Distribution network state deduction system and method based on graph machine learning

Abstract

The invention provides a distribution network state deduction system and method based on graph machine learning. The distribution network state deduction system comprises a data acquisition and preprocessing subsystem, a graph data modeling subsystem, a model training and optimizing subsystem, a real-time state deduction subsystem, a decision support subsystem and the like. The system abstracts a power distribution network into a graph data structure through a graph machine learning method, and automatically learns the characteristics and trend of the network by using a large amount of actual data. Compared with a traditional management method depending on rules and experience, the system is more intelligent, multiple factors can be comprehensively considered, and the performance and reliability of the whole power distribution network are improved. The method can update the model in real time, adapts to the change of the network state, quickly responds to emergencies, and has important value in practical application.

Description

Technical Field

[0001] The present invention relates to a power distribution network state deduction system, and more particularly to a power distribution network state deduction system and method based on graph machine learning. Background Art

[0002] The power industry plays a crucial role in modern society, and the power distribution network state deduction system is a key technology in this industry. The core concept of this system is to obtain the status data of various key devices in the power distribution network, such as transformers, switches, cables, etc., by real-time monitoring, and use this data to predict the current state and possible future situations of the power network, thereby providing strong support for the monitoring and management of the power system.

[0003] First of all, data collection is the basis of the system. Various sensors and monitoring devices are installed at key positions in the power distribution network. They can real-time monitor parameters such as current, voltage, temperature, humidity, etc., and transmit this data to the central processing unit. Subsequently, the central processing unit receives and processes the data from each sensor, and uses algorithms and models to deduce and analyze the network state. In this process, the system can deduce the current state of the power network based on the collected data, including the operating conditions of devices, load conditions, and potential faults, etc. Most importantly, based on the current state, the system can also perform predictive analysis to identify possible future events or problems, such as device failures or overload situations. Finally, the power distribution network state deduction system also has alarm and control functions, which can generate alarms and notify the operator to take necessary measures to maintain the normal operation of the power network.

[0004] Although the power distribution network state deduction system has been widely applied in the power industry, there are still some key technical problems and challenges. First of all, the data quality problem is an important challenge faced by the current system. The performance of this system highly depends on the data obtained from various sensors and monitoring devices, and this data must be accurate and reliable. However, sensors may be affected by factors such as drift, inaccurate calibration, or failure, resulting in a decline in data quality. This situation may mislead the system and make it unable to accurately deduce the state of the power network, thus affecting the reliability and accuracy of the system.

[0005] Secondly, the complexity of data processing is also a problem that needs to be overcome. The power distribution network is a complex system, containing a large number of devices and branches, generating a large amount of data. Processing such a huge and complex amount of data requires powerful computing capabilities and advanced algorithm support. The complexity may lead to a slowdown in data processing speed and may also lead to a decline in the accuracy of the deduction results, which is one of the technical bottlenecks that need to be solved.

[0006] Another key issue is related to the uncertainty of prediction. Although the system can perform state deduction and event prediction, the operation of the power network is affected by various factors, such as meteorological conditions, equipment aging, and user behavior. The uncertainty of these factors makes it complex to accurately predict future events, and improving prediction algorithms to enhance accuracy is a challenging task.

[0007] In response to this, the present application proposes a distribution network state deduction system based on graph machine learning. Summary of the Invention

[0008] In order to overcome the above defects, a distribution network state deduction system based on graph machine learning is proposed.

[0009] In a first aspect, a distribution network state deduction system based on graph machine learning is provided. The distribution network state deduction system based on graph machine learning includes:

[0010] A data acquisition and preprocessing subsystem for acquiring real-time data and / or historical data and preprocessing the real-time data and / or historical data;

[0011] A graph data modeling subsystem for abstracting the distribution network into a graph data structure;

[0012] A graph machine learning model subsystem for selecting and training a graph machine learning model based on the preprocessed historical data and the graph data structure;

[0013] A real-time state deduction subsystem for deducing the current state of the distribution network based on the preprocessed real-time data and the graph machine learning model.

[0014] The data acquisition and preprocessing subsystem further includes acquiring real-time and / or historical data through sensors and monitoring devices distributed in the distribution network, including current, voltage, and load information of power equipment; the specific preprocessing of the data is to clean and process outliers of the data using statistical methods, filters, and interpolation techniques, and perform time-domain and frequency-domain analysis on the current and voltage waveforms to extract amplitude and frequency characteristics.

