Electric energy meter and concentrator data fusion monitoring system and method

Abstract

The invention relates to the technical field of power system monitoring, in particular to an electric energy meter and concentrator data fusion monitoring system and method. Comprising the steps that data of an electric energy meter are periodically collected, integrity verification is conducted on the data, noise interference in the data is removed, then the data are fused to construct a power network diagram structure, and the power network diagram structure comprises node information and edge weight information; a graph convolutional network is used for modeling, a power network graph structure is used as the input of a graph neural network model, an adaptive learning rate adjustment algorithm is combined with an early stop method to prevent overfitting and deep training of the model, and space-time association between nodes is captured through multi-layer space-time graph convolution operation. According to the method, the power network graph structure is constructed by fusing the data of the electric energy meter through the concentrator, nodes and edges are analyzed by using the graph neural network, key nodes and potential fault points are identified, and the network health state is evaluated in real time, so that comprehensive monitoring and optimal management of the power system are realized, and the overall performance and reliability of the power system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and particularly to a data fusion monitoring system and method for an electric energy meter and a concentrator. Background Art

[0002] In the development process of modern power systems, the scale of the power grid continues to expand and the level of intelligence is increasing day by day, making the accurate monitoring and efficient analysis of power data a key factor in ensuring the normal operation of the power system. As a basic power metering device, the electric energy meter records in detail the power consumption of each user and power consumption node; the concentrator is responsible for summarizing the data of multiple electric energy meters.

[0003] However, traditional power monitoring methods have significant drawbacks. Firstly, in terms of data processing, only isolated operations or simple summaries are performed on the electric energy meter data, completely ignoring the overall architecture and dynamic changes of the power network, and it is impossible to build a comprehensive power network analysis system. Secondly, when dealing with large-scale power distribution networks, due to the interference of various factors such as differences in user power consumption behavior, electrical equipment failures, and changes in the power grid topology structure during the operation of the power system, the power metering data is extremely prone to abnormal fluctuations. In the face of such situations in the existing technology, if the data of the electric energy meters in a certain area suddenly becomes abnormal, it is difficult to accurately determine whether the problem is due to abnormal user power consumption or power grid transmission failures because it cannot be associated with the overall operation situation of the entire power network, seriously affecting the stable operation and reasonable distribution of the power system. Thirdly, although some existing monitoring technologies can perform basic data collection and statistical analysis, they lack the ability to analyze the internal connections between power network nodes and edges, and cannot effectively identify key nodes and potential fault points. Especially in the scenarios of smart grid optimization and complex power system management, there is a lack of an effective way to real-time evaluate the health status of the power network, and it cannot provide timely and accurate decision-making support for power distribution optimization and fault handling.

[0004] In summary, the current power industry urgently needs an innovative technology to achieve the deep fusion of electric energy meter and concentrator data, and use advanced graph neural network technology to conduct a comprehensive and in-depth analysis of the power network, thereby enhancing the accuracy and reliability of power system monitoring, and effectively ensuring the safe and stable operation of the power system. Summary of the Invention

[0005] The present invention aims to provide a data fusion monitoring system for an electric energy meter and a concentrator based on a graph neural network. By using the concentrator to fuse the electric energy meter data to construct a power network graph structure, and using the graph neural network to analyze nodes and edges, identify key nodes and potential fault points, and real-time evaluate the network health status, so as to achieve the comprehensive monitoring and optimized management of the power system, provide strong support for power distribution and fault handling, and improve the overall performance and reliability of the power system.

[0006] On the one hand, the technical solution of the present invention: a method for monitoring the data fusion of an electric energy meter and a concentrator, comprising the following steps:

[0007] The concentrator is connected to the electric energy meter by using high-speed power line carrier communication technology, periodically collects the data of the electric energy meter, and uses the cyclic redundancy check algorithm to perform integrity check on the data. After the check, data preprocessing is carried out to remove the noise interference in the data, and then the data is fused to construct a power network graph structure, and the power network graph structure includes node information and edge weight information;

[0008] Modeling is carried out by using a graph convolutional network. The power network graph structure is used as the input of the graph neural network model. The adaptive learning rate adjustment algorithm is combined with the early stopping method to prevent overfitting. The model is deeply trained, the model parameters are optimized, and the spatio-temporal correlation between nodes is captured through multi-layer spatio-temporal graph convolution operations; An evaluation system is constructed through multi-dimensional indicators to identify key nodes and potential fault points. If the node criticality index exceeds the preset threshold, it is determined as a key node, and the deep learning model is used to analyze the node feature vectors to identify potential fault points;

[0009] Integrate the electrical energy data characteristics, connection relationships, edge weights and dynamic change information of the nodes, calculate the network health index by using a method combining a multi-layer perceptron based on deep learning and an attention mechanism, and subdivide the network health status. When an anomaly or potential fault is detected, warning information is sent through multiple channels. At the same time, relying on big data and deep reinforcement learning technologies, fuse multi-party information to construct an intelligent decision-making model, formulate optimization decision-making suggestions, and track and evaluate the implementation effects;

[0010] Based on the long-term accumulated data and operation records, use big data analysis technology and machine learning algorithms to analyze the data, automatically adjust the system parameters and the algorithm model structure, and visually display the output results to the management end.

