Multiple Moving Preys Cooperative Pursuit (MMPCP) problem was discussed where exists multiple random moving preys, each kind of which need different number of pursuers to circle to capture. Following the idea of "ally first, capture later", the pursuer first allies itself to others with same will and then capture the selected prey cooperatively. But in the existing algorithms the alliance, which was formed using a simple algorithm that consider the prey value only, cannot be broken once it has been established no matter how difficult the capture turn out to be and how much time has been consumed. For this problem, the concepts of assistant decision matrix, alliance cohesion, and fine are introduced. Based on these concepts, a changeable alliance based cooperative capture algorithm was proposed that improved both alliance-forming procedure and capturing procedure by allowing pursuer secede from a low-reward alliance to form other possible high-reward one. Simulation results show the feasibility and validity of the given algorithm.