AD-STGNN: adaptive diffusion spatiotemporal GNN for dynamic urban fire vehicle dispatch and emergency response

Due to the increasingly complicated urban building structure and the rising occurrence of emergent events, the on-dutyto-road processing for an urban fire movement has come up against severe challenges for time response and road intelligence. To achieve this, the paper puts forward a new adaptive diffusion spatiotemporal graph neural network framework based on the NFIRS-PDR (National Fire Event Reporting System-Vehicle Scheduling Record) data, to achieve dynamic optimization of vehicle allocation plans of urban fire incidents. Based on the traditional ST-GCN and DCRNN, the model makes the following key contributions: (1) The dual-channel diffusion module is proposed to simultaneously model the directed traffic flow and regional risk in urban roads. (2) The timing attention gating mechanism is used to dynamically capture the law of the time dependence of sudden fire and the law of high-risk period; (3) The scheduling perception reinforcement feedback mechanism is introduced, a dynamic resource constraint learning path based on the Qlearning constraint is learnt in the training process to enhance the model of the accessibility and coverage of firefighting vehicles. The global graph structure is generated dynamically as a function of road connectivity and fire station arrangement and each node contains local history information of events as well as the regional risk obtained in advance from the NFIRS statistical model. The proposed AD-STGNN model has prominent advantages compared with the existing methods including the ASTGCN, ST-MetaNet and GraphSARL, its reduction rate on average scheduling time is up to 14% ), which can greatly improve the efficiency of emergency response.