In traditional traffic signal coordination control, all intersections are shared a common cycle and have fixed offsets in one sub-area. The range of sub-area is normally fixed, plan switching is difficult, and the reaction to the traffic problem is slow. In this paper, a self-organized dynamic control strategy is proposed. it decomposes the network optimization into overlapped and interactive problems between Basic Coordination Units (BCU). It presents an interaction-forecast model among intersections, each intersection not only considers its own benefit, but also takes downstream intersections into account. Multi-scenarios simulation shows it works well in all saturation degrees, reduces total delay around 8% to 12.3%.