[0015] The graph data modeling subsystem further includes defining nodes and edges of the graph data. The nodes represent power equipment, and unique identifiers and attributes are assigned to the nodes. The edges represent the connection relationships between the power equipment, reflecting the flow direction of electricity or the dependency relationships between the equipment. The edges also have attributes to describe the dependency relationships. The topological structure of the power system is constructed by traversing the connection relationships between the power equipment. Attribute and feature information are added to each node, and weights and attributes are added to the edges to represent the importance and characteristics of the connection relationships. The constructed graph data is represented in the form of an adjacency matrix, a node feature matrix, and an edge weight matrix for processing and understanding by machine learning models.

[0016] The graph machine learning model subsystem selects a GNN algorithm that combines multi-layer graph convolutional layers and a multi-head attention mechanism to process the graph data of the power system. Before model training, the graph data structure generated by the graph data modeling subsystem is used as the input of the GNN model. Historical data is used to train the GNN model, and at the same time, a loss function is defined, and an optimization algorithm is used to minimize the loss function.

[0017] The design of the multi-layer graph convolutional layer further includes aggregating neighbor information by weighted averaging or summarizing the features of the neighbor nodes of each node; adopting a weight sharing mechanism, applying the same convolutional kernel on different nodes, making the model have parameter sharing, reducing the number of parameters, and improving the generalization ability of the model; and allowing multi-layer stacking so that the model can learn different levels of abstract features, from local to global, further improving the expressive ability of the model.

[0018] The design of the multi-head attention mechanism further includes a multi-head mechanism that allows the model to learn multiple sets of attentions with different weights. Each head is used to capture different types of relationships or features, improving the adaptability and expressive ability of the model; in each head, the model calculates the attention weights between nodes, which is achieved by calculating the similarity scores between nodes and applying the softmax function to determine the importance of information transmission. A higher weight indicates a closer association between nodes; finally, the weighted information calculated by different heads is merged to generate the final feature representation of the node, emphasizing the information related to the current task to improve the performance of the model.

[0019] The loss function f is:

[0020]

[0021] where y i is the actual label (0 or 1) corresponding to the i-th sample, is the predicted probability value of the model corresponding to the i-th sample, and n is the total number of samples.

[0022] The optimization algorithm passes the input data layer by layer to the output layer, calculates the difference between the model output and the actual labels using the loss function, calculates the gradient of the loss function with respect to the network parameters using the chain rule, and uses the stochastic gradient descent method to update the network parameters. The update rule is: parameter = parameter - learning rate * gradient. Repeat this process multiple times to gradually reduce the loss function. In each iteration, the parameters are updated according to the direction of the gradient, and a learning rate adjustment strategy is adopted to optimize the learning rate through cross-validation.

[0023] The real-time state deduction subsystem receives the real-time operation data collected by the data collection and preprocessing subsystem. The real-time data is stored in a high-performance time-series database and loaded into the trained and optimized GNN model to deduce the power network state, generate the health status of power equipment, fault detection results, and optimization suggestions based on rules and policies, and generate alarms and notifications to respond to emergencies or important events, and perform continuous monitoring and maintenance, including model updates, data quality monitoring, and system troubleshooting, to ensure performance and availability.

[0024] The distribution network state deduction system includes: a decision support subsystem for integrating the deduction results and providing real-time suggestions to the operation and maintenance personnel.

[0025] The decision support subsystem further includes integrating real-time state deduction results, including the current state of equipment, potential fault warnings, and optimization suggestions; generating real-time suggestions based on the deduction results and the power system state, covering equipment maintenance, load adjustment, and alarm handling, and providing an intuitive user interface, including visualization charts, real-time data display, and suggestion reports, to assist the operation and maintenance personnel in understanding and adopting the suggestions; having a decision tracking and feedback mechanism to record decisions and monitor the execution status for system learning and suggestion optimization.

[0026] In a second aspect, a method based on the above-mentioned distribution network state deduction system based on graph machine learning is provided. The method includes:

[0027] The data collection and preprocessing subsystem collects real-time data and / or historical data and preprocesses the real-time data and / or historical data;

[0028] The graph data modeling subsystem abstracts the distribution network into a graph data structure;

[0029] The graph machine learning model subsystem selects and trains a graph machine learning model based on the preprocessed historical data and the graph data structure;

[0030] The real-time state deduction subsystem deduces the current state of the distribution network based on the preprocessed real-time data and the graph machine learning model;

[0031] The decision support subsystem integrates the deduction results and provides real-time suggestions for operation and maintenance personnel.

