A New Learning Model for Swarm Intelligence Based on Q-Learning

Inspired by cooperative transport behaviors of ants, on the basis of Q-learning, a new learning method, Neighbors' Discounted Information (NDI) learning method, is present in the paper. This is a swarm-based learning method, in which principles of swarm intelligence are strictly complied with. In NDI learning, the i-interval neighbor's information, namely its discounted reward, is referenced when an individual selects the next state, so that it can make the best decision in a computable local neighborhood. In application, different policies of NDI learning are recommended by controlling the parameters according to time-relativity of concrete tasks. By applying this learning method, the cooperative transport of ants is simulated. Experiment results show that the transport process in simulation is very similar to the phenomenon in natural world, which proves the designed learning mechanism's rationality.