Guided mutation strategies for multiobjective automotive network architecture

The increasing complexity of electronic functions in cars leads to new challenges in the development of automotive communication networks. A key issue is the mapping of functional software onto hardware nodes, which has a great impact on overall system performance and costs. In this paper we propose two fitness metrics focusing on this mapping process. We further derive guided mutation operators for an application-specific network optimization framework using multiobjective evolutionary algorithms. Our main contribution represents a novel approach to guided evolutionary mutation by interchanging modular operators during execution of the optimization algorithm. We show that our approach outperforms classic random mutation both in terms of convergence behavior and diversity.