Reward Shaping via Diffusion Process in Reinforcement Learning