Q-Learning algorithm for Particle Swarm Optimization

Our developers have implemented the code for simulation of the multi-agent optimization process in the NetLogo environment.

Algorithm is an enhanced version of the particle swarm optimization method where in addition to selection of the appropriate movement velocity in the problem search space, each agent also learns an optimal strategy for internal parameter selection based on his current position.

These parameters include maximum and minimum velocity values, the size of the neighborhood taken into account, trade-off between the effect of global optimum found so far and optimum value in the neighborhood etc.

The algorithm was implemented based on the description in the paper “Intelligent Particle Swarm Optimization Using Q-Learning” by M. Khajenejad, F. Afshinmanesh, A. Mar and B. Araabi. For all the required parameters, the corresponding graphical interface controls was prepared. Code was commented and documented.