An optimized long short-term memory (LSTM)-based approach applied to early warning and forecasting of ponding in the urban drainage system

In this study, we propose an optimized long short-term memory (LSTM)-based approach which is applied to early warning and forecasting of ponding in the urban drainage system. This approach can quickly identify and locate ponding with relatively high accuracy. Based on the approach, a model is developed, which is constructed by two tandem processes and utilizes a multi-task learning mechanism. The superiority of the developed model was demonstrated by comparing with two widely used neural networks (LSTM and convolutional neural networks). Then, the model was further revised with the available monitoring data in the study area to achieve higher accuracy. We also discussed how the number of selected monitoring points influenced the performance of the corrected model. In this study, over 15 000 designed rainfall events were used for model training, covering various extreme weather conditions.