[0011] Optionally, the data preprocessing uses the Z-score standardization method to normalize the voltage, current, and power data of the electric energy meter;

[0012] Suppose the voltage data sequence collected by a certain electric energy meter is V = [v 1 , v 2 , …, v n , its mean is and the standard deviation is σ v , then the standardized voltage data is mapped to the standard normal distribution interval;

[0013] Use wavelet transform to remove the high-frequency noise in the data. Select the db4 wavelet basis function to decompose the data, and process the high-frequency coefficients by setting a threshold to remove the noise;

[0014] The Min-Max normalization technique is used to normalize the data of the electricity meter, mapping the voltage, current, and power data to the interval [0,1]. The Kalman filter algorithm is used to smooth the data and remove noise interference.

[0015] Optionally, the initial feature vector of the node information includes the data of the electricity meter after preprocessing, as well as the geographical location information, operating time information, and device type information of the electricity meter. The weight of the edge is calculated based on the resistance and reactance of the line.

[0016] Optionally, the basic operation of the graph convolutional network model is expressed as:

[0017]

[0018] where H (l) is the node feature matrix of the l-th layer, W (l) is the weight matrix of the l-th layer, σ is the ReLU activation function, is the adjacency matrix A with self-loops added, representing the adjacency matrix of the graph, I is the diagonal matrix, is the normalized degree matrix.

[0019] Optionally, in the multi-dimensional index construction evaluation system, the node criticality index is defined as:

[0020] I nk = α × E var + β × C deg + γ × P sta + δ × T cor

[0021] where E var is the variance of power consumption, reflecting the severity of power fluctuations; C deg is the degree centrality of the node, measuring the connection criticality of the node in the network topology; P sta is the stability index of the power factor, obtained by calculating the standard deviation of the power factor and normalizing it; T cor is the time correlation coefficient of the power data between the node and its surrounding nodes, obtained by calculating the Pearson correlation coefficient, reflecting the co-variation relationship between nodes; α, β, γ, δ are weight coefficients.

[0022] Optionally, the deep learning model uses an autoencoder or a variational autoencoder. The autoencoder uses a large number of node feature vectors in the normal operating state as training samples to train a multi-layer autoencoder network so that it can reconstruct the input feature vector; during detection, the feature vector of the node to be detected is input into the trained autoencoder, and the reconstruction error is calculated:

[0023]

[0024] where \(x\) is the original feature vector, and \(\hat{x}\) is the reconstructed feature vector; when the reconstruction error \(e\) is greater than a preset threshold, it is marked as a potential fault point.

[0025] Optionally, the node electrical energy data features include the electrical energy consumption growth rate, voltage deviation, and current harmonic distortion rate.

[0026] Optionally, the warning information includes the fault location, type, affected range, cause analysis, and emergency measures.

[0027] Optionally, the intelligent decision-making model is constructed by deeply integrating the network health assessment results, warning information, power system operation rules, and historical experience data.

[0028] On the other hand, the present invention proposes an electric energy meter and concentrator data fusion monitoring system, which is applied to the above method and includes:

[0029] An electric energy meter-concentrator data acquisition and graph construction module, which periodically acquires the data of the electric energy meter, uses the cyclic redundancy check algorithm to perform integrity verification on the data, performs data preprocessing after verification to remove noise interference in the data, and then fuses the data to construct a power network graph structure, where the power network graph structure includes node information and edge weight information;

[0030] A graph convolutional network modeling and key node identification module, which takes the power network graph structure as the input of the graph convolutional network model, uses an adaptive learning rate adjustment algorithm combined with early stopping to prevent overfitting, deeply trains the model, optimizes the model parameters, and captures the spatio-temporal correlation between nodes through multi-layer spatio-temporal graph convolutional operations; identifies key nodes and potential fault points through a multi-dimensional index construction evaluation system;

[0031] A power network health assessment and warning decision-making module, which integrates the electrical energy data features, connection relationships, edge weights, and dynamic change information of the nodes, uses a method combining a multi-layer perceptron based on deep learning and an attention mechanism to calculate the network health index, subdivides the network health status, and when an anomaly or potential fault is detected, sends warning information through multiple channels. At the same time, relying on big data and deep reinforcement learning technologies, it fuses multi-party information to construct an intelligent decision-making model, formulates optimized decision-making suggestions, and tracks and evaluates the implementation effects;

[0032] A system performance optimization and intelligent auxiliary decision-making module, which analyzes the data based on long-term accumulated data and operation records, uses big data analysis technologies and machine learning algorithms to analyze the data, automatically adjusts the system parameters and algorithm model structure, and visually displays the output results to the management side.

[0033] Compared with the prior art, the present application includes at least one of the following beneficial technical effects:

[0034] 1. The present invention adopts the method of constructing the power network graph structure by integrating the data of the concentrator and the electricity meter, which can comprehensively and systematically reflect the topological relationship of the power network and the electricity transmission situation, providing a solid foundation for the operation analysis and fault diagnosis of the power system, and effectively improving the integrity and accuracy of power system monitoring.

[0035] 2. By using the graph neural network to analyze the nodes and edges in the power network, through multi-layer spatio-temporal graph convolution operations, it can accurately capture the spatio-temporal correlation relationships and dynamic change trends between nodes, timely discover potential problems and fault hidden dangers in the power network, and significantly improve the reliability and stability of the power system.

[0036] 3. Constructing a multi-dimensional index evaluation system, such as the node criticality index, can comprehensively and objectively evaluate the health status of the power network, provide a scientific basis for optimizing power distribution and fault handling, and ensure the efficient operation of the power system and the rational utilization of power resources.

