Online evaluation method and device for reliability of power distribution network

Abstract

The invention relates to the technical field of power distribution network reliability evaluation, in particular to a power distribution network reliability online evaluation method and device. The method comprises the following steps: establishing a state transition matrix of each element in the power distribution network according to a physical topological structure of the power distribution network, and sampling to obtain a full life cycle state matrix of the elements of the power distribution network; according to the obtained state matrix, power distribution network reliability evaluation considering vehicle network integration grading and source load volatility is carried out, and a first reliability evaluation index is calculated; according to the obtained system state matrix and the first reliability evaluation index, simulating a source load fluctuation scene to construct a data set, and training a graph neural network model; according to the trained target graph neural network model, the load values of the nodes and the real-time state data of the power distribution network are input, and the predicted reliability value of each node is obtained. According to the method, the problems that nonlinearity and dynamic changes in a power distribution network system are difficult to accurately reflect in the prior art and the operation complexity and time are difficult to meet the requirements can be solved.

Description

Technical Field

[0001] This invention relates to the field of distribution network reliability assessment technology, and in particular to an online distribution network reliability assessment method and apparatus. Background Technology

[0002] With the rapid development of distributed energy, electric vehicles (EVs), and user-side response capabilities, the operating environment of power distribution networks is becoming increasingly complex, exhibiting characteristics such as enhanced bidirectional fluctuations between power sources and loads, variable operating states, and high uncertainty. Especially in the context of vehicle-to-grid integration, the large-scale integration of electric vehicles not only causes load fluctuations but also brings randomness to charging and discharging behavior, posing new challenges to the reliability assessment of power distribution networks.

[0003] Currently, distribution network reliability assessment methods mainly fall into two categories: analytical methods and simulation methods. Among them, analytical methods, based on state-space modeling, can quickly obtain reliability indicators, while simulation techniques based on Monte Carlo methods can improve the accuracy of distribution network reliability assessment.

[0004] However, analytical methods are difficult to accurately reflect the nonlinearity and dynamic changes in the distribution network system; although simulation technology based on Monte Carlo methods can improve accuracy, it has high computational complexity and long computation time under high-dimensional uncertainty input, making it difficult to meet the real-time requirements of online evaluation. Summary of the Invention

[0005] This invention provides a method and apparatus for online reliability assessment of power distribution networks, which addresses the problems of existing technologies failing to accurately reflect the nonlinearity and dynamic changes in power distribution network systems, as well as the difficulty in meeting computational complexity and time requirements.

[0006] In a first aspect, embodiments of the present invention provide a method for online reliability assessment of a power distribution network, comprising: Based on the physical topology of the distribution network, establish the state transition matrix of each component in the distribution network; The state changes of each element in the state transition matrix are sampled to obtain the full life cycle state matrix of the distribution network element. Based on the full life cycle state matrix of the distribution network components, a reliability assessment of the distribution network considering vehicle-to-grid integration and source-load fluctuation is performed, and the first reliability evaluation index is calculated. Obtain the system state matrix, and based on the system state matrix and the first reliability evaluation index, simulate the source load fluctuation scenario and construct a dataset, and use the dataset to train a graph neural network model; Based on the trained target graph neural network model, the load values ​​of the input nodes and the real-time status data of the distribution network are used to obtain the predicted reliability values ​​of each node.

[0007] Secondly, embodiments of the present invention provide an online reliability assessment device for a power distribution network, comprising: The state matrix construction module is used to establish the state transition matrix of each element in the distribution network based on the physical topology of the distribution network. The state matrix construction module is also used to sample the state changes of each element in the state transition matrix to obtain the full life cycle state matrix of the distribution network element. The calculation module is used to perform a reliability assessment of the distribution network, taking into account vehicle-to-grid integration and source-load fluctuations, based on the full life-cycle state matrix of the distribution network components, and to calculate the first reliability evaluation index. The model training module is used to obtain the system state matrix, and based on the system state matrix and the first reliability evaluation index, simulate the source load fluctuation scenario and construct a dataset, and use the dataset to train a graph neural network model; The prediction module is used to obtain the predicted reliability value of each node based on the load value of the input node and the real-time status data of the distribution network, according to the trained target graph neural network model.

[0008] This invention provides a method and apparatus for online reliability assessment of a distribution network. The method involves establishing a state transition matrix for each component in the distribution network based on its physical topology; sampling the state changes of each component in the state transition matrix to obtain a full lifecycle state matrix for the distribution network components; performing a distribution network reliability assessment considering vehicle-to-grid integration and source-load fluctuation based on the obtained full lifecycle state matrix, and calculating a first reliability evaluation index; obtaining a system state matrix; simulating source-load fluctuation scenarios and constructing a dataset based on the system state matrix and the first reliability evaluation index; and training a graph neural network model using the dataset; and obtaining the predicted reliability values ​​for each node based on the trained target graph neural network model, inputting the load values ​​of the nodes and real-time distribution network status data. This embodiment constructs a full lifecycle state matrix for distribution network components to simulate the full lifecycle states of these components. This accurately reflects the nonlinearity and dynamic changes in the distribution network system, enhances the ability to model complex stochastic processes, and ensures simulation accuracy under high-dimensional uncertain input conditions. Furthermore, this embodiment achieves significantly faster evaluation while retaining nonlinear modeling capabilities through collaborative training and fusion inference of graph neural network models and Markov Chain Monte Carlo (MCMC) samples. It possesses online rapid evaluation capabilities and is suitable for real-time operational decision-making scenarios. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the implementation of the online reliability assessment method for power distribution networks provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the element state transition provided in an embodiment of the present invention; Figure 3 This is a typical daily curve diagram of a charging and discharging station provided in an embodiment of the present invention; Figure 4 This is a typical daily curve diagram of a battery swapping station provided in an embodiment of the present invention; Figure 5 This is a flowchart illustrating the implementation of the training graph neural network model provided in this embodiment of the invention. Figure 6 This is an improved IEEE-RBTS Bus6 F4 feeder diagram provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of the online reliability assessment device for power distribution networks provided in an embodiment of the present invention. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0012] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0013] Figure 1 The implementation flowchart of the online reliability assessment method for power distribution networks provided in this embodiment of the invention is described in detail below: Step 101: Based on the physical topology of the distribution network, establish the state transition matrix of each component in the distribution network.

[0014] This step may include: determining the connection relationship of each component based on the physical topology of the distribution network; and establishing the Markov chain state transition matrix of each component based on the connection relationship of each component.

