“…A large variety of deep learning modeling approaches for time series analysis have been exploited for a wide range of tasks, such as forecasting, regression, and classification [5, 9, 14 15, 36]. The most common established deep learning models in this area are convolutional neural network (CNN) [13,42,43], recurrent neural networks (RNN) [5,7,8], and attention-based neural networks [10,11,14,15,16,34]. Since CNN-based models can only learn local neighborhood features, recently, RNN-based models and attention-based models which can learn long-range dependencies are increasingly popular for learning from time series data [5].…”