Floating point genetic algorithms for nonconvex nonlinear programming problems: Revised GENOCOP III

In this paper, we focus on the nonconvex nonlinear programming problem and try to perform optimization using a genetic algorithm that can perform the multipoint parallel search. For such type of nonlinear optimization of cons trained problems, Michalewicz and colleagues have recently proposed GENOCOP LII, where individual representation using floating point is adopted along with two groups of constraints; one is the reference group where individua ls satisfy all of the constraints and the other is the search group where i ndividuals satisfy only linear constraints. However, in GENOCOP III, where initial populations are randomly generated, it is very difficult to determi ne at least one reference point that satisfies all of the constraints. More over, since a new search point is randomly generated on the line joining th e search point and the reference point, some problems are encountered that are related to efficient search and processing speed. In this paper, to res olve the problems of GENOCOP III, we propose a method of efficient location of initial reference point by solving the optimization problem where sum o f squares of violated nonlinear constraints is used as the objective functi on. Moreover, the method of search of feasible solution by bisection is pro posed. Finally, the effectiveness and validity of the proposed method are s hown. (C) 2000 Scripta Technica, Electron Comm Jpn Pt 3, 83(8): 1-9, 2000.