Learning Dynamics Model in Reinforcement Learning by Incorporating the Long Term Future