基于小波变换的时间序列相似模式匹配

提出了一种新的时序相似模式匹配方法，它采用小波分析的方法实现时间序列数据的降维，采用小波序列表示原序列，将小波序列组织为多维索引结构R-tree存储，在该索引结构基础上，基于一种表示相似性的距离函数，定义了范围查询和最近邻查询算法，实验结果证明这种方法性能优于传统的基于傅立叶变换的相似模式匹配方法。

Time Series Similar Pattern Matching Based on Wavelet Transform

Similar pattern matching of sequence is an important field in time series data mining. Since time series may be a very long sequence, which results in query performance decreasing sharply when the database is large, therefore, dimension reduction is required before pattern matching. Fourier transform can be used for dimension reduction. But traditional Fourier based matching techniques reflect only the global frequency feature of signals, can't provide any feature in local time interval. While wavelets technique represents the sequence signals from both time field and frequency field, and has multi resolution, shift invariant property, etc. Based on these properties of wavelets methods, this paper proposes a new time series similar pattern matching method. It reduces the dimensionality of time series data with discrete wavelet transform technique, and stores the reduced sequences in a multi dimensional index structure, such as R* tree. The similarity function used this paper is based on Euclidean distance definition, which preserves the distance of two reduced sequences smaller than that in the original space. Based on this property of distance function, this paper proposes the range query algorithm and nearest neighbor query algorithm on the multi dimensional index structure. The experiments show that the performance of this new method has been improved over the Fourier based pattern matching method.