Multi-layer coupling power system risk propagation analysis method based on space-time diagram neural network

Abstract

The invention relates to a multilayer coupling power system risk propagation analysis method based on a space-time diagram neural network, and belongs to the technical field of power systems. According to the method, an information-electric power-traffic three-layer heterogeneous coupling network model is constructed, physical quantities in a heterogeneous coupling network are collected, and graph data are constructed based on neighborhoods and connectivity; a risk propagation prediction model is established based on a space-time multi-graph neural network, a space learning network comprises k parallel graph convolutional layers, and a time learning network is realized by combining a long short-term memory (LSTM) network and an attention mechanism; historical graph data of normal operation of the heterogeneous coupling network and historical graph data of faults are obtained and input into a prediction model for training, and a graph structure can be adaptively updated; and predicting a risk propagation path in the three-layer heterogeneous coupling network by using the trained risk propagation prediction model. According to the method, a risk propagation path in an information-electric power-traffic three-layer heterogeneous complex network structure can be realized, and a real-time space-time relationship can be reflected.

Description

Technical Field

[0001] The present invention belongs to the field of power systems, and specifically relates to a risk propagation analysis method for a multi-layer complex coupled power system based on a spatiotemporal graph neural network. Background Art

[0002] In order to better meet the needs of economic and social development and energy transformation, a high proportion of distributed energy access has gradually become a basic feature and development form of the power system. Among them, electric vehicles (EVs), as a flexible resource on the load side, have developed rapidly in recent years. Massive electric vehicle users make decisions on charging and discharging, and driving under the guidance of various types of information, which will have a certain impact on the operation status of the power grid and the traffic flow of the road network. In addition, with the construction and deployment of 5G communication base stations, considering that base station energy storage can participate in economic dispatch, it provides a more flexible adjustment capability for the power system.

[0003] Electricity, transportation, and communications are important links and basic guarantees for the normal operation of society. Today, these three heterogeneous networks are no longer independently operated individuals, but constitute an information-electricity-transportation ternary coupling network that is mutually penetrating, interconnected, and deeply coupled, involving frequent interactions between information, electricity, and transportation. In this multi-layered complex system, the information and communication network is responsible for the connection of various information and communication equipment (including 5G base stations), and further realizes functions such as computing, sensing, and control. The physical power grid mainly refers to the network connection between various power equipment (including charging piles), and is responsible for functions such as power transmission and supply. The transportation network is closely related to both, mainly through electric vehicles, an important component of traffic flow, such as signal reception and transmission, charging and discharging behavior, etc. The schematic diagram of the information-electricity-transportation coupling network is shown below. Figure 1 The power grid is the power source of the transportation network and the information network; the transportation network is an important load of the power grid and an important user of the information network; the information network is the communication basis for the efficient and stable operation of the power grid and the transportation network.

[0004] It is the strong coupling and dependence between heterogeneous systems that means the failure of one link may affect other networks. The deep coupling and interaction between information, power and transportation networks not only brings convenience to production and life, but also creates favorable conditions for risk transmission and buries safety risks.

[0005] At present, most of the research on complex coupled power systems is focused on two-layer network models such as "cyber-physical power system" and "power + transportation", focusing on analyzing the impact of cyber network attacks on the security performance of power grids and the optimal dispatching and operation of power grids. For example, in Reference 1 (Buldyrev SV, Parshani R, Paul G, et al. Catastrophic cascade offailures in interdependent networks [J]. Nature, 2010, 464 (7291): 1025.), Professor Buldyrev proposed the concept of interdependent networks and the vulnerability analysis framework in 2010, which provided a new idea for analyzing the interactive impact between coupled systems and promoted the trend of studying cyber-physical power systems from the perspective of interdependent networks. Reference 2 (Chen Shiming, Zou Xiaoqun, Lv Hui, et al. Study on the robustness of interdependent networks facing cascading failures [J]. Acta Physica Sinica, 2014, 63 (02): 432-441.) analyzed the critical characteristics of the interdependent network model by adjusting the weight parameters of the two sub-networks and the coupling strength between networks. Reference 3 (Zhang Yiwei, Liu Wenxia, ​​Liu Gengming, Huang Shaofeng. Modeling and vulnerability analysis of power information-physical system considering topological correlation and dual coupling [J]. Proceedings of the CSEE, 2021, 41(16): 5486-5500.) considers the topological correlation between the two-layer networks, establishes a dual-coupled power information-physical system model, and conducts vulnerability assessment. Reference 4 (Ji Xingpei, Wang Bo, Liu Dichen, et al. A review of dependent network theory and its application in structural vulnerability analysis of power information-physical systems [J]. Proceedings of the CSEE, 2016, 36(17): 4521-4532.) considers the actual situation of China's power system and establishes a one-to-one model of information nodes and power nodes. Reference 5 (Wang Zhiwei, Wang Xiuli, Yong Weizhen, et al. Research on the trading strategy of 5G base station virtual power plant with backup energy storage participating in the power market [J]. Power System Technology, 2024, 48(03): 968-982.) proposed the optimal bidding and auxiliary frequency regulation of the virtual power plant (Virtual PowerPlant, VPP) formed by 5G base stations, wind farms and photovoltaic power stations. Reference 6 (Zhang Wei, Zhu Tongtong, Su Jin. Demand response strategy of power-information-transportation coupling network considering electric vehicles and 5G base stations [J / OL]. Automation of Electric Power Systems, 2024.) analyzed the impact of electric vehicles and 5G base stations on the power grid and proposed a two-stage demand response optimization scheduling strategy. Reference 7 (Sheng Yujie, Guo Qinglai, Xue Yixun, et al. Modeling and collaborative optimization of power-transportation coupling network from the perspective of information-physics-society [J / OL]. Automation of Electric Power Systems, 2024.) conducted relevant research on the modeling analysis and collaborative optimization of power-transportation coupling network.

