Power Data Analysis Method and Device Based on Graph Computing

Abstract

The present invention relates to the technical field of power data analysis, and specifically includes a power data analysis method and device based on graph computing, including: constructing a power network graph, performing deep learning processing on multi-dimensional and heterogeneous data, and analyzing dynamic changes to predict trends; optimizing power dispatching in combination with reinforcement learning, aiming to minimize energy loss, matching the optimal power dispatching decision, solving the problems of complex processing of high-dimensional data and heterogeneous data during the operation of the power system, being unable to accurately predict the change trend of the power system, and lacking effective dispatching decision-making technologies, achieving the goal of adaptive strategy optimization learning according to the real-time grid operation status and prediction trends, greatly improving the flexibility of power dispatching, aiming to minimize energy loss, continuously iterating and optimizing to accurately match the optimal power dispatching decision, improving the efficiency of power system data analysis, adjusting strategies to cope with changing supply and demand conditions and grid operation conditions, and maximizing the overall stability and reliability of the power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of power data analysis, and in particular to a power data analysis method and device based on graph computing. Background Art

[0002] Power data analysis methods usually use relational database technology and statistical methods to process data in the power system, and perform well when dealing with relatively simple and small-scale power grids. However, when facing modern complex and large-scale power grid systems, many deficiencies are exposed. Modern power systems include a large number of entities such as power plants, substations, and transmission lines. The logical connection relationships between entities are complex and dynamically changing. The data generated during the operation of the power system has high-dimensional and heterogeneous characteristics, making it difficult to efficiently process high-concurrency complex data streams.

[0003] In summary, in the prior art, there are technical problems such as complex processing of high-dimensional data and heterogeneous data during the operation of the power system, inability to accurately predict the dynamic change trend of the power system, and lack of effective scheduling decisions. Summary of the Invention

[0004] The present application provides a power data analysis device based on graph computing, aiming to solve the technical problems in the prior art, such as complex processing of high-dimensional data and heterogeneous data during the operation of the power system, inability to accurately predict the dynamic change trend of the power system, and lack of effective scheduling decisions.

[0005] In view of the above problems, the technical solution of the present application is as follows:

[0006] On the one hand, this application provides a method for power data analysis based on graph computing. The method includes: constructing a power network structure diagram, where nodes represent entities in the power system, including power plants, substations, and transmission lines, and edges represent the logical connection relationships between entities; extracting features from the nodes and edges in the power network structure diagram to capture the mutual influence relationships and interactive connection relationships between entities in the power system; based on the mutual influence relationships and interactive connection relationships, using a deep learning architecture to perform multi-modal graph fusion calculation to process the high-dimensional data in the power system and collect a first deep dynamic evolution set; based on the mutual influence relationships and interactive connection relationships, using a deep learning architecture to perform multi-modal graph fusion calculation to process the heterogeneous data in the power system and collect a second deep dynamic evolution set; starting from the power grid operation state, analyzing the dynamic changes of the power system in the time dimension and space dimension through the first deep dynamic evolution set and the second deep dynamic evolution set, and extracting the perception prediction trend; on the basis of the power network structure diagram, deploying a reinforcement learning algorithm, and according to the real-time updated power grid operation state, combining the perception prediction trend to perform adaptive policy optimization learning, and aiming at minimizing energy loss, matching the optimal power scheduling decision.

[0007] On the other hand, this application provides a power data analysis device based on graph computing. The device includes: a structure diagram construction module for constructing a power network structure diagram, where nodes represent entities in the power system, including power plants, substations, and transmission lines, and edges represent the logical connection relationships between entities; a feature extraction module for extracting features from the nodes and edges in the power network structure diagram to capture the mutual influence relationships and interactive connection relationships between entities in the power system; a first fusion calculation module for performing multi-modal graph fusion calculation based on the mutual influence relationships and interactive connection relationships using a deep learning architecture to process the high-dimensional data in the power system and collect a first deep dynamic evolution set; a second fusion calculation module for performing multi-modal graph fusion calculation based on the mutual influence relationships and interactive connection relationships using a deep learning architecture to process the heterogeneous data in the power system and collect a second deep dynamic evolution set; a dynamic change analysis module for starting from the power grid operation state, analyzing the dynamic changes of the power system in the time dimension and space dimension through the first deep dynamic evolution set and the second deep dynamic evolution set, and extracting the perception prediction trend; a scheduling decision module for deploying a reinforcement learning algorithm on the basis of the power network structure diagram, and according to the real-time updated power grid operation state, combining the perception prediction trend to perform adaptive policy optimization learning, and aiming at minimizing energy loss, matching the optimal power scheduling decision.

[0008] In summary, one or more technical solutions provided in this application solve the technical problems of complex processing of high-dimensional data and heterogeneous data during the operation of the power system, inability to accurately predict the dynamic change trend of the power system, and lack of effective scheduling decisions. The goal of adaptive strategy optimization learning based on the real-time grid operation status and prediction trend is achieved, greatly improving the flexibility of power dispatching. With the core goal of minimizing energy loss, through continuous iterative optimization, the optimal power dispatching decision is accurately matched, achieving the improvement of the power system data analysis efficiency, automatically adjusting the strategy according to the grid status under different scenarios to cope with the changing supply and demand conditions and grid operation conditions, thereby maximizing the overall stability and reliability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 FIG. is a schematic flowchart of a power data analysis method based on graph computing provided by this application;

[0010] Figure 2 FIG. is a schematic structural diagram of a power data analysis device based on graph computing provided by this application.

[0011] Description of the reference numerals: Structure diagram construction module M100, Feature extraction module M200, First fusion calculation module M300, Second fusion calculation module M400, Dynamic change analysis module M500, Scheduling decision module M600. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] Embodiment 1

[0013] The following specifically describes this application with reference to the drawings. As Figure 1 shown, this application provides a power data analysis method based on graph computing, where the method includes:

[0014] S1: Construct a power network structure diagram, where nodes represent entities in the power system, including power plants, substations, and transmission lines, and edges represent the logical connection relationships between entities; S2: Extract features from the nodes and edges in the power network structure diagram to capture the mutual influence relationships and interactive connection relationships between entities in the power system.

