Traffic flow prediction represents a crucial research avenue within the domain of spatio-temporal data mining.How-ever,prevailing predictions fall short of expectations due to the highly nonlinear and intricate nature of traffic patterns.Exist-ing methodologies often overlook the significance of capturing spatial-temporal dynamics and integrating external environmental variables.A novel approach for traffic flow prediction termed as the spatial-temporal graph traffic flow prediction method,le-veraging generative adversarial network(GAN).This approach entails unsupervised learning of spatial-temporal graph feature representations to facilitate more precise spatial-temporal sequence graph prediction.Utilizing the GAN framework,the method engages in game-based learning to model data distributions and employs variational inference.Extensive experimentation on Chinese Taiwan's high-speed traffic dataset validates the efficacy of the proposed method,demonstrating its superior capability in addressing the issue of prediction result smoothing compared to baseline approaches,obtaining more efficient and accurate forecasting results.
关键词
生成对抗网络/交通流量/时空图/图结构生成
Key words
GAN/traffic flow/spatial-temporal graph/generation of graph structure
维普
万方数据