通用模拟退火用于稳健多元分析校正

模拟退火是一种全局优化算法，具有跨越局部最优点的机制，最小一乘是一种较常用的最小二乘更为稳健的优化准则，更适用于可能偏离正态分布的实际数据集，本文探讨了用最小一乘为准则并利用模拟退火方法同时测定多组分体系的可能性。应用于2－3组分药物体系分析，获得了满意的结果，本文还探讨了改变步长提高模拟退火算法优化精度的方法。

Generalized Simulated Annealing Applied to Robust Multivariate Analytical Calibration

The most commonly used optimization procedure in multivariant calibration is the least squares method. Usually this kind of methods are nonrobust, the normal distribution of the random variables of the regression model are assumed. Searching robust criteria to replace the least squares criterion would reduce the influence of the outliers in the original data on the analytical results obtained in multi-component analysis. Generalized Simulated Annealing (GSA) as an optimization algorithm has the mechanism walking across local optima. The least absolute deviation is a more robust criterion than widely used least squares, and is more suitable for the data set with possible departure from the -assumption of normal distribution. In this paper, the least absolute deviation was taken as the optimization criterion, and GSA with variable step size was used as a tool for the multicomponent determination in spectrophotometry. This method was applied to the analysis of two- and three-component drug mixtures with satisfactory results. The precision of the GSA has been improved by using variable step size in searching process. This method was applied to the analysis of the mixtures of phenol and resorcinol, drug preparations of aminopyrine, antipyrine, barbital and APC tablets. The results were compared with those obtained by target transformation factor analysis and Kalman filtering. Satisfactory agreeable analytical results were obtained. The precision of the GSA has been substantially improved by using variable step size in the searching process.