WP1: Machine learning methods and artificial network architecture
There are many different kinds of machine learning (ML) and many different architectures for artificial networks (AN) . This work-package focusses on selecting appropriate AN structures for the controller and what kind of ML method is most appropriate to tune such a method. to conventional controllers in the design phase. Here we will evaluate the methods on simplified but representative problems. The application to the full-scale problem will take place in WP2 and WP3.
WP2: Improved conventional controller design
The goal of this work package is to improve the baseline controller and obtain a state-of-art baseline controller.
WP3: Machine learning of network architectures
In this work package we will employ the best architectures found in WP1 to benchmark these against existing controllers. We will both evaluate their performance and their stability. The analysis of the results will be part of WP4.
WP4: Machine learning controller design evaluation
The methods for machine learning and the artificial network architectures have been designed and tested in the previous work-packages. This work package focusses on analysing the performance of the control methods and on applying ML methods on different controller architectures. By testing the methods on different controller structures, the robustness of the ML methods and the robustness of their performance is evaluated.