Predictive maintenance is considered to be an effective strategy to optimize system operation. In the execution of increasingly complex tasks, efficient and intelligent management becomes crucial. As the basis of Digital twin (DT), predictive capabilities contribute to the value of systems and help describe their complex behavior. But the challenge in Digital twin model building is still exist, it cannot accurately reproduce the physical resources, and the introduction of error will lead to the differential extension of virtual system from physical space. The challenge is how to build Digital twin capabilities and reduce accumulative error at the same time. Based on the current research progress, this paper analyzed the existing challenges in realizing predictive maintenance capability driven by Digital twin, and then, it described the predictive control process with flow path and layer framework, In addition, the way of inserting the optimization algorithm for Digital twin was explored. Finally, a practical trajectory prediction problem was taken as a case study to effectively utilize the cyclic interaction mechanism and data fusion method of Digital twin, which can consider the offset cumulative signal, and correct the prediction state in real time. This research may provide the reference for Digital twin configuration and further study.