一类两阶段自适应鲁棒多目标规划的对偶性刻画
Duality Characterizations for a Class of Two-Stage Adjustable Robust Multiobjective Programming
通讯作者:
收稿日期: 2023-02-7 修回日期: 2023-10-7
基金资助: |
|
Received: 2023-02-7 Revised: 2023-10-7
Fund supported: |
|
作者简介 About authors
黄嘉译,E-mail:
该文主要研究一类目标函数和约束函数均具有谱面不确定数据的两阶段自适应鲁棒多目标规划问题. 首先, 建立具有仿射自适应变量的两阶段自适应鲁棒多目标规划问题的 Farkas 引理. 随后, 引入该多目标规划问题的半定规划对偶问题. 最后, 借助该 Farkas 引理, 刻画它们之间的对偶性质.
关键词:
In this paper, we deal with a class of two-stage adjustable robust multi-objective programming problems with spectrahedral uncertainty data in both objective and constraints. We first establish a Farkas lemma for a two-stage adjustable robust multiobjective programming with affine adjustable variable. Then, we propose a semidefinite programming dual problem for this multiobjective programming problem. Furthermore, in terms of the obtained Farkas lemma, we explore duality properties between them.
Keywords:
本文引用格式
黄嘉译, 孙祥凯.
Huang Jiayi, Sun Xiangkai.
1 引言
由于自适应鲁棒优化问题的求解非常困难, 即使是简单的两阶段鲁棒优化问题也可能是 NP-Hard的, 因此, 具有特定结构的两阶段自适应鲁棒优化问题的易求解等价模型刻画得到了众多学者的关注. 如 Jeyakumar等[10]通过建立非齐次函数的可分离二次不等式系统的广义的 S-引理, 研究了一类具有不精确数据和二次自适应变量的鲁棒线性优化问题的精确二阶锥规划问题重构. Chuong 等[11]基于可分离二次决策规则, 研究了在盒子不确定集下的两阶段鲁棒线性规划的精确半定规划重构. 借助一类新的二次决策规则, 文献[12]刻画了具有仿射参数化的线性自适应鲁棒优化问题的精确半定规划和二阶锥规划重构. 借助线性矩阵不等式和闭锥约束规范条件, Chuong 等[13]给出了两阶段自适应鲁棒多目标线性规划问题的最优性条件. 借助 Slater 条件, Chuong 等[14] 研究了平方和凸多项式结构的两阶段自适应鲁棒优化问题的精确半定规划对偶. 借助鲁棒对偶方法, 文献[15]将具有多面体不确定集的自适应鲁棒非线性问题转化为自适应鲁棒线性问题, 并刻画了它们之间的性质.
受文献[13]的启发, 本文研究一类目标函数和约束函数均具有谱面不确定数据的两阶段自适应多目标规划问题. 首先, 基于谱面体不确定集, 建立具有仿射自适应变量的鲁棒多目标规划问题的 Farkas 引理. 随后, 引入该两阶段自适应多目标线性规划问题的半定规划对偶问题, 并借助一类闭锥约束规范和 Farkas 引理, 讨论它们之间的对偶性质. 最后, 在盒子和椭球等特定结构的不确定集下, 给出该规划问题的对偶问题, 并刻画它们之间的对偶性.
2 预备知识
假设Rn为n维欧氏空间, 并赋予欧氏范数‖.\mathbb{R}_+^{n}表示\mathbb{R}^{n}的非负象限,\mathrm{int}\mathbb{R}_+^{n}表示\mathbb{R}_+^{n}的拓扑内部. 对任意的x,y\in\mathbb{R}^n,\langle x,y\rangle:=x^{\top}y表示x和y之间的内积. 记0_n为\mathbb{R}^{n}的原点. 对于非空集合\mathrm{\Omega}\subset\mathbb{R}_+^{n},\mathrm{conv\Omega}和\mathrm{cl\Omega}分别表示\Omega的凸包和闭包,\mathrm{coneco}\Omega{:}=\mathbb{R}_+\mathrm{conv}\Omega表示\Omega\cup\{0_n\}的凸锥. 记I_n \in \mathbb{R}^{n \times n}为n \times n阶单位矩阵. 设S^n为所有n \times n阶实对称矩阵所构成的集合. 若对任意的x \in \mathbb{R}^n,x^{\top} M x \geq 0, 则称M \in S^n是半正定矩阵, 记为M \succeq 0. 若对任意的x \in \mathbb{R}^n\setminus\{0_n\},x^{\top} M x > 0, 则称M \in S^n是正定矩阵, 记为M \succ 0.