[0015] Optionally, based on the connection relationships of each element, the Markov chain state transition matrix of each element is established, including: Set the full life cycle operating status of distribution network components and lines as follows: ,in, Represents the runtime state space. Indicates the first time, It is a positive integer greater than or equal to 1. Represents the set of times; for any time... any state , State sequence ,but ; That is, the component is in the next moment. The conditional probability depends only on the current state. Related to, and related to the previous state This property is irrelevant; it is the absence of aftereffects and memorylessness of Markov processes. Therefore, distribution network components have two states: normal operation and fault maintenance. Based on the basic theory of Markov processes, the component state space and state transition probabilities can be established, thereby obtaining a full life-cycle operating state model of the distribution network components.

[0016] Status of power distribution network components It can be divided into express The component is always in working condition. express Since the component is in a fault repair state at any given time, the Markov process for each component can be represented by two states and four state transitions, such as... Figure 2 As shown. Specifically, the transition probability from state 0 to state 1. The failure rate of the corresponding component, and the transition probability of transitioning from state 1 to state 0. To determine the repair rate of the corresponding component, the state transition probability of the component is defined as follows: ; Among them, ignoring The probability of two or more transitions occurring during this period is used, so an approximation sign is adopted; Indicates the first The state transition probability of each component Indicates time interval, The failure rate of the corresponding component refers to the transition probability from state 0 to state 1. The repair rate of the corresponding component refers to the transition probability from state 1 to state 0. Indicates the first Each component in the time interval The probability of transitioning from working state 0 to fault repair state 1. Indicates the first Each component in the time interval The probability of remaining in fault repair state 1 within the system. Indicates the first Each component in the time interval The probability of remaining in a working state of 0 within the specified timeframe. Indicates the first Each component in the time interval Within this context, the probability of transitioning from fault repair state 1 to working state 0 is [not specified]. Based on the state transition probabilities of the element, the state transition matrix of the element is obtained as follows: ; in, This represents the state transition matrix.

[0017] Step 102: Sample the state changes of each element in the state transition matrix to obtain the full life cycle state matrix of the distribution network element.

[0018] In one embodiment, sampling the state changes of each element in the state transition matrix to obtain the full lifecycle state matrix of the distribution network elements may include: make , Indicates the first Each component Always in a state of readiness The probability, ; According to the law of total probability, we have: ; when At that time, the component is The probability of being in state 1 at any given time is: ; when At that time, the component is The probability of being in state 1 at any given time is: ; That is, derived from the law of total probability: ; The probability of the component state is obtained using the law of total probability: ; In the power distribution network, a Markov two-state process is used for modeling. It is assumed that each component exists in only two states: operating and faulty, and that other equipment in the system is continuously considered to be in normal operating condition during the analysis. When a component is in the operating state at the initial moment, its state at any given moment... The downtime probability can be given by the above formula. Given that, under normal circumstances, the time step corresponding to the operating cycle (set to 1 hour) is much smaller than the component's repair time, the equipment can be considered unrepairable within a single time step; that is, if the repair rate... The component downtime probability model is obtained as follows: ; The Markov chain Monte Carlo simulation method is used to sample the state change behavior of each component, and the results are expressed as the state of the distribution network components throughout their entire life cycle.

[0019] The main consideration is the state changes of each line component. It is assumed that the distributed power supply, electric vehicle charging and discharging station, and other equipment are all in normal condition. Random numbers are generated in a uniform distribution of 0-1. This is used to determine whether a state transition should be received. When the simulation time reaches the preset value, the full life cycle state matrix of the distribution network system components is obtained: ,in This represents the total number of components. in, Indicates the first sky Timing element The state.

[0020] This invention uses the Markov chain Monte Carlo method to obtain the full life cycle state matrix of distribution network components, which has the following advantages compared to the random sampling of the sequential Monte Carlo method: The sequential Monte Carlo method simulates the operation of components at each time step by independently and randomly sampling component states based on a fixed failure rate. It is suitable for simulating multiple independent faults in a distribution network over a certain period to evaluate system reliability. However, because the sequential Monte Carlo method does not consider the relationship between the current and historical states of components during state sampling, it is difficult to accurately reflect the dynamic evolution characteristics of component degradation, aging, and fault repair during service. In contrast, the Markov chain Monte Carlo method introduces a state transition probability matrix during sampling, incorporating the transition relationships between different component states into the modeling system. This method can determine the component's state at the next time step based on its current state and a preset state transition probability. This characteristic enables the Markov chain Monte Carlo method to effectively describe the state dependence of components during their service life and their degradation trends over time, demonstrating strong dynamic evolution modeling capabilities. Especially in the scenario of full life cycle modeling, the state of components usually presents an evolution process of "deterioration-failure-repair". It is difficult to reflect this gradual deterioration and cumulative failure characteristics by simply relying on the sequential Monte Carlo method, while the Markov chain Monte Carlo method can realistically reproduce this process by reasonably setting the transition probability matrix.

[0021] Furthermore, the Markov chain Monte Carlo method demonstrates greater flexibility and scalability in multi-state, multi-stage reliability modeling. By adjusting the number of states and transition probabilities, it can be easily extended to multi-state components or consider the dynamic correction of transition probabilities based on service life, making the model more closely reflect actual operating conditions. Because the Markov chain method involves probabilistic constraints during state evolution, it is less dependent on the number of samples compared to the sequential Monte Carlo method, resulting in lower volatility and greater stability in simulation results. Therefore, the Markov chain Monte Carlo method shows significantly superior application value compared to the sequential Monte Carlo method in areas such as full life-cycle reliability assessment of distribution network components, optimization of operation and maintenance strategies, and state prediction analysis.

[0022] Step 103: Based on the full life cycle state matrix of distribution network components, conduct a distribution network reliability assessment that takes into account vehicle-to-grid integration and source-load fluctuation, and calculate the first reliability evaluation index.

[0023] In one embodiment, based on the full lifecycle state matrix of distribution network components, a distribution network reliability assessment considering vehicle-to-grid integration and source-load fluctuations is performed, and a first reliability evaluation index is calculated, which may include: Based on historical data, calculate the load value, distributed photovoltaic output power, and wind power output power; Based on the full life cycle state matrix of distribution network components, the fault data of each node is calculated, including fault time, number of faults and load loss. Using node network analysis, the recoverable load of distributed power sources is calculated; The first reliability evaluation index is calculated based on fault data and recoverable load.

[0024] Optionally, calculating the load value, distributed photovoltaic output power, and wind power output power based on historical data may include: The vehicle-to-everything (V2X) integrated interaction scenarios are classified into different levels, and the objective functions of different levels are quantitatively analyzed. Based on the objective functions corresponding to different levels, the load data of the charging and discharging station is obtained; Calculate the load value based on the load data; Calculate the output power of distributed photovoltaic power based on historical distributed photovoltaic data; Calculate the wind power output based on historical wind power data.

