The route prediction for multiple unmanned aerial vehicles (UAVs) task allocation in dynamic battlefield is a complicated path planning problem characterized by multiple paths and real time demands. Probabilistic Road Maps (PRM) method wais used to plan combinatorial predictive routes for multiple UAVs and multiple tasks in this paper. The route predicting process was split into two phases: the off-line learning phase and the on-line query phase. The paper presented Cost transform based PRM (CTPRM) algorithm, which transforms the influence of enemy threats to the risk cost of the route segments among roadmaps. When battlefield situation was changed, without reconstructing roadmaps, CTPRM could plan new predictive routes rapidly by updating the risk cost of some route segments. According to different planning condition, relevant sampling strategy could be set flexibly, so that the planning time and route quality could be coordinated to achieve tactical goal. Simulation results demonstrate the feasibility of the approach.