弹性粗粒度动态弯曲时序相似性算法

针对动态时间弯曲(DTW)算法在提高计算速度同时不能兼顾分类正确率的问题,提出了一种基于朴素粒计算思想的弹性粗粒度动态时间弯曲(CG-DTW)算法。首先,通过计算时序方差特征的方法来获取较优的时序粒度,用粒度特征代替原始序列;其次,再代入执行DTW算法,允许动态调整被比较时序粒间的弹性大小,从而获得相对最优的时序对应粒;最后,在对应最优粒的情况下计算DTW距离。同时引入下界函数的提前终止策略进一步提高CG-DTW算法效率。实验结果表明,所提算法要比经典算法运行速率提高21.4%左右,比降维策略算法正确率提高近32.3个百分点,尤其是长序列的分类,CG-DTW能够在保持正确率的情况下兼顾较高的运行效率。CG-DTW在实际应用中能适应不确定长序列分类。

Temporal similarity algorithm of coarse-granularity based dynamic time warping

The Dynamic Time Warping (DTW) algorithm cannot keep high classification accuracy while improving the computation speed. In order to solve the problem, a Coarse-Granularity based Dynamic Time Warping (CG-DTW) algorithm based on the idea of naive granular computing was proposed. First of all, the better temporal granularities were obtained by computing temporal variance features, and the original series were replaced by granularity features. Then, the relatively optimal corresponding temporal granularity was obtained by executing DTW with dynamically adjusting intergranular elasticity of granularities compared. Finally, the DTW distance was calculated in the case of the corresponding optimal granularity. During this progress, an early termination strategy of lower bound function was introduced for further improving the CG-DTW algorithm efficiency. The experimental results show that, the proposed algorithm was better than classical algorithm in running rate with increasing by about 21.4%, and better than dimension reduction strategy algorithm in accuracy with increasing by about 32.3 percentage points.Especially for the long time sequences classification, CG-DTW takes consideration into both high computing speed and better classification accuracy. In actual applications, CG-DTW can adapt to long time sequences classification with uncertain length.