- Power System Planning under Multi-Scale Uncertainty: Modeling and Solution Methods
Ángela Flores – DIE, U. Chile | Abstract - Multiperiod, stochastic and dynamic problems in transportation (network design with temporary contract and disruption optimization in picking-packing)
Thierry Pironet – HEC University of Liège | Abstract - Developing decision-making tools for electromobility planning
Mónica Zamora Zapata – DIMEC, U. Chile – CASE | Abstract - Robust network design with equilibrium flow
Fernando Ordóñez – DII U. Chile | Abstract - Contextual Bandits-Guided Local Search for Solving Air Cargo Palletisation Problem
Sabine Limbourg – HEC University of Liège | Abstract - Online Scheduling with Concave Utilities: Simple Algorithms and Bounds
José Soto – CMM-DIM U. Chile | Abstract
Schedule
| time | Talk |
| 09:30 – 10:00 | Welcome Coffee |
| 10:00 – 10:30 | Power System Planning under Multi-Scale Uncertainty: Modeling and Solution Methods (Speaker: Ángela Flores – DIE, U. Chile) |
| 10:40 – 11:10 | Multiperiod, stochastic and dynamic problems in transportation (network design with temporary contract and disruption optimization in picking-packing) (Speaker: Thierry Pironet – HEC University of Liège) |
| 11:20 – 11:50 | Developing decision-making tools for electromobility planning (Speaker: Mónica Zamora Zapata – DIMEC, U. Chile – CASE) |
| 11:50 – 15:00 | Free time |
| 15:00 – 15:30 | Robust network design with equilibrium flow (Speaker: Fernando Ordóñez – DII U. Chile) |
| 15:40 – 16:10 | Contextual Bandits-Guided Local Search for Solving Air Cargo Palletisation Problem (Speaker: Sabine Limbourg – HEC University of Liège) |
| 16:20 – 16:50 | Online Scheduling with Concave Utilities: Simple Algorithms and Bounds (Speaker: José Soto – CMM-DIM U. Chile) |
Talks
Power System Planning under Multi-Scale Uncertainty: Modeling and Solution Methods
Ángela Flores
The deep decarbonization of power systems necessitates coordinated action and substantial investment to adapt to significant changes in the energy matrix and to address uncertainties posed by climate change. On one hand, the successful integration of variable renewable energy requires sufficient operational flexibility to manage short-term variability and uncertainty. On the other hand, rising average temperatures, more variable precipitation patterns, and the increasing frequency and severity of extreme weather events introduce long-term risks that must be anticipated in planning models to ensure a secure and resilient energy supply.
In this talk, we present past and ongoing work in the area of power system planning under uncertainty. First, we introduce a capacity expansion planning model that integrates short-term operational constraints with strategic long-term uncertainty. The resulting formulation is a large-scale multistage stochastic optimization problem, which poses significant computational challenges. To address these, we develop a distributed solution method based on the Column Generation algorithm. Results show that the proposed methodology reduces solution time significantly and allows to solve instances that cannot be handled by existing methods. Finally, we explore ongoing extensions of the model to incorporate short-term operational uncertainty and discuss suitable decomposition methods and enhancements to further improve solution times.
Multiperiod, stochastic and dynamic problems in transportation (network design with temporary contract and disruption optimization in picking-packing)
Thierry Pironet
We will present some past and on-going researches addressing “stochastic and/or dynamic problems”, closer to real-life contexts. The challenges reside in generating and evaluating the performances of a sequence of decisions related to either a multiperiod framework (days-months) or an on-going daily plan facing several disruptions. Firstly, a general introduction will be performed on the values of information in a multi-period stochastic and dynamic problem. Then, a network design problem with temporary capacity commitment contracts with projected demands will be described. Finally, a time-dependent vehicle routing problem focusing on picking and 3D packing operations will be described within a context of disruptions. We will mainly focus on the problem descriptions and managerial insights drawn by the results. This talk aims to give some insights in order to face these kinds of real-life problems instead of their mono-period deterministic version leading to disconnected one-time decision.
Developing decision-making tools for electromobility planning
Mónica Zamora Zapata
Planning future electromobility infrastructure presents several challenges and uncertainties that can be assessed by simulating multiple scenarios. In this talk, we present two works related to electromobility. The first work is a multi-model platform for sustainable electromobility analysis and planning, where different modeling tools are combined to simulate scenarios for whole cities and expand their results to a national level. In this way, projected transportation and decarbonization trends meet city expansion models, agent-based simulations of electric vehicles, and power distribution modeling to deliver national energy system expansion plans with associated economic and environmental indicators. In the second work, we explore the design of a commercial ultra-fast charging station with solar and storage integration, where electric vehicle arrival, charging needs, and waiting time tolerance are specified with parametric distributions, queues form at the chargers, and energy can be bought or sold from the electric grid. Our current analysis remarks on the importance of using realistic charging curves when designing ultra-fast charging stations since simplified charging curves alter the energy balance and economic performance.
Robust network design with equilibrium flow
Fernando Ordóñez
In this work we consider the network design problem where the network flow satisfies a Wardrop Equilibrium and there is demand uncertainty. We present a solution method for the deterministic network design problem with equilibrium flow for both a network with linear travel costs and capacities or a network with BPR latency functions and no capacity constraints. Our solution method formulates this as a mixed integer programming problem that solves a linear approximation of the network design problem. We also show that the robust network design solution, for a polyhedral uncertainty set on the demand, is bounded within the price of anarchy of the maximum network design problem over the extreme points of the demand uncertainty set. Our preliminary computational results show that the proposed model can protect against the worst case outcomes at a small increase in cost on expected demand scenarios.
Contextual Bandits-Guided Local Search for Solving Air Cargo Palletisation Problem
Sabine Limbourg
We address the challenges of efficiently assigning items to Unit Load Devices within the air cargo industry. We present a comprehensive formulation of the three-dimensional air cargo palletisation problem, focusing on cost minimisation and incorporating grouping, positioning, and compatibility constraints. We propose a set of 12 resolution approaches that utilise contextual bandits-guided local search heuristics. We conduct a thorough benchmark experiment to evaluate the performance of our proposed methods. Two objective functions, namely unused volume and costs are employed to underscore the significance of cost minimisation in air cargo palletisation. Furthermore, we address instances encompassing grouping, positioning, and compatibility constraints, enabling us to explore the managerial insights these constraints offer and assess the benefits of integrating cost-reduction strategies. The findings provide valuable insights for decision-makers involved in optimising air cargo palletisation operations.
Online Scheduling with Concave Utilities: Simple Algorithms and Bounds
José Soto
We consider an online scheduling problem where n tasks arrive one by one and must be assigned immediately, without knowledge of future tasks, to one of m identical machines. Each task has a processing time, and each machine’s utility is given by a concave function of its total load. This models diminishing returns—for example, when machines generate revenue over time but future income is discounted, or when the goal is to maximize the amount of work completed before a common deadline. In both cases, spreading work across machines leads to higher total utility.
A natural baseline is to assign each task uniformly at random, which achieves a competitive ratio of at least 1 – 1/e.
In this talk, we’ll discuss some of our results. We show that the classic List Scheduling algorithm achieves a competitive ratio of at least 3/4. We prove that no deterministic online algorithm can exceed φ/2 ≈ 0.809, where φ is the golden ratio, and that no randomized algorithm can exceed 9/10. For the case of two machines, we design an optimal deterministic algorithm, achieving φ/2, and a randomized one achieving 5/6.