A smart grid fault warning and diagnosis system

Abstract

The present invention discloses a smart grid fault warning and diagnosis system, which relates to the field of smart grid technology, including a multimodal data acquisition and fusion module, a deep reinforcement learning warning module, a fault diagnosis module based on a graph neural network, an adaptive diagnosis module integrating multi-task learning, an edge computing and cloud collaboration module, and a system testing and verification module. The system comprises the following steps: collecting multi-source data through power sensors, smart meters and weather stations, and using a deep autoencoder model to reduce the data dimension and fuse features; importing the collected data into the deep reinforcement learning warning module, and dynamically adjusting the warning strategy through interaction with the grid operation environment; analyzing the topological structure of the grid and identifying the fault location by using the fault diagnosis module based on the graph neural network; learning the model and processing the fault type; combining edge computing with the cloud for collaborative work; and testing, verifying and optimizing the model by the system testing and verification module.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a smart grid fault early warning and diagnosis system. Background Art

[0002] With the rapid development of power systems, early warning and diagnosis of power failures play an increasingly important role in smart grids. However, existing technologies have the following shortcomings:

[0003] 1. Current power grid fault diagnosis mostly relies on a single data source (such as voltage, current, and temperature), making it difficult to fully utilize multimodal data. Due to the lack of effective integration of multi-source heterogeneous data, traditional systems cannot comprehensively assess the grid's operating status from multiple perspectives, resulting in limited diagnostic accuracy. This limitation is particularly evident in complex power grid structures.

[0004] 2. Traditional power grid early warning systems often use fixed threshold models. This static warning mechanism cannot effectively respond to environmental changes, such as dynamic factors like meteorological conditions and grid load fluctuations. In severe weather or when loads fluctuate significantly, fixed thresholds are difficult to adjust in a timely manner, significantly reducing the accuracy of fault warnings. This can lead to missed critical warning opportunities and increased risk of grid failure.

[0005] 3. As power grids expand in size and become more complex, traditional fault diagnosis models based on simple topologies are unable to cope with large-scale, complex networks. This is especially true when faced with the complex interdependencies of multiple nodes and paths within the power grid. These models are unable to effectively capture the dependencies between nodes, resulting in inaccurate fault location and diagnosis results.

[0006] 4. Traditional fault diagnosis systems can only handle a single type of fault and are unable to cope with the concurrence or combination of multiple fault modes. In complex power grids, multiple faults may occur simultaneously or be related to each other. Traditional systems are unable to identify multiple fault modes and are often only able to diagnose a part of the fault, failing to fully reflect the health of the power grid.

[0007] 5. Traditional power grid systems typically rely solely on centralized processing methods. When large-scale data analysis or fault diagnosis is required, computing bottlenecks are prone to occur. In addition, centralized processing has large delays and real-time performance is difficult to guarantee. This is particularly insufficient in power grid failure scenarios that require a rapid response, which may delay processing and cause the problem to spread.

[0008] Therefore, to address the above problems, a smart grid fault warning and diagnosis system is urgently needed to improve data processing efficiency, protect data privacy, and achieve more flexible system integration. Summary of the Invention

[0009] The purpose of the present invention is to provide a smart grid fault warning and diagnosis system to solve the problems raised in the above background technology.

[0010] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0011] A smart grid fault warning and diagnosis system, including a multimodal data acquisition and fusion module, a deep reinforcement learning warning module, a graph neural network-based fault diagnosis module, an adaptive diagnosis module integrating multi-task learning, an edge computing and cloud collaboration module, and a system testing and verification module;

[0012] The multimodal data acquisition and fusion module includes a multimodal data acquisition unit and a multimodal data fusion unit. The multimodal data acquisition unit acquires heterogeneous data from multiple sources and collaboratively processes the data acquired by the multimodal data acquisition unit through a multimodal fusion model within the multimodal data fusion unit.

