A study was conducted on the error transmission and accuracy prediction methods during the assembly process of aero-engine rotors, and a digital twin-driven method for dynamic prediction and optimization of rotor assembly accuracy was proposed. This method divides the rotor assembly process into two spaces, physical and virtual, as well as two levels of parts and components. It establishes an assembly deviation characterization model and an assembly accuracy dynamic prediction and optimization model, with the core being the dynamic prediction and optimization technology that integrates virtual and real twin data. It can dynamically monitor any assembly node of the rotor and optimize the process parameters of post-assembly nodes. The experimental results show that this method has a prediction deviation of less than 6% for rotor centroid eccentricity, with an average deviation of 3.61%, and a prediction deviation of less than 5° for eccentricity angle, with an average deviation of 2.94°. The coaxiality of the rotor after optimized assembly using this method is improved by about 16% compared to traditional methods, effectively improving the assembly accuracy of aero-engine rotors and providing theoretical support for the optimization and control of aeroengine rotor assembly processes.