混整數規劃(Mixed Integer Programming) 是線性規劃(Linear Programming),加了整數的決策變數。其化簡標準形式可以寫成
\begin{array}{l}\text{ min } c_{\mathcal{I}}x_\mathcal{I} + c_{\mathcal{R}}x_\mathcal{R} \\ \text{s.t } Ax = b \\ x := (x_\mathcal{I} , x_\mathcal{R}) \geq 0 \\ \end{array}
其中 \mathcal{I} 為整數決策變數集 ,\mathcal{R} 為實數決策變數集,A 為已知係數
此模型已經有很多演算法可以求解詳細可以查關鍵字 branch & cut algorithm
理論上可以找到最佳決策 !! 但實務上整數規劃是 NP-Hard 問題,求解效率是議題!!,所以數學建模上我們希望 |\mathcal{I}| 越小越好!!
而現實問題為何可使用整數規劃來數學建模那麼重要原因有二:
========================================
1. 現實問題物品的數量為整數個 !!
2. 邏輯的 True , False 可以用 0 - 1 決策變數表示!!
========================================
本文介紹一些筆者經驗常見邏輯限制式代表的意義,幫助大家更能理解如何數學建模。而對於有興趣的讀者,可以去閱讀 "Paul Williams : Model Building in Mathematical Programming" 這本書,有詳細的介紹。
+++++++++++++++++++++++++++++++++++
(A) \text{ 從集合裡面挑選元素 }
+++++++++++++++++++++++++++++++++++
Let i \in I , |I| < \infty , x_i \in \{0,1\} , cost : I \longrightarrow \mathbb{R} , denote cost_i \in \mathbb{R}
x_i = 1 \Longleftrightarrow \text{ you select element in } i \in I
(A1) 從集合 I 裡面挑選恰好一個元素
\sum_{i \in I} x_i = 1
========================================
註 : 學術上我們稱 \displaystyle{\sum_{i\in I} x_i =1 } 為 Special Ordered Sets of type 1 constraint ,記做 SOS1(I) ,或寫作 x(I) = 1
========================================
(A2) 從集合 I 裡面挑選至少一個元素
\sum_{i\in I} x_i \geq 1
(A3) 從集合 I 裡面至多挑選 U 個,至少挑選 L個 (L,U\in \mathbb{Z})
L \leq \sum_{i\in I} x_i \leq U
(A4) 如果挑選 i \in I 都需要負擔/獲得 c_i \in \mathbb{R}
則總價格可以寫成 cost(I):= \sum_{i\in I}c_i x_i
(A5) 當然可以考慮子集合 S \subseteq I ,則價格可以寫成
cost(S):= \sum_{i\in S}c_i x_i
(A6) 因為 x_i = 1 代表的是種類 i 有挑選,但如果同種類要計量可以用"整數"決策變數 z_i \in \mathbb{Z^{\geq 0 }}= \{0,1,2,3,4,....\}
qcost(S):= \sum_{i\in S}c_i z_i
當然還可以考慮集合的集合 \mathcal{F}\subseteq 2^{I}= \{S : S\subseteq I \},其中 2^{I} 為 PowerSet of I ,還可以定義 Cost : 2^{I} \longrightarrow \mathbb{R} 類比 (A4) , (A5) 限制式,可以定義
(A7) \quad Cost(\mathcal{F}) := \sum_{S \in \mathcal{F}}cost(S)
(A8) \quad QCost(\mathcal{F}) := \sum_{S \in \mathcal{F}}qcost(S)
如果 z_i \leq U 則我們還可以增加一個維度只用 0-1變數 x_{ij} 來刻劃 z_i , x_i
(A9) \left\{\begin{array}{c} \displaystyle{z_i := \sum^{U}_{j=0} j \cdot x_{ij} \quad \forall i \in I} \\ \displaystyle{x_i:= \sum^{U}_{j=0} x_{ij}\leq 1 \quad \forall i \in I } \\ x_{ij} \in \{0,1\} \quad \forall (i,j) \in I \times \{0,1,2,....U\} \\ \end{array}\right.