[0013] The deep reinforcement learning warning module includes an intelligent agent and reinforcement learning model unit and an adaptive threshold adjustment unit. The intelligent agent and reinforcement learning model unit includes building a deep Q network intelligent agent. Based on the deep Q network intelligent agent system, the multimodal data acquisition and fusion module collects and fuses data to perform real-time dynamic data and environmental interaction processing; the adaptive threshold adjustment unit adaptively adjusts the fault warning threshold according to different environmental conditions;

[0014] The fault diagnosis module of the graph neural network includes a power grid topology modeling unit, a graph convolutional neural network model unit, and a fault location and diagnosis unit. The fault diagnosis module based on the graph neural network utilizes the graph structure properties of the power grid and constructs a complex power grid topology model based on the graph convolutional network to achieve high-precision fault location and diagnosis;

[0015] The adaptive diagnosis module integrated with multi-task learning enhances robustness to different power grid faults and improves diagnostic accuracy by jointly learning multiple fault modes based on the fault diagnosis module data of the graph neural network.

[0016] The edge computing and cloud collaboration module includes an edge computing real-time preprocessing unit and a cloud big data analysis unit. The edge computing and cloud collaboration module preprocesses the adaptive diagnosis module data integrated with multi-task learning in real time at the edge node, and works in conjunction with the cloud big data analysis platform in the cloud collaboration unit to ensure system response speed;

[0017] The system testing and verification module performs system testing and verification based on the data of the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on the graph neural network, the adaptive diagnosis module integrating multi-task learning, and the edge computing and cloud collaboration module.

[0018] A further improvement of the technical solution of the present invention is that the multimodal data acquisition unit includes a power sensor, a smart meter and a weather station to collect multi-source data of the power grid in real time, the power sensor includes a voltage sensor, a current sensor, a frequency sensor and a temperature sensor, and the collected data including voltage, current, frequency and temperature are recorded as ; The smart meter records the user's electricity consumption data as , analyze load fluctuations, collect external environmental parameters such as wind speed and humidity that affect power supply and record them as ;

[0019] Where t represents time, 、 、 are the independent variables of each parameter that changes over time.

[0020] A further improvement of the technical solution of the present invention is that: the multimodal data fusion unit includes a deep autoencoder model, and the multimodal data fusion unit uses the deep autoencoder model in the modal data fusion unit to reduce the dimension and fuse the collected data to generate a unified feature representation, and input They come from different data sources, that is, the collected data is represented as a multi-dimensional state vector X(t), as shown below:

[0021]

[0022] After dimensionality reduction through the deep autoencoder, a unified low-dimensional representation is obtained:

[0023]

[0024] in is the input multimodal feature vector, is the weight matrix, is the bias term, is the activation function.

[0025] A further improvement of the technical solution of the present invention is that the intelligent agent and the reinforcement learning model construct an intelligent agent based on a deep Q network, which dynamically adjusts the early warning strategy by interacting with the power grid operating environment. The agent system selects the optimal action based on the current state and updates the Q value. The specific formula for the updated Q value is as follows:

[0026]

[0027] in Indicates that the status Take action Q value, represents the learning rate, Represents the current reward, represents the discount factor, Indicates the current state, Indicates the current action. Indicates the next state, Represents all possible actions.

[0028] A further improvement of the technical solution of the present invention is that: the threshold value of the adaptive threshold adjustment unit for adaptively adjusting the fault warning is expressed as:

[0029]

[0030] in Indicates the voltage warning threshold, represents the initial voltage threshold, Indicates the voltage value that is adaptively adjusted according to the magnitude of meteorological changes.

[0031] A further improvement of the technical solution of the present invention is that the fault diagnosis module of the graph neural network includes a power grid topology modeling unit, a graph convolutional neural network model unit and a fault location and diagnosis unit. The power grid topology modeling unit represents the power grid as a graph. ,in represents a node in the power grid, Represents the transmission line between nodes, and the characteristic vector of each node is the current voltage, current, frequency and temperature;

[0032] The graph convolutional neural network model unit uses a graph convolutional network to process the graph structure data of the power grid, learn the relationship between node features, and identify potential fault locations. The layer update formula of the graph convolutional network is:

[0033]

[0034] in Represented as the adjacency matrix of the graph, Represented as a degree matrix, Expressed as The node feature matrix of the layer, Represented as a weight matrix, is the activation function;

[0035] The fault location and diagnosis unit classifies each node of the power grid based on a graph convolutional neural network model, and outputs the corresponding fault type and the node location where the fault occurs.

