Deep time series missing data from self-attention-based inference model

Abstract

This study suggests a deep time series missing data inference model based on self-attention, known as AGRU-AE, to fill in the missing data in order to address the issue that data loss influences the impact of data analysis. In order to deal with various missing gaps and achieve the goal of concentrating on highly correlated sequences, the model combines a gated cyclic data unit (GRU) and an autoencoder (AE), adds a self-attention mechanism between the encoder and the decoder, and calculates the relationship weight between the partially generated sequence in the encoder and the partially known sequence in the decoder. The experimental findings demonstrate that the suggested model AGRU-AE, when compared to the conventional approach, can fill and predict the incomplete time series with more accuracy.