Modified particle swarm optimizers and their application to robust design and structural optimization

Many scientific, engineering and economic problems involve optimization. In reaction to that, numerous optimization algorithms have been proposed. Particle Swarm Optimization (PSO) is a new paradigm of Swarm Intelligence which is inspired by concepts from ’Social Psychology’ and ’Artificial Life’. Essentially, PSO proposes that the co-operation of individuals promotes the evolution of the swarm. In terms of optimization, the hope would be to enhance the swarm’s ability to search on a global scale so as to determine the global optimum in a fitness landscape. It has been empirically shown to perform well with regard to many different kinds of optimization problems. PSO is particularly a preferable candidate to solve highly nonlinear, non-convex and even discontinuous problems. The main ambition of this thesis is to propose two enhanced versions of PSO (Modified Guaranteed Convergence PSO (MGCPSO) and Modified Lbest based PSO (LPSO)) and to extend them to different areas of application. MGCPSO is an extension of theGbest based version of PSO and exhibits balanced performance between accuracy and efficiency in empirical numerical tests. It is applied to robust design with metamodel as well as structural sizing optimization in this work. In order to improve the efficiency of computing with regard to robust design, a mixed Fortran-Matlab program is developed as well as its corresponding parallel pattern. It obtains satisfying results for both optimization problems. On the other hand, LPSO constitutes an enhanced Lbest based version of PSO whereby two LPSOs with two and three neighbour links are tested by means of the empirical benchmark test. Both demonstrate excellent global searching ability. For this reason, this algorithm is extended to problems of truss topological design. This type of optimization problems can be characterised as large-scale and non-convex. LPSO is very well suited to this particular field of optimization. Indeed, it achieves solutions that challenge the best ones ever found. Finally, MGCPSO is successfully applied to a structural sizing optimization problem 5.30).