[0025] Optionally, close collaboration among regulatory authorities, manufacturers, grid companies, and users is crucial in promoting the integration and interaction of charging stations with the vehicle-grid network. The specific classification of vehicle-grid integration and interaction scenarios and the quantitative analysis of the objective functions at different levels are shown in Table 1 below.

[0026] Table 1

[0027] Based on the levels and scenarios described in the classification table, the objective functions at different levels are quantitatively analyzed.

[0028] The formula for the objective function at level L0 is: ; In the formula: Decisions based on the user's power consumption; The benefits users gain from electricity use, such as electricity comfort and production revenue; Electricity costs paid on behalf of users.

[0029] Level L1: Based on user self-interest, this level introduces operator (e.g., electricity retailer, virtual power plant) regulation to balance operational revenue and user welfare. The formula for the objective function of Level L1 is shown below: ; In the formula, To help operators adjust power consumption or pricing strategies; This is the operator's revenue function, where the operator can influence users' electricity consumption decisions by adjusting prices, aggregating loads, and implementing demand response.

[0030] Level L2: Building upon the previous level, this level incorporates the power grid company, aiming to achieve a balance between regional power grid security, economic efficiency, and benefits for users and operators. The formula for the objective function at Level L2 is shown below: ; In the formula: These are control variables for power grid companies, such as electricity prices, power limits, and power purchase and sale plans. The costs include power grid operating costs, transmission losses, and penalty fees; it is necessary to meet the power flow balance constraints and safety and stability constraints of the distribution network.

[0031] Level 3: Through dynamic game theory and information exchange among the three parties, the overall utility is maximized, reflecting full interaction and multi-objective collaboration. The formula for the objective function at Level 3 is shown below: ; In the formula: For operator revenue; For the overall benefits of power grid companies (such as power supply reliability, power quality, and operational economy). , , These are the multi-objective weighting coefficients, reflecting the intensity of the interests and demands of each entity.

[0032] Based on the formula for the objective function and its constraints, the typical daily curve of the charging and discharging station is obtained as follows: Figure 3 As shown.

[0033] The direct charging mode of the battery swapping station means that the swapping batteries can be charged in a timely manner regardless of the load status of the distribution network. Peak-shifting mode refers to the practice where, during peak load periods, the CBSS batteries are not charged, but only charged during normal and low load periods to alleviate pressure on the distribution network and prevent further increases in peak load. A typical daily load curve for the battery swapping station is obtained based on historical data, as shown below. Figure 4 As shown. To address the issue that existing assessment methods are unable to accurately reflect the impact of the hierarchical structure of vehicle-to-grid integration, this embodiment establishes a hierarchical modeling framework for electric vehicle access, effectively characterizing load behavior and uncertainty features under different access levels, thereby improving the structured expression capability of reliability assessment.

[0034] Based on the aforementioned vehicle-to-grid (V2G) integration and interaction scenario classification, and considering the impact of wind and solar power processing volatility on the reliability of the new power distribution network, the hourly load data of each node in the system can be calculated using weekly load data (expressed as a percentage of the annual peak load), daily load data for any given week (expressed as a percentage of the weekly peak load), and hourly load data (expressed as a percentage of the daily peak load). The formula is as follows: ; in, Represents a node The hourly load value, Represents a node The annual peak load, This represents weekly load data. This represents daily load data. This represents hourly load data.

[0035] Distributed photovoltaic (PV) panels and distributed wind turbines are installed at designated nodes to supply part of the load to nearby nodes. When a distribution network component fails, causing a line outage, nodes containing distributed power sources form islands, with the distributed power sources replacing the bus to supply load and restore power to the faulty nodes, thereby improving the availability and reliability of the distribution network. However, distributed energy output is easily affected by weather and environmental factors, so the impact of the fluctuations in wind and solar power output on the reliability of the distribution network needs to be considered. Furthermore, considering the significant differences in wind and solar power output across different seasons and times, it is necessary to analyze the characteristics of wind and solar power output and obtain typical daily curves to analyze their impact on distribution network reliability.

[0036] Wind power output is affected not only by its own characteristics but also by the output characteristics of the wind turbine. Therefore, the calculation method for wind power output is as follows: ; In the formula: Indicates the output power of wind power; This indicates the operating status of the wind power system; 0 indicates normal operation and 1 indicates a fault. The output power is determined by the output characteristics. The calculation method is as follows: ; In the formula: This refers to the rated output power of the fan. To cut in wind speed; Rated wind speed; To cut off the wind speed; , , These are the fitting coefficients for the three polynomials.

[0037] A photovoltaic array consists of multiple photovoltaic units, and its output power depends on its own operating status and photovoltaic output characteristics. The calculation method for the output power of distributed photovoltaic systems is as follows: ; In the formula: Indicates photovoltaic output power; This indicates the operating status of the photovoltaic module; 0 indicates normal operation, and 1 indicates a fault. The output power is determined by the output characteristics. according to Calculated; where: Light irradiance; The area of ​​the photovoltaic panel; This refers to the conversion efficiency of the photovoltaic unit.

[0038] The fault data for each node is calculated below.

[0039] The year is divided into hourly time intervals to form a time series. Based on historical data from the same period, wind power output, photovoltaic power output, load information, and charging / swapping station power are determined for each time interval. The operating status of each element in each time interval is determined using the full life-cycle state matrix of distribution network elements obtained by the Markov chain Monte Carlo method.

[0040] Reliability is defined as the level of reliability of components, nodes, and the system under the assumption that the distribution system maintains the current time interval for weather and load data throughout the year, and that the line status and output of renewable distributed power sources remain constant. Distributed power sources include distributed photovoltaic (PV) and wind power. The integration of distributed power sources can improve the reliability level of the corresponding nodes. When a component on the main power supply path of a node fails, it can switch to a backup power supply path powered by a distributed power source. The need for load shedding and the amount of load shedding are determined based on the output power and load value of the distributed power source.

[0041] This embodiment uses node analysis to determine the islanding range formed by distributed power sources when the main power supply path component fails, and to restore the power-outage load. Specifically, when the main power supply path of a node fails, if the backup power supply path also fails, the node's power outage time is the repair time of the failed component; if the output of the distributed power source on the backup power supply path is less than the node's load, the node will shelve its load, resulting in a partial power outage; if the output of the distributed power source on the backup power supply path is greater than the node's load, the node's power outage time is the power supply path switching time.

[0042] The formulas for calculating the fault time of nodes in the backup power supply path that are distributed renewable energy sources at various time intervals are as follows: according to Calculate the failure time of each node; in, Represents a node downtime, Represents a node The failure rate of the main power supply path Represents a node The probability of failure of the backup power supply path; Indicates the repair time of the faulty component; Represents a node The load shedding status is represented by 1, where 0 indicates no load shedding. This indicates the output of distributed energy resources on the backup power supply path; Indicates the switching action time. Represents a node The load.