[0032] In a third aspect, a computer device is provided, including: one or more processors;

[0033] The processor is used to store one or more programs;

[0034] When the one or more programs are executed by the one or more processors, the method of the above-mentioned power distribution network state deduction system based on graph machine learning is implemented.

[0035] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed, the method of the above-mentioned power distribution network state deduction system based on graph machine learning is implemented.

[0036] The method of graph machine learning is used to deduce the state of the power distribution network. Compared with the prior art, it has the following differences and advantages:

[0037] Traditional power distribution network management mainly relies on rules and experience, while graph machine learning can utilize a large amount of actual data to automatically learn the complex features and change trends of the network, thereby achieving more intelligent decision-making. The system can comprehensively consider multiple factors, such as voltage stability, load balance, fault detection, etc., so as to improve the performance and reliability of the entire power distribution network. Graph machine learning can update the model in real time, adapt to the changes in the network state, and quickly respond to emergencies, which has more practical application value than traditional methods. Brief Description of the Drawings

[0038] Figure 1 It is the system composition diagram of the present invention. Detailed Description of the Embodiments

[0039] The following further details the specific embodiments of the present invention with reference to the accompanying drawings.

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0041] As disclosed in the background art, the power industry plays a crucial role in modern society, and the power distribution network state deduction system is a key technology in this industry. The core concept of this system is to obtain the status data of various key devices in the power distribution network, such as transformers, switches, cables, etc., through real-time monitoring, and use this data to predict the current status and future possible situations of the power network, thus providing strong support for the monitoring and management of the power system.

[0042] First of all, data acquisition is the foundation of the system. Various sensors and monitoring devices are installed at key positions in the power distribution network. They can real-time monitor parameters such as current, voltage, temperature, humidity, etc., and transmit this data to the central processing unit. Subsequently, the central processing unit receives and processes the data from each sensor, and uses algorithms and models to deduce and analyze the network state. In this process, the system can deduce the current state of the power network based on the collected data, including the operating conditions of devices, load conditions, and potential faults, etc. Most importantly, based on the current state, this system can also perform predictive analysis to identify possible future events or problems, such as device failures or overload situations. Finally, the power distribution network state deduction system also has alarm and control functions, which can generate alarms and notify operators to take necessary measures to maintain the normal operation of the power network.

[0043] Although the power distribution network state deduction system has been widely applied in the power industry, there are still some key technical problems and challenges. First of all, the data quality problem is an important challenge faced by the current system. The performance of this system highly depends on the data obtained from various sensors and monitoring devices, and this data must be accurate and reliable. However, sensors may be affected by factors such as drift, inaccurate calibration, or faults, resulting in a decline in data quality. This situation may mislead the system and make it unable to accurately deduce the state of the power network, thus affecting the reliability and accuracy of the system.

[0044] Secondly, the complexity of data processing is also a problem that needs to be overcome. The power distribution network is a complex system, containing a large number of devices and branches, generating a large amount of data. Processing such a large and complex amount of data requires powerful computing capabilities and advanced algorithm support. The complexity may lead to a slowdown in data processing speed and may also lead to a decline in the accuracy of the deduction results, which is one of the technical bottlenecks that need to be solved.

[0045] Another key issue is related to the uncertainty of prediction. Although the system can perform state deduction and event prediction, the operation of the power grid is affected by various factors, such as meteorological conditions, equipment aging, and user behavior. The uncertainty of these factors makes it complex to accurately predict future events, and improving prediction algorithms to enhance accuracy is a challenging task.

[0046] In response to this, the present application proposes a distribution network state deduction system based on graph machine learning.

[0047] As Figure 1 shown, the distribution network state deduction system based on graph machine learning of the present invention includes a data acquisition and preprocessing subsystem, a graph data modeling subsystem, a model training and optimization subsystem, a real-time state deduction subsystem, and a decision support subsystem. The specific technical solutions of each subsystem are as follows:

[0048] Among them, the data acquisition and preprocessing subsystem is used to collect real-time or historical data of the distribution network, including current, voltage, load information, etc. And perform data cleaning, denoising, and feature engineering on the data to prepare for the training and deduction of the graph machine learning model.

[0049] Data acquisition is the primary step of the distribution network state deduction system based on graph machine learning, and the key lies in ensuring the reliability, accuracy, and integrity of the data. To this end, real-time data is collected from sensors and monitoring devices distributed throughout the distribution network, including current, voltage, and load information of each power equipment.

[0050] First, plan the deployment locations of data acquisition devices to cover the entire distribution network. These devices include high-precision sensors, current transformers, voltage sensors, etc., and their selection and layout need to be accurately planned according to the network topology and characteristics to ensure the comprehensiveness and representativeness of the data.

