Hydrological Time Series Prediction based on IWOA-ALSTM

Abstract The prediction of hydrological time series is of great significance for the development of flood and drought prevention, and is an important research component of smart water resources. The nonlinear characteristics of the hydrological time series are an important factor affecting the accuracy of prediction. In order to better predict the non-linear component of the hydrological time series, (IWOA) used to tune long-short memory based neural network augmented with attention mechanisms (ALSTM). The proposed model is termed IWOA-ALSTM. The model hyperparameters setting has a significant impact on the prediction accuracy and operation efficiency. So, the improved whale optimization algorithm allows the dynamic adjustment of the required hyperparameters. The used network introduces an attention mechanism between the two layers of the LSTM, capable of adaptively focusing on different features according to each time unit in order to obtain information about the hydrological time series. In this work, we use the information about the non-linear water level, obtained from Hankou station, as experimentatl data. The results achieved by the proposed model, wherein IWOA, is exploited are compared to those yielded with using genetic algorithms, particle swarm optimization and whale optimization algorithm. We demonstrate that IWOA-ALSTM provides a better performance.