[0006] In summary, for the complex system of information-power-transportation, a multi-level coupling network model has not been established from the overall perspective, and the dynamic process of risk evolution brought about by coupling interaction has not been analyzed. Summary of the invention

[0007] In response to the above problems, the present invention proposes a risk propagation analysis method for a multi-layer complex coupled power system based on a spatio-temporal graph neural network. Starting from the complexity of the network structure and the dynamic characteristics of the nodes, a three-layer heterogeneous network model of information, power and transportation is established, and the inter-layer coupling relationship is described by drawing on the idea of ​​graph theory. According to the time-varying characteristics of the data in each layer of the network, the spatio-temporal graph neural network (STGNN) is used to describe the complex spatio-temporal dependency relationship between networks, thereby analyzing the dynamic characteristics of risk propagation.

[0008] A risk propagation analysis method for a multi-layer complex coupled power system based on a spatiotemporal graph neural network of the present invention comprises the following steps:

[0009] Step 1: Establish a three-layer heterogeneous coupling network model of information-power-transportation, which includes an upper-layer information network, an intermediate power network, and a lower-layer transportation network; the information network is constructed based on an Ethernet communication network graph, in which nodes include base stations and communication equipment, and edges are established when nodes are connected for communication; the power network is constructed based on a power system graph, in which nodes are busbars, and electrical lines connecting busbars are constructed as edges in the graph; the transportation network is constructed using a dynamic road network model, in which nodes are intersections, and edges in the graph are established according to road network lanes; if there is a connection relationship between nodes between layers, edges between nodes between layers are established;

[0010] Step 2: Obtain node data in the three-layer heterogeneous coupling network, and construct graph structure data based on neighborhood and connectivity; collect physical quantities in the three-layer heterogeneous coupling network, and construct graph data of the network G, including node features and edge relationships; the collected physical quantities include: signal strength of each node in the information network, current, voltage and voltage phase angle of the bus in the power grid, and traffic volume of each road in the traffic network; normalize the collected data in the same dimension and use a pre-set unified representation method to represent it as node features; establish the adjacency matrix A and topological relationship graph A of G according to the edge relationship between nodes between layers in the network G and the edge relationship between nodes within each layer tp ; Use the adjacency matrix A or topological graph A tp Represent the edge relationship of G;

[0011] Step 3: Build a risk propagation prediction model based on spatiotemporal multi-graph neural network; the spatiotemporal multi-graph neural network that realizes the risk propagation prediction model includes a spatial learning network and a temporal learning network, the spatial learning network includes k parallel graph convolution layers, and the temporal learning network is realized by combining the long short-term memory network LSTM and the attention mechanism; the graph data of the three-layer heterogeneous coupling network collected at the first k moments are input into the risk propagation prediction model after normalization processing, firstly, the k parallel graph convolution layers respectively extract the features of the spatial dimension of the graph data at k moments, and then the extracted feature input is introduced into the LSTM of the attention mechanism in time sequence to extract the features of the temporal dimension, the extracted features are represented by the hidden state input of the LSTM, and finally the features extracted at k moments are fused to output the graph data at k+1 moment; the risk propagation prediction model slides the input graph data with a step size of k, and outputs the graph data of multiple prediction moments in sequence; k≥3;

[0012] Step 4: Obtain historical data of the three-layer heterogeneous coupling network, including historical graph data of each node in the network in normal operation, and historical graph data of the power system, transportation network, and information network in different fault states, and input them into the risk propagation prediction model for training until the model reaches the set accuracy requirement; during the training process, the graph structure of the three-layer heterogeneous coupling network will be adaptively updated according to the collected time data;

[0013] Step 5: When a node in the network fails, the trained risk propagation prediction model is used to predict the risk propagation path in the three-layer heterogeneous coupling network of information, power and transportation.

