【摘要】 研究混沌时间序列预测问题。混沌时间序列同时具有确定性与随机性,传统预测方法精度低,为了提高预测精度,提出基于支持向量机的局域预测法。使用基于演化跟踪的邻近相点选取方法,代替欧几里德距离法,能够有效辨别并剔除伪邻近点。采用基于Hannan-Quinn定阶准则的邻点个数确定法,解决传统方法凭主观经验或多次试验确定邻点数量的不足。最后以Lorenz混沌时间序列为例进行实验分析,结果表明支持向量机方法在预测精度、速率、可预测步数等多个指标上都比传统方法具有更好的性能。更多还原

【Abstract】 This paper made a study on chaotic time series prediction. Chaotic time series have both deterministic and random nature. A local prediction algorithm based on support vector machine was proposed to achieve higher accuracy than traditional prediction methods. A new neighboring phase points selection method based on evolution-tracking instead of the Euclidean distance was used to effectively distinguish and delete pseudo neighboring points. A method based on Hannan- Quinn criterion was adopted to determine the number of neighboring points,instead of based on subjective experience or repeated experiments. An experiment on Lorenz chaotic system was completed and the result shows that the proposed algorithm has better performance in terms of prediction accuracy,rate and predictable steps.更多还原