基于Hadoop的异常传感数据时间序列检测

无线传感器网络中,异常时间序列的研究具有十分重要的意义。针对传统研究在海量数据环境中时间效率低下的问题,提出了基于Hadoop的异常时间序列检测算法。首先对时间序列进行预处理,然后在Hadoop的MapReduce操作中调用动态时间弯曲距离计算算法,实现了DTW距离计算的并行化,从而大大提高检测速度。同时针对传统DTW算法计算复杂度瓶颈问题以及传统约束方法准确率较低问题,提出了基于显著特征匹配的局部约束算法,对弯曲路径进行局部限制,在确保准确性的同时进一步降低了时间、空间复杂度。Hadoop平台下实验结果表明,该方法既提高了检测速度,又保证了检测准确率。

Abnormal Time Series Detection in Wireless Sensor Network Based on Hadoop

In wireless sensor network, the research of abnormal time series detection is of great significance. Due to the poor time efficiency of traditional research under big data, this paper proposes an algorithm about abnormal time series detection based on Hadoop. In this paper, time series are preprocessed firstly and then the DTW algorithm is called during MapReduce operation of Hadoop to realize the parallelization calculation of DTW distance. This measure improves the detection rate greatly. Meanwhile, to solve the bottleneck of computational complexity of classical DTW and the poor precision of the classical constraints, the paper also proposes locally relevant constraints based on salient feature alignments. It constraints the warping path locally to reduce the complexity of time and space further, it also ensures the precision of the algorithm at the same time. The results demonstrate that this algorithm not only decreases the time consumption, but also keeps a high precision.