[0015] Construct a power network structure diagram. Specifically, define nodes, where nodes represent various entities in the power system, including power plants, substations, transmission lines, etc. Each node represents a specific part of the power system and plays an important role in the operation of the power grid. Define edges, where edges represent the logical connection relationships between entities. The connection relationships can be physical connections (such as a transmission line connecting a power plant and a substation), or logical dependency relationships (such as a substation depending on multiple power plants for power supply). The existence of edges reflects the mutual influence and dependence between different parts of the power system; connect the defined nodes and edges to form a power network structure diagram.

[0016] Extract features of the nodes and edges in the power network structure diagram. In order to capture the mutual influence relationships and interactive connection relationships between entities in the power system, specifically, extract node features. For each node, extract a series of features to describe its attributes and states, such as the power generation of a power plant, the voltage level of a substation, the transmission capacity of a transmission line, etc. These features reflect the functions and performances of the nodes in the power system; extract edge features. For the edges connecting different nodes, extract features such as transmission capacity, transmission loss, transmission distance, etc., to reflect the strength and nature of the connection relationships represented by the edges. Clearly showing each part of the power system and their mutual relationships is of great significance for deeply understanding the power system, optimizing power dispatching, and ensuring the stability of the power grid, and provides strong support for subsequent data analysis and decision-making.

[0017] S3: Based on the mutual influence relationships and interactive connection relationships, use a deep learning architecture to perform multi-modal graph fusion calculations to process the high-dimensional data in the power system and collect the first deep dynamic evolution set.

[0018] Based on the mutual influence relationships and interactive connection relationships represented by the nodes and edges in the power network structure diagram, use a deep learning architecture to perform multi-modal graph fusion calculations to process the high-dimensional data in the power system. Further, the power network structure diagram: includes nodes (power plants, substations, transmission lines, etc.) and edges (logical connection relationships); the high-dimensional data are various data generated during the operation of the power system, such as power generation, voltage, current, load changes, etc. These data usually have high dimensions and complexity.

[0019] Select a deep learning model suitable for processing graph-structured data, such as Graph Neural Networks, Graph Convolutional Networks, etc. These models can capture the complex relationships in the graph structure, including the mutual influence between nodes and the interactive connection relationships of edges.

[0020] Since the data in the power system has multiple modalities (such as numerical, text, image, etc.), multi-modal fusion calculation is required. Multi-modal fusion can be achieved by converting data of different modalities into a unified representation form and performing joint learning in a deep learning model.

[0021] The power system is a dynamically changing system, and its operating state changes over time. By using a deep learning model to perform dynamic evolution analysis on the nodes and edges in the power network structure diagram, the changing rules of the power system in the time dimension are captured. After multi-modal graph fusion calculation and dynamic evolution analysis, the collected data set is called the first deep dynamic evolution set. The first deep dynamic evolution set contains the dynamic change information of the power system in the time dimension, as well as the mutual influence and interaction connection relationships between different entities.

[0022] Through multi-modal graph fusion calculation based on mutual influence relationships and interaction connection relationships, high-dimensional data in the power system can be effectively processed, and useful dynamic change information can be extracted. This information is of great significance for subsequent power system analysis, prediction, and optimization decision-making. Using a deep learning architecture for multi-modal graph fusion calculation is a key step in processing high-dimensional data in the power system and capturing dynamic change rules, providing strong support for the intelligent management and optimization of the power system.

[0023] S4: Based on the mutual influence relationship and interaction connection relationship, use a deep learning architecture to perform multi-modal graph fusion calculation to process the heterogeneous data in the power system and collect the second deep dynamic evolution set.

[0024] Heterogeneous data refers to data from different sources with different structures and formats. To effectively process heterogeneous data, a deep learning architecture based on mutual influence relationships and interaction connection relationships is adopted for multi-modal graph fusion calculation to collect the second deep dynamic evolution set. Specifically, heterogeneous data from different sources is integrated, including sensor data, historical records, user feedback, etc. During the data integration process, problems such as inconsistent data formats, data redundancy, and data conflicts need to be solved; the integrated heterogeneous data is cleaned to remove noise, missing values, and inconsistent data, and data preprocessing is performed, such as data standardization and normalization, for subsequent data analysis and processing; for heterogeneous data, features that can reflect the mutual influence relationship and interaction connection relationship between power system entities are extracted, including data of multiple modalities such as numerical, text, and image.

[0025] Based on the already constructed power grid structure diagram, integrate the extracted features. Among them, nodes and edges represent the mutual influence relationships and interactive connection relationships among entities in the power system; select a deep learning model suitable for processing heterogeneous data and graph-structured data. The deep learning model needs to have the ability to process multiple modal data and be able to capture complex relationships in the graph structure; fuse different modal data in the graph model, which can be achieved by converting different modal data into a unified representation form and performing joint learning in the deep learning model. During the fusion process, the mutual influence and dependency relationships between different modal data need to be fully considered.

[0026] Similar to dealing with high-dimensional data, perform dynamic evolution analysis on the nodes and edges in the power grid structure diagram; capture the change laws of the power system in the time dimension through the deep learning model, especially the complexity and uncertainty brought by heterogeneous data.

[0027] After multi-modal graph fusion calculation and dynamic evolution analysis, the collected data set is called the second deep dynamic evolution set. The second deep dynamic evolution set not only contains the dynamic change information of the power system in the time dimension, but also fuses heterogeneous data from different sources, providing a more comprehensive description of the operating state of the power system.

[0028] The multi-modal graph fusion calculation based on the mutual influence relationship and interactive connection relationship processes the heterogeneous data in the power system, comprehensively captures the operating state and change laws of the power system, and collecting the second deep dynamic evolution set provides richer and more accurate data support for subsequent power system analysis, prediction, and optimization decisions.

[0029] S5: Starting from the power grid operating state, analyze the dynamic changes of the power system in the time dimension and space dimension through the first deep dynamic evolution set and the second deep dynamic evolution set, and extract the perception prediction trend.