[0014] The advantages and positive effects of the present invention are:

[0015] (1) The method of the present invention constructs a three-layer heterogeneous network model of information-power-transportation, and fully characterizes the coupling interaction relationship between nodes in the complex network structure. The method of the present invention is extended from the regular data structure to the construction of a dynamic model, realizing the dynamic prediction of risk propagation in the three-layer heterogeneous complex network structure of information-power-transportation.

[0016] (2) The method of the present invention uses multi-dimensional time series to predict the risk propagation path of the three-layer heterogeneous network of information-power-transportation based on STMGCN, so as to achieve the purpose of predicting fault nodes and detecting key nodes. The STMGCN of the present invention introduces time series features on the basis of graph neural network, which can simultaneously obtain the correlation of feature data in time domain and space domain, and is suitable for the scenario of the present invention where the node features change over time. For each graph, the prediction model of the present invention respectively uses graph convolutional neural network (GCN), long short-term memory network (LSTM) and attention mechanism to capture time and space correlation, and finally obtains the final prediction result by fusing the parallelized multi-graph spatiotemporal hidden information.

[0017] (3) The method of the present invention uses graph neural network (GNN) to model and predict the propagation dynamics process on complex networks, designs a training process and an appropriate GNN architecture, and can represent a wide dynamic range with few assumptions. Previously, graph neural network (GNN) has been widely used in various fields such as biological networks, social networks, and recommendation systems. However, these models are limited to static graph data, and the graph structure is fixed. The method of the present invention makes full use of time-varying graph data, mines valuable timing information other than static topology in various systems, and extracts complex spatiotemporal dependencies through the spatiotemporal graph neural network framework, making the prediction results more reasonable and accurate.

[0018] (4) The method of the present invention uses a graph attention mechanism to comprehensively analyze the correlation strength between nodes in each network for the decision-making goals and information interactions of each subject, focusing on the temporal correlation characteristics between data to accurately reflect the real-time spatiotemporal relationship. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is the interactive relationship diagram of information-power-transportation network;

[0020] Figure 2 It is a flow chart for implementing the risk propagation analysis method of the multi-layer complex coupled power system of the present invention;

[0021] Figure 3 It is an overall framework diagram of a risk propagation prediction model constructed by the method of the present invention;

[0022] Figure 4 It is the ST structure diagram of the present invention that captures the spatiotemporal correlation;

[0023] Figure 5 It is a schematic diagram of intra-layer node load redistribution of the present invention. DETAILED DESCRIPTION

[0024] The technical solution of the present invention is described below in conjunction with the accompanying drawings and embodiments.

[0025] The present invention makes full use of the relevant theories of graph neural networks and realizes a risk propagation analysis method for multi-layer complex coupled power systems based on spatiotemporal graph neural networks. The present invention considers that the propagation of fault risks in the power grid not only has temporal characteristics, but also depends on the topological structure of the power grid. In order to make full use of the spatial topological information of the power grid and the temporal characteristics of risk propagation, the present invention proposes a risk propagation prediction method based on spatiotemporal graph convolutional neural networks. First, graph data is constructed according to node characteristics and topological information, and spatial correlation is extracted for each graph through graph convolutional neural networks (GCN). Long Short-Term Memory (LSTM) and attention mechanism are used to capture the temporal association of multiple node data, and spatial and temporal features are fused through spatiotemporal graph convolution units. Then, the risk propagation prediction is modeled as a regression problem, and the spatiotemporal graph convolutional neural network model is trained. The graph neural network is trained based on historical data, the learned propagation dynamic characteristics are generalized, and the existing data set is used for test analysis. The propagation dynamics of complex network structures can be predicted in unknown propagation dynamics and a set of unknown network structures.

