An Improved Method using RBF Neural Networks to Speed up Optimization Algorithms
The paper presents a method using Radial Basis Function (RBF) neural networks to speed up deterministic search algorithms used for the optimization of superconducting magnets for the LHC accelerator project at CERN. The optimization of the iron yoke of the main LHC dipoles requires a number of numer...
Autores principales: | , , |
---|---|
Lenguaje: | eng |
Publicado: |
2001
|
Materias: | |
Acceso en línea: | http://cds.cern.ch/record/517565 |
Sumario: | The paper presents a method using Radial Basis Function (RBF) neural networks to speed up deterministic search algorithms used for the optimization of superconducting magnets for the LHC accelerator project at CERN. The optimization of the iron yoke of the main LHC dipoles requires a number of numerical field computations per trial solution as the field quality depends on the excitation and local iron saturation in the yoke. This results in computation times of about 30 minutes for each objective function evaluation (on DEC-Alpha 600/333). In this paper we present a method for constructing an RBF neural network for a local approximation of the objective function. The computational time required for such a construction is negligible compared to the deterministic function evaluation, and thus yields a speed-up of the overall search process. The effectiveness of this method is demonstrated by means of two- and three-parametric optimization examples. The achieved speed-up of the search routine is up to 30 %. |
---|