Generalized rule accumulation based on Genetic Network Programming considering different population size and rule length

Most evolutionary computation methods such as GA, GP, EP, ES, etc. mainly focus on obtaining the best solution, namely, the elite individual with the optimal gene structure. In this case, only the final result is taken into account rather than the evolutionary process. We noticed that some good experiences generated during the evolutionary period are also valuable for guiding the evolution or directing agent´s actions. This paper concentrates on how to accumulate evolutionary experiences and guide agent´s actions by extracting and using generalized rules based on Genetic Network Programming(GNP), which is a newly developed evolutionary computation method. Each generalized rule is a judgment-action chain which contains the past information and the current information. These generalized rules are accumulated and updated in the evolutionary period and stored in the rule pool which serves as an experience set for guiding new agent´s actions. We extract rules from elite individuals in different generations and how the rule length affects the performance is studied in this paper. Tile-world problem which is a good benchmark for multi-agent systems was chosen as the simulation environment. The effectiveness of the proposed method was verified by the simulation results.