TY - THES A1 - Tahoon, Mohammad T1 - Estimating the State of Charge of Lithium-ion Batteries using Deep Learning for Electric Vehicle Applications N2 - Reliable and safe operation of Li-ion batteries in electric vehicles relies on an accurate estimation of their states, more specifically their state of charge (SOC), which is used for managing charging and discharging, for example. Deep learning is increasingly used for modeling complex systems, which generates more interest in battery state estimation. The theoretical investigation of candidate neural networks for SOC estimation is preceded an experimental approach to evaluate the shortlisted neural networks performance. Publicly available testing datasets representing real driving cycles has been used to train the selected networks, taking into consideration different operating temperatures for the exact cycles. Testing the prediction capability of the models across a realistic range of operating conditions is the driving factor for including testing cycles with a temperature range of [0–40] °C in the training datasets. A demonstration of the full training cycle is conducted, along with the optimization problem of network parameters. This work demonstrates the applicable neural network family used for training this kind of sequential data, followed by the shortlisting of two neural networks based on their performance using literature research. After selecting the networks, finer evaluation criteria is applied, including both their prediction capability and training requirements. The results show that both networks, long short-term memory (LSTM) and gate recurrent units (GRU), offer satisfactory prediction capability with mean absolute error (MAE) of (2.9%, 2.7%) respectively, when averaged across the tested temperatures. The GRU network on the other hand had significantly higher resource requirements and a more complex architecture. Finally, a more recent architecture is briefly discussed, borrowing the strength of a parallel application field. The scope of this work is to highlight both the capabilities and cost of training deep learning networks for SOC estimation. KW - Electric vehicles KW - Lithium-Ion batteries KW - State of charge KW - Deep learning KW - Neural networks Y1 - 2023 ER -