In　this　paper　we　have　studied　a　simple　multi-agents　firm　network　system　which　is　a　competition　simulation　model,　and　it　shows　that　simulating　the　complicated　firm　competition　network　is　feasible.　Because　of　the　complexity　of　the　game　evolutionary　problem,　a　kind　of　learning　classifier　systems　based　on　genetic　programming　is　introduced　instead　of　genetic　algorithms.　The　basic　purpose　to　build　the　model　are　two　aspects:　study　the　evolution　process　of　firm　competition　network　modeled　in　this　paper　and　try　to　compare　the　experiment　results　with　the　firm　behavior　in　real　economy　systems.