The EZopt Optimisation Framework
Tim Hendtlass · 2019
This work demonstrates the advantage that can be obtained by dividing optimisation into two separate activities, one to locate promising areas to search and the other to conduct a local search of such promising areas, and assigning these two activities to different algorithms. Promising areas, once searched, can be marked so that they are not searched again, improving efficiency. A framework is presented that enables an implementation of this approach. In this paper Particle Swarm Optimisation (PSO) is used to find promising areas and Directed Random Search (DRS) is used to perform the local optimisation, but there are many alternate algorithms that could be used. It is demonstrated that this framework is suitable for both static and dynamic problems. Sample results are presented for one problem with a plethora of local optima and a second with irregular and changing optima. The performance on these problems, chosen for their highly varied characteristics, allow the framework's performance on a wide range of problems to be inferred.