关键词: 时间序列;分割;聚类;遗传算法

Abstract:

The fitness value function in the algorithm proposed by  Vincent S. Tseng et al based on cluster-based genetic segmentation of time series with DWT is inadequate. Two points on the calculation of fitness value of each chromosome was proposed to improve this algorithm: data normalization was used to eliminate the influence of amplitude,and the interclass distance was introduced to make distance between classes distinct. Experiments were conducted to compare the  former and improved algorithm,and the results showed that these two improvements improved the fitness value function accuracy which was more beneficial to identify sequence patterns.

Key words:  time series; segmentation; clustering; genetic algorithm