An improved GPU-accelerated heuristic technique applied to the capacitated vehicle routing problem

The capacitated vehicle routing problem (CVRP) is a well-known NP-hard combinatorial problem. Genetic algorithms (GAs) are often used for solving CVRPs. However, the computational effort required to find a feasible solution becomes problematic for very large instances. Parallel-computation technology can significantly improve the performance of CVRP solution algorithms to deal with large problem sets, especially when using GAs. In this paper, an improved genetic algorithm is designed to be entirely executed on an NVIDIA GPU, taking advantage of the special CUDA GPU architecture to solving the CVRP. By distributing array elements over the GPU grid and using GPU kernel functions, the proposed algorithm successfully provides high-quality solutions within reasonable computational times, and near-optimal solutions for smaller benchmark problems. Within this framework, we address the execution speed problem in CVRPs by developing an algorithm that is entirely running on an NVIDIA GPU, investigate how to incorporate local search algorithms with the GA, and develop comparisons between the algorithm performance on both the CPU and the GPU. The utility of improved local search algorithms in the overall performance of the GA is also reported