A data-driven model for water quality prediction in Tai Lake, China, using secondary modal decomposition with multidimensional external features

Abstract

Study regionTai Lake, the third largest freshwater lake in China, with a history of serious ecological pollution incidents. Study focusLake water quality prediction techniques are essential to ensure an early emergency response capability for sustainable water management. Herein, an effective data-driven ensemble model was developed for predicting lake dissolved oxygen (DO) based on meteorological factors, water quality indicators and spatial information. First, variation mode decomposition (VMD) was used to decompose data into multiple modal components and classify them into feature terms and self terms. The feature terms were combined with relevant external features for multivariate prediction by convolutional neural network (CNN) and a bi-directional long and short-term memory (BiLSTM) with attention mechanism (AT), as well as using the whale optimization algorithm (WOA) to optimize the model hyperparameters. The self terms form a secondary modal decomposition model. Finally, the groupings were linearly summed to obtain outcome. New hydrological insights for the region:The proposed model has the highest prediction accuracy in Tai Lake as well as the best prediction effect using 0.5 days as the period. This research also establishes a stepwise water temperature regulation mechanism, where the output of the target DO content value is achieved by changing the magnitude of water temperature and combining it with this prediction model, thereby strengthening the protection of water resources and the management of fishery production.