Urban flow prediction with spatial–temporal neural ODEs

With the recent advances in deep learning, data-driven methods have shown compelling performance various application domains enabling Smart Cities paradigm. Leveraging spatial–temporal data from multiple sources for (citywide) traffic forecasting is a key to strengthen smart city management areas such as urban control, abnormal event detection, etc. Existing approaches of flow prediction mainly rely on development neural networks –e.g., Convolutional Neural Networks ResNet are used modeling spatial dependencies among different regions, whereas recurrent increasingly implemented temporal dynamics modeling. Despite their advantages, existing suffer limitations intensive computations, lack capabilities properly deal with missing values, and simplistic integration heterogeneous data. In this paper, we propose novel framework by generalizing hidden states model continuous-time latent using ordinary differential equations (ODE). Specifically, introduce discretize-then-optimize approach improve balance accuracy computational efficiency. It not only guarantees error but also provides high flexibility decision-makers. Furthermore, investigate factors, both intrinsic extrinsic, that affect volume use separate extract disentangle influencing which avoids brute-force fusion previous works. Extensive experiments conducted real-world large-scale datasets demonstrate our method outperforms state-of-the-art baselines, while requiring significantly less memory cost fewer parameters.