Research And Implementation Of Time Series Classification Algorithm Based On Deep Learning

Time series data are widely found in people's daily life.Time series data usually has a large amount,high feature latitude and updates fast.Time series classification is a branch in the field of time series data mining,which has great research value and significance of application.At present,the main problems of time series classification are as follows:The traditional classifier is ineffective in dealing with such problems,for the large temporal dimension of time series.To achieve more accurate classification results the feature selection and classi-fication algorithm often require a large amount of computation and time,which is difficult to apply to the actual with poor real-time.need to be costly In addition to the exact clas-sification of the results,In addition to accurate classification results,the explanatory of the classification algorithm is also an important index of investigation,the existing classification algorithm is still lack of strong explanatoryThe thesis focuses on the existing research based on time series classification and solves the above problems.The main contributions are as follows:Based on the theory of artificial neural network and deep learning,time series classification algorithm and models are estab-lished,including multi-layer perceptron time series classification model,self-normalized full convolutional time series classification model and memory full convolution network time series classification model of three models of algorithms,and compared with the traditional machine learning method by experimental verification on the classic data set,the results show that the algorithms and models the theseis proposed have higer classification accuracy than the traditional time series algorithms.The result of multi-layer percetron model reaches the baseline of traditional algorithm's classification accuracy and the latter two are proved to have higer accuracy than existing methods in multiple datasets.At last the classification results of the latter two models are visualized by visualization method of neural network,which proves that the classification result of the algorithm model has some explanatory.