Abstract: A novel integrated optimization methodology of global optimization (genetic algorithm) and local (quasi-Newton algorithm) optimization for robust and rapid parameter estimation in the initial ODEs (ordinary differential equations) systems was proposed, and it is of the advantages of both algorithms. This methodology was used successfully to estimate the parameters for dynamic variation process of dissolved oxygen(DO) in the two-step nitrification model suggested, and a high correlation coefficient was reached 0.9955.Accuracy assessment for the results estimated was realized based on a comparison of the confidence regions with those determined by Fisher information matrix and exact directly search.The assessment results indicated that using this method most of the parameters in the two-step model could be reliably estimated and only two did not, showing that it could be a novel inspection method for parameter estimation of dynamic system.Furthermore, DO simulation results could be employed as a tool of soft measurement, which could provide some process information about rapid degradation of COD, oxygen uptake rate, ammonia, nitrite and nitrate in the two-step nitrification model and related parameters estimated from DO data.

Key words: DO simulation, SBR, nitrification model, parameter estimation, soft measurements, confidence region

摘要： 提出了一种全新的针对初值常微分方程组系统的全局最优化(遗传算法)结合局部最优化(拟牛顿法)实现参数的鲁棒、快速估计的算法。利用该算法,对所建两步硝化模型中过程溶解氧(dissolved oxygen, DO)的动态变化成功实现了参数估计,相关度达到了0.9955。采用基于Fisher信息矩阵和直接搜索获得的参数置信区间相比较的方法实现了对估计结果的可靠性分析,结果表明采用该方法大部分参数可实现可靠估计,只有少数两个参数可实现估计却不可靠,为动力学系统的参数估计结果提出了一个全新的检验方法。DO模拟结果可以作为软测量手段,对过程中易生物降解COD、氨氮、亚硝态氮、硝态氮的全程变化情况提供充足的过程信息。

关键词: DO模拟, SBR, 硝化模型, 参数估计, 软测量, 置信区间