此处A_i,i=0,1,\cdots,s,是对称矩阵.显然, (2.1) 式中的谱面体集不仅是闭凸集, 而且包含球、多面体、盒子和椭球等常见的不确定集. 本文如无特殊说明, 总是假定U是非空有界集合.
设v_r:=\left(v_r^1, \cdots, v_r^m\right) \in \mathbb{R}^m,r=1, \cdots, p,和\theta_j:=\left(\theta_j^1, \cdots, \theta_j^m\right) \in \mathbb{R}^m,j=1, \cdots, q,是给定向量.
基于谱面体不确定集U, 本文考虑下述两阶段自适应多目标线性规划问题
其中u:=\left(u_1, \cdots, u_s\right) \in U是不确定参数,x:=(x_1,\cdots,x_n)\in\mathbb{R}^n为第一阶段决策变量,x(\cdot):U\rightarrow\mathbb{R}^m为第二阶段决策变量, 即自适应决策变量, 它依赖于不确定参数u\in U, 并且满足如下仿射决策规则
其中y\in\mathbb{R}^m和W\in\mathbb{R}^{m\times s}.c_r:U\rightarrow\mathbb{R}^n,r=1, \cdots, p,a_j: U\rightarrow\mathbb{R}^n和b_j:U\rightarrow\mathbb{R}^n,j=1, \cdots, q,是仿射映射, 即
其中c_r^i \in \mathbb{R}^n,r=1, \cdots, p,a_j^i \in \mathbb{R}^n,b_j^i \in \mathbb{R}^n,j=1, \cdots, q,i=0,1, \cdots, s.
记\mathcal{F}为鲁棒对等问题 (RTP) 的可行集, 即
受文献[18]的鲁棒弱有效解的概念启发, 引入 (UTP) 的鲁棒弱有效解的定义.
定义 2.1 若\left(\bar{x},\bar{y},\overline{W}\right)\in \mathcal{F}是\mathrm{(RTP)}的弱有效解, 即不存在(x,y,W)\in \mathcal{F},使得
则称\left(\bar{x},\bar{y},\overline{W}\right)是\mathrm{(UTP)}的鲁棒弱有效解.
3 对偶性刻画
本节旨在刻画基于仿射决策规则 (2.2) 式的 (UTP) 的鲁棒对偶性. 为此, 首先建立基于仿射决策规则 (2.2) 式的 (UTP) 的 Farkas 引理. 紧接着, 给出 (UTP) 的半定规划对偶问题. 最后, 利用 Farkas 引理研究问题 (UTP) 与它的对偶问题之间的对偶性质.
引理 3.1 (Farkas 引理)设C{:}=\mathrm{coneco}\{(a_j(u),\theta_j,u_1\theta_j,\cdots,u_s\theta_j,b_j(u)),j=1,\cdots,q \mid u:=(u_1,\cdots,u_s)\in{U}\}\subset\mathbb{R}^{n+m+ms+1}为闭锥. 假设\mathcal{F}\neq\emptyset, 则下列命题等价
(I)(x,y,W)\in \mathcal{F} \Rightarrow \mathop{\mathrm{max}}\limits_{u\in U}\{c_r(u)^{\top}x+v_r^{\top}x(u)\}\geq\xi_r,r=1,\cdots,p.
