The rapid development of 5G has made mobile edge computing technology the first choice for smart highways, but the exponentially growing number of vehicles has also brought great challenges, that is, how to provide vehicle users with low-latency and reliable services has become a key issue. To this end, a multi-level intelligent collaborative computing mechanism based on Double Deep Q-learning(DDQN) for smart highways was proposed. First, a multi-level intelligent collaborative computing model based on RSUs, mobile edge computing (MEC) servers and clouds was established. Further, aiming at the goals of task delay minimization and network load balancing, a multi-level intelligent collaborative computing algorithm based on deep reinforcement learning was proposed to perform collaborative computing on massive tasks generated by high-speed moving vehicles. Simulation results show that the proposed mechanism can reduce service delay and maintain network load balancing while completing service requirements.