Similar Pattern-matching Algorithm of Time Series Based on Improved Empirical Mode Decomposition Method
Huiting Liu · Jisuanji fangzhen · 2011
Overshoot and undershoot problems will occur during the course of obtaining envelopes of time series with spline interpolation.If these problems can not be solved properly,redundant intrinsic mode functions(IMF) will be produced when a time series is decomposed by empirical mode decomposition(EMD),and precision of EMD will become lower.To ameliorate EMD algorithm,an effective method was proposed,which used the means of successive extrema instead of the envelope mean to obtain the mean envelope.In this way,additional boundary and interior data points were created,and only one spline interpolation was required rather than two in each loop of the sifting process.Time complexity was reduced,overshoot and undershoot problems were alleviated,and EMD method was improved.Then dimensionality of time series with the improved EMD technique was reduced,and pattern matching was realized using K-means algorithm.At first,trend series were clustered,which were decomposed by EMD method.And then,accurate similar series patterns were reached by calculating distance of the clustered series in the category the trend of the query belongs to.Experimental results show performance of the new method,based on the improved EMD,is better than that of the wavelet-based pattern matching method.