<p>Accurately predicting the migration and diffusion behavior of elements in water bodies is crucial for the rapid identification and effective response to water pollution incidents. Current research has primarily focused on time series forecasting of single-element concentrations, yet there remains a shortage of predictive methods that comprehensively consider the interactions between multiple elements and their spatiotemporal migration. This study presents a predictive method based on machine learning, which combines the global search capability of a Whale Optimization Algorithm (WOA) with the nonlinear data processing characteristics of the Least Squares Support Vector Machine (LSSVM) to enhance the predictive accuracy of the migration and diffusion behavior of multiple elements in water bodies. The study employs surface water quality monitoring data from the Yalu River for case analysis. The results demonstrate that the WOA-LSSVM method outperforms the LSSVM model in predictive performance, with a 1.19% increase in the R², a 5.89% decrease in the MAE, a 34.5% reduction in the RMSE, a 27.35% increase in the VAF, a 97.12% decrease in the RSD, and a 92.85% decrease in the PI. Compared to hydrological dynamic models, the average error is also reduced by 4.39%. This study not only builds a learner but also provides a scientific method for water pollution early warning through dynamic visualization technology in practical applications.</p>