Research on stage-divided water level prediction technology of rivers-connected lake based on machine learning: a case study of Hongze Lake, China

Abstract

The rivers-connected lake involved in the “River–Lake-Reservoir” hydrological complex system and it's water level fluctuations are more severe than those of other lakes, which challenges the scientific management of lakes. Therefore, to improve the accuracy of water level prediction for the rivers-connected lake, taking Hongze Lake as an example, we used the BFAST algorithm to analyze the inconsistency of the lake's inter-annual water level and selected a stable stage for water level prediction research. Next, considering the lake basin shape, based on the Stage-discharge relationship curve, the fluctuation process of the lake's inter-annual water level was divided into four periods: the discharge period, the early period of storage, the later period of storage, and the balance period. Then, the NARX model was used to build the water level prediction model for different periods. Finally, the wavelet analysis and KNN algorithm were introduced into the water level prediction model for input data pre-process and result post-processing, respectively. The result shows that: (1) There are significant differences in the mechanism of water level regime modification in different periods. The outflowing runoff is the main driving factor for the water level regime modification in most times; (2) Coupling multiple machine learning methods is an effective way to improve the accuracy of the lake water level prediction; (3) The combination of the staged-divided water level prediction method and the hybrid machine learning models can further improve the accuracy of the water level prediction.