[0043] Specifically, when some nodes on the backup power supply path are not distributed power source installation nodes, The remaining output of the distributed power source after supplying the upstream nodes of the evaluation node.

[0044] In one embodiment, the first reliability evaluation index includes the System Average Interruption Frequency Index (SAIFI), the System Average Interruption Duration Index (SAIDI), the Expected Energy Not Supplied (EENS), and the Average Service Availability Index (ASAI). This first reliability evaluation index is a reliability index of the distribution network calculated based on the full lifecycle state sequence of actual components, combined with historical / statistical source-load fluctuation data, under a real simulated operating scenario, and is used to preliminarily assess the reliability level of the distribution network.

[0045] SAIFI represents the system average outage frequency index for load reduction per unit time, measured in times per user. The formula for calculating the year is: ; in This represents the system's average interrupt frequency index. Total number of users It is a node Number of power outages.

[0046] SAIDI represents the average system outage duration index per user per year due to power failures, expressed in hours per user. The formula for calculating the year is: ; in This represents the average downtime index of the system. Represents a node The power outage time, This represents the total number of power outages.

[0047] EENS represents the total electrical energy loss caused by a power outage, i.e., the expected unsupplied energy, used to measure the impact of a power outage on the system's power supply capacity. The calculation formula is: ; in, For the first The load capacity (MW) affected by the power outage. For the first Duration of the power outage.

[0048] ASAI is the Average Service Availability Index, which is the ratio of a user's actual total power supply time to the power supply time required per unit of time. Its calculation formula is: ; in For the first The number of users affected by this power outage. This refers to the total time of the year.

[0049] Step 104: Based on the full life cycle state matrix of distribution network components and the first reliability evaluation index, simulate the source-load fluctuation scenario and construct a dataset, and use the dataset to train the graph neural network model.

[0050] A power distribution network is essentially composed of nodes such as substations, switching stations, feeders, and users, which are interconnected through transmission lines, forming a complex graph network with tightly coupled relationships. Traditional reliability assessment methods based on mathematical statistics or physical models are insufficient to fully characterize the dynamic correlations and complex topological characteristics between nodes, especially when dealing with multi-source heterogeneous data and the volatility of distributed power sources, easily losing the coupling characteristics between local correlations and global states.

[0051] Graph Neural Networks (GNNs), as a deep learning method capable of directly modeling graph-structured data, naturally preserve the topological information of distribution network nodes and lines. Through the information transmission and aggregation mechanism of neighbor node states, they dynamically capture the impact of local structural changes on the overall network reliability. They can integrate multi-dimensional heterogeneous state variables such as real-time load and distributed power output to achieve globally consistent feature extraction and temporal evolution modeling. They support online updates of model parameters or incremental learning to adapt to network structure changes such as renewable energy access and equipment replacement, ensuring the real-time performance and reliability of the evaluation model. Therefore, the online reliability evaluation method for distribution networks based on graph neural networks can fully utilize the graph structure characteristics and real-time operating status data of the distribution network, improving the dynamic response capability and prediction accuracy of reliability evaluation. Thus, this embodiment uses a graph neural network model to perform online reliability evaluation of the distribution network.

[0052] Before using a graph neural network model to conduct online assessment of the reliability of the distribution network, it is necessary to first construct a dataset and train the graph neural network model.

[0053] Considering that the operating status of the distribution network is affected by multiple factors, including fluctuations in distributed generation (DG) output, changes in node load, and fault conditions, and that these conditions are highly random and involve complex combinations of scenarios, making it difficult to fully cover them with actual monitoring data, this embodiment employs the Markov Chain Monte Carlo (MCMC) method to simulate and sample the operating status of distribution network components, simulating the changes in distributed generation output levels and load demands at each node under different operating conditions. Figure 5 As shown, based on the system state matrix and the first reliability evaluation index, simulating source load fluctuation scenarios and constructing a dataset, then training a graph neural network model using the dataset, may include the following steps: Step 501: Construct a fault model for the operation of the distribution network based on the system state matrix.

[0054] Optionally, the system state matrix in this step can be obtained in the same way as the distribution network component lifecycle state matrix in step 102. That is, it is generated by sampling using the temporal Monte Carlo method based on the state transition matrix of each component's Markov chain. However, the purpose of constructing the system state matrix in this step is not to restore the true state of a single full lifecycle, but to construct a fault model for distribution network operation, adapting to subsequent "diverse source-load fluctuation scenarios," such as extreme loads and sudden changes in distributed power output, providing a "fault state basis" for calculating reliability indicators under different scenarios. Therefore, when sampling the state changes of each component in the state transition matrix, it is not necessary to strictly follow the full lifecycle time series, but rather to aim at "covering the possibility of faults under different source-load scenarios."

[0055] Step 502: Random sampling is performed using a grid sampling method to generate a power distribution network state scenario that conforms to the actual operating rules.

[0056] Step 503: In each scenario of the distribution network status scenario, calculate the second reliability evaluation index based on the physical topology and fault model.

[0057] Optionally, the second reliability evaluation index can be calculated in the same way as the first reliability evaluation index in step 103.

[0058] Step 504: Combine the scenario status data in the power distribution network status scenario with the corresponding second reliability evaluation index to form sample data.

[0059] Scene status data can include node voltage, load, DG output power, and line status, etc.

[0060] This method effectively avoids the limitations of relying on a single historical observation data sample, and can comprehensively cover the operating status of the distribution network under different operating conditions, thereby improving the model's generalization ability and adaptability.

[0061] Step 505: Model the graph neural network based on the physical topology to obtain the graph neural network model.

[0062] The node matrix, edge matrix, and adjacency matrix can be determined based on the graph neural network model. Among them, the node feature matrix , The number of nodes in the physical topology of the distribution network. Node characteristics include whether it is under load, node load value, number of users, etc. Edge feature matrix , For the number of lines, These are edge features, including line status, switch type, line length, failure rate, and repair time; Adjacency matrix This is used to confirm the connection relationships between nodes. Represents a node There must be a connection between them, otherwise .

[0063] Step 506: Train the graph neural network model using the training set, and verify the accuracy of the trained graph neural network model using the first reliability evaluation index. If the accuracy does not meet the accuracy threshold, adjust the parameters of the graph neural network model until the accuracy meets the threshold and end the training to obtain the target graph neural network model.

[0064] Multi-layer feature aggregation can be achieved through Graph Convolutional Network (GCN) or Graph Attention Network (GAT).

