Research On Representation And Similarity Measurement Of ECG Time Series

The World Health Organization has revealed that 30 percent of the death is due to cardiovascular disease. It has a very important significance for the diagnosis of cardiovascular disease. Electrocardiogram is a graphical display of cardiac physiological activity signals. Traditional ECG automatic recognition system is based on the characteristics of the extracted waveform, and then uses the mature machine learning algorithm to classify. The method has two major problems: first, it requires the development has a knowledge of medical literacy; second, it needs a large number of data to train and obtain the model, which needs to maintain regularly. In the paper, we hope to find the most similar k sequences of the query sequence in the database without medical knowledge background, and then use the KNN classifier to determine the category of the query sequence. However, due to its complexity and high dimension, the existing methods can not measure their similarity quickly and effectively.The paper studies the representation and similarity measure ment of time series, and proposes the piecewise linear representation method of IPDT_PLR and Seg Mode_DTW similarity measure algorithm respectively.IPDT_PLR method is based on the ITTP_PLR, It inherits the advantages of the ITTP_PLR method to retain the important trend turning point and IPDT_PLR can effectively solve the problems of ITTP_PLR. The IPDT_PLR method overcomes the poor performance in the monotone trend sequence of ITTP_PLR by adding the inflection point information. It can better pick out the important points in the sequence through the top-down method. The fitting error can be further reduced by modifying the segmentation criteria.In the research of similarity measure, Seg Mode_DTW algorithm proposed in this paper improves the original DTW algorithm from three aspects. First, it reduces the searching range and improves the time efficiency by the sub pattern matching. Second, it introduces the concept of line segment mode distance in DTW framework, which effectively overcomes the shortcoming of the DTW method without considering the tending of sample points. Third,it proposes the sub pattern distance based on DTW algorithm, which ensures the alignment of the two strong feature points, so that the trend of each segment is more clear, and the error matching of DTW is reduced effectively. At the same time, it also speeds up the measurement. Experiments show that the Seg Mode_DTW algorithm can be widely and efficiently applied to the similarity query of ECG time series data sets.