[0051] Next, ensure the availability and integrity of real-time data and establish an efficient data transmission system. Adopt high-speed communication protocols and secure data transmission channels to ensure that data can be transmitted to the central data storage system in real time. At the same time, establish a stable data storage system to store historical data for subsequent analysis and training.

[0052] Data quality is the foundation for ensuring the effectiveness of the system. Therefore, data cleaning and outlier handling are indispensable. During the data collection process, outliers, noise, or missing data may occur, which need to be identified and processed in a timely manner. This application uses statistical methods, filters, or interpolation methods to handle abnormal data to ensure the accuracy and integrity of the data. Denoising operation is another important step to reduce noise and interference in the data, thereby improving data quality. The noise level is effectively reduced through digital filters, wavelet transforms, or signal smoothing techniques to ensure that the data is more accurate.

[0053] Finally, in order to better understand the useful information in the data, feature extraction is required. This includes performing time-domain and frequency-domain analysis on the current and voltage waveforms to extract features such as amplitude and frequency for use in subsequent machine learning models. These features will help the model better capture the state changes and characteristics of the power system.

[0054] After data preprocessing, this application uses the graph data modeling subsystem to abstract the distribution network into a graph data structure, which will be the input of the machine learning model to capture the associations and interactions between various components in the power system. In the distribution network state deduction system based on graph machine learning, graph data modeling is a crucial step, which abstracts the power system into a directed graph data structure and provides rich information for subsequent machine learning models.

[0055] First of all, this application clearly defines the nodes and edges of the graph data. Nodes represent various power equipment in the power system, such as transformers, switches, lines, etc. Each node has a unique identifier. Edges represent the physical or logical connection relationships between power equipment, reflecting the power flow or the dependency relationships between equipment. Edges also have attributes to describe these relationships. The clear definition of nodes and edges can ensure the consistency and accuracy of the graph data.

[0056] Next, this application constructs the topological structure of the graph by traversing the connection relationships between power equipment. The complex power system may have a multi-level topological structure, and it is necessary to accurately capture the equipment and connection relationships at all levels. The construction of the network topology needs to consider the physical characteristics of the actual power system to ensure the authenticity of the graph data.

[0057] To better describe the states and characteristics of each power equipment, this application adds attribute and feature information to each node. These attributes may include the rated capacity, working status, health status, etc. of the equipment, while the features can be calculated from historical data, such as statistical features and frequency-domain features of the current and voltage waveforms. These attributes and features will provide useful inputs for the machine learning model.

[0058] Edges also have weights and attributes to represent the importance and characteristics of the connection relationships. For example, the weights can reflect the capacity of current transmission, and the attributes can represent the voltage drop or the current dependence relationship between devices. This information helps to more accurately simulate the operation of the power system.

[0059] The constructed graph data is represented in a way suitable for machine learning algorithms. This application uses the forms of adjacency matrix, node feature matrix, and edge weight matrix. This helps the machine learning model to understand the structure and characteristics of the power system, and thus perform state deduction and prediction. Finally, to help power system engineers and operation and maintenance personnel better understand the graph data, a visualization tool is used to visually present the graph data. Such visualization can provide an intuitive system state and connection relationship, which is helpful for decision-making and operation and maintenance.

[0060] Through the graph data modeling subsystem, this application can effectively model the distribution network into a graph data structure, providing rich input information for the machine learning model, thereby achieving more accurate state deduction and decision support.

[0061] The model training and optimization subsystem selects the Graph Neural Networks (GNNs) algorithm to process the graph data of the distribution network. GNNs can capture the information transmission between nodes and is suitable for network state deduction tasks. Historical data is used to train the graph neural network machine learning model, and the model is optimized to improve the deduction performance. The optimization process includes hyperparameter adjustment and model structure optimization.

[0062] For power system state deduction, this application selects a GNN model that combines multi-layer graph convolutional layers and multi-head attention mechanisms to better capture the relationships between nodes. Before starting model training, the graph data structure generated by the graph data modeling subsystem is sent to the GNN model for processing. Historical data is used to train the selected GNN model. During the training process, an appropriate loss function is defined, which will help the model adjust its parameters to minimize the error between the prediction results and the actual observations.

[0063] Among them, the multi-layer graph convolutional layer and the multi-head attention mechanism are two key components of the graph neural network (GNN). They play important roles in processing graph data and performing information transmission and feature aggregation.

