Abstract
Accurate prediction of water quality is critical for sustainable water resource management, particularly in complex hydrological environments such as the Yellow River Basin. However, existing predictive models often face limitations in capturing complex spatio-temporal features and efficiently optimizing hyperparameters. To address these gaps, this study proposes a hybrid deep learning model integrating Grid Search (GS), an Enhanced Harris Hawks Optimization (EHHO) algorithm, a Convolutional Neural Network (CNN), and Bidirectional Long Short-Term Memory (BiLSTM)—named GS-EHHO-CNN-BiLSTM. Specifically, the model utilizes CNN to effectively extract spatial correlations and BiLSTM to accurately capture temporal dependencies. Additionally, the combined GS-EHHO approach ensures optimal hyperparameter selection, significantly enhancing model performance. Empirical results obtained from extensive testing on water quality datasets collected across multiple monitoring stations in the Yellow River Basin demonstrate that the GS-EHHO-CNN-BiLSTM model outperforms traditional and recently proposed deep learning models, delivering superior predictive accuracy and robustness. The study highlights important practical implications: policymakers and water management institutions can adopt this hybrid model as a reliable tool for proactive water quality monitoring and decision-making, thereby supporting effective management and protection of water resources.

This publication has 24 references indexed in Scilit: