Abstract
Accurate rainfall runoff prediction is crucial for flood forecasting and water resources management, and it remains a challenging issue in hydrological information processing. The most challenging problem is that the hydrological information has a strong locality and nonlinearity, which leads to poor prediction accuracy. Neighbourhood rough sets has strong capability on data classification and reduction, which can reduce those redundant rainfall runoff data. Long short-term memory (LSTM) network is a special recurrent neural network (RNN) that is an excellent variant of RNN, it is good at handling the time series data. In the paper, a hybrid model of neighbourhood rough sets with LSTM is proposed, which is used for the rainfall runoff prediction. The experimental results show that the presented model could improve the training speed of LSTM and achieve much higher prediction accuracy than the conventional rainfall runoff prediction methods.