[0064] The multi-layer graph convolutional layer is the core module of the GNN, mainly including the aggregation of neighbor information, weight sharing, and multi-layer stacking. First, in each layer of graph convolution operation, nodes gather information from their neighbor nodes. By weighted averaging or summarizing the features of neighbor nodes, nodes obtain rich neighbor information. This helps the model better understand the context relationships of nodes. Second, the multi-layer graph convolutional layer uses the method of weight sharing, applying the same convolutional kernel on different nodes to ensure the model has parameter sharing, reduce the number of parameters, and improve the generalization ability of the model. Finally, multi-layer stacking allows the model to learn different levels of abstract features, from local to global, further improving the expressive ability of the model.

[0065] The multi-head attention mechanism is another key component that allows the model to simultaneously focus on different parts of information for weighted information transmission between different nodes. This mechanism includes the multi-head mechanism, attention weight calculation, and weighted information transmission. The multi-head mechanism enables the model to learn multiple groups of attentions with different weights. Each head can be regarded as a sub-model for capturing different types of relationships or features, thus improving the adaptability and expressive ability of the model. In each head, the model calculates the attention weights between nodes by calculating the similarity scores between nodes and applying the softmax function. These weights determine the importance of information transmission, and higher weights mean closer associations between nodes. Finally, the multi-head attention mechanism merges the weighted information calculated by different heads to generate the final feature representation of the nodes. This process emphasizes the information relevant to the current task and improves the performance of the model.

[0066] In model training, the graph data structure generated by the graph data modeling subsystem is divided into a training set, a validation set, and a test set. The training set is used for learning model parameters, the validation set is used for hyperparameter tuning and model structure optimization, and the test set is used for final performance evaluation. The proportion of data division is 70 - 80% for the training set, 10 - 15% for the validation set, and 10 - 15% for the test set.

[0067] In the field of power system state deduction, especially for classification tasks, a reliable and efficient loss function is needed to evaluate the performance of the model. For this, the loss function of this application is:

[0068] The loss function f is:

[0069]

[0070] where y i is the actual label (0 or 1) corresponding to the i-th sample, is the predicted probability value of the model corresponding to the i-th sample, and n is the total number of samples.

[0071] This loss function consists of two terms, corresponding to the cases where the actual label is 1 and 0 respectively. By calculating the logarithmic difference between the model's predicted probability and the actual label, this loss function can effectively measure the model's performance. The use of this loss function is to measure the model's performance by comparing the model's prediction results with the actual labels, aiming to minimize classification errors. It is particularly suitable for classification tasks, which play a crucial role in the state deduction of power systems, including equipment working state classification, event classification, and alarm information classification. These tasks are essential for ensuring the stability and reliability of power systems.

[0072] First, the equipment working state classification aims to determine whether various equipment in the power system, such as transformers, switches, and lines, is operating normally or has faults. By classifying the equipment state as normal or faulty, power system maintenance personnel can quickly identify the problem equipment and take necessary repair or replacement measures to ensure the continuous operation and availability of the system.

[0073] Second, event classification is to distinguish various events that occur in the power system into different types. This helps the operation and maintenance team better understand the problems that occur, respond quickly, and take appropriate measures to deal with different types of events, such as power failures, overload events, short - circuit events, etc. By accurately classifying events, the maintainability and repair efficiency of the power system will be significantly improved.

[0074] Finally, alarm information classification is used to classify and grade the alarm information generated in the power system. This helps to determine the urgency and handling priority of the alarms to ensure that the operation and maintenance team can formulate corresponding response plans according to the importance and urgency of the alarms. By efficiently classifying alarm information, the operation safety of the power system is enhanced, reducing the impact of potential faults on the system.

[0075] To minimize the loss function, the optimization algorithm adopted in this application is a key component in deep learning, which plays a crucial role in the training and optimization of neural networks.

[0076] The optimization algorithm starts from the first step of the neural network. The input data passes through each layer of the network. Each neuron calculates the sum of the weighted inputs and feeds it into the activation function to generate an activation value. The activation function is a common activation function in deep learning, such as the Sigmoid function, ReLU function, etc. This process continues until the output layer of the network is reached, thus obtaining the prediction result of the model.