[0026] First, the coupling relationship between the information-power-transportation network is analyzed. The coupling relationship between the power grid, transportation network and information network is as follows: Figure 1As shown in the figure. In this coupling, the power, information and transportation systems achieve synergy through the coupling between charging stations, 5G base stations and EVs. As the coupling point between the power grid and the transportation network, the charging station provides charging services for electric vehicles EV. The use of EV requires charging stations to provide power supply, and the operation of charging stations also depends on reliable power supply. Therefore, the stability and availability of the power system are crucial to the normal operation of charging stations and transportation systems. At the same time, EV has the dual attributes of source and load based on vehicle-to-grid (V2G) technology, and adjusts its charging and discharging behavior under the influence of charging electricity prices, reducing the impact on the power grid and providing power support. As the coupling point between the power grid and the information network, the 5G base station provides high-speed and reliable communication services. Its coupling with the power system is mainly through the interconnection of power supply and communication. The power supply of 5G base stations requires reliable power system support, and the power system can monitor and control the energy consumption of base stations through the Internet energy management system to optimize energy utilization efficiency. At the same time, the high-speed communication service provided by 5G base stations provides an important foundation for the information transmission between the transportation system and EVs. The intelligent traffic system (ITS) supported by V2G technology enables EVs to share real-time information with charging stations and power companies. As a coupling point between the transportation network and the information network, EVs have a dual role. On the one hand, as mobile energy storage units, EVs can participate in the energy dispatching and energy storage of the power system. At the same time, the popularity and use of EVs have an impact on the transportation system, such as reducing emissions from traditional fuel vehicles and alleviating traffic congestion. On the other hand, EVs can be used as communication loads. By adjusting their driving paths, the communication load of the base station can be transferred, reducing the operating costs on the communication side.

[0027] In order to make full use of the topological structure of the power grid, the present invention uses graph convolution to extract the spatial characteristics of nodes in the power grid to achieve the prediction of risk propagation. When applying graph neural networks to extract features, the data to be processed is first abstracted and converted into graph structured data. In the real world, the structural information and node characteristics of the graph will change over time, which can be reflected in information networks, power networks, and transportation networks. The node state at a certain location is affected by both the nodes in the neighboring locations and the historical state information of the current node. Therefore, the risk propagation model needs to fully consider the dynamic characteristics of the graph structure and node attributes, and predict the network state at time t+1 based on the changes in the network from time 0 to time t.

[0028] The risk propagation analysis method of multi-layer complex coupled power system based on spatiotemporal graph neural network in the embodiment of the present invention is as follows: Figure 2 As shown, the following is explained in five steps.

[0029] Step 1: Establish a three-layer heterogeneous coupling network model of information, power and transportation.

[0030] Based on the above coupling relationship, a mathematical model of the multi-layer coupling system of information-power-transportation is constructed. In a multi-layer coupling system, inter-layer coupling, that is, there is generally a weak correlation between heterogeneous network nodes, and the three-layer network architecture is modeled separately. This step constructs the graph structure of each layer and establishes the node relationship within the layer to describe the spatial dependency relationship of nodes in the same layer.

[0031] (1) Construct an upper-layer information network. The information layer is an Ethernet-based communication network. 5G base stations, communication equipment, data channels, etc. are abstractly represented by information nodes. Signal strength data such as reference signal receiving power and communication range can be obtained as node characteristics. The connection relationship between nodes can be simplified by an adjacency matrix as follows:

[0032] C=(a ij ) k×k (1)

[0033] Among them, when information nodes i and j are connected and communicate, a ij =1, otherwise a ij =0; k is the number of information nodes.

[0034] (2) Construct the intermediate power grid. The buses in the power system are constructed as graph nodes, and the electrical lines connecting the buses are constructed as edges in the graph. The electrical data such as the current, voltage, and voltage phase angle of the bus nodes can be extracted as the features of the nodes in the graph. The power system is abstracted into a network with m bus nodes. The adjacent matrix of each bus node is expressed as:

[0035] P=(a ij ) m×m (2)

[0036] Among them, when busbar nodes i and j are connected, a ij =1, otherwise a ij =0; m is the number of busbar nodes.

[0037] (3) Construct the lower-level transportation network. The dynamic road network model is used for analysis, which is specifically expressed as follows:

[0038]

[0039] Where R is the traffic network, Node is the set of all nodes in the traffic network, assuming that there are n road intersection nodes in total; P is the set of all directed arcs in R, and a directed arc represents a passable road; T is the time series set; U is the set of traffic flows on each road, u ij,t represents the traffic flow of the road from node i to node j at time t; D is the set of road lengths, d ijis the road length from node i to node j. Quantify the value of the traffic network with n intersection nodes and establish the element L in the road adjacency matrix L ij The value of is:

[0040]

[0041] Step 2, obtain node data in the information network, power grid and transportation network, construct multiple graphs based on neighborhood and connectivity, convert network characteristic data into graph data, and abstract the data to be processed into graph structure data to represent various time and space correlations. The above step 1 constructs the graph structure of each layer and establishes the node relationship within the layer to describe the spatial dependency relationship of nodes in the same layer. This step establishes the node relationship between layers based on neighborhood and connectivity, and further describes the spatial dependency relationship between nodes between layers.

