A long Short-Term memory cyclic model with mutual information for hydrology forecasting: A Case study in the xixian basin

Abstract

Floods result in substantial damage throughout the world every year. An accurate predictions of floods can significantly alleviate the loss of lives and properties. However, due to the complexity during flood formation, the accuracy of traditional flood forecasting models suffer from the performance degradation with the increasing of required prediction period. According to the mutual information(MI) analysis to practical hydrologic data of Xixian Basin from year 2011–2018 in China, this paper proposes a long-term cyclic hydrology prediction model with the help of an improved Long Short-Term Memory. Firstly, by flitering, classifying and MI analysis the original data, the hydrological features, e.g., rainfall, reservoir water-level and flow are extracted as the time series features of the long short-term memory cyclic(LSTMC) forecasting model. Next, the structure of the LSTMC model is trained and determined by modeling the rainfall process to reflect the long-term change of flood flow. Finally, the actual flood data is used to verify the output of our model. Compared with some traditional and machine-learning flood forecasting schemes, it can be demonstrated that our model can accurately complete the task of long and short lead time hydrology forecasting.