CSTSTUDIOSUITE2014优化器概述.doc

Optimizer Overview The optimizer may be used to find optimal parameter settings for the modeled structure. Therefore, it is necessary to select the parameters that may be varied during the optimization process. The optimizer tries to minimize a function emdash; the goal function. This function depends on the parameters selected for variation by the optimizer. Therefore, at least one goal must be defined that is evaluated while the optimizer is running. The goals for an optimization are based on results below the 1D Results folder in the tree or on result templates. The optimizer tries to minimize the goals by varying the selected parameters. ? Algorithm type There are four different types of local and three different types of global optimizers available. The local ones are the Trust Region Framework, the Nelder Mead Simplex, the Interpolated Quasi Newton, and the Classic Powell ?optimizer. The global ones are the CMA Evolutionary Strategy, a Genetic and a Particle Swarm optimization algorithm. By using a statistical model, CMA Evolutionary Strategy improves its performance without sacrificing its global optimization approach. The Trust Region Framework can also be set up to be a global optimization technique doing multiple local optimizations at different places in the parameter space. You may select the optimizer type from the ?Optimizer - Settings property page. More information about the optimization strategies can be found at the Optimizer - Interpolation of Primary Data page and on the Optimizer - Global Algorithm Settings page. Parameters The optimizer is a multidimensional optimizer. Thus, it is possible to simultaneously optimize several parameters. The parameters selected for optimization must be bounded by upper and lower limits. You may select the parameters for optimization from the Optimizer - Settings property page. ?You can define a starting parameter combination for algorithms that have a specific initial evaluation point. For algorithms that use