arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2103.11309eess.SYcs.SYstat.MEstat.OT
收藏

用Maple扩展到结构可识别性分析:非受控线性时不变结构的交互式探索

Branching out into Structural Identifiability Analysis with Maple: Interactive Exploration of Uncontrolled Linear Time-Invariant Structures

Jason M. Whyte

摘要:

假设我们希望预测一个物理系统的行为。我们可以选择通过模型结构$S$(由系统变量之间的参数关系定义的一组相关数学模型)和参数集$Θ$来表示系统。$Θ$中的每个参数向量与$S$中完全指定的模型相关联。我们使用$S$与系统的意见,估计“真正的”(未知)参数向量。不方便的是,多个参数向量可能导致$S$同样很好地近似数据。如果我们不能区分这些备选方案,而这些备选方案会导致不同的预测,那么我们就不能自信地在决策中使用$S$。这一结果可能会使数据收集和建模方面的努力徒劳无功。当$S$缺乏结构全局可识别性(SGI)时,会出现这种结果。幸运的是,我们可以在数据收集之前测试SGI的各种结构类。一个非SGI的结果可能会指导我们的结构或实验设计的变化,以获得更好的结果。我们的目标是通过定制的Maple 2020程序协助SGI的结构测试。我们考虑连续时间,不受控制,线性时不变的状态空间结构。在这里,状态变量矢量${\bf x}$的时间演化由常系数常微分方程系统建模。我们利用“传递函数”的方法,这也适用于“房室”子类(质量是保守的)。我们使用Maple的“Explore”可以交互式地考虑父结构及其变体,当用户改变观察到的${\bf x}$的哪些组件或具有非零初始条件时获得。这些变化可能会影响SGI测试可用的理想输出的信息内容,从而影响其结果。我们的方法可以通知其他类的结构的交互式分析。

英文摘要:

Suppose we wish to predict the behaviour of a physical system. We may choose to represent the system by model structure $S$ (a set of related mathematical models defined by parametric relationships between system variables), and a parameter set $Θ$. Each parameter vector in $Θ$ is associated with a completely specified model in $S$. We use $S$ with system observations in estimating the "true" (unknown) parameter vector. Inconveniently, multiple parameter vectors may cause $S$ to approximate the data equally well. If we cannot distinguish between such alternatives, and these lead to dissimilar predictions, we cannot confidently use $S$ in decision making. This result may render efforts in data collection and modelling fruitless. This outcome occurs when $S$ lacks the property of structural global identifiability (SGI). Fortunately, we can test various classes of structures for SGI prior to data collection. A non-SGI result may guide changes to our structure or experimental design towards obtaining a better outcome. We aim to assist the testing of structures for SGI through bespoke Maple 2020 procedures. We consider continuous-time, uncontrolled, linear time-invariant state-space structures. Here, the time evolution of the state-variable vector ${\bf x}$ is modelled by a system of constant-coefficient, ordinary differential equations. We utilise the "transfer function" approach, which is also applicable to the "compartmental" subclass (mass is conserved). Our use of Maple's "Explore" enables an interactive consideration of a parent structure and its variants, obtained as the user changes which components of ${\bf x}$ are observed, or have non-zero initial conditions. Such changes may influence the information content of the idealised output available for the SGI test, and hence, its result. Our approach may inform the interactive analysis of structures from other classes.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