DeepSTF: A Deep Spatial–Temporal Forecast Model of Taxi Flow.

This article focuses on the development and evaluation of DeepSTF, a deep spatial–temporal flow forecast model designed for predicting taxi flow in urban adjacent areas by dividing cities into grid-based graph structures. DeepSTF integrates graph convolutional networks (GCN) to capture spatial correlations and temporal convolutional networks employing dilation and causal convolutions to extract time-dependent features, addressing challenges of rapidly changing traffic data and under-fitting seen in other models. The model also incorporates weather as an implicit factor influencing taxi flow, enhancing prediction accuracy. Experimental results using datasets from Haikou and Chengdu cities, as well as PeMS data, demonstrate that DeepSTF outperforms several existing models in long-term traffic prediction accuracy, computational efficiency, and stability across various evaluation metrics such as MAE and RMSE.