To solve the contradiction between resource utilization and resource waste caused by congestion in the dual-connection offloading process, a dynamic programming algorithm with congestion prediction is proposed. Considering the high intermittent situation of millimeter wave links and fairness between heterogeneous network links under the constraint of total power, taking system weighting and rate maximization as the optimization goal, the non-convex problem is modeled as a finite-horizon discrete-time domain Markov decision process. The proposed algorithm is used to solve the power allocation problem in parallel connection of microwave and millimeter wave. The simulation results show that the algorithm can significantly improve the system performance by learning the power allocation strategy.