一种快速的时间序列相似性算法的研究及其在股票管理中的应用  

A Fast Similarity Search in Stock Market Time Series

作　　者：孟辉[1] 冯志明[2] 王晓晔[1] 

机构地区：[1]河北工业大学电气与自动化学院,天津300130 [2]河北工业大学继续教育学院,天津300130

出　　处：《河北工业大学学报》2005年第5期51-55,共5页Journal of Hebei University of Technology

摘　　要：提出了一种新的股票时间序列相似性的研究方法,该方法与已往的算法相比,具有很高的计算效率.首先将时间序列分段线性化,将近似直线的端点按时间顺序排列形成转折点序列来描述原始序列,从而降低了数据的数量和复杂度,提高了快速性.基于分段线性化,提出了一种相似性计算方法,可以从时间序列中搜索用于传统股票分析的基础模型.将基础模型的谷点和峰点的位置按照某种规律重新排序,然后从转折点序列中寻找重新排序后与基础模型排序相同的子序列(即相似性搜索),用该基础模型可以对股票后续趋势做出预测.与通常的相似性算法相比,该算法不考虑时间尺度和数据的幅值的变化,因此,大大提高了相似性的计算精度.This paper presents a new method of the similarity of stock market time series. Comparing with the existent method, it is robustness and fastness. We represent the stock market time series based on the piecewise linear, which reduces the quantity of the data and the complexity of compute. In the stock market analyses, a usual tool is the chart pattern. The valleys and the peaks of these chart patterns obey some rules about their relative position. We sort the valleys and the peaks based on value, and we sort time series based on the value also. Thus, the position of the data will be permuting. The pattern similarity is to decide if the given pattern is similar to the chart pattern by looking at if they have the same permutation of the data.

关 键 词：股票数据 时间序列 相似性 数据挖掘 模式挖掘 

分 类 号：F830.91[经济管理—金融学]