(II)\exists \alpha_r^0 \in \mathbb{R}_{+}, \sum\limits_{r=1}^p \alpha_r^0=1, \alpha_r^i\in\mathbb{R}, \lambda_j^0 \in \mathbb{R}_{+}和\lambda_j^i\in\mathbb{R}, i=1,\cdots,s, j=1,\cdots,q,使得
证 类似文献[13,定理 2.2] 的证明方法, 将\mathrm{(RTP)}转化为如下凸规划问题\mathrm{(CP)}
其中\widetilde{x}:=\left(x, y, W_1, \cdots, W_s\right),W_i, i=1,\cdots,s, 为矩阵W的列, \widetilde{c}_r(u):=\widetilde{c}_r^0+\sum\limits_{i=1}^s u_i \widetilde{c}_r^i, \widetilde{c}_r^0:=(c_r^0, v_r, \underbrace{0_m, \cdots, 0_m}_s), \widetilde{c}_r^i:=\left(c_r^i, 0_m, \Omega_r^i\right), i=1, \cdots, s, r=1, \cdots, p, \Omega_r^i:=(\underbrace{0_m, \cdots, v_r, \cdots, 0_m}_s), v_r为\Omega_r^i的第i个分量,\widetilde{a}_j(u):=\widetilde{a}_j^0+\sum\limits_{i=1}^s u_i \widetilde{a}_j^i, \widetilde{a}_j^0:=(a_j^0, \theta_j, \underbrace{0_m, \cdots, 0_m}_s), \tilde{a}_j^i:=\left(a_j^i, 0_m, \Psi_j^i\right),i=1, \cdots,s,j=1, \cdots, q, \Psi_j^i:=\underbrace{\left(0_m, \cdots, \theta_j, \cdots, 0_m\right.}_s), \theta_j为\Psi_j^i的第i个分量. 显然,
记\mathcal{F}^\prime:=\{ \widetilde{x} \in \mathbb{R}^{n+m+m s} | \widetilde{a}_j(u)^{\top} \widetilde{x} \leq b_j(u), j=1, \cdots, q, \forall u \in U\}.由于{C}为闭锥, 从而利用文献[19,定理 2.1] 可得下述命题等价
(i)\widetilde{x}\in \mathcal{F}^\prime\Rightarrow\mathop{\mbox{max}}\limits_{u\in U}\widetilde{c}_r(u)^{\top}\widetilde{x}\geq\xi_r,r=1,\cdots,p.
(ii) 存在\alpha_r^0\in\mathbb{R}_{+},\sum\limits_{r=1}^p \alpha_r^0=1,\lambda_j^0 \in \mathbb{R}_{+},\bar{u}^r\in U和u^j\in U,j=1,\cdots,q使得,
进一步地, 通过变量转化以及令\alpha_r^i:=\alpha_r^0\bar{u}_i^r\in\mathbb{R},i=1,\cdots,s,r=1,\cdots,p,\lambda_j^i:=\lambda_j^0u_i^j\in\mathbb{R},i=1,\cdots,s,j=1,\cdots,q,可得式 (3.1)、(3.2)、(3.3) 成立. 最后由\bar{u}^r\in U,r=1,\cdots,p和u^j\in U,j=1,\cdots,q,可知式 (3.4) 成立, 故引理证毕.
接下来, 引入(\mathrm{UTP})的半定规划对偶问题, 并讨论它们之间的弱对偶和强对偶性质.
令\eta:=\left(\eta_1, \cdots, \eta_p\right) \in \mathbb{R}^p, \alpha:=\left(\alpha_1^0, \cdots, \alpha_p^0, \alpha_1^1, \cdots, \alpha_p^1, \alpha_1^2, \cdots, \alpha_p^2, \cdots, \alpha_1^s, \cdots, \alpha_p^s\right) \in \mathbb{R}_{+}^p \times \mathbb{R}^{p s}, \lambda:=\left(\lambda_1^0, \cdots, \lambda_q^0, \lambda_1^1, \cdots, \lambda_q^1, \lambda_1^2, \cdots, \lambda_q^2, \cdots, \lambda_1^s, \cdots, \lambda_q^s\right) \in \mathbb{R}_{+}^q \times \mathbb{R}^{q s}, 则(\mathrm{UTP})的半定规划对偶问题(\mathrm{RTD})定义为
类似于定义2.1, 我们引入,\mathrm{(RTD)},的弱有效解的概念. 方便起见,记,\mathrm{(RTD)},的可行集为,\mathcal{F}_D.
定义 3.1 设(\bar{\eta},\bar{\alpha},\bar{\lambda})\in \mathcal{F}_D. 若不存在\mathrm{(RTD)}的可行解(\eta,\alpha,\lambda)\in \mathcal{F}_D,使得
则称(\bar{\eta},\bar{\alpha},\bar{\lambda})是\mathrm{(RTD)}的弱有效解.