+++++++++++++++++++++++++++++++++++
(B) \text{ if - then constraints}
+++++++++++++++++++++++++++++++++++
考慮 i \in I , y_i \in \{0,1\} ,記作 \bar{y}_i = 1-y_i
(B1) 四大基本,可用來串接邏輯 Proposition
if y_1 = 1 then y_2 = 1 :\qquad y_1 \leq y_2
if y_1 = 1 then y_2 = 0 :\qquad y_1 \leq 1-y_2 \Longrightarrow y_1 \leq \bar{y}_2
if y_1 = 0 then y_2 = 1 :\qquad 1-y_1 \leq y_2 \Longrightarrow \bar{y}_1 \leq y_2
if y_1 = 0 then y_2 = 0 :\qquad 1- y_1 \leq 1- y_2 \Longrightarrow \bar{y}_1 \leq \bar{y}_2
============================================
例如 : P \Longrightarrow Q , P\Longrightarrow R , R\Longrightarrow S , Q \Longrightarrow S 可以寫成
\left\{\begin{array}{c} y_P \leq y_Q \\ y_P \leq y_R \\ y_R \leq y_S \\ y_Q \leq y_S \\ \end{array}\right.
事實上這可以畫成一個邏輯有向網路圖(Directed Graph) G(N,A) 邊的方向性即為 "\leq",如例子: N=\{P,Q,R,S\} , A = \{ (P,Q), (P,R), (R,S) , (Q,S)\}
=============================================
Let a, x \in \mathbb{R}^J , b \in \mathbb{R} , [a,x] := \sum_{j \in J} a_{j}x_j
===================================
註 : [a,x]\geq b 可以改寫成 -b \leq -[a,x]
===================================
再來介紹經典的"開啟"限制式 !!
(B2) if (y = 1) then [a,x] \leq b , else [a,x] < \infty redundant !!
[a,x] \leq b + M(1-y)
註: M 是充分大常數,學術上稱為 Capital M/Big M Constraint,但不能太大(求解上不穩定),最好的 M := -b + \underset{x \in \mathbb{R}^J}{\text{sup }}[a,x]
再來是限制式"開啟" !!
(B3) if ( [a,x] \leq b ) then y=1 , else : redundant !!
b + \epsilon \leq [a,x] + My
========================================
證: 等價於 y = 0 \Longrightarrow [a,x] > b ,右邊引入極小的數 \epsilon , 變成 [a,x] \geq b + \epsilon,根據(B2)最後可寫成 b + \epsilon \leq [a,x] + My
最好的 M:= b + \epsilon - \underset{x \in \mathbb{R}^J}{\text{inf }}[a,x]
========================================
註 : 如果(B2)(B3)條件是 (y = 0) 只要把 y 代換成 \bar{y} 即可 !!
========================================
結合 (B2) (B3) 概念可以有 (B4) , (B5)
(B4) 開關限制式 [a,x] \leq b \Longleftrightarrow y =1
\left\{\begin{array}{c} [a,x] \leq b + M(1-y) \\ b + \epsilon \leq [a,x] + My \\ \end{array}\right.
(B5) 限制式開限制式 [a_1,x_1]\leq b_1 \Longrightarrow [a_2,x_2]\leq b_2
\left\{\begin{array}{c} b + \epsilon \leq [a_1,x_1] + My \\ [a_2,x_2] \leq b + M(1-y) \\ \end{array}\right.
========================================
證: 串聯 [a_1,x_1]\leq b_1 \Longrightarrow y = 1 , y = 1 \Longrightarrow [a_2,x_2]\leq b_2
========================================
+++++++++++++++++++++++++++++++++++
(C) And + Or Constraints !!
+++++++++++++++++++++++++++++++++++
(C1) if y = 1 then \underset{i \in I}{\bigwedge} ([a_i,x_i]\leq b_i) (Right-And Constraints)
\left\{\begin{array}{c} [a_1,x_1]\leq b_1 + M_1(1-y) \\ [a_2,x_2]\leq b_2 + M_2(1-y) \\ [a_3,x_3]\leq b_3 + M_3(1-y) \\ ... \\ [a_{|I|},x_{|I|}]\leq b_{|I|} +M_{|I|}(1-y) \\ \end{array}\right.
Let \forall i \in I \quad y_i \in \{0,1\}
(C2) if y = 1 then \underset{i \in I}{\bigvee}([a_i,x_i]\leq b_i) (Right-Or Constraints)
\left\{\begin{array}{c} [a_1,x_1]\leq b_1 + M_1(1-y_1) \\ [a_2,x_2]\leq b_2 + M_2(1-y_2) \\ [a_3,x_3]\leq b_3 + M_3(1-y_3) \\ ... \\ [a_{|I|},x_{|I|}]\leq b_{|I|} +M_{|I|}(1-y_{|I|}) \\ \displaystyle{y \leq \sum_{i\in I} y_i} \\ \end{array}\right.
