初始模型对含噪动态光散射数据正则化反演结果的影响

发布时间：2019-04-19 18:03

【摘要】：分别采用最小模型矩阵、最平坦模型矩阵、最光滑模型矩阵作为初始化模型,对加入5种不同水平随机噪声的90nm窄单峰、90nm宽单峰和250nm窄单峰、250nm宽单峰颗粒体系的模拟分布进行了正则化反演,并对反演结果进行比较。结果表明:当噪声水平为0时,正则化初始模型的选择对反演结果没有明显影响。随着噪声水平的增加,采用三种初始化模型反演得到的峰值误差和粒度分布误差都随之变大,但采用最平坦模型和最光滑模型反演得到的峰值和粒度分布误差明显小于采用最小初始模型的反演误差。当噪声水平大于0.01时,选择最平坦初始模型获得的粒度分布结果优于采用最光滑初始模型和最小初始模型获得的结果,而采用最光滑初始模型反演得到的峰值优于最平坦初始模型和最小初始模型的反演峰值。因此,采用正则化算法处理含噪动态光散射数据时,为得到最优的粒度分布信息,宜采用最平坦初始模型,若需要获取最准确的峰值信息,则应选择最光滑初始模型。
[Abstract]:The minimum model matrix, the flattest model matrix and the smoothest model matrix are used as initialization models respectively. For the 90nm narrow single peak, the 90nm wide single peak and the 250nm narrow single peak with five kinds of random noise at different levels, The regularized inversion of the simulated distribution of the 250nm wide single peak particle system is carried out and the inversion results are compared. The results show that when the noise level is 0, the selection of the regularized initial model has no obvious effect on the inversion results. With the increase of noise level, the peak error and particle size distribution error obtained from the inversion of the three initialization models increase with the increase of noise level. However, the maximum and particle size distribution errors obtained by using the flattest model and the smoothest model are obviously smaller than those obtained by the minimum initial model. When the noise level is greater than 0.01, the results of particle size distribution obtained by selecting the flattest initial model are better than the results obtained by using the smoothest initial model and the minimum initial model. The peak values obtained by using the smoothest initial model are better than those obtained by the smoothest initial model and the minimum initial model. Therefore, when the regularization algorithm is used to process the noisy dynamic light scattering data, in order to obtain the optimal particle size distribution information, the flattest initial model should be adopted, and if the most accurate peak information is needed, the smoothest initial model should be chosen.
【作者单位】：山东理工大学电气与电子工程学院;
【基金】：国家自然科学基金项目(61205194) 山东省自然科学基金项目(ZR2012FL22,ZR2012EEM028)
【分类号】：TN911.4

【参考文献】