Shapley Value Based Multi-Agent Reinforcement Learning: Theory, Method and Its Application to Energy Network