Comparison of Simulated Annealing, Interval Partitioning and Hybrid Algorithms in Constrained Global Optimization

Abstract

The continuous Constrained Optimization Problem (COP) often occurs in industrial applications. In this study, we compare three novel algorithms developed for solving the COP. The first approach consists of an Interval Partitioning Algorithm (IPA) that is exhaustive in covering the whole feasible space. IPA has the capability of discarding sub-spaces that are sub-optimal and/or infeasible, similar to available Branch and Bound techniques. The difference of IPA lies in its use of Interval Arithmetic rather than conventional bounding techniques described in the literature. The second approach tested here is the novel dual-sequence Simulated Annealing (SA) algorithm that eliminates the use of penalties for constraint handling. Here, we also introduce a hybrid algorithm that integrates SA in IPA (IPA-SA) and compare its performance with stand-alone SA and IPA algorithms. All three methods have a local COP solver, Feasible Sequential Quadratic Programming (FSQP) incorporated so as to identify feasible stationary points. The performances of these three methods are tested on a suite of COP benchmarks and the results are discussed.