[0077] Subsequently, the output of the model is compared with the actual labels to calculate the value of the loss function of this application. Next, the optimization algorithm calculates the gradients of the loss function with respect to each parameter (weights and bias terms) in the network. This is achieved by using the chain rule. The process of calculating the gradients starts from the output layer. First, the gradient of the loss function with respect to the output values is calculated, and then it proceeds layer by layer backward, multiplying the gradient of the previous layer by the weights of the current layer to calculate the gradient of the current layer. Once the gradients are calculated, stochastic gradient descent can be used to update the parameters in the network. The update rule is as follows:

[0078] parameter = parameter - learning_rate * gradient

[0079] This process is repeated multiple times over the entire training dataset to gradually reduce the loss function and make the model predict more accurately. The learning rate is an important hyperparameter that controls the step size of parameter updates.

[0080] The optimization algorithm iterates the above process multiple times until a predetermined number of iterations is reached or the condition to stop training is satisfied. In each iteration, the parameters are updated according to the direction of the gradients to continuously improve the model performance. Further, to improve the model performance, this application adjusts the hyperparameter of the learning rate. The selection of the learning rate can be optimized using the cross-validation method.

[0081] When using cross-validation, first, an initial learning rate range needs to be determined, between 0.001 and 0.1. Next, K-fold cross-validation is selected as the cross-validation strategy to evaluate the performance of the model at different learning rates on different data subsets. K-fold cross-validation divides the dataset into K subsets, and each time K - 1 subsets are used for training, and the remaining one is used for validation.

[0082] At the beginning of the hyperparameter search process, an initial learning rate value is selected from the learning rate range. Then, using the selected cross-validation strategy, the dataset is divided into a training set and a validation set. Then, using the selected learning rate value, the model is trained on the training set, and the performance metric is calculated on the validation set. This process is iterated multiple times, each time using a different learning rate value. The results of the performance metric will be recorded for subsequent analysis.

[0083] Analyzing the results of the performance metric is a crucial step. Find a learning rate value that makes the performance metric reach the best level. This learning rate value should be within a reasonable range, not too large so as to cause the model to diverge, nor too small so as to cause the model to converge slowly.

[0084] The entire hyperparameter tuning process needs to be iterated multiple times until the optimal learning rate value is found. During the actual training process, this optimal learning rate value is used to train the model on the complete training data to obtain the final performance. This process consumes some time and computing resources, but it is a crucial step in optimizing deep learning models and can significantly improve the model's performance and convergence speed. Finding the learning rate that best suits the task and model structure will help achieve better training results.

[0085] Once the training is completed and verified, the model can be saved for subsequent deployment. After deployment, the model can process the data of the actual distribution network in real time, providing information on the status of power equipment, fault detection, and optimization suggestions, thereby improving the reliability and efficiency of the system.

[0086] Through the above technical implementation solutions, this application can establish an efficient GNN-based model, providing reliable support for the state deduction of the distribution network, thereby improving the reliability and efficiency of network operation.

[0087] The real-time state deduction subsystem, during actual operation, inputs real-time data into the trained model to deduce the current state of the distribution network. This includes the health status of power equipment, detection of potential faults, and optimization suggestions.

[0088] First, data collection and transmission are the first step of the deduction subsystem. Through sensors and monitoring devices, real-time data, including current, voltage, frequency, equipment operation status, etc., are collected from the power system and preprocessed and cleaned to ensure the accuracy and consistency of the data. This data is then stored in a high-performance time-series database for subsequent analysis and deduction.

[0089] Secondly, the graph neural network (GNN) model that has been trained and optimized by the model training and optimization subsystem is loaded into the system. Real-time data is input into the model for deduction, generating predictions of equipment status, fault detection results, etc. Finally, the deduction results need to be interpreted and analyzed so that operators can understand the status of the power network. In addition, the real-time state deduction system can also generate optimization suggestions, which include equipment reconfiguration, load adjustment, and fault handling suggestions, etc. These optimization suggestions are generated based on the deduction results and certain rules and strategies.

[0090] The real-time status deduction results, health status, fault detection, and optimization suggestions are presented to the operators through a visual interface so that they can make decisions quickly and generate reports for recording and analysis when needed. At the same time, the real-time status deduction system needs to be able to generate alarms and notifications to handle emergencies or important events, thus ensuring that appropriate measures can be taken promptly. The system also needs to be continuously monitored and maintained, including model updates, data quality monitoring, and system troubleshooting, to ensure its performance and availability.

[0091] Through the implementation plan, the real-time status deduction subsystem helps to ensure the stability, reliability, and efficiency of the power system, providing timely information and decision support for the operation and management of the power system, especially when facing increasingly complex and dynamic challenges.

[0092] The decision support subsystem plays a crucial role in the operation and maintenance of the power system. It not only integrates the results of real-time status deduction but also provides key real-time suggestions for the operation and maintenance personnel to assist them in making wise decisions, such as equipment maintenance and load adjustment.