下述定理建立了\mathrm{(UTP)}和\mathrm{(RTD)}之间的弱对偶关系.
定理 3.1 若(x,y,W) \in \mathcal{F}和(\eta, \alpha, \lambda) \in \mathcal{F}_D, 则
其中{F}_r(x,y,W):=\mathop{\mathrm{max}}\limits_{{u}\in U}\{c_r(u)^{\top}x+v_r^{\top}x(u)\},r=1, \cdots, p.
证 因为(\eta, \alpha, \lambda) \in \mathcal{F}_D, 所以\alpha_r^0 \in \mathbb{R}_{+},\sum\limits_{r=1}^p \alpha_r^0=1,\alpha_r^i\in\mathbb{R},\lambda_j^0\in\mathbb{R}_{+}和\lambda_j^i\in\mathbb{R},i=1,\cdots,s,j=1,\cdots,q,使得式 (3.5)、(3.6)、(3.7)、(3.8) 和 (3.9) 成立. 借助式 (3.9)
以及文献[20,定理 2.3] 的证明方法, 易证存在\bar{u}^r :=(\bar{u}_1^r,\cdots,\bar{u}_s^r) \in U,r=1,\cdots,p,以及\widetilde{u}^j:=(\widetilde{u}_1^j,\dots,\widetilde{u}_s^j)\in U,j=1,\cdots,q,使得
此外,由(x,y,W)\in \mathcal{F}以及\widetilde{u}^j\in U可知
另一方面, 对于\bar{u}^r\in U,
其中第三个等式成立是因为式 (3.5)、(3.6)、(3.10) 和 (3.14) 成立, 第四个等式成立是因为式 (3.7)、(3.11) 和 (3.13) 成立, 第一个不等式成立是因为式 (3.15) 成立以及\lambda_j^0\in\mathbb{R}_+, 最后一个不等式成立是因为式 (3.8) 和 (3.12) 成立.又因为
从而结合\sum\limits_{r=1}^p\alpha_r^0=1与\alpha_r^0\in\mathbb{R}_+可知
定理证毕.
借助引理 3.1 以及一类闭锥约束规范条件, 我们建立\mathrm{(UTP)}和\mathrm{(RTD)}之间的强对偶关系.
定理 3.2 设{C}{:}=\mathrm{coneco}\{(a_j(u),\theta_j,u_1\theta_j,\cdots,u_s\theta_j,b_j(u)),j=1,\cdots,q \mid u:=(u_1,\cdots,u_s)\in{U}\}\subset\mathbb{R}^{n+m+ms+1}为闭锥. 若\left(\bar{x},\bar{y},\overline{W}\right)\in\mathcal{F}是\mathrm{(UTP)}的鲁棒弱有效解, 则存在\bar{\eta}:=\left(\bar{\eta}_1, \cdots, \bar{\eta}_p\right) \in \mathbb{R}^p,\bar{\alpha}:=\left(\bar{\alpha}_1^0, \cdots, \bar{\alpha}_p^0, \bar{\alpha}_1^1, \cdots, \bar{\alpha}_p^1, \bar{\alpha}_1^2, \cdots, \bar{\alpha}_p^2, \cdots, \bar{\alpha}_1^s, \cdots, \bar{\alpha}_p^s\right) \in \mathbb{R}_{+}^p \times \mathbb{R}^{p s},以及\bar{\lambda}:=\left(\bar{\lambda}_1^0, \cdots, \bar{\lambda}_q^0,\right.\left. \bar{\lambda}_1^1, \cdots, \bar{\lambda}_q^1, \bar{\lambda}_1^2, \cdots, \bar{\lambda}_q^2, \cdots, \bar{\lambda}_1^s, \cdots, \bar{\lambda}_q^s\right) \in \mathbb{R}_{+}^q \times \mathbb{R}^{q s},使得(\bar{\eta},\bar{\alpha},\bar{\lambda})\in\mathcal{F}_D为\mathrm{(RTD)}的弱有效解, 并且
其中{F}_r\left(\bar{x},\bar{y},\overline{W}\right){:}=\mathop{\mathrm{max}}\limits_{{u}\in U}\left\{c_r({u})^{\top}\bar{x}+v_r^{\top}\left(\bar{y}+\overline{W}u\right)\right\},r=1, \cdots, p.
