隨機規劃通常會假設一個機率分佈,而魯棒最佳化假設這樣的機率資訊是不知道。魯棒最佳化一般用uncertainty set來characterize不確定性,常用的uncertainty set包括box/ellipsoid uncertaity set, budgeted uncertainty set等。隨機規劃一般最佳化期望值,魯棒最佳化optimized worst-cast cost/profit.
舉例:reliable facility location problem中,facility可能面對一定的失效風險,在stochastic context中,我們可以假設單個facility的失效機率是p;目標函式最小化設施選址成本+期望的運輸成本。在robust opt context中,我們可以說設施的失效機率是未知的(比如自然災害造成的設施失效),最佳化worst-case cost。我們可以用一個budgeted uncertainry set來表示“同一時間最多有幾個設施失效"。
Robust opt (RO)又可以分為static RO, adjustable RO. Static RO makes all the decisions "here-and-now". Adjustable RO makes the first-stage decision "here-and-now", and recourse decisions "wait-and-see" based on observed uncertainty information. Static RO一般用duality technique進行reformulation. Two-stage RO (adjustable RO的一種)可以用benders, column-and-constraint generation求解(精確演算法),也可以用affine policy(近似演算法)。Multi-stage RO一般用affine policy求解。
robust opt 文獻推薦:
(1)Bertsimas, D. and Sim, M., 2004. The price of robustness.Operations research,52(1), pp.35-53.
(2)Atamt ̈urk, A., Zhang, M., 2007. Two-stage robust network flow and design under demand uncertainty. Operations Research 55 (4), 662–673.
(3) Gorissen, B. L., Yanıko ̆glu,I., den Hertog, D., 2015. A practical guide to robust optimization. Omega 53, 124–137.
未完待續,有空再繼續新增DRO的知識。
隨機規劃通常會假設一個機率分佈,而魯棒最佳化假設這樣的機率資訊是不知道。魯棒最佳化一般用uncertainty set來characterize不確定性,常用的uncertainty set包括box/ellipsoid uncertaity set, budgeted uncertainty set等。隨機規劃一般最佳化期望值,魯棒最佳化optimized worst-cast cost/profit.
舉例:reliable facility location problem中,facility可能面對一定的失效風險,在stochastic context中,我們可以假設單個facility的失效機率是p;目標函式最小化設施選址成本+期望的運輸成本。在robust opt context中,我們可以說設施的失效機率是未知的(比如自然災害造成的設施失效),最佳化worst-case cost。我們可以用一個budgeted uncertainry set來表示“同一時間最多有幾個設施失效"。
Robust opt (RO)又可以分為static RO, adjustable RO. Static RO makes all the decisions "here-and-now". Adjustable RO makes the first-stage decision "here-and-now", and recourse decisions "wait-and-see" based on observed uncertainty information. Static RO一般用duality technique進行reformulation. Two-stage RO (adjustable RO的一種)可以用benders, column-and-constraint generation求解(精確演算法),也可以用affine policy(近似演算法)。Multi-stage RO一般用affine policy求解。
robust opt 文獻推薦:
(1)Bertsimas, D. and Sim, M., 2004. The price of robustness.Operations research,52(1), pp.35-53.
(2)Atamt ̈urk, A., Zhang, M., 2007. Two-stage robust network flow and design under demand uncertainty. Operations Research 55 (4), 662–673.
(3) Gorissen, B. L., Yanıko ̆glu,I., den Hertog, D., 2015. A practical guide to robust optimization. Omega 53, 124–137.
未完待續,有空再繼續新增DRO的知識。