Advanced data augmentation techniques coupled with enhanced particle swarm optimization for predicting total phosphorus concentrations in limited transmission spectra samples: A case study on the Yangtze River

Abstract

Accurate prediction of total phosphorus concentration in water bodies is crucial for effective water pollution control and management. However, obtaining real-world water samples presents challenges such as being costly, time-consuming, or difficult. To boost prediction accuracy, this study introduces novel methodologies integrating advanced data augmentation techniques with enhanced optimization algorithms. In the data augmentation techniques, we employ a novel Oversampling Nearest Neighbor Generative Adversarial Network (ONNGAN) to augment the collected sample data, thereby mitigating issues arising from poor model training due to insufficient sample numbers. Regarding the enhanced particle swarm optimization, we introduce the Ant Colony Cooperative Particle Swarm Optimization (ACCPSO) algorithm, which enhances traditional particle swarm optimization by incorporating several strategic improvements. These enhancements reduce the likelihood of the algorithm getting stuck in local optima and accelerate its convergence speed. The ACCPSO algorithm is employed to automatically optimize the hyperparameter combinations for a Convolutional Neural Network (CNN), and the optimized CNN is then used to predict total phosphorus concentrations in water bodies. By combining these two advanced algorithms, we can more effectively explore the complex nonlinear relationships in transmission spectrum data, thereby enhancing prediction accuracy. The experimental results indicate that, following data augmentation, the ACCPSO-CNN model achieves an R2 of 0.9773, an RMSE of 0.0018, and an MAE of 0.0005, outperforming other benchmark models such as Support Vector Regression and Partial Least Squares Regression across all evaluation metrics. In summary, our research provides a powerful and practical tool for water quality monitoring and pollution control, offering broad application prospects.