This paper presents the approach that we developed to solve the ROADEF 2003 challenge problem. This work is part of a research program whose aim is to study the benefits and the computer-aided generation of hybrid solutions that mix constraint programming and meta-heuristics, such as large neighborhood search (LNS). This paper focuses on three contributions that were obtained during this project: an improved method for propagating Hamiltonian chain constraints, a fresh look at limited discrepancy search and the introduction of randomization and de-randomization within our combination algebra. This algebra is made of terms that represent optimization algorithms, following the approach of SALSA [1], which can be generated or tuned automatically using a learning meta-strategy [2]. In this paper, the hybrid combination that is investigated mixes constraint propagation, a special form of limited discrepancy search and large neighborhood search.

Fractionnal mathematical programs appear in numerous operations research, computer science and economic domains. We consider in this paper the problem of maximizing the sum of 0–1 hyperbolic ratios (SRH). In contrast to the single ratio problem, there has been little work in the literature concerning this problem. We propose two mixed-integer linear programming formulations of SRH and develop two different strategies to solve them. The first one consists in using directly a general-purpose mixed-integer programming solver. The second one is based on a specialized branch and bound algorithm that reformulates more precisely the problem at every node of search tree. We also propose a heuristic method and we exploit the obtained solution in order to improve the first strategy. We present computational experiments that allow to compare the different approaches.

In the course of globalization, many enterprises change their strategies and are coupled in partnerships with suppliers, subcontractors and customers. This coupling forms supply chains comprising several geographically distributed production facilities. Production planning in a supply chain is a complicated and difficult task, as it has to be optimal both for the local manufacturing units and for the whole supply chain network. In this paper two analytical models are used to solve the production planning problem in supply chain involving several enterprises. Generally in practice, for competitive and/or practical reasons, frequently each enterprise prefers to optimize its production plan with little care about the other members of the supply chain. This case is presented through a simple model of decentralized optimization. The aim of this study is to analyze and compare the two types of optimization: centralized and decentralized. The initial question is: what are the profit and the optimal policy of global (centralized) optimization in contrast to local (decentralized)? We characterize this gain by comparing the optimal profits obtained in both cases.

In on-line computation, the instance of the problem dealt is not entirely known from the beginning of the solution process, but it is revealed step-by-step. In this paper we deal with on-line independent set. On-line models studied until now for this problem suppose that the input graph is initially empty and revealed either vertex-by-vertex, or cluster-by-cluster. Here we present a new on-line model quite different to the ones already studied. It assumes that a superset of the final graph is initially present (in our case the complete graph on the order n of the final graph) and edges are progressively removed until the achievement of the final graph. Next, we revisit the model introduced in [Demange, Paradon and Paschos, Lect. Notes Comput. Sci.1963 (2000) 326–334] and study relaxations assuming that some paying backtracking is allowed.

A bound for the greedy heuristic applied to the K-facility location problem can be calculated, using values gathered during the calculation of the heuristic. The bound strengthens a well-known bound for the heuristic. Computational experiments show that this bound can be beneficial when the number of facilities is small or close to the total number of potential sites. In addition, it is consistent with previous results about the influence of the data characteristics upon the optimal value.

Periodic Vehicle Routing Problem: classification and heuristic for tactical planning. The Periodic Vehicle Routing Problem (PVRP) consists in assigning customer visits to vehicle routes in some periods of a time horizon so as to satisfy some service level requirements that can take the form of frequency of visit, constraint on time lag between visits, or pre-defined visit patterns. We present different variants of this problem and propose a classification. Then, we consider a model for tactical planning for which we propose a heuristic: we optimise the planning of customer visits to achieve both workload balancing and regionalisation of the routes. The objective of regionalisation reflects a desire to specialize routes to restricted geographical area. The standard minimisation of distance travelled is left for the underlying operational decision making model. Our heuristic achieves practical solutions for an industrial instance with 16658 visits to schedule over a horizon of 20 days.

Markov Decision Processes (MDPs) are a classical framework for stochastic sequential decision problems, based on an enumerated state space representation. More compact and structured representations have been proposed: factorization techniques use state variables representations, while decomposition techniques are based on a partition of the state space into sub-regions and take advantage of the resulting structure of the state transition graph. We use a family of probabilistic exploration-like planning problems in order to study the influence of the modeling structure on the MDP solution. We first discuss the advantages and drawbacks of a graph based representation of the state space, then present our comparisons of two decomposition techniques, and propose to use a global approach combining both state space factorization and decomposition techniques. On the exploration problem instance, it is proposed to take advantage of the natural topological structure of the navigation space, which is partitioned into regions. A number of local policies are optimized within each region, that become the macro-actions of the global abstract MDP resulting from the decomposition. The regions are the corresponding macro-states in the abstract MDP. The global abstract MDP is obtained in a factored form, combining all the initial MDP state variables and one macro-state “region” variable standing for the different possible macro-states corresponding to the regions. Further research is presently conducted on efficient solution algorithms implementing the same hybrid approach for tackling large size MDPs.

In the wine AOC system, the regulation of quantities performed by the professional organizations is aimed to smooth the variations of the quality of the wine due to the variations in the climate that affect the quality of the grapes. Nevertheless, this regulation could be damaging to the consumers due to the price increase resulting from the reduction of the quantities sold on the market. We propose a stochastic control model and a simulation tool able to measure the effects of this mechanism on the producer's profit and the consumer's surplus.