| Large-scale cascading failures that are triggered by some small disturbances would lead to disastrous consequences.Therefore,analysis and prevention of cascading failures become a hotspot of system safety and reliability research in recent years.This work is a systemic research on the mechanism of cascading failures from the point of negative entropy competition and synergic movement among the subsystems.This work also provides a technical support for improving system reliability and preventing the occurrence of cascading failures by analyzing and optimizing system reliability.All proposed methods are applied in a ship automatic fire alarm system.The main work is as follows.First,in the brittleness theory of complex system,the survival of subsystems is affected by their own entropy increase and other crashed subsystems.The influence of other crashed subsystems is described as brittle link entropy.The key point of whether a subsystem would recover from the influence it suffered is the size of negative entropy which is absorbed by the subsystem from the outside world.However,the negative entropy provided by the outside world is limited.A non-cooperative game is proposed to analyze the mechanism of cascading failures in terms of competition among subsystems,and then the main reason of cascading failures is obtained.A concept of interval-valued reliability is presented to further determine strategies that subsystems absorb negative entropy.An example of a ship automatic fire alarm system is applied to verify the effectiveness of the proposed method.Second,the number of state variables which are used to describe the activity of complex system is large.The state variables may change with the change of system structure.Therefore,it is very difficult to control them.To overcome these drawbacks,the evolution process of a complex system is described with two kinds of variables,and a system synergetic evolution equation is proposed based on the concept of order parameter.In order to analyze the development trend of a system as the system evolves to a critical stage and evaluate the influence of system evolution with the modification of order parameters,the system synergetic evolution equation is then solved.The corresponding time of the maximum point in the equation is considered as the time of system instability,and a prediction model is presented.Based on slaving principle,a game-theoretic model among order parameters is developed.It provides for unique perspectives on the mechanism of cascading failures.The proposed methods are applied in the ship automatic fire alarm system.Third,a complex system in reality is closely related to the environment which is uncertain,complex,and even unpredictable.The system could be interfered by nature and human factors at anytime.Consequently,it is impossible to totally prevent cascading failures.In order to reduce the risk of cascading failures,an optimal redundancy allocation design is provided to improve the system reliability.Consideration on the large size of a Pareto solution set to the multi-objective RAP,a multi-objective clustering of solutions based on a game-theoretic framework is described.This algorithm comprises single iteration of k-means and multiplayer normal form of game with Nash Equilibrium.The elegant definition of payoff function and consideration of mixed strategies make the approach escape from local optimum solution as much as possible.By comparing with other methods,it is obvious that the proposed algorithm is capable of determining better clusters in the terms of compaction and equi-partitioning.It is applied to reliability optimization design in the ship automatic fire alarm system.Finally,in practice,the reliability of a component can not be given accurately.It is generally estimated from field or test data which is finite.Therefore,there is some uncertainty associated with the component reliability estimates.The uncertainty at the component level may propagate to the system level.A new three-stage approach is proposed for system reliability optimization.At the first stage,probability distributions of component reliabilities are analyzed based on the maximum entropy principle to reduce the uncertainty of input data.At the second stage,NSGA-? is applied to identify a Pareto optimal solution set.At the third stage,the identified Pareto optimal solutions are classified into several clusters by applying k-means and the optimal clustering number is determined by applying silhouettes.Afterwards,representative efficient solutions are selected.Comparison with other methods,our presented approach considers the system configuration with a lower system reliability variance.The proposed method is then applied in the ship automatic fire alarm system. |
PDF Full Text Request