Distributed Optimization by Using Artificial Life
Daisaku Hayashi, Taiji Satoh, Tsuyoshi Okita · IEEJ Transactions on Electronics Information and Systems · 1996
This paper presents a distributed algorithm for minimizing a nonconvex multimodal function. The local search methods based on the gradient information have efficient convergence property, but these methods have a tendency to get stuck at a local minima. Therefore a global optimization method with good convergence property is desired.Although many global optimization methods such as Simulated Annealing and Tree Annealing have been proposed, these methods require much computational cost. In recent years, new distributed algorithms based on Artificial Life (ALife) is studied and its potential power is reported. In this paper, therefore, the frame work of ALife is employed into a function minimization. Since the proposed method utilizes no gradient information, it can be applied to very wide class of problems.In this paper, firstly, ALife in the two-dimensional discrete system is realized. Then it is extended to the multi-dimensional continuous system. Next, ALife is applied to an optimization problem of function minimization. The effectiveness of the proposed method is demonstrated through some numerical tests on multimodal test functions. The numerical tests also show that the proposed method is superior to the Simplex method and Tree Annealing method.