On The Efficiency Of Bionic Optimisation Procedures
Free (open access)
257 - 269
S. Gekeler, R. Steinbuch & C. Widmann
Bionic optimisation strategies have proven to be efficient in many applications especially if there are many local maxima to be expected in parameter spaces of higher dimensions. In structural mechanics, the central question is whether one particular procedure is to be preferred generally or if there are different problem types where some procedures are more efficient than others. Evolutionary optimisation with some sub-strategies, particle swarm optimisation, and neural nets along with hybrid approaches that couple the aforementioned methods have been investigated to some extent. These approaches are not uniquely defined, but rather imply many variants regarding the definition and selection of nextgeneration members, varying parameters of the underlying processes and the criteria to switch the strategy. To measure the performance of the different approaches some simple test examples have been used. The indicator of the procedures performance was the number of individuals which needed to be studied in order to come up with a satisfactory solution. As our main concern was about problems with many optimisation parameters, artificial neural nets do not show sufficient convergence velocities in our class of optimisation studies. Evolutionary optimisation, its subclass of fern optimisation and particle swarm optimisation prove to be of comparable power when applied to the test problems. It should not be disregarded that for all these approaches some experience about the optimisation parameters has to be gathered. In consequence, the total number of runs or individuals necessary to do the final optimisation is essentially larger than the number of runs during this final optimisation. Good initial proposals prove to be the most important source for all optimisation processes. Keywords: bionic optimisation, evolutionary optimisation, particle swarm optimisation, performance, structural mechanics.
bionic optimisation, evolutionary optimisation, particle swarm optimisation, performance, structural mechanics.