Li-ion Battery State-of-Charge estimation algorithm with CNN-LSTM and Transfer Learning using synthetic training data

Abstract

The development of State-of-Charge (SoC) algorithms for Li-ion batteries involves carrying out different laboratory tests with the money and time that this entails. Furthermore, such laboratory labours must typically be repeated for each new Li-ion cell reference. In order to minimise this issue, this work proposes a new approach for developing SoC algorithms, using a Recurrent Neural Network in combination with a Transfer Learning method. The latter technique will make possible to take advantage of the data generated for previously tested cell references and use it for the development of a SoC estimation algorithm for a new cell reference. This work provides a proof-of-concept for the proposed approach, using synthetic data generated from electrochemical models, which describes the behaviour of different Li-ion cell references.