Showing 1–1 of 1 results for author: Kral, L

Data-Augmented Numerical Integration in State Prediction: Rule Selection

Authors: Jindrich Dunik, Ladislav Kral, Jakub Matousek, Ondrej Straka, Marek Brandner

Abstract: This paper deals with the state prediction of nonlinear stochastic dynamic systems. The emphasis is laid on a solution to the integral Chapman-Kolmogorov equation by a deterministic-integration-rule-based point-mass method. A novel concept of reliable data-augmented, i.e., mathematics- and data-informed, integration rule is developed to enhance the point-mass state predictor, where the trained neu… ▽ More This paper deals with the state prediction of nonlinear stochastic dynamic systems. The emphasis is laid on a solution to the integral Chapman-Kolmogorov equation by a deterministic-integration-rule-based point-mass method. A novel concept of reliable data-augmented, i.e., mathematics- and data-informed, integration rule is developed to enhance the point-mass state predictor, where the trained neural network (representing data contribution) is used for the selection of the best integration rule from a set of available rules (representing mathematics contribution). The proposed approach combining the best properties of the standard mathematics-informed and novel data-informed rules is thoroughly discussed. △ Less

Submitted 14 January, 2025; v1 submitted 9 December, 2024; originally announced December 2024.

Comments: This work has been accepted to IFAC for publication - IFAC SYSID24