1367. Addressing multi-actor decision-making problems under uncertainty: an application on public private partnerships for infrastructure projects
Invited abstract in session HE-15: Multilevel and Stochastic Optimization , cluster Multilevel and Stochastic Optimization Methods.
Thursday, 16:15-17:45Room: FENP208
Authors (first author is the speaker)
| 1. | Samuel Rodriguez Gonzalez
|
| School of Industrial and Systems Engineering, University of Oklahoma | |
| 2. | Juan Beltrán
|
| Universidad de los Andes | |
| 3. | Santiago Bobadilla
|
| Universidad de los Andes | |
| 4. | Andres Gonzalez
|
| School of Industrial and Systems Engineering, University of Oklahoma | |
| 5. | Carlos Lozano
|
| Universidad de los Andes | |
| 6. | Camilo Gomez
|
| Industrial Engineering, Universidad de los Andes |
Abstract
Multi-actor decision-making problems are pervasive in practice, where individuals interact with each other (supply chains, energy systems, finance). Modeling decisions anticipating the reactions of others is challenging for practitioners, especially when considering combinatorial problems and non-deterministic behaviors.
The principal-agent problem is an instance of these problems, occurring when an agent, hired by the principal to perform a task, takes advantage of asymmetries of information to pursue its own goals, potentially in detriment to the principal’s. Public-private-partnerships (PPPs) are prone to these behaviors, as governments rely on contractors to provide goods and services where their supervision capacity is limited. We explore an infrastructure maintenance problem in the context of a PPP to analyze the decision strategies of the involved ‘players’ under diverging objectives. Strategies and outcomes for both ‘players’ being affected by uncertainty and the other player’s decisions is a key challenge for this analysis: the contractor’s cashflows are affected by the government’s inspection policies, while the societal benefit of the system is affected by stochastic deterioration and the contractor’s maintenance policy.
We adopt an approximate dynamic programming approach allowing us to integrate different modeling and solution approaches into a unified framework. The latter is achieved through the implementation of a computational environment based on Markov Decision Processes capturing the decisions of both players and the stochastic deterioration process of the system. The environment provides a general representation of the problem; with it, we can evaluate and compare different mechanisms to generate decision strategies for players (i.e., policies). We evaluate policies based on the following: rules-of-thumb, mixed-integer stochastic bilevel linear programming, and math-heuristic approaches tailored for bilevel problems.
Our contribution is twofold: first, the proposed environment provides the basis for a simulator enabling PPP participants and regulators in assessing the potential outcomes of a specific PPP project (including the evaluation of players’ likely strategies); and, second, the set of policy generating techniques shed light about key tradeoffs in PPP projects and may provide decision-support tools for PPP participants with varying levels of complexity.
Keywords
- Decision Support Systems
- Bilevel programming
- Stochastic Optimization
Status: accepted
Back to the list of papers