Predicting Curb Side Parking Availability for Commercial Vehicle Loading Zones

Commercial fleet management and operations pose distinct challenges compared to regular passenger vehicles. These challenges stem from the varying sizes, shapes, and parking demands of commercial vehicles, requiring specific curbside accommodations. Despite extensive research on smart-parking management for personal vehicles, there has been limited focus on improving parking outcomes for urban freight systems. To address this gap, we have developed a framework that utilizes sensors installed in parking areas to collect occupancy information. This framework predicts parking space availability for commercial vehicles in 10-minute intervals. The current states and the predictions are relayed to the drivers in near real-time through a web-based interface, empowering them to find suitable parking spaces and reducing search time. Our framework incorporates a suite of machine-learning models for predicting curbside parking availability based on real-time sensor data from commercial vehicle loading zones. We evaluated these models in a busy commercial district in the Seattle area, focusing on prediction accuracy and real-world performance. Our study concludes that, for practical use, the convolutional neural network (CNN) model outperforms other architectures, including Spatial Temporal Graph Convolutional Networks (ST-GCN) and Transformer.