[0093] The decision support subsystem first organically integrates the results of real-time status deduction into its decision-making engine. This means it receives and interprets data from the deduction subsystem, including the current status of equipment, any potential fault warnings, and optimization suggestions.

[0094] Based on the deduction results and the current status of the power system, the decision support system generates real-time suggestions. These suggestions can cover multiple aspects, such as equipment maintenance, load adjustment, alarm handling, etc. For example, if the deduction results indicate that a certain device may malfunction, the system will provide corresponding maintenance suggestions to avoid potential faults. In addition, for load adjustment, the system can provide reasonable suggestions according to the change of power demand to ensure the stable operation of the power system.

[0095] To enable the operation and maintenance personnel to easily understand and adopt these real-time suggestions, the decision support system provides an intuitive user interface. This interface usually includes visual charts, real-time data displays, and suggestion reports. Through this interface, the operation and maintenance personnel can quickly obtain key information, make decisions, and monitor their implementation.

[0096] The decision support system also has a decision tracking and feedback mechanism. It can record the decisions of the operation and maintenance personnel and monitor their implementation. This helps the system to continuously learn and optimize suggestions to adapt to the changes in the power system and the evolution of operation and maintenance strategies.

[0097] Through the above implementation solutions, the decision support subsystem provides a powerful tool for the operation and maintenance of the power system, combining the accuracy of real-time deduction and the rapidity of decision-making, which helps to ensure the stability, reliability, and efficiency of the power system. This is crucial for the operation and management of the power system, especially when facing increasingly complex and dynamic challenges.

[0098] Embodiment 2

[0099] Based on the same inventive concept, the present invention also provides a method based on the above-mentioned power distribution network state deduction system based on graph machine learning, and the method includes:

[0100] The data acquisition and preprocessing subsystem acquires real-time data and / or historical data and preprocesses the real-time data and / or historical data;

[0101] The graph data modeling subsystem abstracts the power distribution network into a graph data structure;

[0102] The graph machine learning model subsystem selects and trains a graph machine learning model;

[0103] The real-time state deduction subsystem deduces the state of the current power distribution network;

[0104] The decision support subsystem integrates the deduction results and provides real-time suggestions for the operation and maintenance personnel.

[0105] Embodiment 3

[0106] Based on the same inventive concept, the present invention also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of the method of a power distribution network state deduction system based on graph machine learning as described in the above embodiments.

[0107] Example 4

[0108] Based on the same inventive concept, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device, used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. And, in this storage space, there is also stored one or more instructions suitable for being loaded and executed by the processor. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the steps of the method of a power distribution network state deduction system based on graph machine learning described in the above embodiments.

[0109] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0110] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0111] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one or more of the flows Figure 1 a flow or flows and / or boxes Figure 1 specified in a box or boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the flows Figure 1 a flow or flows and / or boxes Figure 1 specified in a box or boxes.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific embodiments of the present invention. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A power distribution network state deduction system based on graph machine learning, characterized in that: the power distribution network state deduction system includes; a data acquisition and preprocessing subsystem for acquiring real-time data and / or historical data and preprocessing the real-time data and / or historical data; a graph data modeling subsystem for abstracting the power distribution network into a graph data structure; a graph machine learning model subsystem for selecting and training a graph machine learning model based on the preprocessed historical data and the graph data structure; a real-time state deduction subsystem for deducing the state of the current power distribution network based on the preprocessed real-time data and the graph machine learning model.

2. The power distribution network state deduction system based on graph machine learning according to claim 1, characterized in that: the data acquisition and preprocessing subsystem further includes acquiring real-time and / or historical data through sensors and monitoring devices distributed in the power distribution network, including current, voltage and load information of power equipment; the data preprocessing specifically includes cleaning and outlier processing of the data using statistical methods, filters and interpolation techniques, and performing time-domain and frequency-domain analysis on the current and voltage waveforms to extract amplitude and frequency characteristics.

3. The power distribution network state deduction system based on graph machine learning according to claim 1, characterized in that: the graph data modeling subsystem further includes defining nodes and edges of the graph data, where the nodes represent power equipment and are assigned unique identifiers and attributes to the nodes, and the edges represent the connection relationships between the power equipment, reflecting the flow direction of electricity or the dependency relationships between the devices, and the edges have attributes to describe the dependency relationships; constructing the topological structure of the power system by traversing the connection relationships between the power equipment, adding attribute and feature information to each node, adding weights and attributes to the edges to represent the connection relationships, and representing the constructed graph data in the form of an adjacency matrix, a node feature matrix and an edge weight matrix for processing by a machine learning model.