==============================================
註: 學術上稱 (C2) 叫 disjunctive constraints ,可以輕易推廣成
if y = 1 then \displaystyle{\underset{i \in I}{\bigvee}\left\{\underset{j\in J_i}{\bigwedge}([a_j,x_j]\leq b_j)\right\}}
==============================================
(C3) if (\underset{i \in I}{\bigwedge}([a_i,x_i]\leq b_i) then y = 1 (Left-And Constraints)
\left\{\begin{array}{c} b_1 + \epsilon_1 \leq [a_1,x_1] + M_1(1-y_1) \\ b_2 + \epsilon_2 \leq [a_2,x_2] + M_2(1-y_2) \\ b_3 + \epsilon_3 \leq [a_3,x_3] + M_3(1-y_3) \\ ... \\ b_{|I|} + \epsilon_{|I|} \leq [a_{|I|},x_{|I|}] + M_{|I|}(1-y_{|I|}) \\ \displaystyle{ (1-y) \leq \sum_{i\in I} y_i} \\ \end{array}\right.
========================================
證: 等價於 if y = 0 then \underset{i \in I}{\bigvee}([a_i,x_i] > b_i)
========================================
(C4) if (\underset{i \in I}{\bigvee}([a_i,x_i]\leq b_i) then y = 1 (Left-Or Constraints)
\left\{\begin{array}{c} b_1 + \epsilon_1 \leq [a_1,x_1] + M_1 \cdot y \\ b_2 + \epsilon_2 \leq [a_2,x_2] + M_2 \cdot y \\ b_3 + \epsilon_3 \leq [a_3,x_3] + M_3 \cdot y \\ ... \\ b_{|I|} + \epsilon _{|I|}\leq [a_{|I|},x_{|I|}] + M_{|I|}\cdot y \\ \end{array}\right.
========================================
證: 等價於 if y = 0 then \underset{i \in I}{\bigwedge}([a_i,x_i] > b_i)
========================================
(C5) 互斥或,只有 [a_i,x_i] \leq b_i 其中之一會發生 (Disjoint Or Constaints)
\left\{\begin{array}{c} [a_1,x_1]\leq b_1 + M_1(1-y_1) \\ [a_2,x_2]\leq b_2 + M_2(1-y_2) \\ [a_3,x_3]\leq b_3 + M_3(1-y_3) \\ ... \\ [a_{|I|},x_{|I|}]\leq b_{|I|} +M_{|I|}(1-y_{|I|}) \\ \displaystyle{\sum_{i\in I} y_i = 1} \\ \end{array}\right.
+++++++++++++++++++++++++++++++++++
(D) 離散時間狀態網路圖(Time-State-Network)
+++++++++++++++++++++++++++++++++++
考慮離散時間 t \in T := \{0,1,2,........U\} (單位時刻點集)
考慮離散狀態向量集\displaystyle{i\in(i_1,i_2,i_3,....,i_n) \in I = \prod^{n}_{j=1} I_{j}}
註: 狀態向量集可能隨著時間 t 不同而不同,即 I^{(t)}
考慮狀態-時間函數 \displaystyle{v : T \times I \longrightarrow } \{0,1\}
很明顯我們一個時間點,只會有一個狀態
(D1) \quad \forall t \in T \quad \sum_{i\in I} v_{it} = 1 \quad \text{或寫成 } v_t(I) = 1
如果我們看 i 的每個分量,很明顯每個分量也只能選一種
因為(D1)的決策變數非常多,事實上如果 State 間是獨立的,只要定義 X^{jt}_{i_j} = 1 \Longleftrightarrow \text{at time }t , j\text{-th state is } i_j (只需要 |T| \sum^{n}_{j = 1}|I_j| 個決策變數)
(D2) \quad \forall t\in T \quad \forall j \in \{1,2,3,...n\} \quad \sum_{i_j \in I_j} X^{jt}_{i_j} = 1
[小結] 在整數規劃裡,邏輯 0-1 是強大數學建模的工具,但邏輯的串聯是困難的部分使用正確的限制式需要時間練習與經驗,而書本鮮少提到,故筆者以此文分享給需要的讀者,更多限制式技巧如 TakeMin , TakeMax , Absolute , Piecewise Linear 可以參考 FICO Xpress 教學文件 ,關於圖論(Graph) 限制式,如 TSP,VRP,Shortest Path,Multi-commodity Flow 等等日後會再寫一篇 !!
