作者:吴裔,郭棋林,陈颢天,郭乃网
Authors:WU Yi,GUO Qi-lin,CHEN Hao-tian,GUO Nai-wang摘要:摘要:时间序列的研究已经被应用到越来越多的领域中。越来越多的领域应用需要索引和分析海量的时间序列，代表性的比如金融，电力，生物信息等等。这类应用往往面临数以亿计的时间序列的处理，然后从中识别出一些隐藏的模式来。然而目前对时间序列的索引技术都是单机版本，需要用漫长的时间来对大量的时间序列进行索引，限制了时间序列分析的产出率。提出了一种基于Isax表达的分布式时间序列索引算法，并在Spark分布式计算框架下实现算法。首先，给出了基于Isax的分布式索引算法的朴素实现想法，指明了其存在的问题。然后提出一种先建立索引结构，再将时间序列哈希到相应叶子节点的分布式索引算法。最终，构建了一个完整的电力时间序列的近邻近似查询系统，再保证查询精确率的前提下大大提高了计算效率。并在实验数据集上证明了算法的正确性、高效性和可扩展性。
Abstract:Abstract:Time series research has been applied to more and more areas. More and more domain applications need to index and analyze massive time series, such as finance, electricity, bioinformatics, and so on. Such applications are often faced with hundreds of millions of time series of processing, and then identify some hidden pattern from the model. Firstly, we give a simple idea of the distributed indexing algorithm based on Isax, which points out its existing problems. Then we propose a distributed indexing algorithm to establish the index structure and then insert the time series to the corresponding leaf node. Finally, this paper constructs a complete approximation query system of power time series, and greatly improves the computational efficiency under the premise of ensuring the accuracy of query. The correctness, efficiency and expansibility of the algorithm are proved on the experimental data set

PDF全文下载地址:

可免费Download/下载PDF全文