[0065] In each layer, the model is based on the adjacency matrix. Given a specified topological connection relationship, the features of neighboring nodes are weighted and aggregated with the node's own features to extract the high-order structural characteristics of the node and its neighborhood. During the training of the graph neural network model, the feature update process equation is: ; in, Indicates the updated number The node feature matrix of the layer This represents the activation function (such as ReLU, LeakyReLU). This represents the adjacency matrix after adding self-loops and normalizing. Indicates the number before the update The node feature matrix of the layer This represents a trainable weight matrix.

[0066] By using multi-layer graph convolution operations, the local and global structural information of nodes is gradually fused, effectively capturing the complex topological relationships, multi-factor coupling, and abnormal state propagation characteristics of the power distribution network.

[0067] After completing the multi-layer feature extraction of the graph neural network model, global pooling is used to embed the nodes of the entire graph into the representation. The features are aggregated into a unified feature vector. Subsequently, through a multilayer perceptron (MLP) or regression output layer, regression predictions of distribution network reliability indicators (such as SAIFI, SAIDI, and node power supply reliability) are achieved. The prediction function can be expressed as: ;in, This represents the predicted reliability assessment value. This indicates a fully connected regression network.

[0068] Determine whether the upper limit of the number of iterations has been reached or the loss function has converged. If so, save the trained model; otherwise, return to continue training.

[0069] The accuracy of the trained graph neural network model is verified by using the first reliability evaluation index. The first reliability evaluation index can be compared with the index output by the trained graph neural network model to verify the accuracy of the graph neural network model. If the accuracy does not meet the standard, the parameters of the graph neural network model need to be retrained.

[0070] Step 105: Based on the trained target graph neural network model, input the load values ​​of the nodes and the real-time status data of the distribution network to obtain the predicted reliability values ​​of each node.

[0071] By applying the trained graph neural network model, the load value of the load point and the output value of the distributed power source are input, and the reliability of each load point can be evaluated online and vulnerable components can be identified, which can help stabilize the operation of the distribution network and provide key protection for users with low reliability indicators.

[0072] With the large-scale integration of distributed power sources (such as distributed photovoltaic and wind power) and the increasing complexity of load characteristics, the operating status of distribution networks exhibits high dynamism and randomness. Traditional reliability assessment methods mostly rely on historical data and static network models, which are insufficient to reflect the network health status in real-time operating environments. This is especially true in scenarios involving fluctuations in renewable energy output, load surges, and sudden faults, where response lags and assessment accuracy are inadequate. This assessment lag leads to a decline in the ability to predict fault risks, affecting the scientific and timely nature of distribution network dispatching decisions, and consequently impacting power supply reliability and user experience.

[0073] Therefore, establishing an online reliability assessment mechanism oriented towards the real-time operating status of distribution networks can dynamically sense the real-time changes in equipment operating status and power supply and load, predict potential fault risks in advance, assist in adjusting dispatch strategies, and ensure the safe, economical, and reliable operation of the system. Especially in modern distribution networks where the proportion of renewable energy is constantly increasing, online assessment has become an important means to support the coordinated interaction of power sources, grids, and loads and proactive operation and maintenance decisions, possessing significant engineering application value and economic benefits.

[0074] As a crucial aspect of distribution network reliability assessment in distribution system planning and operation management, traditional methods primarily rely on probabilistic statistical methods, sequential Monte Carlo simulations, and state transition models. These methods calculate system-level reliability indices such as SAIFI, SAIDI, ENS, and ASAI by generating component failure probabilities, state transition matrices, and a large number of random samples. While these methods are well-suited for static or offline conditions, they suffer from drawbacks due to the complexity of distribution network structures, rapid dynamic evolution of equipment states, and strong randomness and correlation in source-load characteristics. These traditional methods suffer from high computational overhead, poor real-time performance, and difficulty in fully exploring the correlation between complex network structures and node states.

[0075] Graph Neural Networks (GNNs), a class of deep learning methods that have emerged in recent years, can learn high-dimensional nonlinear features of nodes, edges, and their topological relationships based on graph-structured data. They propagate and aggregate state information of neighboring nodes layer by layer, uncovering hidden high-order dependencies in complex networks. A power distribution network can essentially be abstracted as an undirected or directed graph with a complex topology. Nodes represent buses and load points, while edges represent feeders and switching equipment. Node and edge attributes can correspond to information such as real-time component status, power, and failure probability. GNNs can dynamically perceive the impact of system structural changes and local disturbances on global reliability indicators based on the current power distribution network topology and real-time component status. Through end-to-end learning, they can achieve efficient online prediction of reliability indicators.

[0076] Compared to traditional Monte Carlo simulation methods, the Generative Neural Network (GNN) method offers significant advantages in reliability assessment. First, GNNs can fully leverage the complex topology of distribution networks and the correlation between node neighborhoods to achieve global modeling of the impact of factors such as local faults, islanding, and equipment degradation on the overall network reliability, without relying on the extraction of a large number of independent samples, thus greatly reducing computational overhead. Second, GNNs possess excellent end-to-end generalization capabilities, enabling them to quickly respond to system state changes under limited sample training conditions and adapt to the reliability assessment needs of different operating scenarios, exhibiting high real-time performance and dynamic adaptability. Furthermore, the GNN method can integrate historical operating data, environmental factors, component status monitoring information, and real-time topology to construct a multi-dimensional feature-driven reliability prediction model, demonstrating higher prediction accuracy and robustness compared to traditional methods based on static probability parameters or state transition matrices.

[0077] The following verifies the online distribution network reliability assessment method provided in this embodiment. To verify the effectiveness of the online distribution network reliability assessment method proposed in this invention, an F4 feeder under the improved IEEE-RBTS Bus6 test system is used. The system structure diagram is shown below. Figure 6 As shown, wind power and photovoltaic power generation systems are connected at nodes 27 and 13 respectively, and charging stations and battery swapping stations of different vehicle-to-grid integration levels are connected at nodes 28 and 15 respectively.

[0078] The reliability parameters of the components are shown in the table below:

[0079] Option 1: Connect wind power and distributed photovoltaic power generation systems at nodes 27 and 13 respectively, and connect charging stations and battery swapping stations of different vehicle-to-grid integration levels at nodes 28 and 15 respectively. A typical daily curve is shown below. Figure 3 and Figure 4 As shown. From Figure 3 It can be seen that the initial stage of EV load changes and the charging pattern of residents during travel gradually increases after 7:00 AM, peaks at 12:00 PM, and then gradually decreases, showing the same trend after 5:00 PM when people leave work. The social altruistic stage sees users actively utilizing time-of-use pricing policies to avoid peak loads, resulting in lower peak EV charging loads than the initial stage and better off-peak loads, exhibiting peak-shaving and valley-filling characteristics. The regional altruistic stage employs orderly charging control, accepting dynamic adjustments to charging power to further reduce the peak-valley difference. The fully interactive stage, building on the previous stage, has bidirectional flow characteristics, allowing it to discharge to the grid according to demand and restore power to interrupted loads. Figure 4The results show that direct charging can exacerbate grid load during evening peak hours, posing a risk of overload. Peak shaving and valley filling, through demand-side management strategies, shifts charging load to nighttime or low-load periods, balancing the load curve throughout the day, reducing the peak-valley difference, and improving the economy and safety of grid operation.

