BAFD Algorithm And EGFD Algorithm For System-level Fault Diagnosis

The wide application of network system has promoted the development of the contemporary information society greatly.But at the same time,it is more and more difficult to ensure the system operates safely and efficiently for the expansion of the system.As an important method to maintain the stability and security of the system,the system-level fault diagnosis has become the focus of more and more scholars' attention.Based on the characteristics of the system-level fault diagnosis problem,some measures like speed mapping are taken to break through the limitation of bat algorithm,so that it can be applied to the system-level fault diagnosis,and then bat algorithm for fault diagnosis,designed for the t-diagnosable system,is proposed.During initialization,in order to avoid the initial solution is too centralized or decentralized,the population is divided into two categories: large and small,and is handled in different ways.The fact that the existing fitness function is very difficult to keep the feasible solution from producing redundant compatible symptoms leads to an increase in the amount of calculation.To solve this problem efficiently,the equation-constrained fitness function is designed by analyzing the characteristics of the diagnosis model.Taking into account the precocious phenomenon of bat algorithm,a variable coefficient is introduced into the velocity update formula to balance search and mining capacity.And discretization of addressing is realized by performing binary mapping for bat speed.Simulation results based on experiments show that bat algorithm for fault diagnosis has significant advantages over FAFD which is a typical algorithm among swarm intelligence diagnosis algorithms in the number of iterations,diagnostic accuracy and the adaptability of the optimal solution.Secondly,most of the existing diagnosis algorithms only consider the aspect of "trust most",the diagnostic ability of many algorithms decreases obviously when the number of fault units exceeds t.In order to improve the phenomenon,efficient greedy for fault diagnosis which is combined with the research method used in t-diagnostic by Liu et al and designed for the situation that the number of fault units exceeds t is proposed.In this algorithm,a sufficient condition is given firstly,it enriches existing methods of absolute fault units identification to make sure that more absolute fault units could be found,and it also decreases the scale of the system in the next diagnosis.Then,two kinds of greedy strategies are formulated respectively by using the dispersion degree and the aggregation degree of the group.At last,it willcause misjudgement if the group selected according to greedy strategy is set directly.so,we carry out "check" strategy for group to reduce the diagnostic error rate,and improve the diagnostic effect of the algorithm.Simulation results based on experiments show that EGFD algorithm can be able to diagnose the system efficiently when the fault unit accounts for the most of the whole system.