Metaheuristics for Optimisation of Organic Computing Systems (MOOCS)

点击翻译按钮获取中文摘要

Organic computing (OC) provides techniques for the design of complex autonomous and self-adaptive systems. The approaches are often inspired by natural phenomena mimicking nature’seffectiveness in sustaining systems of high sophistication, though their functionality may even be distributed between several less sophisticated subsystems. To this end, OC utilises machine learning approaches and a central role is assigned to metaheuristics, which are necessary for the internal optimisation, self-adaptation and overall functionality of the system.Metaheuristics are essential when trying to solve complex optimisation problems. They are more versatile in their application compared to simple heuristics, they require less or even no knowledge of the search space and they provide very good results in little computational time. These advantages, combined with the No-free-lunch theorem, which states that no metaheuristic can be equally effective on all optimisation problems, resulted in a vast amount of newly developed metaheuristics.The application of metaheuristics in OC systems still poses difficulties, especially when trying to find the most suitable metaheuristic or metaheuristics. They should be applicable to all optimisation problems in the OC system and, in addition, self-adaptively respond to changes in the system’s environment. Furthermore, they should also be applicable in distributed systems with little computational resources.In this project, we want to improve the application of metaheuristics in OC systems. We will examine which metaheuristics are suitable for specific problems and, thereby, advance the state of the art through analysing the components of the metaheuristic responsible for these relations. We will furthermore develop self-adaptive strategies for applying and exchanging these components, which will lead to a more general optimiser, suitable for many problems common to OC. Additionally, we will facilitate the direct application on distributed systems by analysing adequate strategies for the distribution and parallelisation of the algorithms.