The accurate and reliable estimation of the State-of-Charge (SOC) and State-of-Health (SOH) of Li-ion batteries is paramount to the safe and reliable operation of any electrified vehicle. Not only is accuracy and reliability necessary, but these estimation techniques must also be practical and intelligent since their use in real world applications can include noisy input signals, varying ambient conditions and incomplete or partial sequences of measured battery data. To that end, a novel framework, utilizing deep learning techniques, is considered whereby battery modelling and state estimation are performed in a single unified step.
For SOC estimation, two different deep learning techniques are used with experimental data. These include a Recurrent Neural Network with Long Short-Term Memory (LSTM-RNN) and a Deep Feedforward Neural Network (DNN); each one possessing its own set of advantages. The LSTM-RNN achieves a Mean Absolute Error (MAE) of 0.57% over a fixed ambient temperature and a MAE of 1.61% over a dataset with ambient temperatures increasing from 10°C to 25°C. The DNN algorithm, on the other hand, achieves a MAE of 1.10% over a 25°C dataset while, at -20°C, a MAE of 2.17% is obtained.
A Convolutional Neural Network (CNN), which has the advantage of shared weights, is used with randomized battery usage data to map raw battery measurements directly to an estimated SOH value. Using this strategy, average errors of below 1% are obtained when using fixed reference charge profiles. To further increase the practicality of this algorithm, the CNN is trained and validated over partial reference charge curves. SOH is estimated with a partial reference profile with the SOC ranging from 60% to 95% and achieves a MAE of 0.81%. A smaller SOC range is then used where the partial charge profile spans a SOC of 85% to 95% and a MAE of 1.60% is obtained.
Finally, a fused convolutional recurrent neural network (CNN-RNN) is used to perform combined SOC and SOH estimation over constant charge profiles. This is performed by feeding the estimated SOH from the CNN into a LSTM-RNN, which, in turn, estimates SOC with a MAE of less than 0.5% over the lifetime of the battery. / Thesis / Doctor of Philosophy (PhD)
| Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/23021 |
| Date | January 2018 |
| Creators | Chemali, Ephrem |
| Contributors | Emadi, Ali, Electrical and Computer Engineering |
| Source Sets | McMaster University |
| Language | English |
| Detected Language | English |
| Type | Thesis |
Page generated in 0.0026 seconds