With the successive commissioning of large-capacity DC lines, it is increasingly urgent to improve the cascading failure prediction model for AC-DC systems. In view of the lack of dynamic characteristic simulation in current research on the AC-DC interaction process, a cascading failure prediction method for AC-DC systems based on multi-time scale is proposed. First, the cascading failures are divided into three time-scale processes and corresponding simulation models are established according to different time scales. Among them, DC commutation failure, temporary stability and severe overload of the AC line are considered in the short time scale process. The AC line general overload and heavy load interruption are considered in the long time scale process (long time scale process includes short time scale process). Load fluctuations are considered in the extremely long time scale process (extremely long time scale process includes long time scale process). Secondly, the IEEE39 node system is taken as an example and then the cascading failure process formed by using this prediction method is analyzed and compared with the cascading failure process formed by other methods. The superiority of the prediction method is mainly demonstrated from time-scale characteristics and consistency with the actual situation. Finally, the average failure risk of different control measures at each stage of the accident chain and the predicted path risk with or without DC modulation are analyzed. The simulation results show that different control measures have different risks at different stages and that DC modulation can be introduced into the control measures to reduce the cost of control. This work is supported by the Natural Science Foundation of Anhui Province (No. 1808085ME142).