A data-driven equivalent circuit model's parameter estimation method applied to Lithium-Titanate battery

Lithium battery systems are becoming essential for a variety of applications, such as maximisation of self -produced renewable energy in buildings, grid balances, and electric vehicles. Accurate and equivalent models are required to study and analyse the behaviour of batteries, both static and dynamic. This paper presents a novel data-driven parameter estimation procedure for high-current battery modelling, which has been applied to experimental measurement data conducted on Lithium Titanate batteries. The proposed estimation methodology has been applied to three different equivalent circuit models characterised by increasing modelling complexity. Genetic algorithm optimisation techniques solve the optimisation problem, estimating the model's parameters and a look-up table of battery parameters estimated in discretised State of Charge. Enhance the battery operation modelling at high depth of discharge, the model is applied to the battery operated at high currents, covering its entire domain of state of charge. The fully depleted and charged state is given increased focus to maintain good modelling performance. Unlike previous works, the proposed methodology repeats the parameters estimation procedure for more than thirty-five state of charge ranges using non-equal width ranges to increase model density in critical regions. Based on the estimation results for one C-rate, it was found that the parameter estimation error ranges from 0.17% for the Rint model to 0.03% for the Dual-Polarisation model. The validation results confirmed that the proposed approach had a normalised root mean square error lower than 2%. The proposed methodology encompasses the entire process of parameter estimation, starting from raw experimental data. As such, it can be generalised and applied to other battery technologies.