确定天然河流纵向离散系数的有限差分-单纯形法

Finite difference method-simplex method for determination of logitudinal dispersion coefficient in natural river

针对在确定纵向离散系数的示踪试验中解析法难以适用于天然非均匀河流的不足,提出了采用有限差分法结合Nelder-Mead单纯形法的参数识别算法(FDM-NMS)来反演天然河流的水质参数。以纵向离散系数反演为例,利用该算法重点探讨了初值选取、观测噪声、离散河段数等因素对参数识别结果的影响,结合恒定流和非恒定流2个算例验证了该方法的可靠性。计算结果表明,采用FDM-NMS算法当噪声水平≤10%、离散河段数≤10时能给出较好的参数识别结果,算法具有良好的抗噪性。

Given the difficulties of the analytical method which can＇t be appliced to the non-uniform river in tracertest for determination of longitudinal dispersion coefficient,a parameters identification method based on finite difference method-nelder mead simplex（FDM-NMS） was proposed.Taking longitudinal dispersion coefficient as example,the factors such as the different level of the observation noise,the discreted stream segment number and the initial value selected which affects the results of the parameters estimation were discussed in detail by the given algorithm.The method was validated by two numerical examples： parameters identification in steady and unsteady flow.The computational results indicate that FDM-NMS algorithm can give good identification precision results and shows good noise proof for the water quality model with the noise level ≤10% and discreted stream segment number ≤10.