Home
grayboxes
Implementation of gray box models in process modelling
Data-driven modeling is gaining interest due to the availability of machine learning tools such as TensorFlow, Torch, Neurolab, etc. The success of data-driven tools is bound to the availability of large data sets. Estimations beyond the range of original observations are severely limited.
In contrast, multi-purpose solver packages employing numerical methods for solving sets of differential equations are not dependent on data. However, they are often limited in reproducing the behavior of processes with chaotic elements, as seen in welding, for example.
A remedy is the combination of theoretical (white box) with empirical (black box) models. Such hybrid (gray box) models have the potential to lower the expenses of theoretical model development and can be calibrated to both the quality and quantity of observed data.
However, it is still difficult to estimate the split in the share of the empirical submodel relative to the theoretical one. grayboxes contributes to finding the optimal split and improving model reliability.
A Python framework for the implementation of white box, gray box, and black box models is provided. The generic box model concept, along with the possibility to distribute theoretical submodels across subdomains, supports collaboration.