G2Elab_bandeau_MAGE_AC-1

Thèse de Sharath Ram Kumar

Knowledge-Driven Reinforcement Learning for Demand Side Multi-Building Energy Management



 
Thèse soutenue le 12/11/2025, au G2Elab

https://theses.fr/2025GRALT053

Sharath's research focuses on the integration of domain knowledge into reinforcement learning workflows, and demonstrates how such an approach can enable real-world deployments in the field of building energy management. In particular, this allows us to tackle challenges such as limited data access & infrastructure, and long learning times, while keeping solutions scalable and easy-to-deploy. The work was conducted through a joint degree program between Nanyang Technological University Singapore and Université Grenoble Alpes, as part of the DesCartes research project at CNRS@CREATE.
PhD Sharath2

Publications

  1. Kumar, S. R., Easwaran, A., Delinchant, B., & Rigo-Mariani, R. (2026). Deep reinforcement learning for coordinated air-conditioner control in groups of buildings using smart meter data. Engineering Applications of Artificial Intelligence, 166, 113536. https://doi.org/10.1016/j.engappai.2025.113536
  2. Sharath Ram Kumar, Arvind Easwaran, Benoit Delinchant, Rémy Rigo-Mariani. Improving Demand Response Programs Using Override Signals with Reinforcement Learning. The 16th ACM International Conference on Future and Sustainable Energy Systems (E-Energy '25), Jun 2025, Rotterdam, Netherlands. pp.603-611, ⟨10.1145/3679240.3734657⟩. ⟨hal-05123147⟩
  3. Sharath Ram Kumar, Arvind Easwaran, Benoit Delinchant, and Remy Rigo-Mariani. 2024. Real-time Retail Electricity Pricing Using Offline Reinforcement Learning. In Proceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems (e-Energy '24). Association for Computing Machinery, New York, NY, USA, 454–458. https://doi.org/10.1145/3632775.3661964
  4. Kumar, S. R., Rigo-Mariani, R., Delinchant, B., & Easwaran, A. (2023, November). Action Masking for Safer Model-Free Building Energy Management. In ACM SIGEnergy Workshop on Reinforcement Learning for Energy Management in Buildings & Cities (RLEM) (Best Poster Award) https://hal.science/hal-04299564/document
  5. S. R. Kumar, R. Rigo-Mariani, B. Delinchant and A. Easwaran, "Towards Safe Model-Free Building Energy Management using Masked Reinforcement Learning," 2023 IEEE PES Innovative Smart Grid Technologies Europe (ISGT EUROPE), Grenoble, France, 2023, pp. 1-5, https://doi.org/10.1109/ISGTEUROPE56780.2023.10407781
  6. Sharath Ram Kumar, Arvind Easwaran, Benoit Delinchant, Remy Rigo-Mariani, “Behavioural cloning based RL agents for district energy management” BuildSys '22: Proceedings of the 9th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation. November 2022 Pages 466–470 https://doi.org/10.1145/3563357.3566165

Présentation

 / 1
 Télécharger( sharath-phd-thesis-presentation_1771863358544-pdf?ID_FICHE=835048&INLINE=FALSE )

Jury

Rapporteurs : 

  • Patrick REIGNER, PROFESSEUR DES UNIVERSITES, GRENOBLE INP - UGA 

  • Rui TAN, ASSOCIATE PROFESSOR, NTU SINGAPORE 

  • Dipti SRINIVASAN, PROFESSOR, NATIONAL UNIVERSITY OF SINGAPORE 

  • Gilles GUERASSIMOFF, PROFESSOR, MINES-PARIS PSL, PARIS 

Direction de thèse : 
  • Benoit DELINCHANT, PROFESSEUR DES UNIVERSITES, GRENOBLE INP - UGA Directeur de thèse 
  • Arvind EASWARAN, ASSOCIATE PROFESSOR, NTU SINGAPORE Co-Directeur de thèse 
  • Rémy RIGO-MARIANI, RESEARCH SCIENTIST, CNRS, GRENOBLE INP - UGA Co-Encadrant 
jury