MSRN-Informer: Time Series Prediction Model Based on Multi-Scale Residual Network
Xiaohui Wang, M. Xia, Weiwei Deng
Abstract
Time series is a huge quantity of data related to time sequence in real life and its forecast remains challenging. In this study, we propose a deep learning model to enhance the precision of time series forecast, called MSRN-Informer (Multi-scale Residual Network Improved Informer) model. This model can reduce the waste of significant resources and overfitting caused by increasing the depth of the network in traditional improvement methods. A multi-scale structure is added in Informer model to extract data features of different scales, and a residual network is applied to reduce data loss. To prove the effectiveness of the presented MSRN-Informer model, it is compared with Informer, Informer+ and ARIMA methods on four datasets. The results show that MSRN-Informer has a better prediction ability and show a reduced error. The research findings of this paper can be potentially used as reliable reference and basis for effective time series prediction.