4. The power distribution network state deduction system based on graph machine learning according to claim 1, characterized in that: the graph machine learning model subsystem selects a GNN algorithm that combines a multi-layer graph convolutional layer and a multi-head attention mechanism to process the graph data of the power system. Before model training, the graph data structure generated by the graph data modeling subsystem is used as the input of the GNN model, and the GNN model is trained using historical data. At the same time, a loss function is defined, and an optimization algorithm is used to minimize the loss function.

5. The power distribution network state deduction system based on graph machine learning according to claim 4, characterized in that: the design of the multi-layer graph convolutional layer further includes aggregating neighbor information by weighted averaging or summarizing the features of the neighbor nodes of each node; adopting a weight sharing mechanism to apply the same convolutional kernel to different nodes; and allowing multi-layer stacking.

6. The power distribution network state deduction system based on graph machine learning according to claim 4, characterized in that: The design of the multi - head attention mechanism further includes a multi - head mechanism that allows the model to learn multiple sets of attention with different weights, where each head is used to capture different types of relationships or features; In each head, the model calculates the attention weights between nodes by calculating the similarity scores between nodes and applying the softmax function; finally, the weighted information calculated by different heads is combined to generate the final feature representation of the nodes.

7. A power distribution network state deduction system based on graph machine learning as described in claim 4, characterized in that: The loss function f is: where y i is the actual label (0 or 1) corresponding to the i-th sample, is the predicted probability value of the model corresponding to the i-th sample, and n is the total number of samples.

8. A power distribution network state deduction system based on graph machine learning as described in claim 4, characterized in that: The optimization algorithm passes the input data layer by layer to the output layer, calculates the difference between the model output and the actual label using the loss function, calculates the gradient of the loss function with respect to the network parameters using the chain rule, adopts the stochastic gradient descent method to update the network parameters, and repeats multiple iterations to gradually reduce the loss function; in each iteration, the parameters are updated according to the direction of the gradient, and a learning rate adjustment strategy is adopted to optimize the learning rate through cross - validation.

9. A power distribution network state deduction system based on graph machine learning as described in claim 1, characterized in that: The real - time state deduction subsystem receives the real - time operation data collected by the data collection and pre - processing subsystem. The real - time data is stored in a high - performance time - series database and loads the already trained and optimized GNN model to deduce the power network state, generate the health status of power equipment, fault detection results, and optimization suggestions based on rules and policies, and generate alarms and notifications to respond to emergencies or important events, and conduct continuous monitoring and maintenance, including model updates, data quality monitoring, and system troubleshooting, to ensure performance and availability.

10. A power distribution network state deduction system based on graph machine learning as described in claim 1, characterized in that: The power distribution network state deduction system includes: a decision - making support subsystem for integrating the deduction results and providing real - time suggestions for operation and maintenance personnel.

11. A power distribution network state deduction system based on graph machine learning as described in claim 10, characterized in that: The decision - making support subsystem further includes integrating real - time state deduction results, including the current state of equipment, potential fault warnings, and optimization suggestions; generating real - time suggestions based on the deduction results and the power system state, covering equipment maintenance, load adjustment, and alarm handling, providing an intuitive user interface, including visualization charts, real - time data display, and suggestion reports, to assist operation and maintenance personnel in understanding and adopting suggestions; having a decision - tracking and feedback mechanism to record decisions and monitor the execution status for system learning and suggestion optimization.

12. A method for a power distribution network state deduction system based on graph machine learning according to any one of claims 1 - 11, characterized in that: The method includes: The data collection and pre - processing subsystem collects real - time data and / or historical data and pre - processes the real - time data and / or historical data; The graph data modeling subsystem abstracts the power distribution network into a graph data structure; The graph machine learning model subsystem selects and trains a graph machine learning model based on the preprocessed historical data and the graph data structure; The real-time state deduction subsystem deduces the current state of the distribution network based on the preprocessed real-time data and the graph machine learning model; The decision support subsystem integrates the deduction results and provides real-time suggestions for the operation and maintenance personnel.

13. A computer device, characterized in that, comprising: one or more processors; The processor is used to store one or more programs; When the one or more programs are executed by the one or more processors, the method of the distribution network state deduction system based on graph machine learning as described in claim 12 is implemented.

14. A computer-readable storage medium, characterized in that, There is a computer program stored thereon, and when the computer program is executed, the method of the distribution network state deduction system based on graph machine learning as described in claim 12 is implemented.

Images