03:30 PM - 05:10 PM
Modeling and optimizing operations research problems with LocalSolver
In this tutorial, we introduce LocalSolver, a heuristic solver for large-scale optimization problems. The optimization model can be provided as a classic mathematical problem as LocalSolver is not restricted to a specific structure and supports most of the common mathematical operators. Decisions can be booleans, integers or continuous. The solver is designed to find good solutions for large problems in short running times. The main part of the tutorial deals with the new features of LocalSolver 7.
LocalSolver 6.5 introduced high level decisions based on sets that allow users to write simpler models. These decisions are based on the notion of list inspired from Constraint Programming Set-Based Variables. The new models are more compact for the solver and solutions can be found for larger instances. LocalSolver 7 adds new features to simplify the usage of variable length lists so that fixed arrays and variable lists are similar for the user. These new features are particularly useful to model routing and scheduling problems in just a few lines.
LocalSolver 7 also introduces new large neighborhood for numerical optimization based on a better exploitation of the model. These new intensification moves accelerate the convergence speed to local optima in nonlinear continuous optimization and leave more time for diversification.
The conclusion of the tutorial is about the future developments of LocalSolver, namely the extensions of set based models, the computation of lower bounds on generic models and the detection of global structures.