不同先验信息下D-优化法的估值精度及其适用性

Estimation and Its Applicability of D-Optimality-Based Sensitivity Test under Different Prior Information

采用Monte Carlo模拟仿真方法,研究了感度试验在具有不同的先验信息情况下D-优化法对感度总体参数的估值精度的影响。并将其与升降法和兰利法两种试验方法进行了对比研究。结果表明,D-优化法对初始参数的设置依赖性较小;当具有一定感度总体参数先验信息的情况下,采用升降法进行试验可以得到更准确的均值,而采用D-优化法则可以得到更准确的标准差;在对样品感度信息掌握较少的情况下,D-优化法的估值精度优于其它两种方法,在只掌握试验样品的均值,对标准差不了解的情况下,兰利法可以得到更准确的均值估计,而D-优化法则可以得到更准确的标准差;在任何情况下,D-优化法对标准差及P响应点的估值精度优于升降法和兰利法。

The effects of different prior （known） information on the estimation precision of mean and standard deviation value from D-optimality-based sensitivity test method were studied by Monte Carlo simulation, and the results were eompared with that from the other two sensitivity test methods（ up-and-down method and Langlie method）. The results show that the primary test parameter has less effect on the estimation precision from D-optimality than that from the other two methods. More accurate mean values ean be obtained when enough prior information is available, while better standard deviations can be obtained from D-optimality method. In the case of deficient prior information, the value mean is more precise from D-optimality method than that from the other two methods. When mean is known and standard deviation is unknown, more accurate estimation of mean can be obtained by Langlie method, while more reliable standard deviation can be obtained from D-optimality method. Generally, better estimation of standard deviation and P respond value are obtained by D-optimality than that by the other two methods in any ease,