Robust estimation of water quality model parameters under random noise disturbance

The Dobbins BOD-DO water quality model is applied to study the effect of random noise on parameter estimation by means of numerical simulation. The result shows that the least squares estimation of parameters is not robust in case of colored noises and high-level white noise involved. The random noises result in the estimated parameters drifted away from their true values. In order to overcome the disturbance of noises, a robust identification method of water quality model parameters namely trust region algorithm based on M-estimation is proposed. The calculation results indicate that the M-estimation has strong robustness and the true value of parameters can be reliably and robustly acquired either in condition of non-disturbed data or in case of noise involved. The comparison between two kinds of estimation methods shows that the trust region algorithm possesses high accuracy, excellent performance of noise resistance and strong robustness as well as uniform convergence, which are obviously better than the least squares estimation.