Abstract: This paper proposes a time series segmentation approach by combining the advantages of hierarchical temporal memory (HTM) model and the pattern-based genetic time series segmentation algorithm. The approach is a HTM-based genetic algorithm which replaces the pattern-based fitness value function with the HTM-based fitness value function. The approach can be applied to find segments from a time series and to identify the class of sub series. In addition, a pattern-based algorithm of HTM sample generation is proposed for generating sample set with HTM trait. Experimental results show that the two algorithms are effective on the time series of stocks.


Key words: time series, segmentation, HTM（Hierarchical Temporal Memory）, genetic algorithm