Developing Efficient Cooperative Solvers for Constrained Optimization
Yi Shang, Markus P.J. Fromherz, and Ying Zhang
Abstract
When it is impossible to find an optimal solution of a nonlinear
constrained optimization problems within a limited amount of time, the
goal usually becomes finding a sub-optimal solution as good as
possible. In this paper, we study time and quality trade-offs on
continuous constrained optimization problem. We focus on cooperative
solvers that consist of two optimization methods, sequential quadratic
programming (SQP) and quasi-Newton method, and experiment with a
sequential form of composition: quasi-Newton followed by SQP. In the
experiments, we analyze the performance of the cooperative solvers on
a series of problems with increasing constraint-to-variable ratios,
and compare that with SQP in terms of number of function evaluation
and solution quality. The cooperative solvers generally get much
better solutions than SQP, while spending comparable amount of time.
They are flexible and could achieve good results on time-bounded
applications by setting parameters according to the time limits.
PDF file
Back to the top.