[0030] Starting from the power grid operating state, deeply analyze the dynamic changes of the power system in the time dimension and space dimension through the first deep dynamic evolution set and the second deep dynamic evolution set, and extract the perception prediction trend. Specifically, take the current power grid operating state as the starting point of the analysis. The power grid operating state includes key indicators such as power generation, load level, voltage stability, and frequency; real-time monitor the operating parameters of the power grid through the sensor network, monitoring system, and data acquisition system to ensure the accuracy and timeliness of the data.

[0031] The first deep dynamic evolution set contains dynamic change information obtained by processing high-dimensional data in the power system, reflecting the state changes of the power system at different time points and the mutual influence and interaction connection relationships among entities; the second deep dynamic evolution set further integrates heterogeneous data from different sources, providing a more comprehensive description of the operating state of the power system, including numerical data, and also including information in various modalities such as text and images.

[0032] In the time dimension, a deep learning model is used to analyze the time series data in the first deep dynamic evolution set and the second deep dynamic evolution set, capture the dynamic change laws of the power system in the time dimension, analyze the change trends of the power grid state at different time points, and predict future state changes. In the space dimension, combined with the power network structure diagram, the layout and connection relationships of the power grid in the space dimension are analyzed, the mutual influence and dependence relationships among different regions and different nodes are evaluated, and potential weak points and fault propagation paths are identified.

[0033] By comprehensively analyzing the dynamic change information in the time dimension and the space dimension, key trends in the power system are identified, including load growth trends, fault occurrence probabilities, energy supply and demand changes, etc.; based on the identified trends, a prediction model is established to predict the future state of the power system. The prediction model needs to consider various influencing factors and uncertainty factors to provide accurate prediction results; the prediction results are compared and verified with the actual power grid operating state, and the accuracy and reliability of the prediction model are continuously optimized; through real-time sensing and feedback mechanisms, the power dispatching strategy is adjusted in a timely manner and the power grid operating parameters are optimized to ensure the stability and reliability of the power grid.

[0034] Starting from the operating state of the power grid, through the analysis of the first deep dynamic evolution set and the second deep dynamic evolution set, the dynamic changes of the power system in the time dimension and the space dimension are comprehensively captured, and the perception prediction trend is extracted, providing strong support for the optimal dispatching, fault prediction and stable operation of the power system.

[0035] S6: On the basis of the power network structure diagram, a reinforcement learning algorithm is deployed. According to the real-time updated operating state of the power grid, combined with the perception prediction trend, adaptive strategy optimization learning is carried out, and with the goal of minimizing energy loss, the optimal power dispatching decision is matched.

[0036] To further improve the efficiency and stability of the power system, a reinforcement learning algorithm is deployed based on the power network structure diagram to optimize power dispatching decisions. Specifically, the reinforcement learning algorithm is deployed. Further, the power network structure diagram and its dynamic changes (including the first deep dynamic evolution set and the second deep dynamic evolution set) constitute the environment of reinforcement learning, including various states (such as the load level of the power grid, voltage stability, power generation of power plants, etc.) and actions with a certain probability (such as adjusting the output of generators, switching transmission lines, starting or stopping standby power sources, etc.).

[0037] In a feasible implementation, the real-time monitoring system continuously collects the latest data on the operating state of the power grid and feeds it back to the reinforcement learning algorithm; combined with the extracted perception prediction trend, the reinforcement learning algorithm can more accurately understand the current operating state of the power grid and future change trends; the reinforcement learning algorithm performs adaptive policy optimization learning based on the current state and perception prediction trend. The goal of the reinforcement learning algorithm is to find a series of actions that can minimize energy loss and meet other operating constraints of the power grid (such as voltage stability, frequency control, etc.); according to the optimized policy, corresponding actions are executed to adjust the operating state of the power grid. The real-time monitoring system collects the results after the actions are executed and returns them as feedback signals to the reinforcement learning algorithm; the reinforcement learning algorithm adjusts its internal parameters and policies according to the feedback signals to continuously improve the future decision-making process.

[0038] The objective function of the reinforcement learning algorithm is set to minimize energy loss, which means that when the algorithm optimizes decisions, it will give priority to reducing unnecessary energy waste and improving energy utilization efficiency. In addition to minimizing energy loss, the reinforcement learning algorithm can also consider other objectives (such as improving the stability of the power grid, reducing the probability of faults, etc.) to achieve multi-objective optimization; by introducing weight factors in the objective function or using a hierarchical optimization method, a balance can be achieved between different objectives. The reinforcement learning algorithm has the ability of dynamic adaptation. As the operating state of the power grid changes continuously, it can timely adjust its strategy to adapt to the changing supply and demand conditions and power grid operating conditions. The dynamic adaptability ensures the flexibility and effectiveness of power dispatching decisions, and helps to improve the overall performance and reliability of the power grid.

[0039] In summary, deploying a reinforcement learning algorithm based on the power network structure diagram and performing adaptive policy optimization learning according to the real-time updated operating state of the power grid and perception prediction trend is an effective method to achieve minimum energy loss and match the optimal power dispatching decision, making full use of the complex data and dynamic characteristics of the power system and improving the intelligent management level and operating efficiency of the power grid.

[0040] Furthermore, based on the mutual influence relationship and interactive connection relationship, a deep learning architecture is used to perform multi-modal graph fusion calculation to process the high-dimensional data in the power system, and a first deep dynamic evolution set is collected. The method of this application further includes:

[0041] Obtain the fault history records, mark them on the power network structure diagram, and add fault type marks. The fault types corresponding to the fault type marks include short circuit, open circuit, overload, and equipment failure; based on the fault type marks, with the fault time start point and fault time end point as constraints, perform associated factor extraction to obtain the fault occurrence associated factor set, and the fault occurrence associated factor set corresponds to the weather condition factor, load change factor, and equipment operation state factor during the fault period; construct a fault analysis model, use the fault occurrence associated factor set as the input, and perform in-depth analysis using machine learning algorithms to identify the external fault influence, internal fault influence, and interaction mechanism affecting the fault occurrence; based on the fault analysis model and the external fault influence, internal fault influence, and interaction mechanism affecting the fault occurrence, establish a closed-loop fault management system.