\begin{array}{l}\text{ min } c_{\mathcal{I}}x_\mathcal{I} + c_{\mathcal{R}}x_\mathcal{R} \\ \text{s.t } Ax = b \\ x := (x_\mathcal{I} , x_\mathcal{R}) \geq 0 \\ \end{array}
其中 \mathcal{I} 為整數決策變數集 ,\mathcal{R} 為實數決策變數集,A 為已知係數
此模型已經有很多演算法可以求解詳細可以查關鍵字 branch & cut algorithm
理論上可以找到最佳決策 !! 但實務上整數規劃是 NP-Hard 問題,求解效率是議題!!,所以數學建模上我們希望 |\mathcal{I}| 越小越好!!
而現實問題為何可使用整數規劃來數學建模那麼重要原因有二:
========================================
1. 現實問題物品的數量為整數個 !!
2. 邏輯的 True , False 可以用 0 - 1 決策變數表示!!
========================================
本文介紹一些筆者經驗常見邏輯限制式代表的意義,幫助大家更能理解如何數學建模。而對於有興趣的讀者,可以去閱讀 "Paul Williams : Model Building in Mathematical Programming" 這本書,有詳細的介紹。
+++++++++++++++++++++++++++++++++++
(A) \text{ 從集合裡面挑選元素 }
+++++++++++++++++++++++++++++++++++
Let i \in I , |I| < \infty , x_i \in \{0,1\} , cost : I \longrightarrow \mathbb{R} , denote cost_i \in \mathbb{R}
x_i = 1 \Longleftrightarrow \text{ you select element in } i \in I
(A1) 從集合 I 裡面挑選恰好一個元素
\sum_{i \in I} x_i = 1
========================================
註 : 學術上我們稱 \displaystyle{\sum_{i\in I} x_i =1 } 為 Special Ordered Sets of type 1 constraint ,記做 SOS1(I) ,或寫作 x(I) = 1
========================================
(A2) 從集合 I 裡面挑選至少一個元素
\sum_{i\in I} x_i \geq 1
(A3) 從集合 I 裡面至多挑選 U 個,至少挑選 L個 (L,U\in \mathbb{Z})
L \leq \sum_{i\in I} x_i \leq U
(A4) 如果挑選 i \in I 都需要負擔/獲得 c_i \in \mathbb{R}
則總價格可以寫成 cost(I):= \sum_{i\in I}c_i x_i
(A5) 當然可以考慮子集合 S \subseteq I ,則價格可以寫成
cost(S):= \sum_{i\in S}c_i x_i
(A6) 因為 x_i = 1 代表的是種類 i 有挑選,但如果同種類要計量可以用"整數"決策變數 z_i \in \mathbb{Z^{\geq 0 }}= \{0,1,2,3,4,....\}
qcost(S):= \sum_{i\in S}c_i z_i
當然還可以考慮集合的集合 \mathcal{F}\subseteq 2^{I}= \{S : S\subseteq I \},其中 2^{I} 為 PowerSet of I ,還可以定義 Cost : 2^{I} \longrightarrow \mathbb{R} 類比 (A4) , (A5) 限制式,可以定義
(A7) \quad Cost(\mathcal{F}) := \sum_{S \in \mathcal{F}}cost(S)
(A8) \quad QCost(\mathcal{F}) := \sum_{S \in \mathcal{F}}qcost(S)
如果 z_i \leq U 則我們還可以增加一個維度只用 0-1變數 x_{ij} 來刻劃 z_i , x_i
(A9) \left\{\begin{array}{c} \displaystyle{z_i := \sum^{U}_{j=0} j \cdot x_{ij} \quad \forall i \in I} \\ \displaystyle{x_i:= \sum^{U}_{j=0} x_{ij}\leq 1 \quad \forall i \in I } \\ x_{ij} \in \{0,1\} \quad \forall (i,j) \in I \times \{0,1,2,....U\} \\ \end{array}\right.