[0042] For a three-layer heterogeneous coupling network, if there is a direct connection between nodes in different layers, the edges between the nodes of the two layers are established, so that the adjacency matrix of the nodes between the layers can be established. Assume that the three-layer heterogeneous coupling network established by the present invention is a graph G, and the number of all nodes in G is N. The present invention can use an adjacency matrix, that is, an adjacency matrix A, to represent the connection between the nodes in the three-layer complex coupling system, including the connection between the nodes within the layer and between the layers. The node A in the adjacency matrix A is represented by ij for:

[0043]

[0044] The neighbor graph defined based on the connectivity of neighbor nodes can only reflect the first-order neighbors of the node. A node can be connected to a geographically distant but accessible place. Topological connectivity is used to indicate whether there is a path between a node and other nodes. The topological relationship graph of the graph G defined by the method of the present invention is A tp , node v i With node v j The edge between Internal elements of a topology diagram The calculation is as follows:

[0045]

[0046] In the formula, conn(v i ,v j ) represents the node v in the graph G i With node v j The indicator function of the connectivity between two nodes indicates whether there is a path between them.

[0047] The embodiment of the present invention obtains the signal strength of each information node in the information network, the current, voltage and voltage phase angle of each bus node in the power grid, and the traffic volume of each road in the traffic network. A unified representation method of node characteristics is pre-set, and the characteristics of each node in the three-layer heterogeneous coupling network are represented by a unified method, and the adjacency matrix A and the topological relationship graph A are used. tp Characterize the connection relationship of nodes in a three-layer heterogeneous coupling network. The graph data constructed by the embodiment of the present invention includes node features and edge connection relationships. The edge connection relationship is represented by an adjacency matrix or a topological relationship graph, which can be selected according to the data situation in the following processing.

[0048] The method of the present invention also normalizes the collected node data before inputting the graph data into the risk propagation prediction model, normalizes the data of each dimension, constructs node features, and generates final graph data.

[0049] Step 3: Build a risk propagation prediction model based on the spatiotemporal graph neural network (STGCN).

[0050] STGCN is generally composed of a spatial learning network and a temporal learning network combined together. The risk propagation prediction model of the present invention is implemented using multiple STGCNs (MSTGCNs). MSTGCN introduces dynamic graph structures and learning mechanisms into the spatiotemporal graph neural network (STGCN), that is, it adopts a data-driven approach to obtain in-depth insights into the properties of input graph data from an unknown dynamic process using time series. For each graph, MSTGCN uses graph convolutional neural network GCN, long short-term memory network LSTM and attention mechanism to capture temporal and spatial correlations, learn meta-knowledge of nodes and edges in multi-layer graph G from attribute graphs, and extract spatiotemporal features. The spatial learning network and the temporal learning network are organically combined through the spatiotemporal fusion neural network architecture.

[0051] The risk propagation prediction model constructed by the embodiment of the present invention is as follows Figure 3 As shown, the risk propagation prediction model includes a spatial learning network and a temporal learning network. The present invention introduces a spatio-temporal multigraph convolutional network (STMGCN) as a spatial learning network to extract the temporal correlation of node features in a multi-layer coupled network. Figure 3 The illustrated temporal learning network is implemented as TCN+self-attention mechanism. The present invention combines the long short-term memory network LSTM and the attention mechanism to realize the spatial correlation of node features in the multi-layer coupled network as a temporal learning network, such as Figure 4 As shown. Combined Figure 3 and 4The risk propagation prediction model of the present invention is described. The model space learning network includes k parallel graph convolution layers GCN. The graph data of the three-layer heterogeneous coupling network collected at the first k moments are input into the risk propagation prediction model. The k parallel graph convolution layers first extract the features of the spatial dimension of the graph data at k moments respectively, and then the extracted feature input is introduced into the LSTM of the attention mechanism in time sequence to extract the features of the time dimension. The extracted features are represented by the hidden state input of the LSTM. Finally, the multi-graph spatiotemporal hidden features extracted at k moments are fused, and the graph data at k+1 moment is output. Through the sliding window, the number of graphs of continuous collection time is continuously input, and the graph data of multiple prediction moments is continuously output.

[0052] The GCN operation of the present invention is implemented through spectral graph theory, which learns the spatial information of the network G through the topological connection relationship of the graph data. Spectral graph theory generalizes the convolution operation from grid-based data to graph structure data. In spectral graph analysis, the graph G is represented by the Laplacian matrix L, which is defined as

[0053] L=DA=ID -1 / 2 AD -1 / 2 (8)

[0054] Where I is the unit matrix, D∈R N×N is the degree matrix of graph G, and A is the adjacency matrix of graph G.