[0037] 4. Using the method of combining a multi-layer perceptron based on deep learning with an attention mechanism to calculate the network health index can reflect the health status of the power network in real time and accurately, send out early warning information in a timely manner, enabling the operation and maintenance personnel to quickly take measures for handling, and effectively avoiding the expansion of faults and the occurrence of power accidents.

[0038] 5. The intelligent decision-making model constructed relying on big data analysis and deep reinforcement learning technologies can quickly formulate scientific and reasonable optimization decisions and fault handling strategies according to the operation status and fault information of the power system, improving the operation efficiency and emergency handling ability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is the system architecture diagram of the present invention;

[0040] Figure 2 is the flowchart of data processing of the present invention;

[0041] Figure 3 is the structure diagram of the graph convolution network model of the present invention;

[0042] Figure 4 is the schematic diagram of the key node identification process in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0043] The technical solutions of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments. The components of the embodiments of the present disclosure described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the claimed present disclosure, but merely represents the selected embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without creative efforts belong to the scope of protection of the present disclosure.

[0044] As Figure 1 shows the architecture and workflow of the entire electricity meter and concentrator data fusion monitoring system. Data is collected from the electricity meter, collected and preprocessed through the concentrator, and then the data is input into the graph convolutional network for modeling and analysis. In the graph convolutional network modeling and key node identification module, the graph structure of the power network is modeled and trained through the graph convolutional network (GCN) to identify key nodes and potential fault points. At the same time, the power network health assessment and early warning decision-making module evaluates the health status of the power network and issues early warning information and formulates optimization decisions according to the evaluation results.

[0045] The system performance optimization and intelligent auxiliary decision-making module regularly evaluates and optimizes the system performance to ensure the stable operation of the system. Finally, through the monitoring and management of the operation and maintenance personnel on the system, the comprehensive monitoring and optimized management of the power system are realized.

[0046] The specific implementation process is as follows:

[0047] 1. Electricity meter - concentrator data acquisition and graph construction module

[0048] 1.1 Data acquisition input

[0049] The concentrator uses high-speed power line carrier communication (PLC) technology to establish a connection with the electricity meter, and collects data such as the voltage (unit: volt, accuracy of ±0.5V), current (unit: ampere, accuracy of ±0.1A), power (unit: watt, accuracy of ±5W), power factor (accuracy of ±0.02), etc. of the electricity meter at 15-minute intervals. During the acquisition process, the cyclic redundancy check (CRC) algorithm is used to perform integrity check on the data. If the check fails, the data is automatically retransmitted. For example, when the CRC value of the received data packet does not match the calculated value, the retransmission mechanism is triggered to ensure the accuracy of the data.

[0050] 1.2 Data preprocessing

[0051] Preprocess the collected data. Use the Z-score normalization method to normalize data such as voltage, current, and power. Suppose the voltage data sequence collected by a certain electricity meter is V = [v 1 , v 2 , …, v n , its mean is and the standard deviation is σ v . Then the normalized voltage data is mapped to the standard normal distribution interval.

[0052] Use wavelet transform to remove high-frequency noise in the data. Select the db4 wavelet basis function to decompose the data. By setting an appropriate threshold (such as using the soft threshold method, the threshold where σ is the estimated value of the noise standard deviation and n is the data length) to process the high-frequency coefficients, and then reconstruct after removing the noise to obtain the denoised data set.

[0053] For voltage data, if a certain data point exceeds the range of the mean ± 3 times the standard deviation, or is outside the upper and lower limits of the box plot, it is determined as an outlier and removed. Subsequently, use the Min-Max normalization technique to normalize the data, mapping data such as voltage, current, and power to the [0, 1] interval for subsequent efficient processing. Use an advanced Kalman filter algorithm to smooth the data, effectively removing noise interference and further improving the data quality. When processing current data, the Kalman filter algorithm can dynamically estimate the true value of the current based on historical data and the current measurement value, significantly reducing the impact of measurement noise and making the processed data more stable and reliable.

[0054] 1.3 Node Feature Definition

[0055] When constructing the graph structure, use the electricity meter as a node. In addition to the preprocessed electricity data mentioned above, the initial feature vector of the node also includes the geographical location information of the electricity meter (latitude and longitude coordinates, with an accuracy of 6 decimal places), the operation time information (in seconds, recording the operation duration since installation), and the device type information (such as single-phase electricity meter, three-phase electricity meter, etc.).

[0056] 1.4 Edge Weight Calculation

[0057] The weight of the edge is calculated according to the resistance (unit: ohm, with an accuracy of ±0.01Ω) and reactance (unit: ohm, with an accuracy of ±0.02Ω) of the line. For example, for the edge connecting two nodes i and j, its weight is:

[0058]

[0059] where R ij and X ijThey are the resistance and reactance of the line respectively. The accuracies of the resistance and reactance are ±0.01Ω and ±0.02Ω respectively. These accurate parameters are crucial for accurately calculating the edge weights and can reflect the electrical characteristics of the line and its impact on power transmission.

[0060] 1.5 Graph structure formation

[0061] By combining nodes and edges according to the above rules, the power network graph structure is constructed. In this network, the connection relationship between nodes is reflected by edges, and the weight of the edge reflects the characteristics of the line and its importance for power transmission. Such a graph structure can intuitively display the topological structure of the power network and the path of power transmission, providing a basis for subsequent analysis and processing.

[0062] The final output of this module is a complete power network graph structure, including node information (such as the feature vector of the electricity meter) and edge weight information. This graph structure comprehensively reflects the topological structure of the power network and the relationship between electricity meters, providing input data for subsequent graph neural network analysis.

