Comparative Study Of Cascading Failures In Power Grids And Prediction Of Network Structural Integrity After Cascading Failures

People are increasingly relying on power resources in their daily lives.The power system had gradually developed into a large-scale,cross-regional large-scale internet network,which had become an important part of modern social infrastructure.However,with the development of power grids technology,the instability and security problems brought by large-scale power grids have also received more and more attention.In the past,many blackout accidents not only did the inconvenience caused to the residents of the affected areas,but also caused considerable economic losses to the whole society.Therefore,power system security issues have always been a concern of power suppliers,infrastructure developers and governments,as well as issues of high concern to electrical engineers and researchers.This paper constructs a power information interdependence network model,based on complex network theory,grid structure and load characteristics,combined with information network scheduling functions.The betweenness,power of nodes and the betweenness,weight,initial current absolute value of edges in power grid are taken as input values of Back-Propagation neural network(BP neural network)to predict the network structure integrity after the cascading failure is completed.The main contents include the following aspects:First,Comparison of cascading failures between power information interdependent networks and single-layer power grids.Based on the grid structure and load characteristics,combined with the dispatching function of information network,we construct a "power-information interdependence network" model.Three types of node attack methods are applied to attack a single node of the power grid where the attacked node is the highest load node,the lowest load node or the highest capacity proportion node.The cascade effects are compared with single-layer power grid.Study shows that the robustness of the power information interdependent network is weaker than that of the single-layer power grid under the highest load node attack when the initial load is small.The robustness of the two kinds of networks approaches each other when the initial load is large.The difference in robustness of the two kinds of networks is not obvious at the lowest load node and highest capacity proportion nodeattack.On the interdependent network,the cascading failure caused by the highest load node attack is the most difficult to eliminate completely.Second,the study of prediction network structure integrity after cascade failures by using power grids node attributes.The betweenness and initial power of some nodes of the power grid are taken as the input values of the neural network,and the structural integrity of the network after the failure of the nodes are taken as the output value to train the neural network.Then betweenness and initial power of the untrained power grid nodes are input into the trained neural network to predict the structural integrity of the network after the cascade failures of the power grid when the node fail.Study shows betweenness and initial power attributes of nodes are important to cascade failure of power grid;For a given input value,a trained neural network based on power grids data can accurately predict the network structure integrity after cascade failure;The neural network trained by two or more power grid data can also predict the network structural integrity after cascade failure.Third,the study of prediction network structure integrity after cascade failures by using power grid edges attributes.Neural network is trained by absolute value of the initial current,betweenness of edges and the structural integrity of the power grid after the cascade failure of the edges,and the parameter errors of the prediction are evaluated.Then,the betweenness is replaced by the weight and the simulation is repeated.Finally,the data of three power grids are trained and the error is evaluated.Study shows the result of using betweenness is better than weight;There is a positive correlation between the prediction error and the difference between the maximum and minimum of network structural integrity after cascade failures,whether it is node or edge prediction;Edge prediction can also be accomplished by training neural networks with multiple grid data.