[0042] Process the high-dimensional data in the power system and construct a closed-loop fault management system. Specifically, based on the mutual influence relationship and interactive connection relationship between entities in the power system, a deep learning architecture is used to perform multi-modal graph fusion calculation, aiming to process the high-dimensional data in the power system and collect a first deep dynamic evolution set, and the first deep dynamic evolution set reflects the dynamic change characteristics of the power system in the time dimension and space dimension. Obtain the fault history records of the power system and mark the fault types on the power network structure diagram. These marks include different types of fault types, such as short circuit, open circuit, overload, equipment failure, etc. Through the fault type marks, the common fault modes and their distributions in the power system can be clearly identified.

[0043] Based on the fault type marks, with the fault time start point and fault time end point as constraints, perform associated factor extraction. This step aims to identify various factors closely related to the fault occurrence, such as the weather condition factor, load change factor, equipment operation state factor, etc. during the fault period. By organizing these factors into the fault occurrence associated factor set, rich data support is provided for subsequent fault analysis.

[0044] Construct a fault analysis model. The fault analysis model uses the fault occurrence associated factor set as the input and performs in-depth analysis using machine learning algorithms to further identify the external fault influence, internal fault influence, and their interaction mechanism affecting the fault occurrence, which is of great significance for understanding the root cause of the fault occurrence and formulating effective fault prevention measures.

[0045] Based on the fault analysis model and various factors and mechanisms identified that affect the occurrence of faults, a closed-loop fault management system is established. The closed-loop fault management system includes key links such as quantitatively evaluating the contribution degree of fault impact, predicting fault risks, incorporating dynamic updates of the power network structure diagram, establishing a real-time feedback mechanism, and continuously adjusting fault maintenance strategies, so as to achieve comprehensive monitoring, early warning, and response to power system faults, and ensure the stability and reliability of the power grid.

[0046] Furthermore, based on the fault analysis model and external fault impacts, internal fault impacts, and interaction mechanisms that affect the occurrence of faults, a closed-loop fault management system is established. The method of this application includes:

[0047] Based on the fault analysis model and external fault impacts and internal fault impacts that affect the occurrence of faults, a quantitative evaluation is carried out to determine the contribution degree of factors affecting the occurrence of faults; using the contribution degree of factors affecting the occurrence of faults, under the constraint of the interaction mechanism, combined with the real-time updated power grid operation status, a fault risk prediction is carried out to obtain the fault prediction type and fault prediction probability; according to the fault prediction type and fault prediction probability, incorporate them into the dynamic update of the power network structure diagram. At the same time, a real-time feedback mechanism is established and the fault maintenance strategy is continuously adjusted to form a closed-loop fault management system.

[0048] Through the fault analysis model and the quantitative evaluation of external and internal fault impacts, the prediction of fault risks and the adjustment of dynamic maintenance strategies are realized. Specifically, the fault analysis model is used to deeply analyze the collected fault history records. The fault analysis model can identify external factors (such as weather conditions, load changes) and internal factors (such as equipment operation status, maintenance level) that affect the occurrence of faults, and reveal the interaction mechanism between them. Through quantitative evaluation, the contribution degree of each external fault impact and internal fault impact to the occurrence of faults is determined. Quantifying the contribution degree of fault occurrence is crucial for understanding the root cause of faults and identifying key influencing factors.

[0049] Combined with the real-time updated power grid operation status data, including key indicators such as power generation, load level, and voltage stability, real-time information is provided for fault risk prediction. Based on the known contribution degree of each influencing factor, using the constraint conditions of the interaction mechanism, a fault risk prediction is carried out, including predicting the fault types (such as short circuit, open circuit, overload, etc.) with a certain probability of occurrence and the corresponding fault prediction probability.

[0050] Incorporating information such as fault prediction types and fault prediction probabilities into the dynamic update of the power grid structure diagram means marking potential fault areas and risk points on the power grid structure diagram for real-time monitoring and rapid response. Based on the dynamically updated power grid structure diagram, optimize the operation strategy and dispatching plan of the power grid to reduce the probability of faults and the impact of faults on the operation of the power grid.

[0051] Establish a real-time feedback mechanism to compare and verify the real-time data of the power grid operation status with the fault prediction results, which helps to promptly detect prediction errors and adjust the prediction model to improve the accuracy of prediction. According to the information feedback in real time, continuously adjust the fault maintenance strategy, including measures such as optimizing the maintenance plan, improving the equipment maintenance level, and strengthening the monitoring of key areas.

[0052] By repeatedly executing the above steps, a closed-loop fault management system is formed to achieve comprehensive monitoring, prediction, and dynamic management of power grid faults, ensuring the stability and reliability of the power grid. With the continuous accumulation of power grid operation data and the continuous optimization of the analysis model, the closed-loop fault management system will be able to continuously improve its prediction accuracy and management efficiency, achieving accurate prediction and effective management of power grid faults.

[0053] Furthermore, the method of this application also includes:

[0054] Based on the fault type marking, taking the directly affected nodes and indirectly affected nodes as constraints, extract associated factors to obtain a set of fault distribution associated factors. The set of fault occurrence associated factors corresponds to the fault propagation path factors affecting the power space distribution, the load change factors in the affected area, and the standby power input factors. Using the set of fault distribution associated factors as a constraint condition, apply a graph neural network to simulate the fault propagation process in the power grid and establish a fault propagation architecture. Among them, the fault analysis model is trained based on the fault propagation architecture.

[0055] In-depth research has been carried out on the propagation and impact of faults in the power grid. Further, according to historical fault records and the current operation status of the power grid, mark the fault types of nodes (such as power plants, substations, transmission lines) in the power grid structure diagram. The fault types corresponding to the fault type marking include short circuits, open circuits, overloads, equipment failures, etc.

[0056] Based on the fault type labels, relevant associated factors related to the fault distribution are further extracted. The associated factors include constraint conditions at two levels: directly affected nodes and indirectly affected nodes. Directly affected nodes are directly related to the occurrence of faults, such as the transmission lines or substations where the fault points are located; for directly affected nodes, the associated factors include equipment aging, sudden load changes, etc. Indirectly affected nodes do not directly experience faults but are significantly affected by the faults, such as adjacent substations connected to the fault points or neighboring areas affected by load transfer caused by the faults; for indirectly affected nodes, the associated factors include the fault propagation paths affected by the power spatial distribution, load changes in the affected areas, the situation of standby power supply input, etc. The above-mentioned associated factors are organized into a set of fault distribution associated factors, which not only describes the directly affected areas of the faults but also reveals the propagation paths and potential influence ranges of the faults in the power grid.

