Multi-swarm sea horse optimization-CEEMDAN-GMDH model: Toward accurate and stable prediction of total dissolved solid concentrations

Accurate prediction of total dissolved solids (TDS) concentration is essential for water resource management. Thus, our study develops the multi-swarm sea horse optimization (MSSHO) algorithm- complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)- group method of data handling (GMDH) neural network to accurately predict TDS concentration in the Kasilain river of Iran. The model is constructed through a systematic and multi-stage process. First, the MSSHO algorithm optimizes the key parameters of the hybrid model, including the number of intrinsic mode functions (IMFs) and the regression coefficients of the GMDH neural network. The MSSHO algorithm employs a multi-swarm strategy and an information-sharing mechanism, which helps avoid local optima and enhances optimization performance. The multi-swarm strategy divides the total population of search agents into multiple teams, which simultaneously explore the parameter space and share their experience using a sharing mechanism. Next, the CEEMDAN method is applied to decompose the time series data of relevant water quality variables into a set of more stationary components called intrinsic mode functions (IMFs). Finally, the GMDH model produces the final predictions using IMFs. Our findings indicate that the MSSHO-CEEMDAN-GMDH model improves the Kling–Gupta efficiency (KGE) and Explained Variance Score (EVS) of all other models by 2.12–30% and 2.06–20%, respectively.