Prediction of calcium concentration in circulating seawater in a closed-cycle seawater cooling system using machine learning models

The objective of this investigation is to establish and evaluate the efficacy of two machine learning models, namely random forests (RFs) and support vector machines, in forecasting the calcium concentration within the circulating seawater of a closed-cycle seawater cooling system, thereby replacing conventional and time-consuming laboratory testing. These models were constructed based on daily seawater quality data, and their predictive capabilities were evaluated utilizing metrics such as the coefficient of determination (R-2) and root mean square error. Additionally, a sensitivity analysis employing the Sobol sensitivity analysis technique was performed. The findings indicated that both models effectively forecasted the calcium concentration in the circulating seawater within 1-d intervals. The RF model displayed superior prediction accuracy during the training phase, and it yielded comparable results during the validation phase. Moreover, the sensitivity analysis revealed that the RF model outperformed other models in capturing the causal relationship between calcium concentration and the input variables associated with the closed-cycle seawater cooling system.