In the cutting-edge field of smart manufacturing, accurately predicting the remaining useful life (RUL) of intelligent devices plays a crucial role in enhancing production efficiency and ensuring equipment safety. Digital Twin (DT) represents an emergent technology in equipment health management, where high-fidelity digital twin models facilitate the reflection of device operational states, and dynamically updated data aids in the precise prediction of RUL. This paper introduces a DT-based framework for the intelligent prediction of equipment RUL, utilizing a high-fidelity digital twin system to comprehensively capture the operational data of devices, enabling extensive and multi-level monitoring of device operational states. Building upon this foundation, a RUL prediction model (MSCPS) incorporating Multi-Scale Convolution (MSC) and ProSparse Self-Attention is proposed, significantly enhancing the extraction of key features and thereby improving RUL prediction accuracy. Furthermore, through the implementation of a transfer learning strategy supported by the digital twin system, this study successfully addresses the challenge of data scarcity in the target domain, achieving high-accuracy RUL prediction under conditions of limited data. Extensive experiments conducted on two full-lifecycle bearing datasets validate the effectiveness of the proposed method, with results demonstrating its superiority in RUL prediction compared to existing data-driven technologies. This research not only provides a new perspective for equipment health monitoring and management but also lays a solid foundation for the advancement of health diagnosis and prediction technologies in intelligent systems, indicating new directions for future research.