Combined water quality forecasting system based on multiobjective optimization and improved data decomposition integration strategy

Water quality forecasting has great practical significance for sustainable utilization of water resources and timely pollution prevention and control. However, owing to irregularity and volatility of water quality data, achieving accurate forecasts remains a challenging problem. Existing single forecasting models based on points forecasts fail to capture the uncertainty in water quality data or obtain high forecasting accuracy. To address those disadvantages, an uncertain combined water quality forecasting system based on time-varying filtering empirical mode decomposition strategy and multiobjective immune selection optimization is proposed. Simulations are carried out with actual water quality data obtained during dry and wet periods. Empirical results show that the improved empirical mode decomposition strategy has better data denoising effect. Based on the multiobjective optimization algorithm, not only the Pareto optimal weights of the combined forecasting modules can be obtained but also the distribution function parameters can be effectively optimized. The proposed system can integrate the respective advantages of deep learning and neural network models to improve the forecasting accuracy, and effectively analyze the uncertainty of water quality, and provide technical support for the sustainable development of regional water environment.