[0057] Using the extracted set of fault distribution associated factors as constraint conditions, the propagation process of faults in the power grid is simulated by using a graph neural network (Graph Neural Network). The graph neural network is good at processing graph-structured data, can capture the complex relationships between nodes, and predict their future change trends. Through the simulation calculation of the graph neural network, a fault propagation architecture is established. The fault propagation architecture describes the entire process of faults from the occurrence point to the diffusion path and then to the final influence range, which not only reflects the spatial distribution characteristics of the faults in the power grid but also reveals the impact of the faults on the operating state and stability of the power grid. The established fault propagation architecture provides a model basis for the training of the fault analysis model; the fault analysis model uses this architecture to train and learn a large amount of historical fault data to improve its prediction and analysis capabilities for new faults. Generally speaking, it helps to reduce the impact of faults on the operation of the power grid and improves the stability and reliability of the power grid.

[0058] Furthermore, the method of this application also includes:

[0059] Based on the fault propagation architecture, comparing with the nodes and edges in the power network structure diagram, the power grid vulnerability index under the limitation of the fault type labels is evaluated; the potential influence range is combined with the power grid vulnerability index under the limitation of the fault type labels, and a renewable energy power generation unit is connected. The GAE encoder is used to extract the renewable energy power generation characteristics; through the renewable energy power generation characteristics, the power grid energy combination and energy scheduling strategy are optimized and returned to the renewable energy power generation unit, and the renewable energy power generation unit is bound to the closed-loop fault management system.

[0060] Assessment of power grid vulnerability and access and optimization strategies for renewable energy generation units aim to enhance the stability and reliability of the power grid through the analysis of the fault propagation architecture. Further, based on the established fault propagation architecture, a detailed vulnerability assessment is conducted by comparing the nodes and edges in the power network structure diagram, considering various factors under the limitation of fault type markings, such as fault propagation paths, influence ranges, and the criticality of nodes and edges. By comprehensively considering factors such as the probability of fault occurrence, fault influence range, and the criticality of nodes and edges, the vulnerability index of each node and edge is calculated, and the vulnerability index of each node and edge reflects the vulnerability degree of the power grid under a specific fault type.

[0061] A combined analysis of the potential influence range and the power grid vulnerability index aims to identify the most vulnerable areas and nodes in the power grid with the widest potential influence range. According to the results of the combined analysis, the priorities for power grid maintenance and renovation are determined; priority is given to those nodes and areas with high vulnerability indices and wide potential influence ranges.

[0062] Access renewable energy generation units. Specifically, based on the results of the vulnerability assessment and the analysis of the potential influence range, plan the access locations of renewable energy generation units; preferentially select those locations that can improve the stability of the power grid and reduce vulnerability for access. Integrate the renewable energy generation units with the power grid system to ensure their stable and efficient operation in the power grid.

[0063] Use a GAE encoder to extract renewable energy generation characteristics. Specifically, introduce a GAE encoder (Graph Autoencoder) as a feature extraction tool; the GAE encoder can capture the complex relationships between renewable energy generation units and the power grid and extract their key features. Use the GAE encoder to process the operation data of renewable energy generation units and extract their generation characteristics, stability characteristics, and interaction characteristics with other nodes, etc.

[0064] Based on the renewable energy generation characteristics, optimize the energy mix of the power grid; consider the complementarity between different renewable energy generation units and the coordination relationship between them and traditional energy generation units to achieve efficient utilization of energy and stable operation of the power grid. According to the optimized energy mix, formulate a more reasonable energy dispatching strategy, and the energy dispatching strategy should be able to fully consider factors such as the real-time operation state of the power grid, load changes, and the uncertainty of renewable energy generation to ensure the stable operation of the power grid under various working conditions.

[0065] Return the optimized energy portfolio and scheduling strategy to the renewable energy power generation units and the power grid operation system for real-time application and adjustment. Based on the power grid operation status and the real-time data feedback of the renewable energy power generation units, form a closed-loop control system, and continuously adjust and optimize the energy portfolio and scheduling strategy to significantly improve the stability and reliability of the power grid and promote the effective utilization and development of renewable energy.

[0066] Furthermore, the method of this application includes:

[0067] Introduce power grid security indicators, where the power grid security indicators include the power grid voltage qualification rate, power grid stability indicators, N-1 security verification indicators, N-2 security verification indicators, line thermal stability limit, and static stability limit; use graph embedding technology to integrate the power grid security indicators into the power grid structure diagram for enhancing the security features of nodes and edges; through a graph neural network, conduct risk assessments on key nodes and key edges respectively, identify potential security risks, and adaptively optimize the closed-loop fault management system.

[0068] Introduce power grid security indicators and combine graph embedding technology for security feature enhancement, and use graph neural networks for risk assessment and optimization to improve the security and stability of the power grid. Further, in order to comprehensively evaluate the security performance of the power grid, multiple power grid security indicators are introduced, including but not limited to: the power grid voltage qualification rate (a key indicator for measuring the stability of the power grid voltage), power grid stability indicators (reflecting the ability of the power grid to maintain stable operation after being disturbed), N-1 security verification indicators (evaluating the ability of the power grid to still maintain normal power supply in the case of a single component failure, where a single component is commonly a transmission line or a transformer), N-2 security verification indicators (further evaluating the power supply capacity and stability of the power grid when two component failures occur simultaneously), line thermal stability limit (considering the current-carrying capacity of the line and its stable operation limit in a high-temperature environment), and static stability limit (evaluating the stability boundary of the power grid under static conditions to prevent system collapse), so as to comprehensively evaluate the redundancy and risk resistance ability of the power grid and ensure that the power grid has higher reliability and stability.

