Supply-Chain Network Design
SCND · MULTI-ECHELON STRATEGIC MIP
Logistics · Network & Facility · StrategicSupply-Chain Network Design (SCND) is the archetypal strategic logistics problem: given a set of suppliers, candidate plants, candidate distribution centres, and customers, decide which facilities to open, at what capacity, and how product should flow from tier to tier — all to minimise total fixed + variable cost subject to demand, capacity, and service-level constraints. A decade-long decision that locks in a carrier's or manufacturer's operating cost structure. The foundational paper is Geoffrion & Graves (1974); the best survey is Melo, Nickel & Saldanha-da-Gama (2009) "Facility location and supply chain management".
Problem scope
What SCND decides
SCND integrates facility location (which sites to open) with multi-commodity flow (how product moves) across a multi-echelon network — typically suppliers → plants → DCs → customer zones. Unlike the single-tier CFLP/UFLP, SCND must decide flows through intermediate tiers, respect capacities at each tier, and often handle multiple products with different demand patterns.
Extensions add complexity: multi-period (facility opening/closing over a planning horizon), stochastic demand (two-stage or robust), multi-objective (cost vs. service vs. CO₂), and closed-loop (reverse flows for returns and remanufacturing). The MIRHA / Geoffrion-Graves family of benchmarks remains the reference for deterministic SCND; more recent reviews cover stochastic and sustainable variants.
Formulation sketch
Multi-commodity 2-echelon MIP (Geoffrion-Graves 1974)
Notation
| Symbol | Meaning |
|---|---|
| $I$ | set of candidate plants / DCs (facilities to open) |
| $J$ | set of customers / demand zones |
| $K$ | set of products |
| $d_j^k$ | demand at customer $j$ for product $k$ |
| $f_i$ | fixed opening cost at facility $i$ |
| $u_i$ | capacity (in common unit) at facility $i$ |
| $c_{ij}^k$ | unit flow cost of product $k$ from $i$ to $j$ |
| $y_i \in \{0,1\}$ | 1 if facility $i$ is opened |
| $x_{ij}^k \ge 0$ | flow of product $k$ from $i$ to $j$ |
Objective
Constraints
Extensions. Multi-echelon (plants → DCs → customers) adds intermediate nodes with flow conservation. Multi-period adds $t$ on every variable with open/close transitions. Stochastic SCND (two-stage SP or robust) hedges against demand-scenario uncertainty. Closed-loop SCND (Govindan, Fattahi & Keyvanshokooh 2017 review) adds reverse flows from customers back to remanufacturing facilities.
Solution methods. Commercial MIP (Gurobi, CPLEX) handles deterministic instances up to ~10,000 variables. Decomposition (Benders, Lagrangian) is standard for larger or stochastic cases. See Cordeau, Pasin & Solomon (2006) and Snyder (2006) for stochastic-SCND treatments.
Key references
DOIs & permanent URLs