无线传感器网络中，传感器的采集与无线网络的传输等均可能带来时间序列的不确定性，而大数据时代的到来使得传统不确定异常时间序列检测研究面临时间效率低下的问题，为此提出基于Hadoop的不确定异常时间序列检测算法。首先对不确定时间序列进行压缩变换，使不确定数据量大大减少，然后利用MapReduce架构调用基于期望距离的不确定时间序列下的DTW算法，实现算法的并行化处理，降低算法时间复杂度。同时针对Hadoop集群任务级调度分配方法在运行中负载分配不均现象，提出Hadoop集群优化方法，明显缩减集群总任务时间，使得节点资源的利用更为合理。Hadoop平台下实验结果验证显示，该方法既提高了检测速度，又保证了检测准确率。%In wireless sensor network,data collection of sensors and wireless network transmission all can produce uncertain time series. The arrival of big data age makes the detecting of uncertain abnormal time series face the problem of poor efficiency of time. So this paper proposes an algorithm about uncertain abnormal time series detec⁃tion based on Hadoop. In this paper,uncertain time series are firstly compressed so that the uncertain data can be reduced greatly. Then the DTW algorithm of uncertain time series based on expected distance is called during Ma⁃pReduce operation of Hadoop to realize the parallelization calculation of this algorithm. This measure reduces the time complexity greatly. Meanwhile,to solve the uneven load distribution exists in current Hadoop inherent task-lev⁃el scheduling methods;the paper also proposes a method of Hadoop optimization. It not only reduces the total task completion time,but also makes the node resource utilization more reasonable. The results demonstrate that this al⁃gorithm not only decreases the time consumption,but also keeps a high precision.
展开▼