[0069] Use graph embedding technology for security feature enhancement. The graph embedding technology can embed the complex relationships and high-dimensional data of the power grid into a low-dimensional space while retaining the important information in the original graph structure. By integrating the power grid security indicators into the power grid structure diagram, enhance the security features of nodes and edges. In the embodiments of this application, it can be achieved through calculation, where the original security features of nodes, the dynamic weights of each security indicator, the security features of adjacent nodes, and their network topology relationships are considered.

[0070] Using a graph neural network to deeply analyze the enhanced power grid structure diagram, the GNN can capture the complex relationships between nodes and edges, as well as their mutual influences. By conducting risk assessments on nodes and edges, key nodes and edges in the power grid are identified; these nodes and edges play important roles in the operation of the power grid, and once a failure occurs or they are attacked, it may have a serious impact on the power grid. Based on the risk assessment results, potential security risks in the power grid are identified, and the potential security risks stem from various factors such as equipment aging, load mutation, and external environment changes.

[0071] According to the risk assessment results and the identified potential security risks, corresponding optimization strategies are formulated, including strengthening the monitoring and maintenance of key nodes, optimizing the energy scheduling strategy, and improving the redundancy of the power grid. The optimization strategies are applied to the operation of the power grid, and through a real-time feedback mechanism, the strategies are continuously adjusted and optimized. In this step, the closed-loop fault management system will continuously perform adaptive optimization according to the changes in the operation state and security performance of the power grid to ensure the safe and stable operation of the power grid and provide a strong guarantee for the reliable operation of the power system.

[0072] Furthermore, using graph embedding technology, the power grid security indicators are integrated into the power grid structure diagram to enhance the security features of nodes and edges. The method of this application includes: the power grid security indicator fusion formula: ; where is the updated security feature of node v, is the original security feature of node v, is the dynamic weight of the k-th security indicator, is the value of the k-th power grid security indicator corresponding to node v, N(v) is the set of adjacent nodes of node v, represents the network topology correlation coefficient between node v and adjacent node j, is the security feature of adjacent node j; is the weight function between node v and adjacent node j at time t, is the basic connection weight between node v and adjacent node j, is the attenuation rate, t is the current time, is the time of the previous interaction. Using graph embedding technology to integrate the power grid security indicators into the power grid structure diagram to enhance the security features of nodes and edges involves the power grid security indicator fusion formula, and the power grid security indicator fusion formula details how to combine the original security features of nodes, the values of each security indicator, the security features of adjacent nodes, and their network topology and time relationships to calculate the updated security features of nodes.

[0073] By integrating power grid security indicators into the power grid structure diagram and enhancing the security features of nodes and edges, it can be achieved through calculation in the embodiments of this application. Further, configure the power grid security indicator fusion formula: , where is the dynamic weight of the k-th security indicator, reflecting the importance of different security indicators in evaluating the security features of nodes; is the value of the k-th power grid security indicator corresponding to node v, such as the power grid voltage qualification rate, power grid stability indicator, etc.; N(v) is the set of adjacent nodes of node v, indicating the connection relationship of node v in the network; represents the network topology correlation coefficient between node v and adjacent node j, reflecting the tightness and importance of the connection between node v and adjacent node j. is the weight function between node v and adjacent node j at time t, used to consider the influence of time factors on the relationship between nodes; is the basic connection weight between node v and adjacent node j, indicating their static connection strength; is the attenuation rate, reflecting the speed of change of the relationship between nodes over time; t is the current time, is the time of the previous interaction, used to calculate the time difference and affect the value of the weight function.

[0074] As the original security feature of node v, it is retained in the updated security feature; by weighted summing the values of each security indicator and multiplying by the corresponding dynamic weight , multiple power grid security indicators are integrated into the security feature of node v. Considering the mutual influence between nodes and the time dynamics of their relationships, using the network topology correlation coefficient and the time weight function , the security feature of adjacent node j is fused into the security feature of node v. The weight function introduces time factors, enabling the fusion of security features between nodes to reflect the real-time changes in the operating state of the power grid. By adjusting the attenuation rate and considering the previous interaction time , the influence of time factors on the relationship between nodes is accurately controlled. The power grid security indicators are effectively integrated into the power grid structure diagram, enhancing the security feature representation ability of nodes and edges, and more accurately evaluating the stability and reliability of the power grid.

[0075] In summary, the beneficial effects of the embodiments of this application are:

[0076] 1. Intuitively displaying the entities in the power system and their logical connection relationships helps to understand the operation mechanism of the entire system, and through deep learning technology to process high-dimensional and heterogeneous data, perform multimodal graph fusion calculations, efficiently process high-dimensional and heterogeneous data in the power system, and improve the timeliness of data analysis.

[0077] 2. Based on the real-time update and perception prediction trend of the power grid operation status, adaptively perform policy optimization learning, match the optimal power dispatching decision, so as to achieve the minimum loss of energy and improve the overall performance of the system.

[0078] 3. By introducing power grid security indicators and using graph embedding technology to enhance the security features of nodes and edges, analyze the dynamic changes of the power system in the time and space dimensions, identify and evaluate potential security risks, and adaptively optimize the closed-loop fault management system, effectively predict and manage faults in the power system, achieve accurate prediction and prevention of faults, and improve the stability and reliability of the power grid.

[0079] 4. Due to the adoption of the power grid security indicator fusion formula: ; where is the updated security feature of node v, is the original security feature of node v, is the dynamic weight of the kth security indicator, is the value of the kth power grid security indicator corresponding to node v, N(v) is the set of adjacent nodes of node v, represents the network topology correlation coefficient between node v and adjacent node j, is the security feature of adjacent node j; is the weight function between node v and adjacent node j at time t, is the basic connection weight between node v and adjacent node j, is the attenuation rate, t is the current time, is the time of the previous interaction. Effectively integrating the power grid security indicators into the power network structure diagram enhances the security feature representation ability of nodes and edges, and more accurately evaluates the stability and reliability of the power grid.

