Research on water quality prediction based on CEEMDAN-OVMD-DNN-GRU model

Establishing a high-precision water quality prediction model to predict future water quality conditions has significant social and economic value. Therefore, a combined neural network model based on CEEMDAN-OVMD secondary decomposition and DNN-GRU is proposed. The main pollution indicators of the Pearl River Jiaomen in recent years are used as the prediction object to construct a water quality index prediction model. The model first uses fully adaptive noise ensemble empirical mode decomposition (CEEMDAN) to process the original nonlinear data containing a large amount of noise interference. Then K-means is used to reduce the dimensionality into three sequences: high frequency, low frequency and residual according to the sample entropy value of each decomposition sequence, aiming to improve the computing efficiency. Then, variational mode decomposition (VMD) based on Bayesian optimization is used to further extract effective information of high-frequency sequences. Finally, the high-frequency part and the remaining parts are predicted using the DNN network and the GRU network respectively, and the Sparrow Search Algorithm (SSA) is used to optimize the parameters of the GRU network. The predicted experimental results are stable, with R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.991, IA of 0.950, RMSE of 0.028, MAE of 0.025, and MAPE of 0.346%.