Parameter optimization method for the water quality dynamic model based on data-driven theory

• A new parameter optimization approach is proposed for the water quality model.
• We apply the learning machines LS-SVM to study eutrophication processes in waters.
• The parameter optimization method is successful applied to develop the water quality model in Xiangshan Bay in China.

Parameter optimization is important for developing a water quality dynamic model. In this study, we applied data-driven method to select and optimize parameters for a complex three-dimensional water quality model. First, a data-driven model was developed to train the response relationship between phytoplankton and environmental factors based on the measured data. Second, an eight-variable water quality dynamic model was established and coupled to a physical model. Parameter sensitivity analysis was investigated by changing parameter values individually in an assigned range. The above results served as guidelines for the control parameter selection and the simulated result verification. Finally, using the data-driven model to approximate the computational water quality model, we employed the Particle Swarm Optimization (PSO) algorithm to optimize the control parameters. The optimization routines and results were analyzed and discussed based on the establishment of the water quality model in Xiangshan Bay (XSB).