[0036] A further improvement of the technical solution of the present invention is that the adaptive diagnosis module of the integrated multi-task learning includes a multi-task learning model unit and an adaptive learning rate adjustment unit. The multi-task learning model unit uses a shared feature representation and a task-specific output layer to realize the joint diagnosis of different types of faults, setting multiple tasks. , k represents a positive integer, each task corresponds to a specific failure mode, and the loss function is the weighted sum of all task losses:

[0037]

[0038] in Expressed as the total loss function, Expressed as the number of tasks, Represented as a task The weight of Indicates a task The loss function of .

[0039] A further improvement of the technical solution of the present invention is that the adaptive learning rate adjustment unit dynamically adjusts the learning rate of each task based on the learning progress in the multi-task learning model unit, and the adaptive learning rate update formula is:

[0040]

[0041] in represents the current learning rate, Indicates a task The loss at the tth iteration, Indicates a task The loss at iteration t-1.

[0042] A further improvement to the technical solution of the present invention is that the edge computing and cloud collaboration module includes an edge computing real-time preprocessing unit and a cloud-based big data analysis unit. The edge computing real-time preprocessing unit performs real-time data preprocessing at the edge node of the power grid, including data cleaning, dimensionality reduction, and preliminary diagnosis, reducing data transmission delays and ensuring the real-time performance of the system. The cloud-based big data analysis unit uses high-performance computing resources to perform complex analysis on large amounts of historical data and real-time data, providing more refined fault prediction and diagnosis results. The edge and cloud work together to achieve efficient resource utilization.

[0043] The system testing and verification module performs system testing and verification based on the data of the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on the graph neural network, the adaptive diagnosis module integrating multi-task learning, and the edge computing and cloud collaboration module.

[0044] A smart grid fault early warning and diagnosis method, used to implement a smart grid fault early warning and diagnosis system according to any one of claims 1 to 6, comprising the following steps:

[0045] Step 1: The power sensors, smart meters, and weather station equipment in the modal data acquisition and fusion module collect multi-source data in real time. The deep autoencoder model in the modal data acquisition and fusion module then performs dimensionality reduction and fusion on the collected multi-source data, representing it with unified features.

[0046] Step 2: Import the collected data into the deep reinforcement learning early warning module. The deep reinforcement learning early warning module builds an intelligent agent based on the deep Q network. It dynamically adjusts the early warning strategy by interacting with the power grid operating environment. The agent can adaptively adjust the fault warning threshold based on real-time data.

[0047] Step 3: Based on the deep reinforcement learning early warning module warning strategy, when the system detects a potential fault, the system will analyze the topology of the power grid based on the graph neural network fault diagnosis module to identify the location of the fault;

[0048] Step 4: To cope with various failure modes, the system also integrates a multi-task learning model that can handle multiple failure types simultaneously, improving the ability to identify complex failures.

[0049] Step 5: The system combines edge computing with cloud computing to work together. The edge nodes are responsible for real-time data preprocessing and transmit data to the cloud for more complex data analysis.

[0050] Step 6: The system testing and verification module performs system testing and verification based on the data from the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on graph neural network, the adaptive diagnosis module integrated with multi-task learning, and the edge computing and cloud collaboration module.

[0051] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0052] 1. This invention provides an innovative smart grid fault warning and diagnosis system. By integrating multimodal data (including voltage, current, temperature, meteorological data, and historical fault records), and using a deep autoencoder model for feature fusion and dimensionality reduction, it effectively solves the problem of traditional single data source. The collaborative analysis of multi-source data not only provides more comprehensive information on the grid's operating status, but also significantly improves the accuracy of fault diagnosis, demonstrating greater robustness, especially in complex grid environments.

[0053] 2. The smart grid fault warning and diagnosis system provided by this invention introduces an intelligent agent system based on deep reinforcement learning, enabling the fault warning mechanism to adaptively adjust thresholds. Through real-time interaction with the environment, the intelligent agent can dynamically and adaptively adjust the fault warning strategy based on external conditions (such as weather changes and grid load fluctuations). This solves the problem that traditional fixed threshold models cannot flexibly respond to complex environmental changes, greatly improving the timeliness and accuracy of warnings.

