Heuristics for Multi-Population Cultural Algorithm

Cultural Algorithm (CA) is one of the Evolutionary Algorithms (EAs) which de- rives from the cultural evolution process in nature. As an extended version of the CA, the Multi-population Cultural Algorithm (MPCA) has multiple population spaces. Since the evolutionary information can be exchanged among the sub-populations, the MPCA can obtain better results than the CA in optimization problems. In this thesis, we introduce heuristics to improve the MPCA. The heuristic strate- gies target the existing weaknesses in MPCAs. Four strategies are developed address- ing these weaknesses, including the individual memory heuristic, the social interaction heuristic, the dynamic knowledge migration interval heuristic and the population dis- persion based knowledge migration interval heuristic.Five standard benchmark opti- mization functions with di erent characteristics are taken to test the e ciency of the heuristics. Simulation results show that each heuristic, to varying degrees, improves the MPCA in convergence speed, stability and precision. We compared di erent combinations of the strategies, and the results show that the MPCAs with social interaction based knowledge selection, as well as dynamic knowledge migration inter- val/population dispersion based knowledge migration interval, outperform the other combinations in both low-dimension functions and high-dimension functions