A Deep Q-Network Method Based on Upper Confidence Bound Experience Sampling