Intelligent short-term prediction method for campus roads traffic flow based on two-growth convolution mechanism

This paper proposes an intelligent short-term traffic flow prediction method for campus roads based on two-growth convolutional mechanism. The proposed methodology integrates multi-sensor data collection from various traffic participants including pedestrians, non-motorized vehicles, and motor vehicles. By employing spatiotemporal alignment and advanced feature extraction techniques, the system constructs comprehensive spatial-temporal correlation maps. Furthermore, we introduce an innovative gating attention mechanism to enable dynamic fusion of spatiotemporal features, thereby enhancing prediction accuracy. Experimental evaluations demonstrate the superior performance of our approach, achieving a remarkably low spatiotemporal consistency error of 0.036 during peak hours. The method maintains excellent prediction stability with multi-step consistency reaching 0.89 during off-peak periods. Notably, the system attains a peak smoothness index of 0.93, conclusively validating that our design successfully meets all specified performance objectives.