【摘要】 复杂系统中,异常检测的最终目标是预测系统的涌现形态。传统的故障诊断主要是寻找输出信号中的反常特征。本文在复杂性科学的基础上,探讨了异常检测新的发展趋势。首先,回顾了几种典型的检测算法。然后介绍了一种新的检测理念———通过理解系统特征和捕捉隐含模式来提高异常检测算法。分析结果表明:符号化及其相关分析为达到这一目的提供了新的途径,它可以帮助我们采用更多的智能算法来探索系统潜在的规律;不过,如何评价这些算法,如何估计系统参数,尽管很有意义,但非常困难,这将是未来工作的重点。更多还原

【Abstract】 Traditional fault diagnosis is to detect abnormal characters in the output signal,but in complex system,the final goal of anomaly detection is to forecast emerging behavior.The new trends of anomaly detection were discussed based on the development of complexity science.Three classification methods were presented and a series of detection algorithms were reviewed.A new idea of detection was introduced which is to improve the anomaly detection algorithms by understanding the system dynamics and capturing its hidden pattern.Result shows that symbolization and relational analysis is a novel approach which can provide many intelligent methods to explore the potential rules.But how to evaluate these methods and identify the parameters of system is very meaningful and difficult,which is the focus in the future work.更多还原