Research And Application Of Feature Extraction Of Time Series Data Based On TCN

Time series data refers to special data objects with time series characteristics.A set of sequences is the data obtained by a series of sampling according to a specific time interval.Time series data has a wide range of sources,and also has the characteristics of large data volume,high data dimensions,and complex data structure.Compared with other data,this type of data is more difficult to research and faces more challenges.Traditional time series feature extraction methods usually require artificial design of features,which is not efficient and may cause features to be lost.In recent years,feature extraction based on deep learning has achieved excellent results.Through model training,features can be extracted effectively,and then tasks such as time series prediction,classification,and clustering can be achieved.Based on temporal convolutional network(TCN),this paper first proposes a temporal convolutional network with an attention mechanism,which combines causal convolution,dilation convolution,residual connection,and two-layer attention mechanism.The skip connection realizes the extraction of features with different levels of abstraction.In the experiment,the classification task verifies the model's feature extraction ability.Further,for more complex multi-dimensional time series data,a temporal convolutional network based on multi-feature fusion is proposed.Taking into account the differences and connections between different variables,the method first extracts the features of each sub-variable separately.Using the novel attention mechanism feature fusion method to integrate features,the overall process of the algorithm is designed as a unified end-to-end model.Compared with the current advanced algorithms,the method in this paper has higher classification accuracy on multiple public data sets.Finally,the method in this paper is applied to the problem of axle fault detection based on acoustic emission data.Compared with TCN,the method in this paper has a higher detection accuracy,which proves the algorithm's ability to identify axle faults.This paper uses the advantages of TCN,such as parallel computing,parameter sharing,and efficient time series memory ability.By introducing improved methods such as attention mechanism,the article enhances the temporal convolutional network's ability to perceive and recognize sequence data,and improves the performance of the model.