[0055] The result of characteristic decomposition of L is:

[0056] L=UΛU T (9)

[0057] In the formula, Λ is a diagonal matrix composed of eigenvalues; U is the Fourier basis; and the superscript T represents the transpose. Taking the graph node feature x at time t as an example, the graph Fourier transform of the signal is defined as:

[0058]

[0059] Graph convolution is achieved by replacing the classical convolution operator with a diagonalized linear operator in the Fourier domain. Therefore, the convolution formula of the graph signal is as follows:

[0060] g θ * x=g θ (L)x=g θ (UΛU T )x=Ug θ (Λ)U T x (11)

[0061] In the formula, g θ Represents the convolution operation on a graph signal.

[0062] like Figure 4 As shown in the figure, the graph data of the multi-layer heterogeneous coupling network is collected in time series and normalized, and then the normalized data is input into the risk propagation prediction model implemented by MSTGCN. In the prediction model, the graph data is first convolved to extract spatial features, and the extracted graph features are input into the time series prediction module in time series to model the time correlation. The time series prediction model is implemented by combining the long short-term memory network LSTM and the attention mechanism, and the long-term and short-term time correlation patterns of the time series are captured by LSTM. The combination of the two can obtain a spatiotemporal graph neural network model with both spatial and temporal modeling capabilities.

[0063] The time series prediction module of the present invention, i.e., the time learning network, has the same input gate, forget gate, and output gate as LSTM, but they are all obtained by graph convolution operators and an attention mechanism is introduced. The purpose of introducing the attention mechanism is to enhance the information of key nodes, as shown in formula (12):

[0064]

[0065] Where σ(·) is the sigmoid function, ⊙ is the exclusive-OR operator, i, f, o, c represent the input gate, forget gate, output gate and cell state vector respectively. When each of them is updated, there is a corresponding trainable weight W and bias vector b, such as the weight W corresponding to f f and the deviation vector b f , f att Represents the attention network, which can enhance the key node information while ensuring the integrity of the information. The result is the attention matrix, is the hidden state of the LSTM output at time t, h t is the hidden state at time t of the LSTM output with the attention mechanism introduced.

[0066] The attention matrix is ​​set to V = (V1, V2, V3, ... V N ,), V t is a column vector, and the calculation formula is shown in formula (13).

[0067]

[0068] In formula (13), the softmax(·) activation function is used for normalization, and the attention matrix V is finally obtained. t Semantically, it is understood as the vector of interdependence between nodes at the output time t, K and b are the weight matrix and bias vector of the attention network, and tanh is the activation function.

[0069] Step 4: Use multidimensional time series to train risk propagation prediction models based on historical data for dynamic behavior prediction. The fault propagation process in heterogeneous networks is described by coupling node dynamic propagation models. That is, the training network can adaptively update the graph structure according to the changes in spatiotemporal data, and then analyze the dynamic propagation process between heterogeneous networks.

[0070] The load in many actual systems is in a process of constant dynamic change. In particular, when nodes are added or removed and the network structure changes, the load in the network will be redistributed. In the nodes within the layer, the load-capacity model is selected as the basis for research. The carrying capacity C(i) of a node i can be expressed as:

[0071] C(i)=(1+α)L0 (14) In the formula, α is the redundancy parameter, and L0 represents the initial load at node i. The larger the redundancy α value is, the stronger the ability to resist risks and failures is.

[0072] Assume that a node i in network A is attacked and faces load redistribution after the node fails, such as Figure 5 As shown, node i will adjust its load according to the proportion in the formula Assigned to the adjacent node j in the same network A as follows

[0073]

[0074] Where n∈T i represents n as the neighboring node of node i, T i is the set of adjacent nodes of node i, are the carrying capacities of nodes j and n in network A respectively.

[0075] Based on the virus propagation model, the dynamic process of risk propagation between nodes in heterogeneous network layers is described as follows:

[0076] p 1,i (t+1)=(1-p 1,i (t))(1-q 1,i (t))+(1-β1)p 1,i (t)+γ1p 2,i (t)(1-p 1,i (t))(16)

[0077] p 2,i (t+1)=(1-p 2,i (t))(1-q 2,i (t))+(1-β2)p 2,i (t)+γ2p 1,i (t)(1-p 2,i (t))(17)