[0054] 3. The smart grid fault warning and diagnosis system provided by this invention achieves accurate fault location by constructing a grid graph structure and utilizing a graph convolutional network (GCN) model to process the complex correlated data in the grid topology. This solution overcomes the shortcomings of traditional diagnostic models in complex grids. By capturing the deep dependencies between nodes, it can achieve accurate fault location and diagnosis in large-scale, complex grids, significantly improving system reliability.

[0055] 4. The smart grid fault warning and diagnosis system provided by this invention uses a multi-task learning model that can simultaneously learn and diagnose multiple different types of fault modes, overcoming the inability of traditional systems to cope with multiple fault modes. By sharing feature representations and jointly learning different tasks, this system significantly improves its ability to recognize multiple concurrent or combined fault modes, ensuring that the system can comprehensively diagnose grid faults and provide more reliable solutions.

[0056] 5. The smart grid fault warning and diagnosis system provided by the present invention combines edge computing and cloud collaboration, making up for the problems of high processing delay and computing bottlenecks in traditional centralized computing. It performs real-time data preprocessing at the edge nodes of the power grid, reduces transmission delays, and provides efficient fault diagnosis services through large-scale data analysis in the cloud, ensuring the real-time and efficiency of the system, and can quickly respond to power grid faults and reduce the spread of power grid damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0058] Figure 1 This is a flow chart of a smart grid fault early warning and diagnosis system according to the present invention;

[0059] Figure 2 This is a structural diagram of a smart grid fault early warning and diagnosis system of the present invention. DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0061] Examples, such as Figure 1-2 As shown, a smart grid fault warning and diagnosis system includes a multimodal data acquisition and fusion module, a deep reinforcement learning warning module, a graph neural network-based fault diagnosis module, an adaptive diagnosis module integrating multi-task learning, an edge computing and cloud collaboration module, and a system testing and verification module;

[0062] The multimodal data acquisition and fusion module includes a multimodal data acquisition unit and a multimodal data fusion unit. The modal data acquisition unit acquires heterogeneous data from multiple sources and collaboratively processes the data collected by the multimodal data acquisition unit through a multimodal fusion model within the multimodal data fusion unit.

[0063] The deep reinforcement learning early warning module includes an intelligent agent and reinforcement learning model unit and an adaptive threshold adjustment unit. The intelligent agent and reinforcement learning model unit includes building a deep Q network intelligent agent. Based on the deep Q network intelligent agent system, the multimodal data acquisition and fusion module collects and fuses data to perform real-time dynamic data and environmental interaction processing; the adaptive threshold adjustment unit adaptively adjusts the fault warning threshold according to different environmental conditions;

[0064] The fault diagnosis module of the graph neural network includes a power grid topology modeling unit, a graph convolutional neural network model unit, and a fault location and diagnosis unit. The fault diagnosis module based on the graph neural network utilizes the graph structure properties of the power grid and builds a complex power grid topology model based on the graph convolutional network to achieve high-precision fault location and diagnosis.

[0065] The adaptive diagnosis module with integrated multi-task learning and the fault diagnosis module based on graph neural network jointly learn multiple fault modes to enhance the robustness to different power grid faults and improve the accuracy of diagnosis;

[0066] The edge computing and cloud collaboration module includes an edge computing real-time preprocessing unit and a cloud big data analysis unit. The edge computing and cloud collaboration module will integrate the adaptive diagnosis module data of multi-task learning for real-time preprocessing at the edge node, and work together with the cloud big data analysis platform in the cloud collaboration unit to ensure system response speed.

[0067] The multimodal data acquisition unit includes power sensors, smart meters and weather stations to collect multi-source data of the power grid in real time. The power sensors include voltage sensors, current sensors, frequency sensors and temperature sensors. The collected data include voltage, current, frequency and temperature, which are recorded as ; Smart meters record user electricity consumption data as , analyze load fluctuations, collect external environmental parameters such as wind speed and humidity that affect power supply and record them as ;

[0068] Where t represents time, 、 、 are the independent variables of each parameter that changes over time.

[0069] High-precision voltage sensors, current sensors, temperature sensors, smart meters and weather stations are installed at key nodes of the power grid, and each sensor regularly collects corresponding operating data.

