Neural network approach to the selection of an observed system model

This paper deals with a mathematical model for the deformation of a solid body, wheresensor-based observational data serve as input, and the system's internal properties andexternal influences constitute the output. The finite element method is used to numericallysolve the direct problem, determining system parameters at discrete points. Due to the ill-posed nature of the inverse problem, a quasi-solution approach is utilized. A multilayerneural network methodology is proposed, with each network trained on a specific class ofmodels. During real-time parameter identification, the reality model is treated assituational, meaning it is defined for a specific moment based on recorded input data. Theselection of the most appropriate model is guided by an equivalence criterion rooted in thereciprocity theorem.