证 令\xi_r:={F}_r\left(\bar{x},\bar{y},\overline{W}\right),r=1,\cdots,p.因为\left(\bar{x},\bar{y},\overline{W}\right)\in\mathcal{F}是\mathrm{(UTP)}的鲁棒弱有效解, 所以
又因为{C}是闭的, 从而由引理 3.1 可知, 存在\bar{\alpha}_r^0 \in \mathbb{R}_{+}, \sum\limits_{r=1}^p \bar{\alpha}_r^0=1, \bar{\alpha}_r^i\in\mathbb{R}, \bar{\lambda}_j^0 \in \mathbb{R}_{+}和\bar{\lambda}_j^i\in\mathbb{R}, i=1,\cdots,s, j=1,\cdots,q,使得
令\bar{\eta}_r:=F_r\left(\bar{x},\bar{y},\overline{W}\right), r=1,\cdots,p, \bar{\eta}:=\left(\bar{\eta}_1, \cdots, \bar{\eta}_p\right) \in \mathbb{R}^p, \bar{\alpha}:=\left(\bar{\alpha}_1^0, \cdots, \bar{\alpha}_p^0, \bar{\alpha}_1^1, \cdots, \bar{\alpha}_p^1, \bar{\alpha}_1^2, \cdots,\right.\\ \left.\bar{\alpha}_p^2, \cdots, \bar{\alpha}_1^s, \cdots, \bar{\alpha}_p^s\right) \!\in \!\mathbb{R}_{+}^p \!\times \!\mathbb{R}^{p s}, \bar{\lambda}\!:=\!\left(\bar{\!\lambda}_1^0, \!\cdots, \bar{\!\lambda}_q^0, \bar{\!\lambda}_1^1, \!\cdots, \bar{\!\lambda}_q^1, \bar{\!\lambda}_1^2, \!\cdots, \bar{\!\lambda}_q^2, \!\cdots, \bar{\!\lambda}_1^s, \!\cdots, \bar{\!\lambda}_q^s\!\right) \in \mathbb{R}_{+}^q \times \mathbb{R}^{q s}. 显然, (\bar{\eta},\bar{\alpha},\bar{\lambda}) 为 \mathrm{(RTD)} 的可行解.
下证(\bar{\eta},\bar{\alpha},\bar{\lambda})为\mathrm{(RTD)}的弱有效解. 假设(\bar{\eta},\bar{\alpha},\bar{\lambda})不是\mathrm{(RTD)}的弱有效解, 则存在(\hat{\eta},\hat{\alpha},\hat{\lambda})\in \mathcal{F_D},使得
又因为\bar{\eta}_r=F_r\left(\bar{x},\bar{y},\overline{W}\right),r=1,\cdots,p,所以
这与定理 3.1 矛盾. 故,(\bar{\eta},\bar{\alpha},\bar{\lambda})为\mathrm{(RTD)}的弱有效解. 证毕.
本文的最后, 我们考虑\mathrm{(RTP)}中谱面体不确定集U的几类特殊情形. 首先, 若谱面体不确定集U为盒子不确定集B, 即
则, 易建立如下强对偶定理.
