Optimization of intra-vehicle architecture using amulti-objective genetic algorithm

Abstract

Using problem-specific genetic operators, the multi-objective genetic algorithm Non-dominated sorting genetic algorithm II (NSGA-II) is adapted to the allocation of software components (SWCs) to electric control units (ECUs) within an automotive architecture. A simulation environment is developed in order to assess the performance of the allocation within an architecture, and thereby provide the genetic algorithm with objective and constraint values. The validity of the optimization method is evaluated by generating artificial software and hardware architectures, and allowing the genetic algorithm to optimize the software allocation. A novel algorithm for routing signals within the software architecture, based on forming and connecting vehicle features, is presented. The optimized Pareto-fronts of small-scale (17 SWCs and 4 ECUs) automotive architectures are compared to the ground truth through exhaustive search. The average hypervolume ratio is 98.9%, computed over 10 architectures and 100 optimizations, and 48% of the performed optimizations successfully found the entire true Pareto-front. For the large-scale (250 SWCs and 25 ECUs) architectures, no ground truth can obtained, and the optimizations are instead evaluated with regard to consistency. In general, the optimization method quickly finds feasible solutions. However, discrepancies between the approximated Pareto-fronts suggest that premature convergence sometimes occurs. Even though the results indicate that the optimization method works as intended and yields satisfactory results with respect to formulated aims, it is not evident that this method is applicable to the optimization of real automotive architectures. The true nature of these architectures may be too complicated to be able to be compressed into a feasibly low number of objectives, which the developed optimization method requires to perform well.