A generative adversarial network model integrating self-attention mechanism is proposed to address the issues of low efficiency and poor performance in traffic flow missing data repair.In order to fully utilize the timestamp and periodic informa-tion contained in the traffic data flow,self-attention mechanism and location coding are adopted to improve the model learning performance.To improve the model training performance,spectral normalization and time scale updating rules are proposed to accelerate learning efficiency.The experimental results show that,compared with KNN,HA and GAN models,the proposed model has the best comprehensive index performance,and has a good repair effect for traffic flow data in high miss rate scenari-os.
维普
万方数据