pymodconn: A python package for developing modular sequence-to-sequence control-oriented deep neural networks

Abstract

This paper introduces ''pymodconn'', a comprehensive python package developed for constructing modular sequence-to-sequence control-oriented deep neural networks. These deep neural networks (DNNs) are designed to predict the future dynamics of complex time-dependent systems for given known future data, e.g., control inputs, using past known system dynamics and control inputs. The strength of DNNs in modeling complex systems is well known, but developing an optimal deep learning-based model can be a resource-intensive task. This package streamlines this process, simplifying model architecture selection and fine-tuning. The key strength of pymodconn lies in its high-level modularity, enabling users to design their DNN architectures in a flexible manner via a simple text-based configuration file. This flexibility and the comprehensive nature of pymodconn considerably reduce the development efforts and time for applications where precise control over system dynamics is necessary.

Keywords

• Control-oriented deep neural networks
• System dynamics prediction
• Deep neural networks