A novel method for similarity search over electric time series data

This paper proposes a novel similarity search method based on wavelet packet and average approximation for electric time series data. Our method consists of two stages, in the first preprocessing procedure, wavelet packet is applied to electric time series, and average approximation is used to extract feature vectors from decomposed coefficients, then multidimensional index structure is built using these feature vectors; in the second querying procedure, the correct answer sets are obtained by candidate selection and post processing. The method makes use of the properties of multidimensional and multi-scaling decomposition of wavelet packet. Experimental results on real electric time series data show that the proposed method outperforms the conventional method and is promising for electric time series similarity search.