Modeling Parameter Sensitivities via Equation-based Algorithmic Differentiation Techniques: The ADMSL.Electrical.Analog Library
Atiyah Elsheikh · Linköping electronic conference proceedings · 2014
Parameter sensitivities of mathematical models play a vital rule in many applications of sensitivity analysis.The availability of algorithmic capabilities for representing and computing these quantities is surly advantageous.In this work it is shown how to systematically transform a Modelica library to another library that describes the desired models together with derivatives of model variables w.r.t.model parameters.The produced library remains with the same structure and the underlying models keep the same interface and outlook.The proposed approach relies on novel equation-based algorithmic differentiation techniques that are especially designed for Modelica.The illustration is rather done via a compact library, the opensource ADMSL library, but rich enough to facilitate a lot of representative Modelica language constructs.The ADMSL library is the algorithmically differentiated version of the standard Modelica library subpackage Modelica.Electrical.Analog.Basic.