PARAMETER FITTING IN DYNAMIC-MODELS
PARAMETER FITTING IN DYNAMIC-MODELS
复制标题
DOI：
10.1016/0304-3800(79)90029-2
复制
发表时间：
1979-01-01
影响因子：
3.1
通讯作者：
BENSON, M
中科院分区：
文献类型：
--
作者：
BENSON, M
本文讨论了当向量y（t）的部分或全部分量有观测时，初值问题（y“）中参数向量p的最小二乘估计的数值方法。特别注意的是技术的发展，虽然不是全球性的，是不太敏感的初始参数估计比采用敏感性方程的标准方法。交互式的方法可能是非常有价值的，当好的起始参数近似不可用。讨论了交互式参数拟合包装PARFIT的主要特点。该软件包包含采用灵敏度方程的标准技术和旨在改善不良参数估计的特殊算法。这些特殊的算法是在考虑用户交互的情况下选择和开发的。一个特殊的方法设计的情况下，观察是不是所有的状态变量的解释。一个例子（使用计算机生成的意见）来说明这种方法。一个互动的方法的力量证明了2个例子[浮游植物和Rangifer tarandus]涉及试图模拟物理观察到的现象。
The numerical approaches for the least squares estimation of the parameter vector p in the initial value problem (y'') when observations are available on some or all components of the vector y(t) are considered. Special attention is paid to the development of techniques which, although not global, are less sensitive to initial parameter estimates than the standard approach employing the sensitivity equations. Interactive approaches may be very valuable when good starting parameter approximations are unavailable. The main features of an interactive parameter fitting packing PARFIT are discussed. This package contains standard techniques employing the sensitivity equations and special algorithms designed to improve poor parameter estimates. These special algorithms were selected and developed with user interaction in mind. One special approach designed for the case when observations are not available on all state variables is explained. An example (using computer generated observations) is presented to illustrate this approach. The power of an interactive approach is demonstrated with 2 examples [of phytoplankton and Rangifer tarandus] involving attempts to model physically observed phenomena.