[0063] 2. Graph Neural Network (GNN) Modeling and Key Node Identification Module

[0064] 2.1 Power network graph structure

[0065] Figure 3 The model structure of the Graph Convolutional Network (GCN) is presented. The input includes node features, which are processed through the adjacency matrix and degree matrix, and then graph convolution operations are performed. The activation function performs a non-linear transformation on the convolution result, and finally the feature representation of the node is output. In the model, through multi-layer convolution operations, the feature information of nodes and their neighbor nodes in the spatio-temporal dimension is deeply aggregated, so as to accurately capture the complex spatio-temporal correlation relationship and dynamic change trend between nodes. At the same time, the parameters and functions of each part are also marked in the figure, such as the node feature matrix, weight matrix, activation function, etc., which helps to deeply understand the working principle and feature extraction method of GCN.

[0066] The power network graph structure output by the electricity meter-concentrator data acquisition and graph construction module is used as the input of the GNN model. This graph structure contains the feature information of nodes (such as the preprocessed data of electricity meters, geographical location, running time, device type, etc.) and the edge weight information, which are the basis for the GNN model to perform analysis and learning.

[0067] 2.2 Graph Neural Network (GNN) Modeling

[0068] The Graph Convolutional Network (GCN) is used for modeling, and the basic operation of GCN is defined as:

[0069]

[0070] where H (l) is the node feature matrix of the l-th layer, W (l) is the weight matrix of the l-th layer, σ is the ReLU activation function, is the adjacency matrix A with self-loops added, representing the adjacency matrix of the graph, and I is the diagonal matrix, is the normalized degree matrix.

[0071] 2.3 Training Process

[0072] During the training process, an adaptive learning rate adjustment algorithm (such as the AdamW optimizer, with an initial learning rate set to 0.0005 and a weight decay coefficient of 0.01) is used, combined with early stopping (patience set to 10) to prevent overfitting. Through a large number of training samples (such as more than 10,000 different working condition data samples), deep training is carried out to continuously optimize the model parameters. For example, during the training process, every 100 training batches, the AdamW optimizer will automatically adjust the learning rate according to the current gradient information and parameter status to ensure that the model can quickly converge to the optimal solution while avoiding getting stuck in local optima. Through multi-layer spatio-temporal graph convolution operations, the feature information of nodes and their neighbor nodes in the spatio-temporal dimension is deeply aggregated, thereby accurately capturing the complex spatio-temporal correlation relationships and dynamic change trends between nodes. For example, after 5 layers of ST-GCN operations, the node feature vector can comprehensively reflect key information such as the co-variation law of power consumption and the voltage fluctuation propagation characteristics with its surrounding neighbor nodes at different time scales.

[0073] 2.4 Key Node Identification

[0074] To more accurately identify key nodes and potential fault points, an evaluation system integrating multi-dimensional indicators is constructed. The node criticality index is defined as:

[0075] I nk = α × E var + β × C deg + γ × P sta + δ × T cor

[0076] where E var is the variance of power consumption, reflecting the severity of power fluctuations; C deg is the degree centrality of the node, measuring the connection criticality of the node in the network topology; P sta is the stability index of the power factor, obtained by calculating the standard deviation of the power factor and normalizing it; T coris the time correlation coefficient of electrical energy data between a node and its surrounding nodes, which is obtained by calculating the Pearson correlation coefficient and reflects the co-variation relationship between nodes. α, β, γ, δ are weight coefficients (determined through in-depth analysis and experiments using a large amount of historical fault data and actual power grid operation data, e.g., α = 0.3, β = 0.25, γ = 0.2, δ = 0.25).

[0077] When I nk is greater than the preset threshold T nk (e.g., T nk = 0.8), it is determined as a key node. For potential fault points, an anomaly detection model based on deep learning, such as an autoencoder (AE) or a variational autoencoder (VAE), is used to conduct in-depth analysis on the node feature vectors. Taking the autoencoder as an example, a large number of node feature vectors in the normal operating state are used as training samples to train a multi-layer autoencoder network so that it can reconstruct the input feature vectors. During detection, the feature vectors of the nodes to be detected are input into the trained autoencoder, and the reconstruction error is calculated:

[0078]

[0079] where x is the original feature vector, is the reconstructed feature vector. When the reconstruction error e is greater than the preset threshold T e (e.g., T e = 0.5), it is marked as a potential fault point. In this way, it is possible to more sensitively capture the subtle abnormal changes in node features and effectively improve the recognition accuracy of potential fault points.

[0080] For example Figure 4 shows the process of key node recognition. First, according to the calculation formula of the node criticality index, factors such as the variance of power consumption, node degree centrality, power factor stability, and time correlation coefficient of electrical energy data are comprehensively considered to calculate the criticality index of each node. Then, the calculated criticality index is compared with the preset threshold. When the index is greater than the threshold, the node is determined as a key node. For potential fault points, deep learning models such as autoencoders are used to conduct in-depth analysis on the node feature vectors and calculate the reconstruction error. When the reconstruction error is greater than the preset threshold, the node is marked as a potential fault point. Finally, the information of key nodes and potential fault points is output. The output of this module is the information of key nodes and potential fault points in the power network, including the identification, location, and related feature indicators of the nodes. This information has important guiding significance for the operation and maintenance and fault handling of the power system, and can help the operation and maintenance personnel to discover and solve problems in a timely manner and ensure the safe and stable operation of the power system.

