Water Heavy Metal Prediction Using a CBM Model Based on Hybrid Metaheuristic Algorithms

Precise monitoring and forecasting of contaminants in river water quality are essential for the scientific management of aquatic ecosystems. Nonetheless, deep learning (DL) models encounter difficulties adapting to dynamic contexts when managing intricate hyperparameters and high-dimensional time-series data. This leads to inadequate generalization abilities for cross-regional multisample predictions. We propose a collaborative optimization mechanism that integrates the whale optimization algorithm (WOA) and gray wolf optimization (GWO) to facilitate the convolutional neural network (CNN)-bidirectional long short-term memory (BiLSTM) model in attaining optimal parameter selection for five heavy metal pollutants—Pb, Hg, As, Cr(VI), and Cd—in the Ningxia region of the Yellow River in China. Furthermore, we integrate a multihead attention mechanism to diminish computational complexity, thereby developing the WOA-GWO-CNN-BiLSTM-multihead attention (WGCBM) model for accurate multiclass heavy metal cross-sample predictions. Ultimately, we assess the model’s generalization capability in multisample cross-regional predictions employing a multiscale feature aggregation attribute classification approach. The experimental results indicate that: 1) in the Ningxia segment of the Yellow River, when the proportion of outliers ranges from 5% to 15%, the proposed method attains accuracy, recall, and $F1$ score between 0.75 and 0.90, 0.74 and 0.89, and 0.75 and 0.88, respectively, showcasing exceptional predictive performance and generalization capability; 2) the WGCBM model enhances accuracy, recall, and $F1$ score by 12.00%–25.48%, 10.34%–23.45% and 11.89%–21.97%, respectively, compared to the CBM and BiLSTM models; and 3) the WOA-GWO collaborative optimization algorithm adeptly harmonizes global search with local exploitation, markedly decreasing the time needed for conventional parameter tuning methods, circumventing redundant searches, and leveraging distributed computing for parallel processing, thus enhancing tuning speed and computational resource efficiency. This study confirms the efficacy of integrating the WOA-GWO collaborative optimization mechanism with the CBM model. Integrating the WGCBM model into the Internet of Things (IoT) water quality monitoring network enables rapid generation of water quality predictions by aggregating data from sensors at multiple monitoring stations. This facilitates early warning notifications and decision-making assistance for regulatory bodies, thereby serving a crucial function in automated water monitoring systems and markedly improving the timeliness and efficacy of water quality management.