OPTIMAL MULTI-ENERGY MANAGEMENT IN SMART ENERGY SYSTEMS: A DEEP REINFORCEMENT LEARNING APPROACH AND A CASE-STUDY ON A FRENCH ECO-DISTRICT - Mines Paris Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

OPTIMAL MULTI-ENERGY MANAGEMENT IN SMART ENERGY SYSTEMS: A DEEP REINFORCEMENT LEARNING APPROACH AND A CASE-STUDY ON A FRENCH ECO-DISTRICT

Pilotage Optimal Multi-Energies dans les Systèmes Energétiques Intelligents: une Approche Basée sur de l'Apprentissage par Renforcement Profond et une Etude de Cas sur un Ecoquartier Français

Résumé

This research work introduces a novel approach to energy management in Smart Energy Systems (SES) that leverages the power of Deep Reinforcement Learning (DRL) algorithms. We propose a Smart Energy Management System (SEMS) that optimizes the management of flexible energy systems in SES, including heating, cooling and electricity storage systems as well as District Heating and Cooling Systems (DHCS) such as district-level Thermo-Refrigerating Heat Pumps. The proposed approach is applied on Meridia Smart energy (MSE), a new demonstration project for SES that is currently under construction within the Nice Meridia eco-district in the south of France. The eco-district involves 50 buildings, many of which are equipped with photovoltaic (PV) panels. The occupants will be supplied with heat and cold produced locally in the eco-district thanks to a geothermal Fourth Generation DHCS. In addition to local electricity, heating and cooling production, the SES also integrates multi-energy storage systems, namely an innovative heat storage system by phase-changing materials, a cold storage by ice storage tanks and a battery storage offering additional flexibility. The energy management systems developed within the present research work aim at reaching strategic objectives set by the French Environmental and Energy Management Agency (ADEME), funder of this demonstrator project. Among these strategic objectives, we consider raising the renewable energy share to more than 70%, maximizing the energy autonomy of the eco-district and minimizing its carbon footprint. The decision making problem is addressed using a DRL approach. The proposed DRL framework is based on actor-critic architecture and is benchmarked against Model Predictive Control (MPC), which is one of the most widely used methods for advanced process control in both industrial and academic level. Simulation results demonstrate that the proposed DRL approach achieves rewards close to the theoretical MPC optimum in terms of energy cost reduction and environmental sustainability. This work represents one of the first studies in the literature to benchmark DRL and MPC approaches for multi-energy management in SES and suggests that DRL is a promising technique for energy-efficient and sustainable management of SES.
Dhekra_Bousnina.pdf (2.55 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04323498 , version 1 (05-12-2023)

Identifiants

  • HAL Id : hal-04323498 , version 1

Citer

Dhekra Bousnina, Gilles Guerassimoff. OPTIMAL MULTI-ENERGY MANAGEMENT IN SMART ENERGY SYSTEMS: A DEEP REINFORCEMENT LEARNING APPROACH AND A CASE-STUDY ON A FRENCH ECO-DISTRICT. Smart Energy Systems, International Conference, Energy Cluster Denmark; AAlborg University Denmark, Sep 2023, Copenhaguen, Denmark. ⟨hal-04323498⟩
33 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More