Content area
Full Text
Abstract
Purpose - The paper presents an hybrid optimization technique which couples the artificial immune system (AIS) algorithm with a zeroth order deterministic method.
Design/methodology/approach - AIS has been developed to tackle multi-modal optimization problems and it has shown a great ability to explore the objective function space. The algorithm is subdivided into two phases: an outer and an inner cycle. The outer cycle is devoted to the exploration of the space while the inner is a local exploration of the objective function. The new hybrid method proposes to replace the local search by a zeroth order deterministic search to speed up the overall convergence.
Findings - Results on two multi-modal analytical objective functions show an increase of speed of the new procedure with respect to the standard AIS. The method is also tested on the TEAM 22 numerical problem and some a posteriori techniques for the analysis of multimodal blind objective functions are discussed.
Originality/value - The new Multimodal optimization algorithm has allowed to explore thoroughly feasibility space giving rise to a partition of the whole space, the use of hybrid technique increases the performances of standard AIS increasing the convergence to the optimal points.
Keywords Optimization techniques, Body systems and organs, Numerical analysis
Paper type Research paper
1. Introduction
Optimization by means of mathematical procedures is becoming a standard tool in the engineering design process. In this area the use of classical optimization algorithms, based on regularity properties of the objective function, can be frustrating. In fact, functions to be optimized are often ill-conditioned and there is a particular need for multimodal optimization for the following main reasons:
* industrial problems have often multiple solutions which can fulfill objectives and constraints, thus multiple optimal points arise;
* in industrial applications often some solutions can be more feasible than others, so it is better to get from the algorithm not only a feasible point but instead a spread of solutions;
* with a series of solutions the final decision is left to an external agent (the designer) who can express the final preference.
Besides the multimodality issues, some general considerations have emerged in the last few years about optimization in industrial problems which are the rationale for the choice of the best suited...