推论 3.1 设\widehat{C}{:}=\mathrm{coneco}\{(a_j(u),\theta_j,u_1\theta_j,\cdots,u_s\theta_j,b_j(u)),j=1,\cdots,q \mid u:=(u_1,\cdots,u_s)\in{B}\}\subset\mathbb{R}^{n+m+ms+1}为闭锥. 若\left(\bar{x},\bar{y},\overline{W}\right)\in\mathcal{F}是\mathrm{(UTP)}的鲁棒弱有效解, 则存在\bar{\eta}:=\left(\bar{\eta}_1, \cdots, \bar{\eta}_p\right) \in \mathbb{R}^p,\bar{\alpha}:=\left(\bar{\alpha}_1^0, \cdots, \bar{\alpha}_p^0, \bar{\alpha}_1^1, \cdots, \bar{\alpha}_p^1, \bar{\alpha}_1^2, \cdots, \bar{\alpha}_p^2, \cdots, \bar{\alpha}_1^s, \cdots, \bar{\alpha}_p^s\right) \in \mathbb{R}_{+}^p \times \mathbb{R}^{p s},以及\bar{\lambda}:= \left(\bar{\lambda}_1^0, \cdots, \bar{\lambda}_q^0,\right.\left. \bar{\lambda}_1^1, \cdots, \bar{\lambda}_q^1, \bar{\lambda}_1^2, \cdots, \bar{\lambda}_q^2, \cdots, \bar{\lambda}_1^s, \cdots, \bar{\lambda}_q^s \right) \in \mathbb{R}_{+}^q \times \mathbb{R}^{q s},使得(\bar{\eta},\bar{\alpha},\bar{\lambda})\in\mathcal{F_D}为\mathrm{(\widehat{RTD})}的弱有效解, 且
其中{F}_r\left(\bar{x},\bar{y},\overline{W}\right){:}=\mathop{\mathrm{max}}\limits_{{u}\in B}\left\{c_r({u})^{\top}\bar{x}+v_r^{\top}\left(\bar{y}+\overline{W}u\right)\right\},r=1,\cdots,p, 以及
证 设A_i\in S^{2s},i=0,1,\cdots,s, 即
其中E_i是第(i,i)项为 1, 其余项为 0 的s\times s阶对角矩阵. 则式 (3.16) 定义的盒子不确定集B与式 (2.1)定义的谱面体不确定集U等价. 此外, 对于\bar{\alpha}\in \mathbb{R}_{+}^p \times \mathbb{R}^{p s},\bar{\lambda}\in \mathbb{R}_{+}^q \times \mathbb{R}^{q s},有
因此, 借助定理3.2可得该推论成立.
若\mathrm{(RTP)}中谱面体不确定集U为椭球不确定集T, 即
其中M\in S^s,M\succ0,M=H^{\top}H,H是s\times s阶的矩阵, 则有如下强对偶定理.
推论3.2 设\widetilde{C}{:}=\mathrm{coneco}\{(a_j(u),\theta_j,u_1\theta_j,\cdots,u_s\theta_j,b_j(u)),j=1,\cdots,q \mid u:=(u_1,\cdots,u_s)\in{T}\}\subset\mathbb{R}^{n+m+ms+1}为闭锥. 若\left(\bar{x},\bar{y},\overline{W}\right)\in\mathcal{F}是\mathrm{(UTP)}的鲁棒弱有效解, 则存在\bar{\eta}:=\left(\bar{\eta}_1, \cdots, \bar{\eta}_p\right) \in \mathbb{R}^p,\bar{\alpha}:=\left(\bar{\alpha}_1^0, \cdots, \bar{\alpha}_p^0, \bar{\alpha}_1^1, \cdots, \bar{\alpha}_p^1, \bar{\alpha}_1^2, \cdots, \bar{\alpha}_p^2, \cdots, \bar{\alpha}_1^s, \cdots, \bar{\alpha}_p^s\right) \in \mathbb{R}_{+}^p \times \mathbb{R}^{p s},以及\bar{\lambda}:=\left(\bar{\lambda}_1^0, \cdots, \bar{\lambda}_q^0,\right.\left.\bar{\lambda}_1^1, \cdots, \bar{\lambda}_q^1, \bar{\lambda}_1^2, \cdots, \bar{\lambda}_q^2, \cdots, \bar{\lambda}_1^s, \cdots, \bar{\lambda}_q^s\right) \in \mathbb{R}_{+}^q \times \mathbb{R}^{q s},使得(\bar{\eta},\bar{\alpha},\bar{\lambda})\in\mathcal{F_D}为\mathrm{(\widetilde{RTD})}的弱有效解, 且
其中{F}_r\left(\bar{x},\bar{y},\overline{W}\right){:}=\mathop{\mathrm{max}}\limits_{{u}\in T}\left\{c_r({u})^{\top}\bar{x}+v_r^{\top}\left(\bar{y}+\overline{W}u\right)\right\},r=1,\cdots,p, 以及
证 设A_i\in S^{s+1},i=0,1,\cdots,s, 即
其中e_i\in\mathbb{R}^s是第i项为 1, 其余项为 0 的向量. 利用 Schurb 定理, 有
上式表明, 式 (3.17) 定义的椭球不确定集T与式 (2.1) 定义的谱面体不确定集U等价. 此外, 对于\bar{\alpha}\in \mathbb{R}_{+}^p \times \mathbb{R}^{p s}和\bar{\lambda}\in \mathbb{R}_{+}^q \times \mathbb{R}^{q s},有
从而, 借助定理3.2 可得该推论成立.
