[Paper Review] Robust Optimization for Tree-Structured Stochastic Network Design

Xiaojian Wu, Akshat Kumar+2 more|arXiv (Cornell University)|Dec 1, 2016

Fish Ecology and Management Studies11 references4 citations

TL;DR

This paper proposes a robust optimization framework for tree-structured stochastic network design under interval uncertainty in edge probabilities, specifically applied to river network barrier removal for fish conservation. It introduces a bilevel optimization approach combining mixed-integer linear programming and a fully polynomial-time approximation scheme (FPTAS) via rounded dynamic programming (RDP) to maximize robust ratio and minimize regret, demonstrating superior robustness on large-scale river networks compared to baseline methods.

ABSTRACT

Stochastic network design is a general framework for optimizing network connectivity. It has several applications in computational sustainability including spatial conservation planning, pre-disaster network preparation, and river network optimization. A common assumption in previous work has been made that network parameters (e.g., probability of species colonization) are precisely known, which is unrealistic in real- world settings. We therefore address the robust river network design problem where the goal is to optimize river connectivity for fish movement by removing barriers. We assume that fish passability probabilities are known only imprecisely, but are within some interval bounds. We then develop a planning approach that computes the policies with either high robust ratio or low regret. Empirically, our approach scales well to large river networks. We also provide insights into the solutions generated by our robust approach, which has significantly higher robust ratio than the baseline solution with mean parameter estimates.

Motivation & Objective

• To address the limitation of prior stochastic network design methods that assume precisely known network parameters, which is unrealistic in ecological applications.
• To develop a scalable planning approach for river network design where fish passage probabilities are only known within interval bounds.
• To compute policies that are robust to parameter uncertainty by maximizing the robust ratio and minimizing regret under adversarial parameter selection.
• To enable management actions that modify both interval bounds and network structure, unlike prior work that only altered network structure.
• To provide a practical computational tool for environmental scientists facing imprecise ecological models in conservation planning.

Proposed method

• Formulates the robust network design problem as a bilevel optimization: the outer level maximizes the robust ratio over decision policies, and the inner level minimizes this ratio by selecting adversarial parameters within interval bounds.
• Encodes the decision optimization problem as a mixed-integer linear program (MILP) to find the best policy under fixed adversarial parameters.
• Develops a rounded dynamic programming (RDP) algorithm as an FPTAS for the inner ratio minimization problem, which is computationally harder than the outer problem.
• Uses a constraint generation procedure that iteratively adds the most adversarial parameter sets and policies to refine the solution space.
• Employs a rounding technique to approximate the optimal adversarial policy and probabilities efficiently, ensuring polynomial-time convergence with bounded error.
• Applies the framework to tree-structured river networks where nodes represent river segments and edges represent barriers with uncertain passage probabilities.

Experimental results

Research questions

• RQ1 Can a robust optimization framework be developed for tree-structured stochastic network design when edge probabilities are only known within interval bounds?
• RQ2 How can management actions that modify both interval bounds and network structure be effectively integrated into robust network design?
• RQ3 Is the Sample Average Approximation (SAA) approach applicable to this richer action space, or is a new method required?
• RQ4 Can a bilevel optimization approach with FPTAS-based inner maximization produce policies with higher robust ratio and lower regret than baseline methods?
• RQ5 How does the robustness of the computed policy compare to policies based on mean estimates or worst-case assumptions?

Key findings

• The proposed RDP algorithm produces robust ratios greater than the optimal value (OPT) and regrets smaller than OPT, indicating near-optimality in adversarial settings.
• The RDP algorithm runs 100–600 times faster than standard dynamic programming on a 22-node network, with negligible loss in solution quality.
• On a 2028-node river network, robust ratios converge within 2 minutes, indicating that a moderate value of K (set to K_u = K) is sufficient for near-optimal performance.
• The MRR policy (maximizing robust ratio) achieves the highest robust ratio and the lowest regret, demonstrating a strong correlation between the two robustness metrics.
• Visualization shows that the MRR policy allocates budget across multiple critical areas, preventing adversarial exploitation, unlike the midpoint policy that concentrates actions along the main stem.
• On a 9335-node network with 5% budget, the MRR policy maintains high robustness even without tuning K to the convergence area, confirming scalability and robustness.