Battery cell optimisation using time– and parameter–adaptive reduced order models

Efficient and precise optimisation methodologies, when applied to improve the real performance of engineering devices, must balance the need to reduce high computational cost and maintain accuracy in the treatment of first principles and physicochemical properties. This is the aim of the time– and parameter–adaptive model order reduction technique presented in this article. In contrast to (standard) pre–processed reduced order models (ROMs), the proposed adaptive strategy presents two principal advantages: (i) it does not require any previous information to derive an accurate ROM, which provides a substantial acceleration of solution calculations, and (ii) the periods of full order model calculations can be further reduced when information needed by the optimisation algorithms to derive ROM updates is provided by the previous simulations. Its implementation within the framework of efficient gradient–like optimisation methods requires a careful analysis of the sensitivities computation in order to resolve certain issues inherent to adaptive ROMs. The proposed technique is applied to the optimisation of the performance of a three–dimensional lithium–ion cell described by the electrochemical pseudo–four–dimensional battery model (p4D model). In particular, the results obtained when maximising the energy delivered by a graphite–NMC811 cell and a graphite–LFP cell demonstrate the robustness and efficacy of the method. The presented methodology has been implemented in a software tool named OptiBat, developed in the framework of a H2020 project, and available in the project website https://defacto-project.eu/defacto-rom-optimisation-tool/.