基于智能集成架构的时间序列数据挖掘算法研究

Research on Time Series Data Mining Algorithm Based on Intelligent Integrated Architecture

针对单一算法在处理复杂时间序列数据时存在缺陷以致无法挖掘全部信息的问题,提出了智能集成架构,给出了4种集成结构,并分析了它们的适用情况。针对一类随机噪声干扰的时间序列数据,采用并联嵌套建模结构,提出嵌套双种群粒子群算法的自回归滑动平均(ARMA)模型,用于挖掘数据中的随机性趋势;提出基于概率密度控制(PDF)的最小二乘支持向量机(LSSVM),用于挖掘数据中的确定性趋势,两种模型并联补集成实现对数据信息的充分挖掘。通过一组实验验证了所提方法的效果。

Aiming for the setbacks that a single algorithm can＇t dig all information in dealing with complex time-series data defects, the intelligent integrated architecture is proposed, providing four kinds of integration architecture, and analyzing their application. Time-series data for one category of random noise, utilizing series nested modeling structure, proposes Auto Regressive Moving Average model （ARMA） nested with double population particle swarm optimization algorithm for date mining, and figures out its stochastic trends; a probability density control based on support vector machine is provided, aimed to determine the trend of data mining, two categories of model of parallel compensation are set to implement the objective of thoroughly data mining, via a series of experiments that revealed the effectiveness of the proposed method.