[0081] 3. Power Network Health Assessment and Early Warning Decision-making Module

[0082] 3.1 Data Input

[0083] Integrate the electrical energy data characteristics of the nodes (such as the growth rate of electrical energy consumption r E , voltage deviation ΔV, current harmonic distortion rate THD i , etc.), the connection relationship of the nodes (deeply reflected by the spatio-temporal adjacency matrix), and the weight and dynamic change information of the edges (including the influence of factors such as the line temperature and load change rate monitored in real time on the edge weight).

[0084] 3.2 Evaluation Model Construction and Calculation

[0085] Adopt a method that combines a multi-layer perceptron (MLP) based on deep learning with an attention mechanism to calculate the network health index H I . Specifically, first input the above multi-dimensional feature data into the MLP network for feature extraction and preliminary mapping to obtain an intermediate feature representation. Then, use the attention mechanism to dynamically weight the importance of different features and nodes, enabling the network to focus on key information. Finally, output the network health index H through a fully connected layer I . For example, the attention mechanism assigns different weights to different nodes and features according to the key degree of the nodes and the abnormality degree of the data, making the calculation of the network health index more accurate and targeted

[0086] 3.3 Health Status Evaluation and Grading

[0087] According to the value of, the health status of the power network is subdivided into multiple fine levels such as healthy, slightly sub-healthy, moderately sub-healthy, severely sub-healthy, fault warning, and fault. For example, when H I > 0.9, it is in a healthy state, and when 0.7 < H I < 0.9, it is in a slightly sub-healthy state, and so on

[0088] Each level corresponds to different maintenance and treatment strategies, providing more targeted guidance for power operation and maintenance. For example, for a healthy power network, only regular monitoring and maintenance are required; for a slightly sub-healthy network, monitoring and analysis need to be strengthened to detect potential problems in a timely manner; for a network in a fault warning or fault state, emergency measures need to be taken for treatment, such as fault troubleshooting and equipment repair, to ensure the safe and stable operation of the power system

[0089] 3.4 Early Warning Decision-making Process

[0090] When the network health assessment module detects an anomaly or potential fault, warning messages are quickly sent through multiple channels. The SMS content not only includes the fault location (accurate to the electricity meter number or specific line branch), fault type (detailed descriptions such as voltage dip, current overload, abnormal power factor fluctuation, etc.), and the possible affected scope (specific user areas involved, types and estimated quantities of electrical equipment), but also provides a preliminary analysis of the fault cause and recommended emergency handling measures. The email sends a detailed fault analysis report, including detailed data charts before and after the fault (such as voltage, current, power curves), comparative analysis with historical fault data, fault prediction trends based on models, and information on possible fault root cause investigation paths, etc., providing comprehensive and in-depth decision-making support for operation and maintenance personnel.

[0091] Meanwhile, a system pop-up window prominently displays the fault information on the monitoring interface of the power dispatching center. The pop-up window also provides one-key operation buttons, facilitating operation and maintenance personnel to quickly activate the emergency plan or view detailed information. Relying on big data analysis and deep reinforcement learning technologies, the network health assessment results, warning messages, and rich power system operation rules and massive historical experience data are deeply integrated to build an intelligent decision-making model. For example, using reinforcement learning algorithms such as Deep Q-Network (DQN) or Proximal Policy Optimization (PPO), taking the operating state of the power system as the environmental state and different decision-making strategies (such as power distribution adjustment plans, equipment maintenance schedules, load transfer strategies, etc.) as the action space, through continuous interaction and learning with the environment, to find the optimal decision-making strategy.

[0092] The output of this module is detailed warning messages and optimized decision-making suggestions, including fault handling solutions, power distribution adjustment measures, equipment maintenance plans, etc. These messages can help operation and maintenance personnel take effective measures in a timely manner to deal with abnormal situations in the power network and ensure the safe and stable operation of the power system. At the same time, the system will continuously monitor the implementation effect of the decision and dynamically adjust the decision-making strategy according to the actual situation, continuously improving the operation efficiency and reliability of the power system.

[0093] 4. System Performance Optimization and Intelligent Assistant Decision-making Module

[0094] 4.1 Data Foundation and Analysis Methods

[0095] Based on the massive power data and system operation records accumulated over a long time, these data include the acquisition data of electricity meters, the summary data of concentrators, the operation status data of equipment, etc. These data provide rich materials and a basis for system performance evaluation and optimization.

[0096] Using big data analysis techniques and machine learning algorithms, such as the combination of principal component analysis (PCA) and random forest (RF) algorithms, the data is analyzed. PCA is used to reduce the dimensionality of high-dimensional data, extract key features, and reduce data complexity; the RF algorithm is used to build a prediction model to evaluate the impact of different parameters and configurations on the accuracy, real-time performance, and stability of the system.

[0097] 4.2 Performance Optimization Process

[0098] According to the analysis results, the parameter settings and algorithm model structure of the system are automatically adjusted. For example, optimize the data acquisition frequency, and reasonably adjust the time interval of data acquisition according to the load changes and data importance of the power grid to improve the efficiency and quality of data acquisition; adjust the number of layers and nodes of the graph neural network, and select an appropriate neural network structure according to the scale and complexity of the power grid to improve the fitting ability and generalization ability of the model; improve the setting of the warning threshold, and determine a more accurate warning threshold through continuous learning and optimization to improve the accuracy and timeliness of warnings.

[0099] Through these optimization measures, continuously improve the performance and adaptability of the system, so that it can better adapt to different power system operating environments and working conditions changes. For example, during the peak power load period, the system can quickly and accurately monitor and analyze the operating status of the power grid, issue warning information in a timely manner, and provide effective decision-making support to ensure the safe and stable operation of the power system.

