Spatial-Temporal Fusion Graph Neural Networks for Traffic Flow Forecasting

Abstract

Spatial-temporal data forecasting of traffic flow is a challenging task because of complicated spatial dependencies and dynamical trends of temporal pattern between different roads. Existing frameworks usually utilize given spatial adjacency graph and sophisticated mechanisms for modeling spatial and temporal correlations. However, limited representations of given spatial graph structure with incomplete adjacent connections may restrict effective spatial-temporal dependencies learning of those models. Furthermore, existing methods were out at elbows when solving complicated spatial-temporal data: they usually utilize separate modules for spatial and temporal correlations, or they only use independent components capturing localized or global heterogeneous dependencies. To overcome those limitations, our paper proposes a novel Spatial-Temporal Fusion Graph Neural Networks (STFGNN) for traffic flow forecasting. First, a data-driven method of generating “temporal graph” is proposed to compensate several genuine correlations that spatial graph may not reflect. STFGNN could effectively learn hidden spatial-temporal dependencies by a novel fusion operation of various spatial and temporal graphs, treated for different time periods in parallel. Meanwhile, by integrating this fusion graph module and a novel gated convolution module into a unified layer parallelly, STFGNN could handle long sequences by learning more spatial-temporal dependencies with layers stacked. Experimental results on several public traffic datasets demonstrate that our method achieves state-of-the-art performance consistently than other baselines.

Towards Learning in Grey Spatiotemporal Systems: A Prophet to Non-consecutive Spatiotemporal Dynamics

“…Analysis on factor decoupling. In fact, the spatiotemporal elements are regularly varied with various exogenous factors for their tidal patterns and weather-induced variations, while it also reveals heterogenous region-wise proximity patterns due to different time, weather and holidays [8], [9], [25], [39]. In this way, the instantaneous variations influenced by exogenous factors are further decoupled into inherent intensity regularity and the factor-induced dynamic aggregations.…”

“…Given the intrinsic challenge of the unavailability of partial observations in grey spatiotemporal systems, the key is to perform status inference under expected conditions by resorting to proxy. Fortunately, it is interesting to show that multiple exogenous factors like weather, locations, timestamps can be exploited to vicariously estimate the expected statuses of spatiotemporal elements [2], [8], [25], [39]. Hence, by resorting to available historical observations and informative exogenous factors, we can turn predictions of non-consecutive future statuses into inferring status under combined expected exogenous factors.…”

“…(2) STFGNN: A framework jointly learns localized heterogeneity and global homogeneity with data-driven graph generation [25]. (3) STG2Seq: A hierarchical convolutional framework to capture spatial and temporal dependencies of passenger demands with awareness of context factors [8].…”

“…Unfortunately, in more practical scenarios, early pre-arrangements of both large-scale urban activities and individual traveling require the model to foresee non-consecutive series, where fragments of observations are unavailable. Promisingly, recent proposals rise the attention of contexts to realize more customized predictions [2], [8], [9], [25], [39]. To this end, the techniques of non-consecutive spatiotemporal forecasting and strategies for decoupling and gathering multi-level factors are highly desired to empower more intelligent urban life.…”

“…To empower an adaptive topology, AGCRN [31] learns a dynamic topology based on feature-wise product, while STAR-GNN [46] and L2P framework [47] respectively design a mutual-information based strategy and a generative process, to estimate the optimal receptive fields for GNNs. More recently, exogenous context factors those are out-of-graph have been demonstrated to potentially interfere the intrinsic graph topology and have complicated interactions on element-wise aggregations [25], [27], [48]. Unfortunately, existing literature has barely discussed this issue.…”