Research On Underload Cascading Failures In Supply Chain Networks

Supply chain is the product of economic globalization and industrial division of labor,which realizes the efficient flow of resources such as materials,information and capital.With the deepening of cooperation,the supply chain has gradually changed from a single chain to a network composed of multiple supply chains,where enterprise nodes and the supply-demand relationship between them forms a supply chain network.Enterprises in the supply chain network are closely connected.When an enterprise cannot make ends meet,it goes bankrupt and its representative node fails correspondingly.This will bring reduction in demand for its upstream and supply in downstream,affecting the operations of upstream and downstream enterprises.Failures will spread and amplify on the supply chain network,forming underload cascading failure phenomenon.Underload cascading failure often brings catastrophic consequences and huge costs,so it is necessary to deeply study the problem of underload cascading failure in supply chain networks.How to determine nodes that need to be defended so as to improve supply chain security is the current research focus,but there are still two deficiencies in current research:(1)Current underload cascading failure model adopts the linear assumption of load and capacity,which cannot well reflect the complex relationship between them,and most of the work is carried out based on synthetic networks,lacking actual data support,and most of the work is carried out based on synthetic networks,lacking realworld data support;(2)The attack-defense confrontation scenario hasn’t been considered,so supply chain networks cannot specifically defend against the attacker’s strategic attack.To address the above two problems,the main work and innovations of this dissertation are as follows:(1)Supply chain robustness analysis.Based on the improved underload cascading failure model and an appropriate index,this thesis studies the robustness of supply chain network in non-adversarial scenarios.First,this thesis defines the problem of Cascading Failure Robustness Evaluation(CFRE);and this thesis proposes an improved underload cascading failure model based on the nonlinear assumption between load and capacity.Second,this thesis starts the cascading process by simulating the attack on a real data set,and introduces the relative size of the largest functional subgraph as the robustness index.Third,this thesis compares the degree that five strategies of random,large(small)degree,and large(small)load reduce the robustness measure.The experimental results show that:when the model parameters are 0.4:5≤θ1≤2 and 0≤θ2≤1,the enterprise nodes with small degrees should be defended first.(2)Research on supply chain attack and defense game.This thesis focuses on the underload cascading failure attack-defense game in the adversarial scenario,and explores the heuristic algorithm by solving the strong Stackelberg equilibrium.First,this thesis defines the problem of Adversarial Underload Cascading Suppression(AUCS)based on the aforementioned model and the robustness measure,and transforms it into a two-player zero-sum Stackelberg security game.Secondly,in order to solve the NP-hard game equilibrium,this thesis proposes two symmetrical heuristic algorithms by improving Memetic algorithm:the defender GM-SCF(Genetic Memetic-based Suppressing Cascading Failures)algorithm and the attacker GM-SCF algorithm.These algorithms use search instead of traversal,which improves the efficiency of solving the game equilibrium.Then,this thesis takes the above algorithm as the specific implementation of the better oracle in the Double Oracle framework,constructing the DO-SCF(Double-Oracle based Suppressing Cascading Failures)algorithm.Finally,this thesis evaluates the effectiveness,scalability and parameter nonsensitivity of the DO-SCF algorithm:in terms of effectiveness,compared with the baseline method,the algorithm in this paper can still achieve the highest benefit with the maximum increase of 104%;in terms of scalability,the DO-SCF algorithm has much less running time than Core-LP,the maximum difference between the two is 161650 times;in terms of parameter sensitivity,DO-SCF algorithm’s payoff is not affected by the parameters of the underload cascade failure model,and its performance is stable.