Strategy system of small-size RoboCup robot is a Multi-Agent System of coordination and control . In order to solve the delay problem of small-size RoboCup competition system, this paper applies BP Neural Network to the situation prediction of strategy system. A linear prediction model based on BP Neural Network is established and Neural Network topology is ascertained. Then trained network is applied to the existing competition system and the efficiency in position, coordination and cooperation capability are greatly improved. The experiments show that the method can predict position and direction of robot correctly, thus proving the feasibility and superiority of prediction algorithm in the system.