[0080] Embodiment 2

[0081] Based on the same inventive concept as the power data analysis method based on graph calculation in the foregoing embodiment, as Figure 2 shown, the embodiment of the present application provides a power data analysis device based on graph calculation, wherein the device includes:

[0082] The structure diagram construction module M100 is used to construct the structure diagram of the power grid. Among them, the nodes represent the entities in the power system, including power plants, substations, and transmission lines, and the edges represent the logical connection relationships between the entities;

[0083] The feature extraction module M200 is used to extract features from the nodes and edges in the power grid structure diagram, and capture the mutual influence relationships and interactive connection relationships between the entities in the power system;

[0084] The first fusion calculation module M300 is used to perform multi-modal graph fusion calculation based on the mutual influence relationships and interactive connection relationships, using a deep learning architecture, to process the high-dimensional data in the power system and collect the first deep dynamic evolution set;

[0085] The second fusion calculation module M400 is used to perform multi-modal graph fusion calculation based on the mutual influence relationships and interactive connection relationships, using a deep learning architecture, to process the heterogeneous data in the power system and collect the second deep dynamic evolution set;

[0086] The dynamic change analysis module M500 is used to start from the power grid operation state, and analyze the dynamic changes of the power system in the time dimension and space dimension through the first deep dynamic evolution set and the second deep dynamic evolution set, and extract the perception prediction trend;

[0087] The scheduling decision module M600 is used to deploy a reinforcement learning algorithm on the basis of the power grid structure diagram, and perform adaptive policy optimization learning according to the real-time updated power grid operation state, combined with the perception prediction trend, and match the optimal power scheduling decision with the goal of minimizing energy loss.

[0088] Furthermore, the first fusion calculation module M300 is used to execute the following method:

[0089] Obtain the fault history record, mark it on the power grid structure diagram, and add the fault type mark. The fault types corresponding to the fault type marks include short circuit, open circuit, overload, and equipment failure;

[0090] Based on the fault type mark, with the fault time start point and fault time end point as constraints, perform associated factor extraction to obtain the fault occurrence associated factor set. The fault occurrence associated factor set corresponds to the weather condition factor, load change factor, and equipment operation state factor during the fault period;

[0091] Construct a fault analysis model, use the fault occurrence associated factor set as input, and perform in-depth analysis using machine learning algorithms to identify the external fault influence, internal fault influence, and interaction mechanism that affect the fault occurrence;

[0092] Based on the fault analysis model, external fault impacts, internal fault impacts, and interaction mechanisms that affect the occurrence of faults, a closed-loop fault management system is established.

[0093] Furthermore, the first fusion calculation module M300 is also used to execute the following method:

[0094] Based on the fault analysis model, external fault impacts, and internal fault impacts that affect the occurrence of faults, a quantitative assessment is carried out to determine the contribution degree that affects the occurrence of faults;

[0095] Using the contribution degree that affects the occurrence of faults, under the constraint of the interaction mechanism, combined with the real-time updated power grid operation status, a fault risk prediction is carried out to obtain the fault prediction type and fault prediction probability;

[0096] According to the fault prediction type and fault prediction probability, incorporate them into the dynamic update of the power network structure diagram. At the same time, establish a real-time feedback mechanism and continuously adjust the fault maintenance strategy to form a closed-loop fault management system.

[0097] Furthermore, the first fusion calculation module M300 is also used to execute the following method:

[0098] Based on the fault type label, with the directly affected nodes and indirectly affected nodes as constraints, relevant factors are extracted to obtain a set of fault distribution correlation factors. The set of fault occurrence correlation factors corresponds to the fault propagation path factors that affect the power space distribution, the load change factors in the affected area, and the standby power input factors;

[0099] Using the set of fault distribution correlation factors as a constraint condition, a graph neural network is used to simulate the fault propagation process in the power grid, and a fault propagation architecture is established. Among them, the fault analysis model is trained based on the fault propagation architecture as the model basis.

[0100] Furthermore, the first fusion calculation module M300 is also used to execute the following method:

[0101] Based on the fault propagation architecture, comparing the nodes and edges in the power network structure diagram, evaluate the power grid vulnerability index under the limitation of the fault type label;

[0102] Combine the potential impact range with the power grid vulnerability index under the limitation of the fault type label, connect to the renewable energy power generation unit, and use the GAE encoder to extract renewable energy power generation characteristics;

[0103] Through the renewable energy power generation characteristics, optimize the power grid energy combination and energy dispatching strategy, and return to the renewable energy power generation unit. The renewable energy power generation unit is bound to the closed-loop fault management system.

[0104] Furthermore, the first fusion calculation module M300 is also used to execute the following method:

[0105] Introduce power grid security indicators, which include the qualified rate of power grid voltage, power grid stability indicators, N-1 security verification indicators, N-2 security verification indicators, line thermal stability limit, and static stability limit;

[0106] Use graph embedding technology to integrate the power grid security indicators into the power grid structure diagram to enhance the security features of nodes and edges;

[0107] Through graph neural networks, perform risk assessments on key nodes and key edges respectively, identify potential security risks, and adaptively optimize the closed-loop fault management system.

[0108] Furthermore, the first fusion calculation module M300 is also used to execute the following method: Power grid security indicator fusion formula: ; where is the updated security feature of node v, is the original security feature of node v, is the dynamic weight of the kth security indicator, is the value of the kth power grid security indicator corresponding to node v, N(v) is the set of adjacent nodes of node v, represents the network topology correlation coefficient between node v and adjacent node j, is the security feature of adjacent node j; is the weight function between node v and adjacent node j at time t, is the basic connection weight between node v and adjacent node j, is the attenuation rate, t is the current time, is the time of the previous interaction. In summary, any step can be stored as computer instructions or programs in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor, without further limitation here.

[0109] Furthermore, the above technical solutions only reflect the preferred technical solutions of the technical solutions of the embodiments of the present application. Some changes that those skilled in the art may make to some parts thereof all reflect the principles of the novel embodiments of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application.

