Multi-Attention Generative Adversarial Network for Multivariate Time Series Prediction

Multivariate Time series data play important roles in our daily life. How to use these the process of prediction is a highly attractive study for many researchers. To achieve this goal, paper, we present novel multivariate time method based on multi-attention generative adversarial network. This includes three phases explore prediction. Firstly, encoder stage consists two modules, from which input-attention and self-attention can encode exogenous sequence into latent space. Secondly, decoder temporal-convolution-attention module, extract long-term temporal patterns. solve problem low accuracy prediction, inspired by weight clipping method, design an improved discrimination network finally. The experiment results indicate that mechanism useful improve performance We also tested extensive empirical studies with five real world datasets (NASDAQ100, SML2010, Energy, EEG Air Quality) demonstrate effectiveness robustness proposed approach.