<p>Multivariate Time Series (MTS) acquisition processes often exhibit irregularities, making accurate MTS forecasting challenging. Previous researches focused on interpolation approaches to address data completeness in irregular MTS, but these approaches may introduce noise, thereby altering the feature distributions of irregular MTS. Recent researches trend advocate embedding the missing temporal information through position encoding for forecasting irregular MTS. However, these position encodings were typically designed for text sequences and assumed fixed time intervals, which lead to the loss or distortion of temporal information when applied to irregular MTS. Moreover, they struggled to capture the temporal dynamic information in irregular MTS. To address these challenges, we propose a novel approach called TITD (Time Interval and Temporal Decay), which utilizes time interval and temporal decay information to enhance irregular MTS forecasting. TITD optimizes position encoding to effectively capture both local time interval features and long-term temporal decay patterns, breaking the limitations of static and fixed interval position encoding on time dynamic representation. Simultaneously, TITD integrates multi-view input information from irregular MTS to enhance the representation learning of the relationships across different views, thereby achieving superior forecasting performance without interpolation. Extensive experiments on three real-world time series datasets have demonstrated that TITD provides significant improvements over state-of-the-art methods in irregular MTS forecasting.</p>