MICHAEL A. ZMUDA
Department of Computer Science and Systems Analysis, Miami University, Oxford, OH 45056, USA
Department of Computer Science and Engineering, Wright State University Dayton, OH 45435, USA
Department of Computer Science and Engineering, Wright State University Dayton, OH 45435, USA
MATEEN M. RIZKI
Department of Computer Science and Systems Analysis, Miami University, Oxford, OH 45056, USA
Department of Computer Science and Engineering, Wright State University Dayton, OH 45435, USA
Department of Computer Science and Engineering, Wright State University Dayton, OH 45435, USA
LOUIS A. TAMBURINO
Department of Computer Science and Systems Analysis, Miami University, Oxford, OH 45056, USA
Department of Computer Science and Engineering, Wright State University Dayton, OH 45435, USA
Department of Computer Science and Engineering, Wright State University Dayton, OH 45435, USA

Received 1 September 2003
Revised 7 October 2003

A new self-adaptive mutation operator, Angular Displacement, for optimizing real-valued vectors is presented. This is designed for applications, called directional problems, where the quality of a solution vector is based exclusively on the direction of the vector and not the length of the vector. Angular Displacement maintains one control parameter that stochastically governs the amount of angular displacement, θ, induced by a single mutation. After θ is chosen, a random vector is selected that forms an angle of θ radians with the input vector. This approach contrasts the standard mutation operator that maintains one extra parameter for each vector component to control the displacement of the vector's head. Experiments show Angular Displacement is superior to the standard mutation operator in a directional problem involving the optimization of a hyperplane's parameters.

Mutation operator; evolution strategy; optimization; real-valued vectors