Growth prediction of Microcystis aeruginosa based on a secondary decomposition integration model

Abstract

Abstract Microcystis aeruginosa is the dominant species in the blooms of eutrophic lakes such as Taihu Lake in China. Chlorophyll-a is one of the most common indicators to characterize its biomass. The nonlinearity and unsteadiness of the chlorophyll-a sequence decrease the prediction accuracy. In this paper, a secondary decomposition prediction method based on the integration of wavelet decomposition, variational modal decomposition, and gated recurrent unit (WD–VMD–GRU) is proposed. First, the original sequence is decomposed once using wavelet decomposition (WD). Then, the components with higher sample entropy values are decomposed using variational modal decomposition (VMD). Finally, each component is predicted using a gated recurrent unit (GRU), and the final prediction results are obtained by reconstructing each component result. The decomposition effect is ranked as VMD > WD > CEEMDAN > EMD. The WD–VMD–GRU model has a significant advantage compared to the basic model, with an increase of over 6.5% in R2. The secondary decomposition reduces the difficulty of predicting GRU components and has better prediction performance. The RMSE, MAE, and R2 were 1.752, 1.450, 0.969 at 2-day prediction, and 3.169, 2.711, 0.908 at 6-day prediction. Therefore, the WD–VMD–GRU model is superior in accuracy to other methods and can provide a scientific basis for the growth prediction research of M. aeruginosa.