Towards an Agent-Based Architecture using Deep Reinforcement Learning for Intelligent Internet of Things Applications