[0080] The impact on the reliability of the distribution network needs to be evaluated through reliability index calculation. At the same time, in order to verify the accuracy of the Markov chain Monte Carlo simulation method established in this invention, an analytical model based on the minimum path analysis method is used as a comparison method. When the model performance index reaches the predetermined standard, that is, the error between the reliability evaluation index and the analytical model result does not exceed 2%, it indicates that the model meets the application requirements and can provide data support for online identification.

[0081] Option 2: Maintaining the system node model described in Option 1, the system load fluctuation is set within the range of [0.1, 1], wind and solar power fluctuations within the range of [0, 1], and charging / swapping station load fluctuations within the range of [0.2, 0.8]. The step size is set to 0.1. Using a network search method, 8470 sets of source-load data combination samples were retrieved. After dividing the dataset into a 6:2:2 ratio, the graph neural network method proposed earlier was used to input the above data samples for offline training. Three different source-load fluctuation scenarios not covered in the samples were used to test the model. Markov chain Monte Carlo simulation and an analytical model based on minimum path analysis were used as comparison methods.

[0082] After verification under different source-load fluctuation scenarios, when the model's performance indicators meet predetermined standards—for example, the error between the reliability assessment indicator and the Markov chain Monte Carlo simulation result does not exceed 2%, and the assessment time does not exceed that of the analytical method—it is indicated that the model meets the application requirements and can be used for online reliability assessment of real-time data. Specific evaluation indicators are shown in Table 2 below: Table 2

[0083] This invention provides an online reliability assessment method for distribution networks. The method involves establishing a state transition matrix for each component in the distribution network based on its physical topology. Then, it samples the state changes of each component in the state transition matrix to obtain a full lifecycle state matrix for the distribution network components. Based on this full lifecycle state matrix, a distribution network reliability assessment considering vehicle-to-grid integration and source-load fluctuations is performed, and a first reliability evaluation index is calculated. A system state matrix is ​​obtained, and based on the system state matrix and the first reliability evaluation index, a source-load fluctuation scenario is simulated and a dataset is constructed. A graph neural network model is trained using this dataset. Based on the trained target graph neural network model, the load values ​​of the nodes and real-time distribution network status data are input to obtain the predicted reliability values ​​for each node. This embodiment effectively integrates topology connectivity, electrical parameters, and operational status data by constructing a state feature extraction and reliability index prediction based on a graph neural network model. This achieves deep modeling of the interdependencies between distribution network nodes, improving the accuracy and generalization ability of the prediction results.

[0084] To address the issues of low evaluation efficiency and difficulty in online application of traditional Monte Carlo simulation-based methods, this invention achieves significantly faster evaluation while retaining nonlinear modeling capabilities through collaborative training and fusion inference of samples generated by GNN models and Markov chain Monte Carlo methods. This enables rapid online evaluation and is suitable for real-time decision-making scenarios.

[0085] In this embodiment, to address the problem of high system state uncertainty caused by enhanced source load volatility, the present invention introduces the Markov chain Monte Carlo method to jointly sample and simulate multi-source load disturbances, thereby improving the ability to model complex stochastic processes and ensuring simulation accuracy under high-dimensional uncertain input conditions.

[0086] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0087] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above. Figure 7 A schematic diagram of an online reliability assessment device for a power distribution network provided by an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 7 As shown, the online reliability assessment device 7 for power distribution networks includes: a state matrix construction module 71, a calculation module 72, a model training module 73, and a prediction module 74.

[0088] The state matrix construction module 71 is used to establish the state transition matrix of each element in the distribution network according to the physical topology of the distribution network. The state matrix construction module 71 is also used to sample the state changes of each element in the state transition matrix to obtain the full life cycle state matrix of the distribution network element. Calculation module 72 is used to perform a reliability assessment of the distribution network taking into account vehicle-to-grid integration and source-load fluctuation based on the full life cycle state matrix of distribution network components, and to calculate the first reliability evaluation index. The model training module 73 is used to obtain the system state matrix, and based on the system state matrix and the first reliability evaluation index, simulate the source load fluctuation scenario and construct a dataset, and use the dataset to train the graph neural network model; The prediction module 74 is used to obtain the predicted reliability value of each node based on the load value of the input node and the real-time status data of the distribution network, according to the target graph neural network model that has been trained.

[0089] In one possible implementation, when the state matrix construction module 71 establishes the state transition matrix of each element in the distribution network based on the physical topology of the distribution network, it is used for: Determine the connection relationships of each component based on the physical topology of the power distribution network; Based on the connection relationships of each component, establish the Markov chain state transition matrix for each component.

[0090] In one possible implementation, when the state matrix construction module 71 establishes the Markov chain state transition matrix of each element based on the connection relationship between the elements, it is used for: Set the full life cycle operating status of distribution network components and lines as follows: ,in, Represents the runtime state space. Indicates the first time, It is a positive integer greater than or equal to 1. Represents the set of times; express The component is always in working condition. express If a component is in a fault repair state at any given time, then the state transition probability of the component is defined as: ; in, Indicates the first The state transition probability of each component Indicates time interval, The failure rate of the corresponding component refers to the transition probability from state 0 to state 1. The repair rate of the corresponding component refers to the transition probability from state 1 to state 0. Indicates the first Each component in the time interval The probability of transitioning from working state 0 to fault repair state 1. Indicates the first Each component in the time interval The probability of remaining in fault repair state 1 within the system. Indicates the first Each component in the time interval The probability of remaining in a working state of 0 within the specified timeframe. Indicates the first Each component in the time interval Within this context, the probability of transitioning from fault repair state 1 to working state 0 is [not specified]. Based on the state transition probabilities of the element, the state transition matrix of the element is obtained as follows: ; in, This represents the state transition matrix.