+++++++++++++++++++++++++++++++++++
(B) \text{ if - then constraints}
+++++++++++++++++++++++++++++++++++
考慮 i \in I , y_i \in \{0,1\} ,記作 \bar{y}_i = 1-y_i
(B1) 四大基本,可用來串接邏輯 Proposition
if y_1 = 1 then y_2 = 1 :\qquad y_1 \leq y_2
if y_1 = 1 then y_2 = 0 :\qquad y_1 \leq 1-y_2 \Longrightarrow y_1 \leq \bar{y}_2
if y_1 = 0 then y_2 = 1 :\qquad 1-y_1 \leq y_2 \Longrightarrow \bar{y}_1 \leq y_2
if y_1 = 0 then y_2 = 0 :\qquad 1- y_1 \leq 1- y_2 \Longrightarrow \bar{y}_1 \leq \bar{y}_2
============================================
例如 : P \Longrightarrow Q , P\Longrightarrow R , R\Longrightarrow S , Q \Longrightarrow S 可以寫成
\left\{\begin{array}{c} y_P \leq y_Q \\ y_P \leq y_R \\ y_R \leq y_S \\ y_Q \leq y_S \\ \end{array}\right.
事實上這可以畫成一個邏輯有向網路圖(Directed Graph) G(N,A) 邊的方向性即為 "\leq",如例子: N=\{P,Q,R,S\} , A = \{ (P,Q), (P,R), (R,S) , (Q,S)\}
=============================================
Let a, x \in \mathbb{R}^J , b \in \mathbb{R} , [a,x] := \sum_{j \in J} a_{j}x_j
===================================
註 : [a,x]\geq b 可以改寫成 -b \leq -[a,x]
===================================
再來介紹經典的"開啟"限制式 !!
(B2) if (y = 1) then [a,x] \leq b , else [a,x] < \infty redundant !!
[a,x] \leq b + M(1-y)
註: M 是充分大常數,學術上稱為 Capital M/Big M Constraint,但不能太大(求解上不穩定),最好的 M := -b + \underset{x \in \mathbb{R}^J}{\text{sup }}[a,x]
再來是限制式"開啟" !!
(B3) if ( [a,x] \leq b ) then y=1 , else : redundant !!
b + \epsilon \leq [a,x] + My
========================================
證: 等價於 y = 0 \Longrightarrow [a,x] > b ,右邊引入極小的數 \epsilon , 變成 [a,x] \geq b + \epsilon,根據(B2)最後可寫成 b + \epsilon \leq [a,x] + My
最好的 M:= b + \epsilon - \underset{x \in \mathbb{R}^J}{\text{inf }}[a,x]
========================================
註 : 如果(B2)(B3)條件是 (y = 0) 只要把 y 代換成 \bar{y} 即可 !!
========================================
結合 (B2) (B3) 概念可以有 (B4) , (B5)
(B4) 開關限制式 [a,x] \leq b \Longleftrightarrow y =1
\left\{\begin{array}{c} [a,x] \leq b + M(1-y) \\ b + \epsilon \leq [a,x] + My \\ \end{array}\right.
(B5) 限制式開限制式 [a_1,x_1]\leq b_1 \Longrightarrow [a_2,x_2]\leq b_2
\left\{\begin{array}{c} b + \epsilon \leq [a_1,x_1] + My \\ [a_2,x_2] \leq b + M(1-y) \\ \end{array}\right.
========================================
證: 串聯 [a_1,x_1]\leq b_1 \Longrightarrow y = 1 , y = 1 \Longrightarrow [a_2,x_2]\leq b_2
========================================
+++++++++++++++++++++++++++++++++++
(C) And + Or Constraints !!
+++++++++++++++++++++++++++++++++++
(C1) if y = 1 then \underset{i \in I}{\bigwedge} ([a_i,x_i]\leq b_i) (Right-And Constraints)
\left\{\begin{array}{c} [a_1,x_1]\leq b_1 + M_1(1-y) \\ [a_2,x_2]\leq b_2 + M_2(1-y) \\ [a_3,x_3]\leq b_3 + M_3(1-y) \\ ... \\ [a_{|I|},x_{|I|}]\leq b_{|I|} +M_{|I|}(1-y) \\ \end{array}\right.
Let \forall i \in I \quad y_i \in \{0,1\}
(C2) if y = 1 then \underset{i \in I}{\bigvee}([a_i,x_i]\leq b_i) (Right-Or Constraints)
\left\{\begin{array}{c} [a_1,x_1]\leq b_1 + M_1(1-y_1) \\ [a_2,x_2]\leq b_2 + M_2(1-y_2) \\ [a_3,x_3]\leq b_3 + M_3(1-y_3) \\ ... \\ [a_{|I|},x_{|I|}]\leq b_{|I|} +M_{|I|}(1-y_{|I|}) \\ \displaystyle{y \leq \sum_{i\in I} y_i} \\ \end{array}\right.
