A deep learning-based predictive controller for the optimal charging of a lithium-ion cell with non-measurable states. (May 2023)

• Record Type: Journal Article
• Authors: Pozzi, Andrea; Moura, Scott; Toti, Daniele
• Is Part Of: Computers & chemical engineering. Volume 173(2023)
• Publication Date: 2023-05
• Subjects: Machine learning -- Model predictive control -- Battery management systems -- Computational complexity -- Deep learning; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2023.108222
• Languages: English
• ISSNs: 0098-1354

Abstract

Abstract: Battery charging is a complex task, which needs to be addressed by a proper control methodology to find the highest charging current while guaranteeing safety. Among the different approaches, model predictive control appears particularly suitable due to its ability in dealing with nonlinear systems and constraints. However, its use in a realistic scenario is limited due to the high computational burden required by the online solution of an optimal control problem. A neural network-based algorithm is here proposed to significantly reduce the real-time computational effort by approximating the predictive control law. In addition, for the first time to the authors' knowledge, an adaptation of the proposed deep learning-based algorithm is presented for the case in which the battery's internal states are not measurable. The superiority of proposed methodology is highlighted in simulation by comparing it with a predictive controller coupled with a properly designed state observer. Highlights: Batteries require suitable management systems to maintain high performance over time. The use of a model-based controller is limited due to high computational cost. Neural networks can be used to significantly reduce computational effort. A deep learning-based approximation of the predictive control law is proposed. The algorithm relies only on output measurements when states are not available.