Urban Waterlogging Prediction Based on Time Lag Correlation Analysis and Multi-model Coupling

Abstract

Abstract As extreme weather becomes more frequent and the damage caused by urban waterlogging is increasing, it is important to establish a fast and accurate model of waterlogging disasters. However, the smartization of most cities starts relatively late, and the types and quality of monitoring data are uneven. Therefore, there has been a focus on researching and developing a reasonable, fast, and accurate urban waterlogging prediction model that can effectively utilize limited data. Based on this situation, a method of time lag correlation analysis considering the mechanism of regional physics (PTLC) is proposed in this paper. Combined with spatial decoupling, a prior analysis is provided for model prediction. At the same time, a deep learning model (Poar_LSTM) with automatic optimization function is proposed and coupled with hydrodynamic model (Poar_DHC). Based on the verification of Doumen area in Fuzhou, the typical rainfall process from 2021 to 2022 is reviewed. The results indicate that Poar_LSTM shows obvious advantages in the river level prediction during the same rainfall period. The Nash efficiency coefficients in the verification reach 0.969 and 0.971 respectively. Different data-driven models have little influence on the overall prediction effect of waterlogging coupling model. Poar_DHC has the highest accuracy in the prediction of underground liquid level. According to the different rainfalls, PTLC plays a good guiding role in the prior analysis and later evaluation of the early prediction of the model. This study can provide a scientific reference for the analysis of flood mechanism and the rapid and accurate prediction of rainstorms.