Improving the Effectiveness of Potential-Based Reward Shaping in Reinforcement Learning