参考文献
Robust optimization-methodology and applications
DOI:10.1007/s101070100286 URL [本文引用: 2]
Robust linear semi-infinite programming duality under uncertainty
DOI:10.1007/s10107-013-0668-6 URL [本文引用: 1]
Characterizations of approximate duality and saddle point theorems for nonsmooth robust vector optimization
DOI:10.1080/01630563.2019.1660891 URL [本文引用: 1]
鲁棒复合优化问题的 Lagrange 对偶
Lagrange duality of robust composite optimization problems
Adjustable robust solutions of uncertain linear programs
DOI:10.1007/s10107-003-0454-y URL [本文引用: 1]
Robust multistage decision making
Robust optimization of uncertain multistage inventory systems with inexact data in decision rules
DOI:10.1007/s10287-016-0253-6 URL [本文引用: 1]
Adjustable robust optimization via Fourier-Motzkin elimination
DOI:10.1287/opre.2017.1714
URL
[本文引用: 1]
We demonstrate how adjustable robust optimization (ARO) problems with fixed recourse can be cast as static robust optimization problems via Fourier–Motzkin elimination (FME). Through the lens of FME, we characterize the structures of the optimal decision rules for a broad class of ARO problems. A scheme based on a blending of classical FME and a simple linear programming technique that can efficiently remove redundant constraints is developed to reformulate ARO problems. This generic reformulation technique enhances the classical approximation scheme via decision rules, and it enables us to solve adjustable optimization problems to optimality. We show via numerical experiments that, for small-sized ARO problems, our novel approach finds the optimal solution. For moderate- or large-sized instances, we eliminate a subset of the adjustable variables, which improves the solutions obtained from linear decision rules.
Quadratically adjustable robust linear optimization with inexact data via generalized S-lemma: Exact second-order cone program reformulations
DOI:10.1016/j.ejco.2021.100019 URL [本文引用: 1]
Exact SDP reformulations of adjustable robust linear programs with box uncertainties under separable quadratic decision rules via SOS representations of non-negativity
DOI:10.1007/s10898-021-01050-x [本文引用: 1]
Exact conic programming reformulations of two-stage adjustable robust linear programs with new quadratic decision rules
DOI:10.1007/s11590-020-01595-y [本文引用: 1]
Adjustable robust multiobjective linear optimization: Pareto optimal solutions via conic programming
Exact dual semi-definite programs for affinely adjustable robust SOS-convex polynomial optimization problems
DOI:10.1080/02331934.2021.1902521 URL [本文引用: 1]
Dual approach for two-stage robust nonlinear optimization
DOI:10.1287/opre.2022.2289
URL
[本文引用: 1]
In “Dual Approach for Two-Stage Robust Nonlinear Optimization,” de Ruiter, Zhen, and den Hertog study adjustable robust minimization problems where the objective or constraints depend in a convex way on the adjustable variables. They reformulate the original adjustable robust nonlinear problem with a polyhedral uncertainty set into an equivalent adjustable robust linear problem, for which all existing approaches for adjustable robust linear problems can be used. The reformulation is obtained by first dualizing over the adjustable variables and then over the uncertain parameters. The polyhedral structure of the uncertainty set then appears in the linear constraints of the dualized problem, and the nonlinear functions of the adjustable variables in the original problem appear in the uncertainty set of the dualized problem. The authors show how to recover linear decision rules to the original primal problem and how to generate bounds on its optimal objective value.
Some geometric results in semidefinite programming
DOI:10.1007/BF01100204 URL [本文引用: 1]
On robust multiobjective optimization
Robust alternative theorem for linear inequalities with applications to robust multiobjective optimization
DOI:10.1016/j.orl.2017.09.002 URL [本文引用: 3]
Linear matrix inequality conditions and duality for a class of robust multiobjective convex polynomial programs
DOI:10.1137/17M1143484 URL [本文引用: 1]
/
〈 |
|
〉 |