[0070] The multimodal data fusion unit includes a deep autoencoder model. The multimodal data fusion unit uses the deep autoencoder model in the modal data fusion unit to reduce the dimension and fuse the features of the collected data to generate a unified feature representation. They come from different data sources, that is, the collected data is represented as a multi-dimensional state vector X(t), as shown below:

[0071]

[0072] After dimensionality reduction through the deep autoencoder, a unified low-dimensional representation is obtained:

[0073]

[0074] in is the input multimodal feature vector, is the weight matrix, is the bias term, is the activation function.

[0075] All collected data uses a unified data format (such as JSON or XML) to ensure compatibility and readability between data from different sensors. At the same time, multimodal data is fused into a low-dimensional feature representation through a deep autoencoder to form a unified feature vector to ensure the effectiveness of subsequent analysis.

[0076] The intelligent agent and reinforcement learning model builds an intelligent agent based on a deep Q network. By interacting with the power grid operating environment, the early warning strategy is dynamically adjusted. The agent system selects the optimal action based on the current state and updates the Q value. The specific formula for the updated Q value is as follows:

[0077]

[0078] in Indicates that the status Take action Q value, represents the learning rate, Represents the current reward, represents the discount factor, Indicates the current state, Indicates the current action. Indicates the next state, Represents all possible actions.

[0079] The adaptive threshold adjustment unit adaptively adjusts the threshold of the fault warning as follows:

[0080]

[0081] in Indicates the voltage warning threshold, represents the initial voltage threshold, Indicates the voltage value that is adaptively adjusted according to the magnitude of meteorological changes.

[0082] The fault warning module builds an intelligent agent based on a deep Q network to monitor the power grid status in real time and adjust the warning strategy based on feedback. The intelligent agent continuously optimizes its behavior through online learning, enabling fault warnings to adaptively respond to environmental changes, thereby improving the safety and response speed of the power grid.

[0083] The fault diagnosis module of the graph neural network includes a power grid topology modeling unit, a graph convolutional neural network model unit, and a fault location and diagnosis unit. The power grid topology modeling unit represents the power grid as a graph. ,in represents a node in the power grid, Represents the transmission line between nodes, and the characteristic vector of each node is the current voltage, current, frequency and temperature;

[0084] The graph convolutional neural network model unit uses the graph convolutional network to process the graph structure data of the power grid, learn the relationship between node features, and identify potential fault locations. The layer update formula of the graph convolutional network is:

[0085]

[0086] in Represented as the adjacency matrix of the graph, Represented as a degree matrix, Expressed as The node feature matrix of the layer, Represented as a weight matrix, is the activation function;

[0087] The fault location and diagnosis unit classifies each node of the power grid based on the graph convolutional neural network model, and outputs the corresponding fault type and the node location where the fault occurs.

[0088] During the fault diagnosis phase, the system uses graph neural networks to analyze the topology of the power grid to identify the location of the fault. By modeling the power grid as a graph structure, the GCN model can deeply learn the complex relationships between nodes, ensure the accuracy of fault location, and provide maintenance personnel with timely fault information.

[0089] The adaptive diagnosis module integrating multi-task learning includes a multi-task learning model unit and an adaptive learning rate adjustment unit. The multi-task learning model unit uses shared feature representation and task-specific output layer to achieve joint diagnosis of different types of faults and set multiple tasks. , k represents a positive integer, each task corresponds to a specific failure mode, and the loss function is the weighted sum of all task losses:

[0090]

[0091] in Expressed as the total loss function, Expressed as the number of tasks, Represented as a task The weight of Indicates a task The loss function of .

[0092] The adaptive learning rate adjustment unit dynamically adjusts the learning rate of each task based on the learning progress within the multi-task learning model unit. The adaptive learning rate update formula is:

[0093]

[0094] in represents the current learning rate, Indicates a task The loss at the tth iteration, Indicates a task The loss at iteration t-1.

[0095] The integrated multi-task learning model enables the system to handle multiple fault modes simultaneously, ensuring the accurate identification of various types of faults. By sharing feature representations and dynamically adjusting the learning rate, the system effectively improves its ability to identify complex faults, making it more flexible and robust when facing different types of faults.

