MultiGran-STGCNFog: Towards Accurate and High-Throughput Inference for Multi-Granular Spatiotemporal Traffic Forecasting (2505.01279v1)

Abstract: Accurate traffic forecasting and swift inference provision are essential for intelligent transportation systems. However, the present Graph Convolutional Network (GCN)-based approaches cannot extract and fuse multi-granular spatiotemporal features across various spatial and temporal scales sufficiently, proven to yield less accurate forecasts. Besides, additional feature extraction branches introduced in prior studies critically increased model complexity and extended inference time, making it challenging to provide fast inference for traffic forecasting. In this paper, we propose MultiGran-STGCNFog, an efficient fog distributed inference system with a novel traffic forecasting model that employs multi-granular spatiotemporal feature fusion on generated dynamic traffic graphs to fully capture interdependent traffic dynamics. The proposed scheduling algorithm GA-DPHDS, optimizing layer execution order and layer-device scheduling scheme simultaneously, contributes to considerable inference throughput improvement by leveraging heterogeneous fog devices in a pipelined manner. Extensive experiments on real-world datasets demonstrate the superiority of the proposed method over selected baselines.

MultiGran-STGCNFog: An Efficient Fog Computing System for Traffic Forecasting

In the field of intelligent transportation systems, efficient traffic forecasting plays a pivotal role in enhancing urban mobility and mitigating congestion. The integration of IoT technology within smart cities has enabled the accumulation of vast volumes of traffic data. However, leveraging this data for real-time insights requires innovative models capable of accurately predicting complex traffic patterns with minimal latency. This paper presents MultiGran-STGCNFog, an advanced fog computing framework designed to deliver precise and swift spatiotemporal traffic forecasts.

MultiGran-STGCNFog introduces a novel Graph Convolutional Network (GCN)-based model that performs dynamic spatiotemporal feature fusion across multiple scales. Unlike conventional GCN approaches that struggle with feature extraction complexity and delayed inference, this model constructs dynamic traffic graphs to capture intricate traffic dependencies effectively. The multi-granular strategy integrates hierarchically clustered spatial features and temporally scaled data to enhance the forecasting accuracy significantly.

The system architecture is underpinned by an efficient scheduling algorithm, GA-DPHDS, which orchestrates layer execution and device allocation in the fog network. As fog computing avoids the latency issues associated with centralized systems, leveraging distributed pipeline parallelism across heterogeneous devices further accelerates the inference process. By optimizing layer-device scheduling, GA-DPHDS enhances throughput and achieves a balanced workload distribution, overcoming traditional hardware heterogeneity challenges.

The experimental evaluation on real-world datasets, PEMS04, PEMS07, and PEMS08, underscores the model's superiority over previous techniques, including HA, LSTM, GRU, and various GCN frameworks. MultiGran-STGCNFog consistently yields lower MAE, MAPE, and RMSE metrics across diverse prediction horizons, indicating a robust capacity to capture long-term traffic dependencies. Moreover, the ablation paper reveals the transformative impact of multi-spatiotemporal feature fusion, elucidating substantial performance gains over single-scale approaches.

From an implementation perspective, the system's reliance on fog computing for distributed parallel inference is noteworthy. This design choice not only accelerates computational processes but also aligns with the increasing scalability demands in urban transportation networks. The proposed methodology could serve as a blueprint for future models embracing interactive and responsive AI-driven traffic management systems.

In terms of future developments, further exploration could examine the adaptability of the multi-granular spatiotemporal fusion framework to transformer-based models, potentially advancing the boundaries of prediction accuracy in dynamic environments. Additionally, optimizing runtime strategies to accommodate fluctuating network loads and device availability could bolster the resilience and efficiency of fog computing schemes.

Ultimately, MultiGran-STGCNFog represents a significant step toward realizing high-throughput traffic forecasting systems, addressing real-world constraints while paving the way for smarter urban mobility solutions. The implications of this research extend beyond theoretical advancements, offering practical insights into the deployment and scalability of next-generation AI models in intelligent transportation systems.