to build multiple isolated trees; anomaly scores are calculated based on path lengths to obtain anomaly detection results. The topology network is optimized based on the anomaly detection results, and abnormal nodes are removed. The shortest path of the optimized topology network is calculated based on the energy metering path efficiency requirements to obtain the optimized edge weight matrix and node list. The optimization effect was verified by simulating the injection of abnormal data, and the anomaly detection accuracy and path delay were calculated to obtain optimized power topology network data.

6. The real-time metering method for the intelligent modular energy metering box according to claim 5, characterized in that, Edge computing nodes are used to map actual metering values ​​to standard energy units to obtain standardized energy metering data, including: Deploy a unit conversion engine on the edge computing node to receive the compensated actual power metering value; identify the power parameter type through a pre-trained neural network model to obtain initial standardized data; A dynamic calibration strategy is deployed, and the deviation between the initial standardized data and the standard value is calculated by combining the historical data of the standard energy meter stored at the edge node to obtain intermediate standardized data. A time-series model within edge computing nodes is used to perform nonlinear error correction on the intermediate standardized data; by capturing the time correlation of power parameters, the cumulative error in the unit conversion process is predicted and corrected to obtain standardized power metering data. The standardized electricity metering data is synchronized to the cloud blockchain platform, and a data fingerprint is generated through a hash algorithm and stored on the blockchain to obtain a standardized electricity metering data stream.

7. The real-time metering method for the intelligent modular energy metering box according to claim 6, characterized in that, The standardized electricity metering data is input into a pre-trained fusion model, and electricity usage pattern labels are obtained through temporal feature extraction and spatial feature fusion, including: The standardized electricity metering data is divided into multi-scale time windows to obtain short-time, medium-time, and long-time time series segments; frequency domain features of each segment are extracted by wavelet transform, and a multimodal feature matrix is ​​constructed by combining time domain statistics to obtain enhanced data; A parallel LSTM and CNN architecture is deployed, with the LSTM branch capturing the dynamic changes of the enhanced data and the CNN branch extracting the spatial distribution patterns of frequency domain features. The outputs of the two branches are fused through a cross-attention mechanism to obtain a fused feature vector.

8. The real-time metering method for the intelligent modular power metering box according to claim 7, characterized in that, The standardized electricity metering data is input into a pre-trained fusion model, and electricity usage pattern labels are obtained through temporal feature extraction and spatial feature fusion. The model also includes: The fused feature vector is input into a pre-trained classification layer and combined with an energy usage scenario label library for pattern matching; a reinforcement learning algorithm is used to dynamically adjust the classification threshold to obtain energy usage pattern labels. The accuracy of the labels is verified by backtracking historical electricity data, and the classification F1 value is calculated to obtain the electricity usage pattern label stream.

9. A real-time metering system for an intelligent modular energy metering box, applicable to the real-time metering method for the intelligent modular energy metering box as described in any one of claims 1-8, characterized in that, include: The data filtering unit (1) deploys a sensor array inside the power metering box; The sensor array is used to collect raw power data and environmental parameters to obtain raw power metering data stream; Dynamic threshold filtering is applied to the raw energy metering data stream to obtain clean energy metering data; The power topology unit (2) performs principal component analysis on the clean power metering data to obtain key power parameters; A power topology model is constructed based on a graph neural network. The key power parameters are input into the power topology model to obtain initial power topology network data. Anomaly nodes in the power topology model are detected by the isolated forest algorithm to obtain optimized power topology data. The data correction unit (3) acquires environmental data and performs environmental effect compensation on the optimized power topology data based on the environmental data to obtain the compensated optimized power topology data. The clean energy metering data is dynamically corrected based on the compensated optimized power topology data using the Kalman filter algorithm to obtain the actual power metering data. The actual electricity metering data is mapped to standard electricity units using edge computing nodes to obtain standardized electricity metering data; The electricity usage specification unit (4) inputs the standardized electricity metering data into a pre-trained fusion model, extracts time-series features and fuses spatial features to obtain an electricity usage mode label; compares the electricity usage mode label with the preset electricity usage specification data, calculates the electricity usage compliance score, and obtains a smart metering report; wherein, the smart metering report includes real-time metering values, electricity usage mode and compliance score.

Images