[0100] 4.3 Intelligent Assistant Decision-making Tools

[0101] Provide intelligent decision-making support tools for power operation and maintenance personnel, such as a web-based visual decision-making platform. This platform displays information such as the real-time operating status of the power system, the distribution of key nodes and potential fault points, the results of network health assessment, and historical data comparative analysis through intuitive charts (such as bar charts, line charts, network diagrams, etc.). These charts can help operation and maintenance personnel quickly understand the overall operating situation of the power system and discover potential problems and risks.

[0102] The platform integrates a variety of decision-making simulation and evaluation functions. Operation and maintenance personnel can input different decision-making schemes (such as equipment replacement plans, power dispatching strategy adjustments, etc.) on the platform, and the platform uses built-in simulation models and historical data to quickly simulate and evaluate the decision-making effects. For example, when formulating a power dispatching strategy adjustment plan, operation and maintenance personnel can set different load distribution ratios and power generation plans on the platform, and the platform shows the changes in system operating indicators (such as line load rate, power loss, voltage stability, etc.) under different schemes through simulation calculations, helping operation and maintenance personnel select the optimal decision-making scheme and improve the efficiency and scientificity of power operation and maintenance management.

[0103] The output of this module is optimized system performance and intelligent auxiliary decision-making support, including higher system accuracy, real-time performance, and stability, as well as more scientific and reasonable decision-making schemes. These output results can help power operation and maintenance personnel better manage and maintain the power system, improve the operation efficiency and reliability of the power system, and provide strong support for the development of the power industry.

[0104] Figure 2 It details the entire process of data from collection to processing, analysis, evaluation, and decision-making. First, the concentrator collects data such as voltage, current, power, and power factor of the electricity meters through high-speed power line carrier communication (PLC) technology and performs cyclic redundancy check (CRC). Then, preprocessing operations such as Z-score normalization and wavelet transform are carried out on the data to improve data quality. Next, a power network graph structure is constructed, and graph neural network training is performed to identify key nodes and potential fault points. After that, information such as node electrical energy data characteristics and connection relationships is integrated, network health indicators are calculated, and the health status of the power network is evaluated. According to the evaluation results, early warning information is issued, and an optimization decision is made. Finally, the implementation effect of the decision is tracked and evaluated, and the decision-making strategy is continuously adjusted and improved to improve the operation efficiency and reliability of the power system.

[0105] It should be noted in this embodiment that in a certain urban power network, which covers multiple commercial areas, residential areas, and industrial areas, the electricity meter and concentrator data fusion monitoring system of the present invention is deployed.

[0106] Electricity meters are widely installed at each user end and important power nodes to accurately record electricity consumption data. The concentrator stably collects key data such as voltage, current, power, and power factor of the electricity meters through high-speed power line carrier communication (PLC) technology according to a fixed 15-minute cycle. For example, in a data collection, the electricity meter in a commercial area shows a voltage of 220V (fluctuating within the range of ±0.5V in accuracy), a current of 50A (with an accuracy of ±0.1A), a power of 11kW (with an accuracy of ±5W), and a power factor of 0.9 (with an accuracy of ±0.02). During the collection process, the cyclic redundancy check (CRC) algorithm strictly guarantees the integrity of the data. When the CRC value of a collected data packet does not match the calculated value, the system immediately automatically retransmits the data to ensure the accuracy of the data.

[0107] The collected data is then preprocessed. Taking a sequence of 100 data points of voltage data as an example, its mean is calculated to be 220.3V and the standard deviation is 1.2V. After Z-score normalization, the data is mapped to the standard normal distribution interval, effectively removing the influence of the data dimension and highlighting the relative change characteristics of the data. At the same time, wavelet transform is used to remove high-frequency noise. The db4 wavelet basis function is selected to decompose the data. After processing the high-frequency coefficients by the soft threshold method (the threshold is calculated according to the estimated value of the noise standard deviation and the data length), the data set is reconstructed, further improving the data quality. Then, the Min-Max normalization technique is used to map data such as voltage, current, and power to the interval [0,1], facilitating subsequent data analysis and processing. When processing current data, the Kalman filter algorithm dynamically estimates the true value of the current based on historical data and current measurement values, making the processed data more stable and reliable. For example, after filtering the current data in a certain period, the fluctuation range is significantly reduced, better reflecting the actual power consumption trend.

[0108] Based on the preprocessed data, the system constructs the power network graph structure. Taking the electricity meters as nodes, the initial feature vectors of the nodes not only include the processed electricity data mentioned above, but also cover the geographical location information of the electricity meters (accurate to 6 decimal places of the longitude and latitude coordinates, such as the location of an electricity meter in a residential area is 30.123456° north latitude and 120.654321° east longitude), the operation time information (recorded as 365 days, 12 hours, 30 minutes, and 15 seconds since installation), and the device type information (such as most of the electricity meters in this area are single-phase electricity meters). The weights of the edges are accurately calculated based on the resistance (such as the resistance of a section of the line connecting two nodes is 0.5Ω, with an accuracy of ±0.01Ω) and reactance (0.3Ω, with an accuracy of ±0.02Ω) of the line, comprehensively reflecting the electrical characteristics of the line and its influence on power transmission.

