Abstract:
The development of theory and applications of multi-agent systems
determined in the last years a real revolution regarding the modeling of
complex systems. The structure of any agent-based computational model
contains the next elements: parameters, variables, algorithms and agents. We
make a comparative analysis of three multi-agent computational algorithms
(RAND, NMEM and WMEM) used to harvest grain in a bi-dimensional
lattice. The first algorithm RAND is the simplest because the agent sets
randomly a certain harvesting direction and harvests all the grain he can
find. The second algorithm NMEM uses 8 searching directions. From these 8
variants, the agent selects the alternative that gives the maximum amount of
grain. The searching is repeated for vision times. If this search fails to find
any grain than the agent pass to RAND algorithm. The third algorithm
WMEM uses the same searching algorithm as NMEM algorithm, but this one
memorizes the patches that the agents have discovered. When this searching
algorithm fails to find grain, than the agent uses its memory to find the
nearest patch with available grain. The algorithms are implemented using
NetLogo. This software platform was designed by Uri Wilensky it in the year
1999. NetLogo is in a process of development and modernization in the frame
of Center for Connected Learning and Computer-Based Modeling -
Northwestern University, Illinois, USA. NetLogo is written in Java language
and can be run on all major platforms (Mac, Windows, Linux etc.). In
addition, individual models can be run as Java applets inside web pages. We
did three computational experiments and we observe that the best results are
obtained when we used WMEM algorithm. In this case, the grain was
harvested in a period of 1141 simulation steps. On the second place was
NMEM algorithm with a harvesting period of 6982 simulation steps and on
the last place was RAND algorithm with a harvesting period of 18183
simulation steps.
