Abstract:

Water quality prediction is an important component in water pollution prevention and control, To improve the accuracy of surface water quality prediction, this paper proposes a new novel hybrid model for water quality prediction. Firstly, the original water quality sequence is decomposed by Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), and then the components are divided into three kinds of high, medium and low complexity components using the fuzzy dispersion entropy (FuzzDE). Secondly, the high, medium, and low complexity components are predicted using the Improved Mantis Search Algorithm (IMSA) optimized the Long Short-Term Memory (BiLSTM),Least Square Support Vector Regression (LSSVR), and Extreme Learning Machines (ELM) respectively, and the prediction results are combined and reconstructed, finally, a BiLSTM error correction model is established to correct the errors. In this paper, the Dissolved Oxygen concentration in two sections of Youshui, a tributary of Yuanjiang River, and the PH value in one section of Xiangjiang River Basin were utilized for simulation verification, and the R² can reach more than 90%, the results show that the accuracy of the hybrid model prediction is better than other comparative prediction models.

Key words: water quality prediction, CEEMDAN decomposition, fuzzy dispersion entropy, mantis search algorithm, hybrid model