[0109] The graph neural network (GNN) models and analyzes the constructed graph structure. Using the graph convolutional network (GCN), during the training process, the adaptive learning rate adjustment algorithm (AdamW optimizer, with an initial learning rate set to 0.0005 and a weight decay coefficient of 0.01) is combined with the early stopping method (patience is 10), and a large number of historical data samples (more than 10,000) under different working conditions are used for in-depth training. For example, in the early stage of training, the model gradually adjusts the weight parameters by continuously learning the relationship between node features and edges. After 500 iterations, the ability to capture the association between nodes has been significantly improved. Through multi-layer spatio-temporal graph convolutional operations, the spatio-temporal association relationship and dynamic change trend between nodes are accurately captured. For example, it is found that during the peak power consumption period in an industrial area, the power consumption of multiple surrounding electricity meters shows a coordinated growth trend, which is closely related to the load change of the line, providing an important basis for subsequent power dispatching.

[0110] An evaluation system constructed by integrating multi-dimensional indicators such as the variance of comprehensive power consumption, node degree centrality, power factor stability, and the time correlation coefficient of power data can effectively identify key nodes and potential fault points. In one analysis, the key degree index of a power meter node in a commercial area reached 0.85 (the preset threshold is 0.8), and it was determined as a key node because its power consumption variance was large and the time correlation coefficient of its power data with surrounding nodes was high, indicating that it has an important impact on the power supply stability of the entire area. For potential fault points, an autoencoder is used to deeply analyze the node feature vectors. When the reconstruction error of a certain node reaches 0.55 (the preset threshold is 0.5), it is marked as a potential fault point. Further inspection found that there was a slight poor contact problem with the power meter corresponding to this node, and it was repaired in time to avoid the occurrence of potential faults.

[0111] The power network health assessment and early warning decision-making module runs in real time. By integrating the power data characteristics of nodes (such as the power consumption growth rate in some areas reaching 10% during a certain period, the voltage deviation fluctuating by ±5V, and the current harmonic distortion rate being 5%, etc.), connection relationships, edge weights, and dynamic change information (such as a certain line's resistance increasing due to temperature rise and the corresponding change in edge weight), a method combining a multi-layer perceptron (MLP) based on deep learning with an attention mechanism is used to calculate the network health index HI. According to the HI value, the system can accurately judge the health status of the power network. When HI drops to 0.75 (in a mild sub-healthy state), the system responds quickly. Early warning information is sent in a timely manner through multiple channels such as text messages, emails, and system pop-ups. The text message content details the fault location (accurate to the specific power meter number, such as the area where the 00123 power meter is located), the fault type (abnormal voltage fluctuation), the possible affected range (50 surrounding households and 3 small shops), provides a preliminary analysis of the fault cause (possibly due to nearby construction interfering with the line), and recommended emergency handling measures (such as temporarily adjusting the power distribution in the surrounding area). The email sends a detailed fault analysis report, including voltage, current, and power curves before and after the fault, a comparative analysis with historical fault data showing that this fluctuation is similar to a previous one caused by weather reasons, a model-based fault prediction trend indicating that if not handled in time, it may lead to a local power outage, and possible paths for troubleshooting the root cause of the fault (first check the line connection points, then detect nearby power equipment), etc., providing comprehensive and in-depth decision-making support for operation and maintenance personnel. At the same time, the system pop-up prominently displays the fault information on the monitoring interface of the power dispatching center. The pop-up provides one-button operation buttons, facilitating operation and maintenance personnel to quickly start the emergency plan or view detailed information.

[0112] An intelligent decision-making model built based on big data analysis and deep reinforcement learning technologies quickly formulates optimized decision-making suggestions. For example, when there is a shortage of power supply and potential fault risks in a certain area, the decision-making model is based on the real-time electricity consumption demand in that area (through the analysis of electricity consumption data in the past week, it is obtained that the electricity consumption in that area increases by 15% per hour during the peak electricity consumption period on weekdays), the priority of electricity consumption demand (industrial users have the highest priority, followed by residential users, and commercial users are relatively low), the real-time topological structure of the power grid (including the connection status of nodes and edges, load conditions, such as the load rate of the connection lines between a certain substation and multiple surrounding areas reaches 80%), and the adjustable power resources (there is a standby power station nearby with a power generation capacity of 5 MW, and the access point is 2 km away from the fault area), etc. to formulate a scientific and reasonable power distribution adjustment plan. It decides to start the standby power station and adjust the power distribution in some areas, giving priority to ensuring the basic production electricity consumption demand of key industrial users, and at the same time adopting a time-of-use power supply strategy for residential users to reduce the non-essential electricity consumption of commercial users. During the implementation process, the system continuously monitors and evaluates the decision-making effect, and dynamically adjusts the decision-making strategy according to the actual situation. If it is found that starting the standby power station causes a certain impact on the surrounding lines, the power generation power and grid connection strategy are adjusted in a timely manner to ensure the stable and reliable operation of the power system, effectively improving the operation efficiency and reliability of the urban power grid.

