Time Series Forecasting for Charging Stations via Joint Learning of Missing Data and Temporal Dynamics

Time series forecasting plays a crucial role in various real-world applications. However, the pervasive missing data caused by sensor failures, communication interruptions, and system malfunctions poses significant challenges to accurate forecasting. Existing forecasting methods that rely on imputation techniques often struggle to effectively preserve temporal dependencies and capture underlying patterns of missing data, thereby compromising forecasting accuracy and robustness. To address this issue, we propose a novel dual-stage framework that jointly learns missing data patterns and time series dynamics. It consists of (1) a pattern-aware encoder that captures missing value distributions and (2) a dual-forecasting module to enhance forecasting accuracy. Experimental results on real-world electric power data from charging stations demonstrate that our approach outperforms several baseline models, achieving superior forecasting performance under missing data conditions.