Abstract: Robocup is an international academic competition which focuses on artificial intelligence and robotics. The 2D simulation is one of the earliest and most influential projects in Robocup. Attacking is the core behaviour of the simulated football game, as well as the attack recognition is considered as an important part in team-confrontations. This paper selects some active and contribution index of attacking, extracts lots of attacking behaviour data of the key agents, proposes two kinds of attacking patterns of 2D simulation, as ‘separate attack’ and ‘cooperative attack’, according to the human-player actions. The following simulation tests give the accuracy of ‘separate attack’ behaviour detection as 93%, while the ‘cooperative attack’ behaviour detection as 80%, which proves the existence of two attacking behaviour model and the rationality of the key parameters threshold setting.

Key words: Robocup, simulation, model, data mining, behaviour detection