[0091] In one possible implementation, when the state matrix construction module 71 samples the state changes of each element in the state transition matrix to obtain the full lifecycle state matrix of the distribution network elements, it is used for: make , Indicates the first Each component Always in a state of readiness The probability, ; Based on the law of total probability, we derive: ; The probability of the component state is obtained using the law of total probability: ; If the repair rate The component downtime probability model is obtained as follows: ; The state change behavior of each component is sampled using the Markov chain Monte Carlo simulation method, and random numbers are generated in a 0-1 uniform distribution. ,but When the simulation time reaches the preset value, the full life cycle state matrix of the distribution network system components is obtained: ,in This represents the total number of components. in, Indicates the first sky Timing element The state.

[0092] In one possible implementation, the calculation module 72 performs a distribution network reliability assessment considering vehicle-to-grid integration and source-load fluctuations based on the full lifecycle state matrix of distribution network components. When calculating the first reliability evaluation index, it is used for: Based on historical data, calculate the load value, distributed photovoltaic output power, and wind power output power; Based on the full life cycle state matrix of distribution network components, the fault data of each node is calculated, including fault time, number of faults and load loss. Using node network analysis, the recoverable load of distributed power sources is calculated; The first reliability evaluation index is calculated based on fault data and recoverable load.

[0093] In one possible implementation, when the calculation module 72 calculates the load value, distributed photovoltaic output power, and wind power output power based on historical data, it is used for: The vehicle-to-everything (V2X) integrated interaction scenarios are classified into different levels, and the objective functions of different levels are quantitatively analyzed. Based on the objective functions corresponding to different levels, the load data of the charging and discharging station is obtained; Calculate the load value based on the load data; Calculate the output power of distributed photovoltaic power based on historical distributed photovoltaic data; Calculate the wind power output based on historical wind power data.

[0094] In one possible implementation, when calculating the fault time of each node based on the full life-cycle state matrix of the distribution network components, the calculation module 72 is used for: according to Calculate the failure time of each node; in, Represents a node downtime, Represents a node The failure rate of the main power supply path Represents a node The probability of failure of the backup power supply path; Indicates the repair time of the faulty component; Represents a node The load shedding status is represented by 1, where 0 indicates no load shedding. This indicates the output of distributed energy resources on the backup power supply path; Indicates the switching action time. Represents a node The load.

[0095] In one possible implementation, the first reliability evaluation index includes: system average outage frequency index, system average outage duration index, expected unsupplied energy, and average service availability index. according to: Calculate the system average outage frequency index for load reduction per unit time; in, This represents the system's average interrupt frequency index. Indicates the total number of users. Represents a node Number of power outages; according to: Calculate the average annual system outage duration index for each user due to power failure; in, This represents the average downtime index of the system. Represents a node The power outage time, Indicates the total number of power outages; according to Calculate the expected unsupplied energy due to the power outage; in, This indicates an expectation that no energy will be supplied. Indicates the first The load capacity (MW) affected by the power outage. Indicates the first Duration of the power outage; according to Calculate the average service availability index for users; in, This represents the average service availability index. Indicates the first The number of users affected by this power outage. It indicates the total time of the year.

[0096] In one possible implementation, the model training module 73 simulates source load fluctuation scenarios and constructs a dataset based on the system state matrix and the first reliability evaluation index. When training the graph neural network model using the dataset, it is used for: Based on the system state matrix, a fault model for the operation of the distribution network is constructed. Random sampling is performed using a grid sampling method to generate power distribution network state scenarios that conform to actual operating patterns; In each scenario of the distribution network status, a second reliability evaluation index is calculated based on the physical topology and fault model. The scenario status data in the power distribution network status scenario is combined with the corresponding second reliability evaluation index to form sample data; The graph neural network model is obtained by modeling the graph neural network based on the physical topology. The graph neural network model is trained using a training set, and the accuracy of the trained graph neural network model is verified using the first reliability evaluation index. If the accuracy does not meet the accuracy threshold, the parameters of the graph neural network model are adjusted until the accuracy meets the threshold, and the training ends to obtain the target graph neural network model.

[0097] In one possible implementation, the feature update process equation during the training of the graph neural network model is: ; in, Indicates the updated number The node feature matrix of the layer This represents the activation function. This represents the adjacency matrix after adding self-loops and normalizing. Indicates the number before the update The node feature matrix of the layer This represents a trainable weight matrix.

[0098] The above embodiment provides an online reliability assessment device for a distribution network. A state matrix construction module establishes state transition matrices for each component in the distribution network based on its physical topology. Then, it samples the state changes of each component in the state transition matrices to obtain a full lifecycle state matrix for the distribution network components. A calculation module performs a distribution network reliability assessment, taking into account vehicle-to-grid integration and source-load fluctuations, based on the obtained full lifecycle state matrix, and calculates a first reliability evaluation index. A model training module acquires the system state matrix and, based on the system state matrix and the first reliability evaluation index, simulates source-load fluctuation scenarios and constructs a dataset. The dataset is then used to train a graph neural network model. A prediction module, based on the trained target graph neural network model, inputs the load values ​​of nodes and real-time distribution network status data to obtain predicted reliability values ​​for each node. This embodiment, by constructing a state feature extraction and reliability index prediction based on a graph neural network model, effectively integrates topology connections, electrical parameters, and operating status data, achieving deep modeling of the interdependencies between distribution network nodes and improving the accuracy and generalization ability of the prediction results.

[0099] To address the issues of low evaluation efficiency and difficulty in online application of traditional Monte Carlo simulation-based methods, this invention achieves significantly faster evaluation while retaining nonlinear modeling capabilities through collaborative training and fusion inference of samples generated by GNN models and Markov chain Monte Carlo methods. This enables rapid online evaluation and is suitable for real-time decision-making scenarios.

[0100] In this embodiment, to address the problem of high system state uncertainty caused by enhanced source load volatility, the present invention introduces the Markov chain Monte Carlo method to jointly sample and simulate multi-source load disturbances, thereby improving the ability to model complex stochastic processes and ensuring simulation accuracy under high-dimensional uncertain input conditions.

[0101] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A power distribution network reliability online evaluation method, characterized in that, include: Based on the physical topology of the distribution network, establish the state transition matrix of each component in the distribution network; The state changes of each element in the state transition matrix are sampled to obtain the full life cycle state matrix of the distribution network element. Based on the full life cycle state matrix of the distribution network components, a reliability assessment of the distribution network considering vehicle-to-grid integration and source-load fluctuation is performed, and the first reliability evaluation index is calculated. Obtain the system state matrix, and based on the system state matrix and the first reliability evaluation index, simulate the source load fluctuation scenario and construct a dataset, and use the dataset to train a graph neural network model; Based on the trained target graph neural network model, the load values ​​of the input nodes and the real-time status data of the distribution network are used to obtain the predicted reliability values ​​of each node.

2. The power distribution grid reliability online evaluation method of claim 1, wherein, The step of establishing the state transition matrix of each component in the distribution network based on the physical topology of the distribution network includes: Determine the connection relationships of each component based on the physical topology of the power distribution network; Based on the connection relationships of each component, establish the Markov chain state transition matrix for each component.

