An Integrated Multisensor and LSTM-Based Deep Learning Storm Sewer System for Real-Time Flood Risk Prediction and Mitigation
摘要
Storms occurring in urban areas lead to runoff that overwhelms storm sewer systems. This paper discusses a system for flood risk prediction system in real-time using environmental sensors and deep learning. Synthetic data of temperature, humidity, soil moisture, rainfall, water level, and flow rate were used to train several different models. The LSTM model achieved the highest accuracy. Once the model was trained, it was deployed on an Arduino and ESP32. The environmental sensors send the data to a Flask server which is hosting the LSTM prediction model. The prediction model classifies the flood risk into 3 categories: LOW–MODERATE–HIGH. Once the model makes a prediction, the system alerts and triggers a response for gate control. The LSTM pre-dictions are displayed locally and logged. The system is a low-cost and scalable solution for smart flood management systems.