Claims

1. A method for power data analysis based on graph computing, characterized in that, The method includes constructing a power network structure diagram, where nodes represent entities in the power system, including power plants, substations, and transmission lines, and edges represent the logical connection relationships between entities; extracting features of the nodes and edges in the power network structure diagram to capture the mutual influence relationships and interactive connection relationships between entities in the power system; based on the mutual influence relationships and interactive connection relationships, using a deep learning architecture to perform multi-modal graph fusion calculations to process the high-dimensional data and heterogeneous data in the power system, and collecting the first deep dynamic evolution set and the second deep dynamic evolution set; starting from the power grid operation state, through the first deep dynamic evolution set and the second deep dynamic evolution set, analyzing the dynamic changes of the power system in the time dimension and space dimension, and extracting the perception prediction trend; on the basis of the power network structure diagram, deploying a reinforcement learning algorithm, and according to the real-time updated power grid operation state, combining the perception prediction trend to perform adaptive policy optimization learning, and aiming at minimizing energy loss, matching the optimal power dispatch decision; based on the mutual influence relationships and interactive connection relationships, using a deep learning architecture to perform multi-modal graph fusion calculations to process the high-dimensional data in the power system, and collecting the first deep dynamic evolution set. The method further includes: obtaining the fault history records and marking them on the power network structure diagram, adding fault type marks, and the fault types corresponding to the fault type marks include short circuit, open circuit, overload, and equipment failure; based on the fault type marks, with the fault time start point and fault time end point as constraints, extracting associated factors to obtain a set of fault occurrence associated factors, and the set of fault occurrence associated factors corresponds to weather condition factors, load change factors, and equipment operation state factors during the fault period; constructing a fault analysis model, using the set of fault occurrence associated factors as input, and using a machine learning algorithm to perform in-depth analysis to identify external fault impacts, internal fault impacts, and interaction mechanisms that affect fault occurrence; based on the fault analysis model and the external fault impacts, internal fault impacts, and interaction mechanisms that affect fault occurrence, establishing a closed-loop fault management system.

2. The power data analysis method based on graph calculation according to claim 1, wherein Based on the fault analysis model and the external fault impacts, internal fault impacts, and interaction mechanisms that affect fault occurrence, establishing a closed-loop fault management system. The method includes: performing quantitative evaluation based on the fault analysis model and the external fault impacts and internal fault impacts that affect fault occurrence to determine the contribution degree that affects fault occurrence; using the contribution degree that affects fault occurrence, under the constraint of the interaction mechanism, combining the real-time updated power grid operation state to perform fault risk prediction, and obtaining the fault prediction type and fault prediction probability; according to the fault prediction type and fault prediction probability, incorporating them into the dynamic update of the power network structure diagram. At the same time, establishing a real-time feedback mechanism and continuously adjusting the fault maintenance strategy to form a closed-loop fault management system.

3. The power data analysis method based on graph calculation according to claim 2, wherein The method further includes: Based on the fault type markers, with the directly affected nodes and indirectly affected nodes as constraints, relevant factors are extracted to obtain a set of fault distribution correlation factors. The set of fault occurrence correlation factors corresponds to the fault propagation path factors affecting the power spatial distribution, the load change factors in the affected area, and the standby power input factors. Taking the set of fault distribution correlation factors as a constraint condition, a graph neural network is used to simulate the fault propagation process in the power grid, and a fault propagation architecture is established. Among them, the fault analysis model is trained based on the fault propagation architecture.

4. The power data analysis method based on graph calculation according to claim 3, wherein The method further includes: Based on the fault propagation architecture, comparing the nodes and edges in the power network structure diagram, the power grid vulnerability index under the limitation of the fault type markers is evaluated. Combining the potential impact range with the power grid vulnerability index under the limitation of the fault type markers, a renewable energy power generation unit is connected, and a GAE encoder is used to extract the renewable energy power generation characteristics. Through the renewable energy power generation characteristics, the power grid energy combination and energy scheduling strategy are optimized and returned to the renewable energy power generation unit, and the renewable energy power generation unit is bound to the closed-loop fault management system.

5. The power data analysis method based on graph calculation according to claim 4, wherein The method includes: Introducing power grid security indicators, which include the power grid voltage qualification rate, power grid stability indicators, N-1 security verification indicators, N-2 security verification indicators, line thermal stability limit, and static stability limit. Using graph embedding technology, the power grid security indicators are integrated into the power network structure diagram to enhance the security features of nodes and edges. Through the graph neural network, risk assessments are respectively carried out on key nodes and key edges to identify potential security risks, and the closed-loop fault management system is adaptively optimized.

6. The power data analysis method based on graph computing according to claim 5, wherein, Using graph embedding technology, the power grid security indicators are integrated into the power network structure diagram to enhance the security features of nodes and edges. The method includes: Power grid security index fusion formula: ; Among them, is the updated security feature of node v, is the original security feature of node v, is the dynamic weight of the k-th security index, is the value of the k-th power grid security index corresponding to node v, and N(v) is the set of adjacent nodes of node v, represents the network topology correlation coefficient between node v and adjacent node j, is the security feature of adjacent node j; is the weight function between node v and adjacent node j at time t, is the basic connection weight between node v and adjacent node j, is the attenuation rate, t is the current time, is the time of the previous interaction.

7. The power data analysis device based on graph computing is characterized in that, For implementing the graph-computing-based power data analysis method according to any one of claims 1-6, the device includes: A structure diagram construction module for constructing a power network structure diagram, where nodes represent entities in the power system, including power plants, substations, and transmission lines, and edges represent the logical connection relationships between entities. A feature extraction module for extracting features from the nodes and edges in the power network structure diagram to capture the mutual influence relationships and interactive connection relationships between entities in the power system. A fusion calculation module for performing multi-modal graph fusion calculations based on the mutual influence relationships and interactive connection relationships, using a deep learning architecture to process the high-dimensional data and heterogeneous data in the power system, and collecting the first deep dynamic evolution set and the second deep dynamic evolution set. A dynamic change analysis module for starting from the power grid operation state, analyzing the dynamic changes of the power system in the time dimension and space dimension through the first deep dynamic evolution set and the second deep dynamic evolution set, and extracting the perception prediction trend. A scheduling decision-making module, which is used to deploy a reinforcement learning algorithm based on the power grid structure diagram, perform adaptive policy optimization learning according to the real-time updated power grid operation status in combination with the perceived prediction trend, and match the optimal power scheduling decision with the goal of minimizing energy loss.

Images