STGAN: Spatial-temporal Graph Autoregression Network for Pavement Distress Deterioration Prediction

TL;DR

This work introduces STGAN, a Spatial-Temporal Graph Autoregression Network designed to predict pavement distress deterioration from irregular, asynchronous data. By embedding time into the spatial domain and framing prediction as a graph autoregression on a growing spatiotemporal graph, STGAN captures complex spatial-temporal dependencies with a novel TOP-augmented connection scheme and a spatiotemporal attention mechanism. Experiments on the ConTrack Shanghai dataset show STGAN outperforms baselines and benefits substantially from TOP connections, the temporal-aware attention, and time-difference features; ablations confirm the necessity of these components. The approach enables proactive road maintenance decisions by delivering accurate short-term distress forecasts, though challenges remain due to data imbalance, environmental noise, and label uncertainties. Future work will expand datasets with finer granularity and traffic factors and seek architectures that scale efficiently with smaller sample sizes.

Abstract

Pavement distress significantly compromises road integrity and poses risks to drivers. Accurate prediction of pavement distress deterioration is essential for effective road management, cost reduction in maintenance, and improvement of traffic safety. However, real-world data on pavement distress is usually collected irregularly, resulting in uneven, asynchronous, and sparse spatial-temporal datasets. This hinders the application of existing spatial-temporal models, such as DCRNN, since they are only applicable to regularly and synchronously collected data. To overcome these challenges, we propose the Spatial-Temporal Graph Autoregression Network (STGAN), a novel graph neural network model designed for accurately predicting irregular pavement distress deterioration using complex spatial-temporal data. Specifically, STGAN integrates the temporal domain into the spatial domain, creating a larger graph where nodes are represented by spatial-temporal tuples and edges are formed based on a similarity-based connection mechanism. Furthermore, based on the constructed spatiotemporal graph, we formulate pavement distress deterioration prediction as a graph autoregression task, i.e., the graph size increases incrementally and the prediction is performed sequentially. This is accomplished by a novel spatial-temporal attention mechanism deployed by STGAN. Utilizing the ConTrack dataset, which contains pavement distress records collected from different locations in Shanghai, we demonstrate the superior performance of STGAN in capturing spatial-temporal correlations and addressing the aforementioned challenges. Experimental results further show that STGAN outperforms baseline models, and ablation studies confirm the effectiveness of its novel modules. Our findings contribute to promoting proactive road maintenance decision-making and ultimately enhancing road safety and resilience.

Figures (16)

• Figure 1: Visualization of the data and timestamps at different locations in Shanghai. We selected $5$ different locations as examples. For each location, the solid line represents the timeline of all timestamps across all locations, while the red marks indicate the timestamps of the data points collected at the specific location. It can be seen that the collection patterns of timestamps vary substantially across locations, with some timestamps being quite sparse (e.g., the most right point has only one record).
• Figure 2: The statistical behavior of the timestamps of all data points, which reveals the sparsity and asynchronicity properties of the pavement distress dataset in the temporal domain.
• Figure 3: Visualization of the spatial-temporal graph built based on the TOP and hard connections. Here different nodes correspond to the different locations (e.g., $U, I, P, H, Q$) or the same location with different timestamps (e.g., $t_I^{(a)}$ and $t_I^{(o)}$). The node in red denotes the newly added one, which describes the graph expansion of the autoregression process.
• Figure 4: Diagram of the spatial-temporal graph autoregression network model.
• Figure 5: Samples in the ConTrack Dataset. The orginal data is represented as a image collected by the camera on a car and environment conditions collected by the sensors.