Application Research Of Multiple Methods Integration Feature Extraction In Different Time Series Classification

Time series data exists widely in many important application fields.Most of time series problems can be attributed to time series data mining problems,and time series classification is one of the important research contents in time series data mining.One of the most important indexes to evaluate classification performance is classification accuracy.There are two ways to improve classification accuracy:one is to improve the classifier,and the other is to improve the feature extraction method of time series classification.Aiming at time series classification problem,a multiple methods integration feature extraction of time series classification is proposed to improve classification accuracy by improving feature extraction method in this paper.The performance of proposed method is verified by several simulation experiments.And it is applied to time series classification in different fields and improves the accuracy of time series classification.On the basis of analyzing the characteristics of time series data,a feature extraction method of time series classification combining wavelet,fractal and statistical methods is proposed.First of all,the original time series is de-noised by using wavelet transform,the de-noised and reconstructed time series is decomposed and the average high frequency coefficients in each scale space are calculated to constitute the feature vectors as the first part of time series classification features.Secondly,the multi-fractal spectrum of the de-noised and reconstructed time series at multiple scales is analyzed,and according to the characteristics of specific time series data and classification need,the relevant parameters of multi-fractal spectrum are extracted as the second part of time series classification features.And then according to different characteristics of time series data,the relevant statistical characteristics of time series are extracted as the third part of time series classification features.Finally,combining the characteristics of time series and experimental results,the extracted features by using the above method are analyzed,and the final classification features of time series are identified.Then,the performance of proposed method is verified by simulation experiments of time series classification data in different fields.By comparison with other feature extraction methods from multi-angle,the effectiveness and superiority of proposed method are demonstrated using Japanese Vowels data and Synthetic Control data from UCI dataset.Finally,the application of multiple methods integration feature extraction in different time series classification is researched.The proposed method is applied to different time series problems,and the practicability and superiority of proposed method in different time series classification problems are verified.Applying the multiple methods integration feature extraction to EEG signals classification,the EEG classification problem of Colorado released is researched by comparison with the previous classification results of feature extraction using the same data,it shows that the classification accuracy of proposed method is significantly higher than that of the previous method.At the same time,compared with the classification results of proposed method and the other two methods combination,the results show that the classification results of EEG using the multiple methods integration in all classification are higher than that of the combination of other two feature extraction methods.Then the method is applied to the hand movements sEMG signals classification.The classification for time series data of six basic hand movements sEMG signals corresponding to 6 categories of two channels is achieved.The classification results are compared and analyzed from multiple perspectives and different levels.The results show that the proposed method not only can achieve higher classification accuracy,but also get good classification results in each movement classification.The applications of proposed method in different time series classification are made a preliminary exploration.