Aiming at the complex and changeable flight operation environment， a dynamic prediction model of flight chain delay based on digital twin is proposed to improve the accuracy and adaptability of the traditional prediction methods. The model is constructed based on the digital twin flight chain system. Through the model， the unit-level delay prediction is accomplished by the multi-channel feature modeling method under the sliding windows， and the parameter optimization is completed by a hybrid optimization strategy. Moreover， the whole chain’s delays are analyzed and corrected with the help of the twin data-driven strategy. Flight data in China are used to conduct the study. The mean absolute error （MAE） of flight delay prediction is 11.79 min， which is lower than those of other baseline models or static models， and the forecast error of subsequent flight delay can be reduced by 6.44% after twin data-driven analysis is implemented. The results show that the model is beneficial for the digital twin system to realize the virtual reality interaction， and has excellent prediction accuracy and adaptability.