基于嵌入式索引的水文时间序列相似性搜索模型

Embedding-based Index Model for Hydrological Time Series Similarity Searching

相似模式挖掘已成为水文领域一个重要研究方向。对水文数据的相似性挖掘,有利于掌握水文数据变化规律和趋势,为洪水预报、防洪调度提供支持,是具有重要意义的工作。为此,在引入时间序列嵌入索引的基础上,结合水文时间序列的特点提出水文时间序列的快速搜索方法。该方法通过序列分割、聚类和参考集训练从原始序列中获取参考序列集,在此基础上通过索引计算方法,将相似性搜索过程映射到欧氏向量空间的搜索,从而提高了搜索效率。

Similar pattern mining has become an important research direction in the field of Hydrology. It is a significant work to process similarity mining in historical data that can be conducive to recognize the trend pattern of hydrological data and provide technical support for the flood forecasting and flood control. Thus, this paper proposed a quick similarity search model according to hydrological sequence features. This model employed series segment, serial cluster and reference training method to generate reference set, and transferred similarity search to European vector space search with indexed by reference set so as to improve the searching efficiency.