A two-level ALife system with predator

This paper proposes a method of deriving a global solution for a nonlinear optimization problem with a multipeak function as the objective function. When the objective function in the nonlinear optimization problem is differentiable, a gradient method such as the quasi-Newton method can be used to derive the solution with high speed. When the function is multipeaked, however, there is a high probability that the procedure may fall in a local solution. Methods that can derive a global solution for a multipeaked function without using a gradient include simulated annealing (SA) and tree annealing (TA), in which the probability of arriving at a global solution is increased by including probabilistic operations in the mechanism of search for the solution. A problem in these methods, however, is that considerable computation time is required. In this study, artificial life (ALife) is implemented on a computer in order to realize global optimization by a distributed optimization technique. Nonlinear optimization is considered, the ALife technique is extended to a multidimensional continuous system, and an evaluation function is proposed so that emergent colonization is generated in a promising area including the global solution. In this study, a predator is introduced in order to reduce the time until the emergent colonization is generated. This situation is supported by a rapid decrease in the variation of the Artorg, which is a form of ALife. © 2004 Wiley Periodicals, Inc. Electron Comm Jpn Pt 2, 87(8): 53–60, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjb.20108