Dynamic Attention-Gated Transformer Network Model for Multi-Time-Step Time Series Forecasting

Time series forecasting is a critical task in many fields, with wide applications in industries such as energy, retail, and finance. Traditional time series forecasting methods often fail to effectively capture complex nonlinear patterns and interdependencies among variables. In recent years, deep learning techniques have made significant progress in time series forecasting. This paper proposes the DAGNet model to address the need for multi-time-step time series forecasting. DAGNet effectively handles various input types, including static covariates, prior known inputs, and observed inputs, through a dynamic attention mechanism and gating mechanism. Meanwhile, DAGNet also enhances the interpretability of forecasts, enabling users to identify globally important variables and duration patterns. Experimental results show that DAGNet outperforms state-of-the-art (SOTA) models in forecasting performance.