3. The power distribution grid reliability online evaluation method of claim 2, wherein, The step of establishing the Markov chain state transition matrix for each element based on the connection relationship of each element includes: Setting the full life cycle operation state of power distribution network elements, lines is , wherein represents the operation state space, represents the first moment, is a positive integer greater than or equal to 1, represents the time set; represents the working state of the element at the moment, represents the fault maintenance state of the element at the moment, and the state transition probability of the element is defined as: ； in, Indicates the first The state transition probability of each component Indicates time interval, The failure rate of the corresponding component refers to the transition probability from state 0 to state 1. The repair rate of the corresponding component refers to the transition probability from state 1 to state 0. Indicates the first Each component in the time interval The probability of transitioning from working state 0 to fault repair state 1. Indicates the first Each component in the time interval The probability of remaining in fault repair state 1 within the system. Indicates the first Each component in the time interval The probability of remaining in a working state of 0 within the specified timeframe. Indicates the first Each component in the time interval Within this context, the probability of transitioning from fault repair state 1 to working state 0 is [not specified]. Based on the state transition probabilities of the element, the state transition matrix of the element is obtained as follows: ； in, This represents the state transition matrix.

4. The online reliability assessment method for distribution networks according to claim 3, characterized in that, By sampling the state changes of each element in the state transition matrix, a full lifecycle state matrix of the distribution network elements is obtained, including: make , Indicates the first Each component Always in a state of readiness The probability, ; Based on the law of total probability, we derive: ； The element state probability is obtained according to the total probability formula: ； If the repair rate The component downtime probability model is obtained as follows: ; The state change behavior of each component is sampled using the Markov chain Monte Carlo simulation method, and random numbers are generated in a 0-1 uniform distribution. ,but When the simulation time reaches the preset value, the full life cycle state matrix of the distribution network system components is obtained: ,in This represents the total number of components. in, Indicates the first sky Timing element The state.

5. The online reliability assessment method for distribution networks according to claim 4, characterized in that, Based on the full lifecycle state matrix of the distribution network components, a distribution network reliability assessment considering vehicle-to-grid integration and source-load fluctuations is performed, and a first reliability evaluation index is calculated, including: Based on historical data, calculate the load value, distributed photovoltaic output power, and wind power output power; Based on the full life cycle state matrix of the power distribution network components, the fault data of each node is calculated, including fault time, number of faults and load loss. Using node network analysis, the recoverable load of distributed power sources is calculated; Based on the fault data and the recoverable load, a first reliability evaluation index is calculated.

6. The online reliability assessment method for distribution networks according to claim 5, characterized in that, The calculation of load value, distributed photovoltaic output power, and wind power output power based on historical data includes: The vehicle-to-everything (V2X) integrated interaction scenarios are classified into different levels, and the objective functions of different levels are quantitatively analyzed. Based on the objective functions corresponding to different levels, the load data of the charging and discharging station is obtained; Calculate the load value based on the load data; Calculate the output power of distributed photovoltaic power based on historical distributed photovoltaic data; Calculate the wind power output based on historical wind power data.

7. The online reliability assessment method for distribution networks according to claim 5, characterized in that, Based on the full lifecycle state matrix of the distribution network components, the fault time of each node is calculated, including: according to Calculate the failure time of each node; in, Represents a node downtime, Represents a node The failure rate of the main power supply path Represents a node The probability of failure of the backup power supply path; Indicates the repair time of the faulty component; Represents a node The load shedding status is represented by 1, where 0 indicates no load shedding. This indicates the output of distributed energy resources on the backup power supply path; Indicates the switching action time. Represents a node The load.

8. The online reliability assessment method for distribution networks according to claim 5, characterized in that, The first reliability evaluation index includes: system average outage frequency index, system average outage duration index, expected unsupplied energy, and average service availability index; according to: Calculate the system average outage frequency index for load reduction per unit time; in, This represents the system's average interrupt frequency index. Indicates the total number of users. Represents a node Number of power outages; according to: Calculate the average annual system outage duration index for each user due to power failure; in, This represents the average downtime index of the system. Represents a node The power outage time, Indicates the total number of power outages; according to Calculate the expected unsupplied energy due to the power outage; in, This indicates an expectation that no energy will be supplied. Indicates the first The load capacity affected by this power outage Indicates the first Duration of the power outage; according to Calculate the average service availability index for users; in, This represents the average service availability index. Indicates the first The number of users affected by this power outage. It indicates the total time of the year.

9. The method for online reliability assessment of distribution networks according to any one of claims 1-8, characterized in that, Based on the system state matrix and the first reliability evaluation index, a source load fluctuation scenario is simulated and a dataset is constructed. A graph neural network model is trained using the dataset, including: Based on the system state matrix, a fault model for the operation of the distribution network is constructed; Random sampling is performed using a grid sampling method to generate power distribution network state scenarios that conform to actual operating patterns; In each of the power distribution network status scenarios, a second reliability evaluation index is calculated based on the physical topology and the fault model. The scenario status data in the power distribution network status scenario is combined with the corresponding second reliability evaluation index to form sample data; The graph neural network is modeled based on the physical topology to obtain the graph neural network model; The graph neural network model is trained using the training set, and the accuracy of the trained graph neural network model is verified using the first reliability evaluation index. If the accuracy does not meet the accuracy threshold, the parameters of the graph neural network model are adjusted until the accuracy meets the threshold, and the training ends to obtain the target graph neural network model. During the training process of the graph neural network model, the feature update process equation is as follows: ； in, Indicates the updated number The node feature matrix of the layer This represents the activation function. This represents the adjacency matrix after adding self-loops and normalizing. Indicates the number before the update The node feature matrix of the layer This represents a trainable weight matrix.

10. An online reliability assessment device for power distribution networks, characterized in that, include: The state matrix construction module is used to establish the state transition matrix of each element in the distribution network based on the physical topology of the distribution network. The state matrix construction module is also used to sample the state changes of each element in the state transition matrix to obtain the full life cycle state matrix of the distribution network element. The calculation module is used to perform a reliability assessment of the distribution network, taking into account vehicle-to-grid integration and source-load fluctuations, based on the full life-cycle state matrix of the distribution network components, and to calculate the first reliability evaluation index. The model training module is used to obtain the system state matrix, and based on the system state matrix and the first reliability evaluation index, simulate the source load fluctuation scenario and construct a dataset, and use the dataset to train a graph neural network model; The prediction module is used to obtain the predicted reliability value of each node based on the load value of the input node and the real-time status data of the distribution network, according to the trained target graph neural network model.

Images