Research on parameter inversion methods for water quantity and quality models based on FCLPSO

Abstract

[Objective] Complex river network water quantity and quality models involve numerous parameters and high dimensionality, making parameter inversion challenging. An in-depth analysis is required to investigate how the selection of optimization objective functions and different single-parameter and multi-parameter inversion method affect the accuracy of parameter inversion. [Methods] A parameter inversion method for water quantity and quality models was proposed based on the Fast Comprehensive Learning Particle Swarm Optimization(FCLPSO). Numerical experiments for parameter inversion were designed, and the LH-OAT global sensitivity analysis method was used to optimize the objective function for seven model performance evaluation indicators. Furthermore, the inversion result using single-parameter and multi-parameter inversion method were analyzed, and the differences between different inversion method were examined. [Results] The result showed that NSE^* had the highest sensitivity as the objective function. Parameters of different types achieved high accuracy, with the single-parameter inversion having a mean relative error(MRE) of 5.2% and a coefficient of variation(CV) of 7.2%. The multi-parameter inversion result had an MRE of 13.5% and a CV of 14%. In the multi-parameter inversion, the inversion result of hydrodynamic parameters were better than those of water quality parameters, and the multi-parameter “layered inversion”method outperformed the “simultaneous inversion” method. [Conclusion] The result indicate that the proposed model parameter inversion method achieves high accuracy. It can help improve the timeliness and accuracy of parameter estimation for complex river network water quantity and quality models, providing technical support for improving the accuracy of numerical simulation of complex river networks.

Keywords

water quantity and quality model / parameter inversion / fast comprehensive learning particle swarm optimization / objective function / sensitivity analysis