Influence of initial model on regularized inversion of noisy dynamic light scattering data
-
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90 nm窄单峰、90 nm宽单峰和250 nm窄单峰、250 nm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: Regularization algorithm is a common method to recover the particle size distributions (PSDs) from dynamic light scattering (DLS) data. The initial regularization model has important influence on the inversion results. In this paper, the narrow and wide simulation distributions of 90 nm and 250 nm particles were inversed by the smallest, the flattest and the smoothest initial model respectively. The inversion results show that the initial model has almost no influence on the inversion results under the noise level of 0. With the increase of noise level, although the errors of peak and PSD value inversed by the three initial models are all increased, the increases by using flattest model and smoothest model are obviously less than that using the smallest model. When the noise level is greater than 0.01, the better particle size distribution results can be obtained by using the flattest model than the smallest and the smoothest models, and more accurate particle peak values can be got by using the smoothest model than the flattest and the smallest models. For obtaining the optimal PSDs inversed by regularization from noisy DLS data, the flattest model is recommended, and for getting the optimal peak value, the smoothest model is the best choice.
点击查看大图
计量
- 文章访问数: 1342
- HTML全文浏览量: 292
- PDF下载量: 352
- 被引次数: 0