2025_Barbay_SMART_multi-agent

Dataset for the conference report: Barbay A., Pisarski D., Mikułowski G., Błachowski B., Jankowski Ł., A case study in semi-active structural control based on multi-agent reinforcement learning, SMART 2025, XI ECCOMAS Thematic Conference on Smart Structures and Materials, 3-5 July, 2025, Linz (Austria). The publiction is available as: https://hal.science/hal-05358218

This research has been supported by the National Science Centre, Poland, under grant agreement 2020/39/B/ST8/02615. The title of the project is "Reinforcement learning for semi-active structural control and decentralized mitigation of vibrations: development of new algorithms and assessment of of their efficiency". The aim of the project is the development and application of the machine learning techniques of reinforcement learning (RL) in tasks of semi-active structural control. The ultimate goal is the design of a framework that learns quasi-optimal control by itself in a repeated trial-and-error interactions with simulated structures.

The data files are in the txt/CSV format. They contain the control results defined in terms of the total mechnical energy, maximal inter-story displacement, and RMS top floor displacement in the RL-controlled structure, normalized with respect to the same quantities obtained for the structure equipped with an optimally tuned passive TMD. The structure is subjected to a seismic-like excitation, and the means are computed over 1000 independent episodes. Two control modes are compared (seven cases of independently rained single-agent RL, one case of majority-based multi-agent RL), as described in the original publication.