Real-time prediction and correction of regional flooding by combining artificial intelligence and IoT sensors.

Taiwan is located in the subtropical region and is hit by an average of 3 to 5 typhoons yearly. In recent years, due to climate change, short-term and heavy rainfall has occurred, resulting in frequent flooding. Therefore, the government has invested a lot of resources in using early warning and forecast systems. Deploy relevant disaster prevention measures and personnel to reduce the impact of disasters caused by extreme rainfall. In the past, to analyze the flooding caused by rains, it was necessary to simulate the flooding distribution through a two-dimensional hydraulic model. However, due to factors such as many parameters of the hydraulic model and long calculation time, it isn't easy to carry out a real-time two-dimensional flooding forecast. Only applicable to the pre-disaster preparation phase.

This study proposes to combine AI and the hydraulic model with IoT flooding sensors for regional flooding prediction and correction. First, the two-dimensional hydraulic simulation data were used as the training data for configuring the AI models. Then, the constructed RNARX model is implemented for the T+1 to T+3-hour average flooding depth forecast (time distribution), and the SOM model is used to cluster the flooding distribution (spatial distribution). Finally, the IoT sensor data are used to update the regional flooding depths and predict the flooding depth of each grid in the next 1 to 3 hours. The results show that the real-time monitoring data of 25 actual flooding sensors have significantly improved the forecast value of each grid point. For example, in the T+1~T+3 mode, the RMSE improvement rate can reach 25%, 30%, and 36%. Therefore, the combination of the two-dimensional hydraulic simulation model and the IoT sensor through AI can make the forecast results close to the actual flooding situation to improve model forecast accuracy.