In usual, the flow of a football game have different phase, and change from one to another, and the coach is due to observe them, understand and solve the tasks in the game by using appropriate structural strategies. Therefore, it is a critical issues for a coach to decide what kind of structural strategies have been effective for their own team. Therefore, we propose 3 different views to help to the coach to make decisions. First of views, we formulate the passing ball path as a network (passing net-work. More specific, we utilize clustering coeffcient to determine the relations between players. It turnouts that core player will have a strong cluster ability. And our propose network focus not only on local network, but global passing relations.-Final of views, we propose a novel reinforcement learning based Graph-to-Graph framework to decide structure of team. We formulate the positions of players as a graph, and we use the current graph as input, while our deigns return award will effect the structure of team by change the positions step by step. In experiment, we simulate the result of our team versus 3 different level team [1]