DNN Partition and Offloading Strategy with Improved Particle Swarm Genetic Algorithm in VEC

Vehicular edge computing (VEC) is a novel computing paradigm, which is designed to satisfy the growing computation and communication needs of vehicle systems. With the assistance of VEC, vehicles can execute artificial intelligence (AI) tasks based on deep neural network (DNN), which are compute-intensive and delay-sensitive. However, it is difficult to deploy large-scale and compute-intensive DNN on resource-constrained terminal devices. Therefore, DNN model partition and offloading strategy have received a lot of attention, however, most of the researches have not taken into account the problem that the optimal partition point of a DNN model changes with the allocated computing resources. To address this problem, we propose a computing offloading strategy based on DNN model partition. This strategy selects the optimal DNN model partition points based on the computing capability of the vehicle, and then develops the optimal task offloading strategy to realize the effective distribution and execution of tasks between the edge server and the service vehicle. To minimize the task offloading delay, we propose an improved particle swarm genetic algorithm (IPSGA) to achieve the optimal offloading strategy. The algorithm uses the variable acceleration coefficient with the number of iterations and the inertia weight with the success rate as the feedback parameters to improve the particle swarm optimization algorithm (PSO), and the genetic algorithm (GA) is improved with the adaptive crossover probability and the adaptive mutation probability. Experimental results show that compared to the baselines, the IPSGA can reduce the overall system delay and increase the task completion rate.