Due to the issues of significant power consumption and extra hops from source to destination in the Dragonfly topology of network-on-chip (NoC) system, it becomes increasingly important to develop a power-aware routing algorithm for the Dragonfly topology. We propose a dynamic adaptive threshold routing algorithm (DATRA) by applying dynamic adaptive threshold and its own dynamic threshold step to balance the load of network on chip and achieve lower power-delay-product. In order to evaluate DATRA, we propose an evaluation model by extending terms Ptotal, Lavg, VAR(load), and Havg, etc. The experimental results verified that DATRA can respectively save 6.53% (n=4) and 5.93% (n=8) on power-delay product on average by comparing to the UGAL-LVC-H, which implies that DATRA is a promising alternative strategy of routing algorithm for the Dragonfly topology.