In order to meet the explosive growth of vehicle networking communication demand, millimeter wave communication has attracted wide attention. However, millimeter wave is susceptible to blocking and interference, and some scholars have proposed using intelligent reflectors to build new channels to improve communication quality. In order to solve the problem that mmwave communication between vehicle and road test unit is easily interrupted in V2X communication, intelligent reflector is deployed in relevant position and beamforming design is carried out to improve communication quality. In this paper, the sparse channel is recovered by compressed sensing algorithm, and a two-level complex neural network with attention mechanism is designed for joint beamforming design. The simulation results show that compared with the traditional beamforming algorithm, the proposed algorithm can improve the sum rate of vehicle communication, avoid communication interruption, and effectively improve signal transmission quality.