[0113] The above specific embodiments are only several alternative embodiments of the present invention. Based on the technical solution of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for monitoring electric energy meter and concentrator data fusion, characterized in that: The following steps are involved: The concentrator uses high-speed power line carrier communication technology to connect with the electric energy meter, periodically collects data from the electric energy meter, and uses a cyclic redundancy check algorithm to check the integrity of the data. After the check, the data is pre-processed to remove noise interference in the data, and then the data is integrated to construct a power network graph structure, which includes node information and edge weight information; The graph convolutional network is used for modeling. The power network graph structure is used as the input of the graph neural network model. The adaptive learning rate adjustment algorithm is combined with the early stopping method to prevent overfitting, deeply train the model, optimize the model parameters, and capture the spatiotemporal correlation between nodes through multi-layer spatiotemporal graph convolution operations. An evaluation system is built through multi-dimensional indicators to identify key nodes and potential fault points. If the node criticality index exceeds the preset threshold, it is determined to be a key node, and a deep learning model is used to analyze the node feature vector to identify potential fault points. Integrate the power data characteristics, connection relationships, edge weights and dynamic change information of the nodes, use a method based on deep learning combined with a multi-layer perceptron and an attention mechanism to calculate network health indicators, and subdivide the network health status. When an abnormality or potential failure is detected, send warning information through multiple channels. At the same time, relying on big data and deep reinforcement learning technology, integrate multi-party information to build an intelligent decision-making model, formulate optimized decision-making suggestions, and track and evaluate the implementation effect; Based on long-term accumulated data and operation records, big data analysis technology and machine learning algorithms are used to analyze data, automatically adjust system parameters and algorithm model structure, and visualize the output results to the management end.

2. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 1 is characterized in that: The data preprocessing uses the Z-score standardization method to normalize the voltage, current and power data of the electric energy meter; Suppose the voltage data sequence collected by a certain electric energy meter is V = [v1, v2, ..., v n ], whose mean is The standard deviation is σ v , then the standardized voltage data Map it to the standard normal distribution interval; Use wavelet transform to remove high-frequency noise in the data, select db4 wavelet basis function to decompose the data, and process the high-frequency coefficients by setting thresholds to remove noise; The Min-Max normalization technology is used to normalize the data of the electric energy meter, and the voltage, current and power data are mapped to the [0,1] interval. The Kalman filter algorithm is used to smooth the data and remove noise interference.

3. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 2 is characterized in that: The initial feature vector of the node information includes the pre-processed data of the electric energy meter, and also includes the geographical location information, operating time information and equipment type information of the electric energy meter. The weight of the edge is calculated according to the resistance and reactance of the line.

4. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 3 is characterized in that: The basic operation of the graph convolutional network model is represented as follows: Among them, H (l) is the node feature matrix of the lth layer, W (l) is the weight matrix of the lth layer, σ is the ReLU activation function, is the adjacency matrix with self-loops added, A represents the adjacency matrix of the graph, I is a diagonal matrix, is the normalized degree matrix.

5. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 4 is characterized in that: In the multi-dimensional indicator construction evaluation system, the node criticality indicator is defined as: I nk =α×E var +β×C deg +γ×P sta +δ×T cor Where E var is the variance of power consumption, reflecting the severity of power fluctuation; C deg is the degree centrality of the node, which measures the connection criticality of the node in the network topology; P sta It is the stability index of the power factor, which is obtained by calculating the standard deviation of the power factor and normalizing it; T cor is the time correlation coefficient of the power data between the node and the surrounding nodes, which is obtained by calculating the Pearson correlation coefficient and reflects the coordinated change relationship between the nodes; α, β, γ, δ are weight coefficients.

6. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 5 is characterized in that: The deep learning model uses an autoencoder or a variational autoencoder, in which the autoencoder uses a large number of node feature vectors in normal operation as training samples to train a multi-layer autoencoder network so that it can reconstruct the input feature vector; during detection, the feature vector of the node to be detected is input into the trained autoencoder to calculate the reconstruction error: Where x is the original eigenvector, is the reconstructed feature vector; when the reconstruction error e is greater than the preset threshold, it is marked as a potential fault point.

7. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 1 is characterized in that: The node power data characteristics include power consumption growth rate, voltage deviation, and current harmonic distortion rate.

8. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 7 is characterized in that: The warning information includes fault location, type, impact range, cause analysis and emergency measures.

9. The method for monitoring the data fusion of electric energy meter and concentrator according to claim 1, characterized in that: The intelligent decision-making model is constructed by deeply integrating network health assessment results, early warning information, power system operation rules and historical experience data.

10. An electric energy meter and concentrator data fusion monitoring system, applied to the method according to any one of claims 1 to 9, characterized in that: include: The electric energy meter-concentrator data collection and graph construction module periodically collects the data of the electric energy meter and uses the cyclic redundancy check algorithm to check the integrity of the data. After the check, the data is preprocessed to remove the noise interference in the data, and then the data is integrated to construct the power network graph structure, which includes node information and edge weight information; The graph convolutional network modeling and key node identification module uses the power network graph structure as the input of the graph convolutional network model, adopts an adaptive learning rate adjustment algorithm combined with an early stopping method to prevent overfitting, deeply trains the model, optimizes model parameters, and captures the spatiotemporal correlation between nodes through multi-layer spatiotemporal graph convolution operations; Identify key nodes and potential failure points by building an evaluation system through multi-dimensional indicators; The power network health assessment and early warning decision-making module integrates the power data characteristics, connection relationships, edge weights and dynamic change information of the nodes, and uses a method based on deep learning combined with a multi-layer perceptron and an attention mechanism to calculate network health indicators and subdivide the network health status. When an abnormality or potential fault is detected, early warning information is sent through multiple channels. At the same time, relying on big data and deep reinforcement learning technology, it integrates multi-party information to build an intelligent decision-making model, formulates optimized decision-making suggestions, and tracks and evaluates the implementation effect; The system performance optimization and intelligent decision-making support module analyzes the data based on long-term accumulated data and operation records using big data analysis technology and machine learning algorithms, automatically adjusts system parameters and algorithm model structures, and visualizes the output results to the management end.

Images