IoT-Enabled Dynamic HydroGraph Transformer Flood Prediction Network for Early Urban Flood Warning

Abstract

Urban flood prediction needs faster and clearer warning because water levels rise within short periods. The aim of this research is to predict flood risk through real-time IoT sensing. The study addresses sudden flood formation in rivers, canals, drains, roads and low-lying regions. Existing flood models often depend on rainfall and river gauge values alone. Such models fail when drainage flow, blockage and road water depth change quickly. Sensor noise, missing values and delayed readings also reduce warning quality during heavy rainfall. This research proposes DHT-FloodNet for early urban flood prediction using IoT data. DHT-FloodNet stands for Dynamic HydroGraph Transformer Flood Prediction Network. The system uses water level, flow speed, rainfall intensity, soil moisture, humidity, pressure and drainage depth. A sensor recovery module first repairs missing and noisy IoT readings through temporal consistency. This step reduces false warning caused by faulty sensors and unstable communication. The dynamic hydrograph construction module then connects sensor points using water movement and elevation. These links describe possible flood spread between rivers, canals, drains and road zones. The transformer module studies sudden changes in each sensor stream over short time windows. It captures fast changes in water depth, flow, pressure and rainfall intensity. The hydrograph layer studies how floodwater travels between linked sensing points. This design improves flood prediction because it reads both time change and water movement. The final prediction layer estimates flood risk for one hour, three hours, six hours and twenty-four hours. The model gives early warning levels for safe drainage control and public alert planning. Experimental results show better flood risk prediction than CNN, LSTM, GRU and standard transformer models. DHT-FloodNet reduces missed flood cases and gives stronger warning under noisy sensor conditions. The method supports smart city flood monitoring and rapid disaster response using live IoT data.