[0078] Among them, q 1,i (t), q 2,i (t) represent the probability that node i will not be infected by neighboring nodes in network A and network B that are in a faulty state, respectively. Then 1-q 1,i (t), 1-q 2,i (t) represents the probability of node i being infected by neighboring nodes in network A and network B respectively; p 1,i (t+1), p 2,i (t+1) represents the probability that node i in network A and network B is in an abnormal state at time t+1, p 1,i (t), p 2,i (t) represents the probability that node i in network A and network B is in a faulty state at time t. Therefore, (1-p 1,i (t))、(1-p 2,i (t)) represent the probability that node i is in normal operation and the probability that it is infected by neighboring nodes in the same layer are (1-q 1,i (t)) and (1-q 2,i (t)), β1, γ1, β2, γ2 are all weight parameters, (1-β1)p 1,i (t), (1-β2)p 2,i (t) represent the probability that node i is not cured and restored to operation at time t after being infected, γ1p 2,i (t)(1-p 1,i (t)),γ2p 1,i (t)(1-p 2,i (t)) represent the probability that a node i is infected by a neighboring node in another layer of the network that is in a faulty state. In the formula, the definition

[0079]

[0080] Among them, α1 and α2 represent the influence factors of network A and network B respectively, a ij represents the edge relationship between node i and node j in the adjacency matrix of network A, b ij Represents the edge relationship between node i and node j in the adjacency matrix of network B. N1 and N2 are the number of nodes in network A and network B respectively.

[0081] Therefore, for different faults, the propagation dynamics process in heterogeneous networks can be analyzed by setting different parameters.

[0082] Step 5: The trained risk propagation prediction model can generalize the learned propagation dynamic characteristics and predict the propagation dynamics of complex systems in unknown propagation dynamics and a set of unknown network structures, such as predicting the charging and discharging behavior of electric vehicles based on graph timing networks and predicting the risk propagation paths between complex coupled networks.

[0083] An implementation process of the multi-layer coupled power system risk propagation prediction method of the present invention includes the following four steps:

[0084] (1) Sample acquisition: The voltage, current, phase angle, traffic volume, signal strength and other information of each bus node in the whole network are obtained when the power system, transportation network and information network are in normal operation and various fault states. R ), R is the node feature dimension.

[0085] (2) Graph data processing: Based on the analysis of network topology, network coupling relationship, time dependency, and spatial dependency mentioned above, the data collected by the three-way coupling system is converted into graph data information. When each information is converted into node information of graph data, it needs to be normalized, and Z-score standardization is used.

[0086]

[0087] Where: Carrying data information X for a node r The mean of r is the standard deviation of information X; r∈[1,R]. After the node features are normalized, they are combined with the adjacency matrix to form the graph data required by the model.

[0088] (3) Model training: In the training phase, a risk propagation prediction model is built and the hyperparameters of the training are set, including batch size, learning rate, and number of iterations. After the training is completed, the accuracy and prediction error of the propagation model are verified. In the present invention, convolution is used to achieve parallelization at the input end, with fewer parameters and faster training speed.

[0089] The model training target is the true value Y pred and predicted values The error between them is the smallest, and the corresponding loss function is

[0090]

[0091] In the formula, the first term To ensure the minimization of the error, the second term L reg is a regularization term that can effectively avoid overfitting, and λ is a hyperparameter.

[0092] (4) Model application: In the model application stage, for actual faults in the power grid, the voltage phase angle data of each bus node is first obtained and the fault location is input. Then, the obtained voltage phase angle data of each bus node is input into the corresponding risk propagation prediction model to predict the propagation path of the fault risk in the system in the future and locate the key risk nodes.

[0093] Except for the technical features described in the specification, all of them are known technologies to those skilled in the art. The present invention omits the description of known components and known technologies to avoid redundancy and unnecessary limitation of the present invention. The implementation methods described in the above embodiments do not represent all implementation methods consistent with the present application. On the basis of the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative labor are still within the protection scope of the present invention.

