Python Package for Balanced NeuralODEs

The package builds on top of the Python package torchdiffeq and allows for the training of balanced NeuralODEs, a method to train surrogates. It allows to either perform model reduction, i.e. to reduce the states of the systems, or to train a linear surrogate in the sense of the Koopmann theory. While reduced order models can be used to speed up simulation, linear models are in particular useful in control applications.

The package contains methods for data generation from FMUs, training of the balanced NeuralODEs and analysis of the surrogates,

Since it is a Python package it is platform independent.

See star gazers. Balanced NeuralODEs will be used within the Bosch Heatpump Use Case.

The package is available open source under the MIT licence.