==============================================
註: 學術上稱 (C2) 叫 disjunctive constraints ,可以輕易推廣成
if y = 1 then \displaystyle{\underset{i \in I}{\bigvee}\left\{\underset{j\in J_i}{\bigwedge}([a_j,x_j]\leq b_j)\right\}}
==============================================
(C3) if (\underset{i \in I}{\bigwedge}([a_i,x_i]\leq b_i) then y = 1 (Left-And Constraints)
\left\{\begin{array}{c} b_1 + \epsilon_1 \leq [a_1,x_1] + M_1(1-y_1) \\ b_2 + \epsilon_2 \leq [a_2,x_2] + M_2(1-y_2) \\ b_3 + \epsilon_3 \leq [a_3,x_3] + M_3(1-y_3) \\ ... \\ b_{|I|} + \epsilon_{|I|} \leq [a_{|I|},x_{|I|}] + M_{|I|}(1-y_{|I|}) \\ \displaystyle{ (1-y) \leq \sum_{i\in I} y_i} \\ \end{array}\right.
========================================
證: 等價於 if y = 0 then \underset{i \in I}{\bigvee}([a_i,x_i] > b_i)
========================================
(C4) if (\underset{i \in I}{\bigvee}([a_i,x_i]\leq b_i) then y = 1 (Left-Or Constraints)
\left\{\begin{array}{c} b_1 + \epsilon_1 \leq [a_1,x_1] + M_1 \cdot y \\ b_2 + \epsilon_2 \leq [a_2,x_2] + M_2 \cdot y \\ b_3 + \epsilon_3 \leq [a_3,x_3] + M_3 \cdot y \\ ... \\ b_{|I|} + \epsilon _{|I|}\leq [a_{|I|},x_{|I|}] + M_{|I|}\cdot y \\ \end{array}\right.
========================================
證: 等價於 if y = 0 then \underset{i \in I}{\bigwedge}([a_i,x_i] > b_i)
========================================
(C5) 互斥或,只有 [a_i,x_i] \leq b_i 其中之一會發生 (Disjoint Or Constaints)
\left\{\begin{array}{c} [a_1,x_1]\leq b_1 + M_1(1-y_1) \\ [a_2,x_2]\leq b_2 + M_2(1-y_2) \\ [a_3,x_3]\leq b_3 + M_3(1-y_3) \\ ... \\ [a_{|I|},x_{|I|}]\leq b_{|I|} +M_{|I|}(1-y_{|I|}) \\ \displaystyle{\sum_{i\in I} y_i = 1} \\ \end{array}\right.
+++++++++++++++++++++++++++++++++++
(D) 離散時間狀態網路圖(Time-State-Network)
+++++++++++++++++++++++++++++++++++
考慮離散時間 t \in T := \{0,1,2,........U\} (單位時刻點集)
考慮離散狀態向量集\displaystyle{i\in(i_1,i_2,i_3,....,i_n) \in I = \prod^{n}_{j=1} I_{j}}
註: 狀態向量集可能隨著時間 t 不同而不同,即 I^{(t)}
考慮狀態-時間函數 \displaystyle{v : T \times I \longrightarrow } \{0,1\}
很明顯我們一個時間點,只會有一個狀態
(D1) \quad \forall t \in T \quad \sum_{i\in I} v_{it} = 1 \quad \text{或寫成 } v_t(I) = 1
如果我們看 i 的每個分量,很明顯每個分量也只能選一種
因為(D1)的決策變數非常多,事實上如果 State 間是獨立的,只要定義 X^{jt}_{i_j} = 1 \Longleftrightarrow \text{at time }t , j\text{-th state is } i_j (只需要 |T| \sum^{n}_{j = 1}|I_j| 個決策變數)
(D2) \quad \forall t\in T \quad \forall j \in \{1,2,3,...n\} \quad \sum_{i_j \in I_j} X^{jt}_{i_j} = 1
[小結] 在整數規劃裡,邏輯 0-1 是強大數學建模的工具,但邏輯的串聯是困難的部分使用正確的限制式需要時間練習與經驗,而書本鮮少提到,故筆者以此文分享給需要的讀者,更多限制式技巧如 TakeMin , TakeMax , Absolute , Piecewise Linear 可以參考 FICO Xpress 教學文件 ,關於圖論(Graph) 限制式,如 TSP,VRP,Shortest Path,Multi-commodity Flow 等等日後會再寫一篇 !!
[以上純為學術經驗交流知識分享,如有錯誤或建議可留言~~]
by Plus & Minus 2017.09
留言
張貼留言