[0096] The edge computing and cloud collaboration module includes an edge computing real-time preprocessing unit and a cloud-based big data analysis unit. The edge computing real-time preprocessing unit performs real-time data preprocessing at the edge nodes of the power grid, including data cleaning, dimensionality reduction, and preliminary diagnosis, reducing data transmission delays and ensuring the real-time performance of the system. The cloud-based big data analysis unit uses high-performance computing resources to perform complex analysis on large amounts of historical data and real-time data, providing more sophisticated fault prediction and diagnosis results. The edge and cloud work together to achieve efficient resource utilization.

[0097] The system testing and verification module performs system testing and verification based on data from the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on graph neural network, the adaptive diagnosis module integrating multi-task learning, and the edge computing and cloud collaboration module.

[0098] Edge computing nodes are responsible for the initial processing and fault detection of real-time data, while cloud services perform more in-depth data analysis and storage. Through the collaborative work of edge and cloud, the system achieves the best combination of real-time response and complex analysis capabilities, effectively improving the operational efficiency and fault handling capabilities of the power grid.

[0099] The system testing and verification module performs system testing and verification based on data from the edge computing and cloud collaboration modules. After the system is deployed, comprehensive functional testing and on-site debugging will be carried out to ensure the stability and reliability of each module in actual application. At the same time, through long-term monitoring and feedback collection, the system will continuously optimize algorithms and models to improve its intelligence level to adapt to the ever-changing power grid environment.

[0100] A smart grid fault early warning and diagnosis method, used to implement a smart grid fault early warning and diagnosis system according to any one of claims 1 to 6, comprising the following steps:

[0101] Step 1: The power sensors, smart meters, and weather station equipment in the modal data acquisition and fusion module collect multi-source data in real time. The deep autoencoder model in the modal data acquisition and fusion module then performs dimensionality reduction and fusion on the collected multi-source data, representing it with unified features.

[0102] Step 2: Import the collected data into the deep reinforcement learning early warning module. The deep reinforcement learning early warning module builds an intelligent agent based on the deep Q network. It dynamically adjusts the early warning strategy by interacting with the power grid operating environment. The agent can adaptively adjust the fault warning threshold based on real-time data.

[0103] Step 3: Based on the deep reinforcement learning early warning module warning strategy, when the system detects a potential fault, the system will analyze the topology of the power grid based on the graph neural network fault diagnosis module to identify the location of the fault;

[0104] Step 4: To cope with various failure modes, the system also integrates a multi-task learning model that can handle multiple failure types simultaneously, improving the ability to identify complex failures.

[0105] Step 5: The system combines edge computing with cloud computing to work together. The edge nodes are responsible for real-time data preprocessing and transmit data to the cloud for more complex data analysis.

[0106] Step 6: The system testing and verification module performs system testing and verification based on the data from the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on graph neural network, the adaptive diagnosis module integrated with multi-task learning, and the edge computing and cloud collaboration module.

[0107] Working principle: First, the multi-source data of the power grid is collected in real time through power sensors, smart meters, and weather station equipment. The collected data including voltage, current, frequency and temperature are recorded as , smart meters record user electricity consumption data , analyze load fluctuations, collect external environmental parameters such as wind speed and humidity that affect power supply and record them as , the collected data consists of a feature vector composed of multiple sensors Preprocessing is performed, and the deep autoencoder model is used to reduce the dimension and fuse the data. The fused feature representation ,in is the input multimodal feature vector, is the weight matrix, is the bias term, is the activation function. The data collected and fused by the multimodal data acquisition and fusion module is then processed by the fault warning module of deep reinforcement learning. In the fault warning stage, the system builds an intelligent agent based on the deep Q network (DQN). The agent interacts with the power grid operating environment and updates its strategy to maximize the warning effect. The specific formula for the updated Q value is as follows:

[0108]

[0109] in Indicates that the status Take action Q value, represents the learning rate, Represents the current reward, represents the discount factor, Indicates the current state, Indicates the current action. Indicates the next state, Represents all possible actions, based on the early warning strategy of the deep reinforcement learning early warning module. When the system detects a potential fault, the system will analyze the topology of the power grid based on the fault diagnosis module of the graph neural network and identify the location of the fault. In order to deal with various fault modes, the system also integrates a multi-task learning model, which can handle multiple fault types at the same time and improve the ability to identify complex faults. The model realizes joint learning of different fault modes through shared feature representation to ensure that the system comprehensively diagnoses the health status of the power grid. Finally, the system combines edge computing with cloud computing to work together. The edge node is responsible for real-time data preprocessing, thereby reducing delays and transmitting data to the cloud for more complex data analysis. This collaborative mechanism ensures the efficiency and real-time response capability of the system. In summary, this system can effectively realize fault early warning and diagnosis of the power grid through advanced technical means to ensure the safety and stability of the power system.

