Internet worms increasingly threaten the Internet hosts and services. More terribly, good point set scanning-based self-learning worms can reach a stupendous propagation speed in virtue of the non-uniform vulnerable-host distribution. In order to combat self-learning worms, this paper proposes an interaction model. Using the interaction model, we obtain the basic reproduction number. The impact of different parameters of predators is studied. Simulation results show that the performance of our proposed models is effective in combating such worms, in terms of decreasing the the number of hosts infected by the prey and reducing the prey propagation speed.