Optimization Software and Systems for Operations Research: Best Practices and Current Trends

Abstract

For a great variety of large-scale optimization problems arising in Operations Research applications, it has become practical to rely on "off-the-shelf" software, without any special programming of algorithms. As a result the use of optimization within business systems has grown dramatically in the past decade.

One key factor in this success has been the adoption of model-based optimization. Using this approach, an optimization problem is conceived as a particular minimization or maximization of some function of decision variables, subject to varied equations, inequalities, and other constraints on the variables. A range of computer modeling languages have evolved to allow these optimization models to be described in a concise and readable way, separately from the data that determines the size and shape of the resulting problem that may have thousands (or even millions) of variables and constraints.

After an optimization problem is instantiated from the model and data, it is automatically put into a standard mathematical form and solved by sophisticated general-purpose algorithmic software packages. Numerous heuristic routines embedded within these packages enable them to adapt to many problem structures without any special effort from the model builder.

The evolution and current state of both modeling and solving software for optimization will be presented in the main part of this talk. The presentation will then conclude with a consideration of current trends and likely future directions.