11.07.2015 Views

View - Universidad de Almería

View - Universidad de Almería

View - Universidad de Almería

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

An Adaptive Radial Basis Algorithm (ARBF) for Mixed-Integer Expensive Constrained Global Optimization137Solve this problem formulating the global non-convex optimization problem:IS2. The maximal distance ∆ from any point in the feasible region Ω to the closest samplepointminx,∆−∆x L ≤ x ≤ x Us/t 0 ≤ ∆ ≤ ∞0 ≤ ||x − x i || 2 − ∆ 2 ≤ ∞, i = 1, ..., n.(10)3.2 The Global Search (G-step) in ARBFTo find a good point first a sequence of global optimization problems are solved. If this phasefails, the global solution x Globs nof the RBF surface is consi<strong>de</strong>red. If rejected, the point mostdistance to all the sampled points, as well as to the boundary, is computed by solving a globaloptimization problem.G-step Phase 1. For a sequence of <strong>de</strong>creasing target values f ∗ n k, solve the global optimizationproblemg n (x k g n) =minx∈Ωgiving the global minimum g n (x k g n) for each f ∗ n k. Findf ˆkn =µ n (x) [ s n (x) − f ∗ n k] 2(11)minα(x k gn ))=0 f ∗ n kIf f ˆkn is non-empty, pick the global minimum corresponding to this target value, xˆkgn, as theG-step. Otherwise turn to G-step Phase 2. If α(x k g n)) = d during the search, there is no pointin <strong>de</strong>creasing f ∗ n kfurther, and the sequence is interrupted. If a minimum with α(x k g n)) = 0 isfound, a refined set of target values are tried in or<strong>de</strong>r to try to <strong>de</strong>crease the target values f ∗ n kfurther.G-step Phase 2. Select the global RBF solution x Globs nvalue and sufficiently far from currently best point x Min . If α(x Globs n0.01 max(1, f Min ) and ‖x GlobG-step Phase 3.s n− x Min ‖ > 0.2∆ pick x Globs nif it is interior, with sufficiently low) = 0, f Min − s n (xs Globn) ≥as the G-step point. Otherwise turn toG-step Phase 3. Find the point which have the maximal distance to any sample point as wellas the boundary, solving the global optimization problemδ IP =s/tminx ∈ Ω, δ−δx Li ≤ x i − δ/d , i = 1, ..., n.x i + δ/d ≤ x Ui , i = 1, ..., n.0 ≤ δ ≤ ∆0 ≤ ||x − x i || 2 − δ 2 ≤ ∞ , i = 1, ..., n.(12)3.3 The Global-Local Search (GL-step) in ARBFTo find a good point first a sequence of global optimization problems are solved. If this phasefails, the global solution x Globs nof the RBF surface is consi<strong>de</strong>red. If rejected, the InfStep point= −∞) is used.(f ∗ n

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!