[0110] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A smart grid fault warning and diagnosis system, characterized by: It includes a multimodal data acquisition and fusion module, a deep reinforcement learning early warning module, a fault diagnosis module based on graph neural networks, an adaptive diagnosis module integrating multi-task learning, an edge computing and cloud collaboration module, and a system testing and verification module; The multimodal data acquisition and fusion module includes a multimodal data acquisition unit and a multimodal data fusion unit. The multimodal data acquisition unit acquires heterogeneous data from multiple sources and collaboratively processes the data acquired by the multimodal data acquisition unit through a multimodal fusion model within the multimodal data fusion unit. The deep reinforcement learning warning module includes an intelligent agent and reinforcement learning model unit and an adaptive threshold adjustment unit. The intelligent agent and reinforcement learning model unit includes building a deep Q network intelligent agent. Based on the deep Q network intelligent agent system, the multimodal data acquisition and fusion module collects and fuses data to perform real-time dynamic data and environmental interaction processing; the adaptive threshold adjustment unit adaptively adjusts the fault warning threshold according to different environmental conditions; The fault diagnosis module based on the graph neural network includes a power grid topology modeling unit, a graph convolutional neural network model unit, and a fault location and diagnosis unit. The fault diagnosis module based on the graph neural network utilizes the graph structure properties of the power grid and builds a complex power grid topology model based on the graph convolutional network to achieve high-precision fault location and diagnosis. The fault diagnosis module of the graph neural network includes a power grid topology modeling unit, a graph convolutional neural network model unit and a fault location and diagnosis unit. The power grid topology modeling unit represents the power grid as a graph. ,in represents a node in the power grid, Represents the transmission line between nodes, and the characteristic vector of each node is the current voltage, current, frequency and temperature; The graph convolutional neural network model unit uses a graph convolutional network to process the graph structure data of the power grid, learn the relationship between node features, and identify potential fault locations. The layer update formula of the graph convolutional network is: in Represented as the adjacency matrix of the graph, Represented as a degree matrix, Expressed as The node feature matrix of the layer, Represented as a weight matrix, is the activation function; The fault location and diagnosis unit classifies each node of the power grid based on a graph convolutional neural network model and outputs the corresponding fault type and the node location where the fault occurred; The adaptive diagnosis module integrated with multi-task learning enhances robustness to different power grid faults and improves diagnostic accuracy by jointly learning multiple fault modes based on the fault diagnosis module data of the graph neural network. The adaptive diagnosis module of the integrated multi-task learning includes a multi-task learning model unit and an adaptive learning rate adjustment unit. The multi-task learning model unit uses a shared feature representation and a task-specific output layer to achieve joint diagnosis of different types of faults, setting multiple tasks , k represents a positive integer, each task corresponds to a specific failure mode, and the loss function is the weighted sum of all task losses: in Expressed as the total loss function, Expressed as the number of tasks, Represented as a task The weight of Indicates a task The loss function of The adaptive learning rate adjustment unit dynamically adjusts the learning rate of each task based on the learning progress in the multi-task learning model unit. The adaptive learning rate update formula is: in represents the current learning rate, Indicates a task The loss at the tth iteration, Indicates a task The loss at the t-1th iteration; The edge computing and cloud collaboration module includes an edge computing real-time preprocessing unit and a cloud big data analysis unit. The edge computing and cloud collaboration module preprocesses the adaptive diagnosis module data integrated with multi-task learning in real time at the edge node, and works in conjunction with the cloud big data analysis platform in the cloud collaboration unit to ensure system response speed; The system testing and verification module performs system testing and verification based on the data of the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on the graph neural network, the adaptive diagnosis module integrating multi-task learning, and the edge computing and cloud collaboration module.

