大虚警率下的多故障诊断算法

主要研究不可靠测试下多信号模型的多故障诊断问题。最优的多故障诊断是计算复杂度完全类(NP-Complete)问题,因此大型系统的诊断一般只能用次优的随机搜索算法。次梯度优化算法能够在虚警概率较小时给出较好的结果,但如果测试个数很多且虚警概率较大时,该算法就不能消除虚警的影响,会使估计的故障覆盖所有失败的测试,而不是找到系统真实的故障。针对这一问题,提出了能够同时考虑虚警和误警的目标函数,使算法能排除虚警的测试准确定位故障,并用改进的遗传算法搜索故障部件提高诊断速度。仿真诊断结果表明,同时发生故障的部件个数较少时,遗传算法的诊断速度明显优于次梯度优化算法,而且能够更有效地抑制虚警的影响。

Multiple Fault Diagnosis Algorithm Under Large False-alarm Rate Tests

Multiple fault diagnosis under unreliable tests using the multi-signal model was studied.The optimal multiple fault diagnosis problem is NP-complete,so sub-optimal algorithm was used for large scale system diagnosis.A good fault estimation under low false-positive rate was obtained by the subgradient optimization method.But if the size of the graph is big and the false-alarm rate is high,this algorithm will not be able to eliminate the influence of false-positive,and the resulting estimation would cover every failed test instead of finding the true fault position of the system.A new kind of discriminate function considering both false and true positive tests was used,and an improved genetic algorithm was proposed to search the failure components.Simulation results show that the new method is faster than subgradient optimization and has a higher correct isolation rate when intercurrent faults are few.