Claims

1. A risk propagation analysis method for multi-layer complex coupled power systems based on spatiotemporal graph neural network, comprising the following steps: Step 1: Establish a three-layer heterogeneous coupling network G of information-power-transportation, including an upper-layer information network, an intermediate power network, and a lower-layer transportation network; the information network is constructed based on an Ethernet communication network graph, in which nodes include base stations and communication equipment, and edges are established when nodes are connected for communication; the power network is constructed based on a power system graph, in which nodes are busbars, and electrical lines connecting busbars are constructed as edges in the graph; the transportation network is constructed using a dynamic road network model, in which nodes are intersections, and edges in the graph are established according to road network lanes; if there is a connection relationship between nodes between layers, edges between nodes between layers are established; Step 2: Collect physical quantities in the three-layer heterogeneous coupling network and construct the graph data of the network G, including node features and edge relationships; the physical quantities collected include: signal strength of each node in the information network, current, voltage and voltage phase angle of the bus in the power grid, and traffic volume of each road in the traffic network; normalize the collected data in the same dimension and use a pre-set unified representation method to represent it as node features; According to the edge relationship between the nodes in the layer and the edge relationship between the nodes in each layer of the network G, the adjacency matrix A and the topological relationship graph A of G are established. tp ; Use the adjacency matrix A or topological graph A tp Represent the edge relationship of G; Step 3: Build a risk propagation prediction model based on spatiotemporal multi-graph neural network; the spatiotemporal multi-graph neural network that realizes the risk propagation prediction model includes a spatial learning network and a temporal learning network, the spatial learning network includes k parallel graph convolution layers, and the temporal learning network is realized by combining the long short-term memory network LSTM and the attention mechanism; the graph data of the three-layer heterogeneous coupling network collected at the first k moments are input into the risk propagation prediction model, and the k parallel graph convolution layers are used to extract the features of the spatial dimension of the graph data at k moments respectively, and then the extracted feature input is introduced into the LSTM of the attention mechanism in time sequence to extract the features of the temporal dimension, and the extracted features are represented by the hidden state input of the LSTM, and finally the features extracted at k moments are fused to output the graph data at k+1 moment; k≥3; Step 4: Obtain historical data of the three-layer heterogeneous coupling network, including historical graph data of each node in the network in normal operation, and historical graph data of the power system, transportation network, and information network in different fault states, and input them into the risk propagation prediction model for training until the model reaches the set accuracy requirement; during the training process, the graph structure of the three-layer heterogeneous coupling network will be adaptively updated according to the collected time data; Step 5: When a node in the network fails, the trained risk propagation prediction model is used to predict the risk propagation path in the three-layer heterogeneous coupling network of information, power and transportation.

2. The method according to claim 1, characterized in that: In the above step 1, when constructing the traffic network, the set P of all roads and the set U of road lengths are obtained, and the element L in the adjacency matrix L is calculated. ij The values ​​of are as follows: Among them, i and j represent the numbers of the two intersections, ij represents the road from intersection i to intersection j, and d ij Represents the length of road ij.

3. The method according to claim 1, characterized in that In the step 2, the collected data is normalized in the same dimension, including: assuming that the R physical quantity information of the acquired node is X, for the rth physical quantity X r The Z-score was used for standardization. Expressed as and σ r are the mean and standard deviation of the rth physical quantity respectively; r∈[1,R].

4. The method according to claim 1, characterized in that: In step 3, the temporal learning network is implemented by combining the long short-term memory network LSTM and the attention mechanism. The information of key nodes is enhanced by introducing the attention mechanism. Let f att Represents the attention network, and the time learning network outputs the hidden state h at time t t It is expressed as: in, is the hidden state of the LSTM output at time t, h t is the hidden state at time t of the LSTM output with the attention mechanism introduced; is the attention matrix, let V represent it, the column vector V in V t The calculation is as follows: Among them, softmax and tanh are activation functions, K is the weight matrix of the attention network, and b is the bias vector of the attention network.

5. The method according to claim 1, characterized in that In the step 4, the training network adaptively updates the graph structure according to the changes in the spatiotemporal data, and describes the fault propagation process in the heterogeneous network by coupling the node dynamics propagation model, including: (1) When a node is added or removed from the network, the load in the network will be redistributed and the graph structure will be updated; when a node i in a certain layer of network A is attacked and fails, the load will be redistributed. Node i will redistribute its own load in proportion to the Assigned to adjacent node j in the same network A as follows: Among them, T i is the set of adjacent nodes of node i, are the carrying capacities of node j and node n in network A respectively; (2) The fault propagation process in heterogeneous networks is described by the coupled node dynamics propagation model as follows: p 1,i (t+1)=(1-p 1,i (t))(1-q 1,i (t))+(1-β1)p 1,i (t)+γ1p 2,i (t)(1-p 1,i (t)); p 2,i (t+1)=(1-p 2,i (t))(1-q 2,i (t))+(1-β2)p 2,i (t)+γ2p 1,i (t)(1-p 2,i (t)); Among them, p 1,i (t+1), p 2,i (t+1) represents the probability that node i in network A and network B is in a fault state at time t+1, p 1,i (t), p 2,i (t) represents the probability that node i in network A and network B is in a faulty state at time t, q 1,i (t), q 2,i (t) represents the probability that node i will not be infected by neighboring nodes in network A and network B that are in a faulty state; β1, γ1, β2, γ2 are all weight parameters; q 1,i (t), q 2,i (t) is calculated as follows: Among them, α1 and α2 represent the influence factors of network A and network B respectively, a ij represents the edge relationship between node i and node j in the adjacency matrix of network A, b ij Represents the edge relationship between node i and node j in the adjacency matrix of network B.

Images