In this paper, we develop a novel incremental learning scheme for reinforcement learning (RL) in dynamic environments, where the reward functions may change over time instead of being static. The proposed incremental learning scheme aims at automatically adjusting the optimal policy in order to adapt to the ever-changing environment. We evaluate the proposed scheme on a classical maze navigation problem and an intelligent warehouse system in simulated dynamic environments. Simulation results show that the proposed scheme can greatly improve the adaptability and applicability of RL in dynamic environments compared to several other direct methods.