2. The smart grid fault early warning and diagnosis system according to claim 1, characterized in that: The multimodal data acquisition unit includes power sensors, smart meters and weather stations to collect multi-source data of the power grid in real time. The power sensors include voltage sensors, current sensors, frequency sensors and temperature sensors. The collected data including voltage, current, frequency and temperature are recorded as ; The smart meter records the user's electricity consumption data as , analyze load fluctuations, collect external environmental parameters such as wind speed and humidity that affect power supply and record them as ; Where t represents time, 、 、 are the independent variables of each parameter that changes over time.

3. The smart grid fault early warning and diagnosis system according to claim 2, characterized in that: The multimodal data fusion unit includes a deep autoencoder model. The multimodal data fusion unit uses the deep autoencoder model in the modal data fusion unit to reduce the dimension and fuse the features of the collected data to generate a unified feature representation. They come from different data sources, that is, the collected data is represented as a multi-dimensional state vector X(t), as shown below: After dimensionality reduction through the deep autoencoder, a unified low-dimensional representation is obtained: in is the input multimodal feature vector, is the weight matrix, is the bias term, is the activation function.

4. The smart grid fault early warning and diagnosis system according to claim 1, characterized in that: The intelligent agent and reinforcement learning model construct an intelligent agent based on a deep Q network. By interacting with the power grid operating environment, the early warning strategy is dynamically adjusted. The agent system selects the optimal action based on the current state and updates the Q value. The specific formula for the updated Q value is as follows: in Indicates that the status Take action Q value, represents the learning rate, Represents the current reward, represents the discount factor, Indicates the current state, Indicates the current action. Indicates the next state, Represents all possible actions.

5. The smart grid fault early warning and diagnosis system according to claim 1, characterized in that: The adaptive threshold adjustment unit, the adaptive adjustment fault warning threshold is expressed as: in Indicates the voltage warning threshold, represents the initial voltage threshold, Indicates the voltage value that is adaptively adjusted according to the magnitude of meteorological changes.

6. The smart grid fault early warning and diagnosis system according to claim 1, characterized in that: The edge computing and cloud collaboration module includes an edge computing real-time preprocessing unit and a cloud-based big data analysis unit. The edge computing real-time preprocessing unit performs real-time data preprocessing at the edge node of the power grid, including data cleaning, dimensionality reduction and preliminary diagnosis, to reduce data transmission delays and ensure the real-time performance of the system. The cloud-based big data analysis unit uses high-performance computing resources to perform complex analysis on large amounts of historical data and real-time data, providing more refined fault prediction and diagnosis results. The edge and cloud work together to achieve efficient resource utilization. The system testing and verification module performs system testing and verification based on the data of the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on graph neural network, the adaptive diagnosis module integrated with multi-task learning, and the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on graph neural network, the adaptive diagnosis module integrated with multi-task learning, and the edge computing and cloud collaboration module.

7. A smart grid fault early warning and diagnosis method, for implementing a smart grid fault early warning and diagnosis system according to any one of claims 1 to 6, comprising the following steps: Step 1: The power sensors, smart meters, and weather station equipment in the modal data acquisition and fusion module collect multi-source data in real time. The deep autoencoder model in the modal data acquisition and fusion module then performs dimensionality reduction and fusion on the collected multi-source data, representing it with unified features. Step 2: Import the collected data into the deep reinforcement learning early warning module. The deep reinforcement learning early warning module builds an intelligent agent based on the deep Q network. It dynamically adjusts the early warning strategy by interacting with the power grid operating environment. The agent can adaptively adjust the fault warning threshold based on real-time data. Step 3: Based on the deep reinforcement learning early warning module warning strategy, when the system detects a potential fault, the system will analyze the topology of the power grid based on the graph neural network fault diagnosis module to identify the location of the fault; Step 4: To cope with various failure modes, the system also integrates a multi-task learning model that can handle multiple failure types simultaneously, improving the ability to identify complex failures. Step 5: The system combines edge computing with cloud computing to work together. The edge nodes are responsible for real-time data preprocessing and transmit data to the cloud for more complex data analysis. Step 6: The system testing and verification module performs system testing and verification based on the data from the edge computing and cloud collaboration module. The system will continuously optimize the deep reinforcement learning early warning module, the fault diagnosis module based on graph neural network, the adaptive diagnosis module integrated with multi-task learning, and the edge computing and cloud collaboration module.

Images