15h30 - 15h55
Optimal design of data centers.
We present our current research on data centers design. Our goal is to find a network on 64 vertices with a small average distance and a bounded maximum degree, and some robustness properties. First, we establish some mathematical properties and then we find some interesting candidates by optimization.
15h55 - 16h20
Integer column generation using decomposition
Integer column generation using decomposition (ICG) is a new primal method that aims to solve the popular set partitioning problem. This method finds a sequence of integer solutions, with non-increasing cost, leading to optimal or near-optimal solutions in reasonable time. Potential columns favoring integrality are generated using a suited dual vector. Some acceleration strategies improving the effectiveness of ICG will be discussed. Computational experiments on some large-scale bus drivers scheduling and aircrew pairing problems will be presented. The results obtained demonstrate the efficiency of ICG.
16h20 - 16h45
ANNULÉ / Pick-up and delivery with complex loading constraints: application to the gasoline distribution
In this work, we present a Branch & Price method to solve a real-world pick-up and delivery problem arising in the sector of the distribution of gasoline. The underlying network consists of four distinct depots, a group of private carriers with heterogeneous tank trucks and five types of gasoline to replenish three groups of customers on a weekly basis. Complex loading and routing rules are handled in the sub-problem, a very difficult shortest path problem with resource constraints. Acceleration strategies will be discussed. Numerical results on real data show the high potential of the proposed approach.
16h45 - 17h10
Primal Neighbourhood Search Algorithm for solving the Shortest Path Problem with Resource Constraints
We propose a new exact primal method for solving the shortest path problem with resource constraints. Our algorithm performs a search in the neighbourhood of a set of source-task paths. We first define the notion
of adjacency in the context of the SPPRC. Then, we develop some polyhedral properties that are useful in the definition and exploration of the neighbourhood. Computational results on the VCSP show that the proposed solution approach is more efficient than classical dynamic programming algorithm.