The development of smart connected vehicle technology and in-vehicle applications generates many heterogeneous latency-sensitive and compute-intensive tasks, and the way of offloading processing tasks brings opportunities and challenges to reduce the computational burden of vehicles.Due to the differences in the demand for communication and computational resources for different types of tasks, smart vehicles made by different manufacturers are equipped with different amounts of computational resources, and how to make full use of the resources of heterogeneous vehicles in order to realize the computational offloading and resource allocation for heterogeneous tasks is a major challenge.In light of this, a model for a vehicle edge computing system that combines the use of cloud, edge, and vehicle nodes is proposed. This model is built as a semi-Markov decision process model, and an intelligent algorithm with reinforcement learning determines the best allocation strategy to achieve a reasonable resource allocation. The simulation results demonstrate that the suggested scheme's long-term average benefit is higher than that of the adaptive threshold algorithm and greedy algorithm by 43.9% and 86.52%, respectively. This effectively reduces processing task delays and energy consumption while improving user service quality.