地下工程多时间序列监测数据异常检测算法

An Abnormality Detection Algorithm for Multi-time Series Monitoring Data of Underground Engineering

为实现地下工程监测数据中的异常信息检测,提出了一种多传感器关联的异常检测及识别算法。首先,通过分段线性表示算法提取原始监测序列数据特征,结合局部异常因子检测算法实现单维监测时序数据的异常值识别;然后,基于动态时间弯曲距离分析,进行多传感器间关联异常检测,识别各传感器的监测异常区段;最后,综合两步计算结果,判断异常值是由环境突变引起还是由传感器监测误差引起。算法有效利用了不同传感器间的关联信息,实现对异常来源的甄别。通过对人为施加的异常干扰的识别,验证了所提出算法的有效性。将算法应用于某实际隧道穿越工程多传感器系统的异常识别,进一步证明了本算法的可行性和适用性。

In order to realize the detection of abnormal information in the monitoring data of underground engineering,an abnormality detection and recognition algorithm based on multi-sensor correlation is proposed.Firstly,the data characteristics of the original monitoring sequence are extracted through piecewise linear representation method,and combined with the local outlier factor detection algorithm,to identify the abnormality from the single-dimensional monitoring data series.Secondly,based on the dynamic time warping distance calculation,abnormality detection is performed among multiple interrelated sensors to detect the abnormal time zone of each sensor.Finally,the results of the two calculation steps are combined to determine whether the abnormal values are caused by environmental sudden change or caused by monitoring errors.The algorithm effectively uses the associated information between different sensors to realize the identification of the abnormality sources.The effectiveness of the proposed algorithm was verified through identifying the artificial disturbances from a normal data series.The feasibility and applicability of the algorithm was further verified by applying the developed algorithm to detect the abnormality of monitoring data for a real tunnel crossing project.