Deep Reinforcement Learning-Based Battery Management System for Electric Vehicles

Description du projet

Un nouveau modèle pour de meilleures batteries de VE

Les véhicules électriques (VE) représentent l’avenir. Alors, qu’est-ce que l’avenir réserve aux batteries de VE? Afin de rendre ces batteries plus simples et faciles à recharger, il est important d’améliorer le système de gestion de batteries (BMS) qui surveille, contrôle et protège les blocs de batterie lithium-ion des VE. Tenant compte du fait que les batteries de VE sont soumises à des cycles de fonctionnement rigoureux et à une variation des conditions environnementales qui compliquent considérablement les interactions, le projet DeepBMS, financé par l’UE, développera un nouveau modèle pour améliorer la fonctionnalité BMS dans les VE. Des algorithmes efficaces fondés sur l’apprentissage par renforcement profond mesureront le comportement complexe variable dans le temps de la batterie. DeepBMS améliorera également la fiabilité des batteries et étendra leur durée de vie grâce à des évaluations plus précises réalisées dans une large gamme de températures et sur l’ensemble de leur cycle de vie.

Objectif

Battery Management System (BMS) plays a pivotal role in monitoring, control, and protecting the Electric Vehicle (EV) Lithium-ion battery packs. In vehicular applications, batteries are usually subjected to harsh operating cycles and varying environmental conditions leading to very complicated interactions of different aging factors and unforeseeable modeling uncertainties. Therefore, the classical model-based techniques cannot completely handle the foregoing factors, which always leave an unwanted state estimation error in the BMS. This project intends to apply a multidisciplinary approach by combining the advantages of deep reinforcement learning and classical model-based techniques to improve the BMS functionality in EVs. Specifically, DeepBMS aims to: 1-Develop efficient deep reinforcement learning-based algorithms which are able to capture the convoluted time-varying behavior of battery and can gradually improve themselves by learning in real-time 2- Combine the beneficial features of model-based and data-driven techniques to improve the state estimation accuracy in a wide temperature range and over the full life span of the batteries, thereby increasing the reliability and extending the battery lifetime. The interdisciplinary nature of DeepBMS is very strong, involving a combination of control and state estimation theory, power electronics, battery storage systems, and machine learning. The supervisor and candidate have excellent complemental research experiences in these fields providing the necessary competencies to bring the project to successful completion. The project ensures two-way transfer of knowledge including training of the candidate in cutting-edge advanced techniques in a state-of-the-art laboratory, which improves his future career prospects. Likewise, DeepBMS is in line with the EU strategic action plan on batteries and its results have a great potential to be further developed at the fundamental and applied levels through follow-up research.