A transportation programming model considering project interdependency and regional balance The 20th International Symposium on Transportation and Traffic Theory Noordwijk, the Netherlands July 18, 2013 Kuancheng Huang* and Yi-Ming Kuo Dept. of Transportation and Logistics Management National Chiao Tung University, Taiwan 1 Outlines • • • • • Introduction Problem Background and Literature Review Mathematical Model Solution Algorithm Conclusions 2 Introduction Transportation Programming (TP) Given a fixed budget, how to select the projects to be implemented? 25 transportation projects in Taiwan with a cost larger than 1 billion NTD in 2010 3 Introduction Knapsack Problem (KP) Approach G (budget) Issue I: Project Inter-dependency -bj (benefit of project j) Issue II: Regional Balance mj (cost of project j) 4 Literature Review & Background Example of Political Influence • The districts receiving the most funds are … (Spanish data for the period 1964–2004) – The incumbent’s vote margin is low. – Few votes are needed to win an additional seat. – The central and regional governments are controlled by the same party. – Regional parties play a pivotal role in the legislature. Solé-Ollé, A. (2013), Inter-regional redistribution through infrastructure investment: tactical or programmatic, Public Choice, 156:229–252. 5 Literature Review & Background TP Works w/ Dependent Projects Independent Complementary Substitutive Teng & Tzeng (1996), A multiobjective programming approach for selecting 6 non-independent transportation investment alternatives, TR-B, 30: 291-307. Literature Review & Background Iniestra and Gutiérrez (2009) • Follow the multi-objective context and project-interdependent features – The average benefit/cost ratio of project portfolio – The relevance to network connectivity – State authorities’ preferences (an assigned importance rank) – The total level of congestion in the region – The social impact (# of people who benefit) • Solve the problem by Genetic Algorithm Li et al. (2012), New Methodology for Transportation Investment Decisions 7 with Consideration of Project Interdependencies, TRR, 2285:36-46. Literature Review & Background Modeling Inter-dependency • Basic project • Joint project: the overall benefit (or cost) is different from the sum of the individual benefits (or costs) of the basic projects. x1 x2 x3 x1 x2 1 x1 x3 1 x2 x3 1 8 Literature Review & Background Other Exclusive Relations • “Intrinsic” exclusive relations (beyond the issue of joint projects) – For one single initiative, the various versions (e.g., a one-lane or two-lane expansion of a highway section). – For multiple initiatives, the corresponding options (e.g., a public bus terminal vs. a commuter parking facility) cannot be selected simultaneously (e.g., utilizing the same piece of land). 9 Mathematical Model Integer Programming Model Minimize s.t. π b x jJ m x jJ to be relaxed f jJ λi j j j j Minimize negative benefit (Maximize benefit) G Budget constraint Exclusive Relations, e.g., rj x j 1, r R x j {0, 1}, j J x1 x2 1 x1 x3 1 x2 x3 1 Regional balance: project j a x 1 , i I ij j jJ contributes aij to region i. A modification to SCP (Set Covering Problem) 10 Significance (Relevance) Level A 51.03-km MRT line (140 billion NTD) connecting Taipei city center and the major int’l airport Industry Airport (Aerotropolis) 1.00 Heavily Populated Suburban 0.75 Capital City Economic Center 0.25 11 Mathematical Model Lagrangian Relaxed Model c j (, ) b j i aij m j iI L(, ) Min c j (, ) x j ( i G) jJ f jJ rj x j 1, r R x j {0, 1}, j J 0 x j 1, j J Linearly relax the LR problem. iI Exclusive Relations, e.g., x1 x2 1 x1 x3 1 x2 x3 1 Constraints similar to those of the classic assignment problem (AP), which holds the Total Unimodularity property. 12 1. 2. Set the iteration counter n = 0 Determine the initial values of the Lagrangian multipliers (Sub-section 3.4) 1. 2. 3. Increment the counter from n to n +1 Compute Lagrangian cost cj(λ, π) (12) Solve linearized Lagrangian-relaxed problem (Sub-section 3.3) Compute gradients (15) and (17) 4. Solution Algorithm No Solution feasible? Follow the 3-step procedure for the feasible solution (Sub-section 3.4) Yes 1. 2. Compute the upper bound Update Lagrangian multipliers (14) and (16) n = N or gap < ? No Yes Derive solution Flowchart of Algorithm 13 Mathematical Model Feasible Solution c j (, ) b j i aij m j iI • STEP 1: If the significance constraint (5) is violated, add the unselected option based on the Lagrangian cost cj(, π) in “ascending” order until Constraint (5) is satisfied. • STEP 2: If the budget constraint (6) is violated, remove the selected options (a) based on the Lagrangian cost cj(, π) in “descending” order and (b) under the condition of satisfying Constraint (5), until the overall budget is under the limit. • STEP 3: If there is a slack for Constraint (6), further add the option with the most negative Lagrangian cost cj(, π) until the budget is fully utilized. • Note: For STEP 1 and STEP 3, an option is added only if the mutually exclusive constraint (7) is not violated. 14 Mathematical Model Lagrangian Multiplier Update Sub-gradient Method (Held & Karp, 1970) si (t ) 1 t t ), i I , ( x j j J i , i I s( t ) B c j x j (t , t ) j J i t t t ) , ( L UB t 0 ), ( s ti 1 max ti , i I i t 2 s ( ) t t t ) , ( L UB t 0 ), ( s t 1 max t t 2 s( ) 15 Numerical Experiment Illustrative Example International Airport RWY #1 Improvement International Airport MRT System Local Railway MRT Transformation (Section I) National Freeway #4 Extension National Freeway #6 Extension National Freeway #6 Extension West Coast Highway Improvement Chiayi City Local Railway Overheadization Highway #.9 Mountain Section Improvement Highway #11 Pacific Coast Highway Improvement Highway #9 Phase III Improvement Local Railway MRT Transformation (Section IV) 考慮涵蓋限制 16 Benefit maximization only預算值=2000 Imposing coverage constraint 未考慮涵蓋限制 Numerical Experiment Results of the Illustrative Example Price for compromising! 17 Numerical Experiment Hypothetical Test Problems Significance Level Setting Project Interdependence 18 Numerical Experiment Results of Hypothetical Problems Number of basic projects 50 100 150 200 Number of regions Budget Average number of options Average number of constraints 25 49 81 121 10000 20000 30000 40000 265 570 821 1012 455 993 1424 1745 Number of IP time (s) LR time (s) basic projects 50 100 150 200 0.9 11.5 66.5 225.0 3.4 5.9 12.4 14.7 Solution quality Mean Std. Dev. 0.53% 0.63% 0.20% 0.15% 0.32% 0.17% 0.63% 0.60% Bound quality Mean Std. Dev. 1.76% 1.27% 1.07% 0.65% 1.67% 0.54% 1.90% 0.80% 30 test problems of each problem scale 19 Conclusions Conclusions • An IP model addressing two practical TP features: project interdependency and regional balance. – Two Things You Shouldn't Watch Being Made: Sausage and Legislation (Mark Twain) • A Lagrangian-relaxed solution algorithm to derive a promising approximate solution and a tight bound . • Future Research Extensions – A multi-stage model to address the issue of consistency – A stochastic model for the issue of uncertainty – A multi-objective model, given the diversified stakeholders 20