One of the essential indicators of surface water quality is total dissolved solids (TDS) that is prominent for both short- and long-term water resources management. In this study, in order to resolve the non-stationar...
详细信息
One of the essential indicators of surface water quality is total dissolved solids (TDS) that is prominent for both short- and long-term water resources management. In this study, in order to resolve the non-stationarity challenges faced by water quality parameters (WQPs), variational mode decomposition (VMD) combined with two unsupervised artificial intelligence (AI) models, artificial neural network (ANN) and extreme learning machine (ELM), was used. Crow search algorithm (CSA) was recruited to train the ANN and ELM networks and find the optimum weight and bias values. To validate the constructed hybrid VMD-ANN/ELM-CSA models, Molasani station in Karoon River was considered to model TDS at monthly timescale. The WQPs time-series data are simultaneously factorized into respective intrinsic mode functions (IMFs) using VMD algorithm. The proposed VMD-ELM-CSA model shows the best performance at considered station, in comparison with the comparative standalone and hybrid models, with maximum values of R-2 = 0.97 and E-ns = 0.96 at testing period. Results show that using VMD could reduce RMSE about 37 and 12% when coupled with ANN-CSA and ELM-CSA, respectively, at Molasani station. It was concluded that the CSA optimization algorithm and VMD datadecomposition technique could improve the prediction accuracy of the AI models and may be recommended for accurate WQPs prediction.
暂无评论