(A) study on AI-based aging estimation algorithms of lithium polymer batteries with phase change materials

다국어 초록 (Multilingual Abstract)

Thermal-run away is the major issue in the lithium battery of the Electric Vehicle (EVs). Battery will subjected to temperature small changes if operating for charge and discharge conditions. This small temperature will lead to the bigger risks such as explosions. In this point of view, we proposed novel effective thermal management system with the Phase Change Materials (PCMs) instead of traditional methods of liquid and air cooling methods. PCMs are very simple to install and integrate in the EV battery. Additionally machine learning such as Long Short Term Memory (LSTM) and Random Forest (RF) methods are suggested for the fast and accurate prediction of the battery maximum temperature () and the thermal gradient △T.
Initially lithium polymer pouch battery was selected to conduct this study. Hot soaking at 50℃ and cold soaking at -10℃ experiments were conducted to evaluate the PCM characteristic with the heating and cooling curves. From this experiment the expanded graphite PCMs are able to protect the battery for a long time at high extreme weather conditions.
Second the battery pack of 500W, PCMs were incorporated on both sides of the battery pack. This battery pack was placed in the aluminum case. Experiments were conducted with PCMs and NoPCMs battery module at 25℃, 50℃ and –10℃, the data was collected and analyzed to identify the best PCM that suitable for any temperature conditions based on thermal metrics of and △T.
Third, the experiment data were send to train with the commonly used machine learning methods. The data were normalize, cleaned and trained with the selected networks of LSTM and RF and evaluated the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) to find the best PCMs out of all selected PCMs by calculating the predicted accuracy of and △T.
From the results it was identified that expanded graphite PCMs are protecting longer time during hot and cold soaking periods. Overall prediction accuracy obtained with LSTM is 99.7% and with the RF method the predicted accuracy is slightly lower than LSTM of 98.4%. With the experiment results and the machine learning results, the EG PCMs strongly recommended for EV batteries at any weather conditions.

목차 (Table of Contents)

• List of Figures
• List of Tables
• List of Abbreviations
• Abstract
• 1. Introduction 1
• 2. Electric Vehicle Battery Cooling Methods 4
• 2.1 Traditional battery cooling methods 4
• 2.2 Effect of Battery Aging with High Temperature 6
• 3. Thermal Management with PCMs for EV Battery 9
• 3.1 Optimization of PCM Characteristics 10
• 3.1.1 Selection of PCMs at different temperatures 11
• 3.1.2 PCM Hysteresis Characteristics Evaluation 12
• 3.1.3 Proposed Battery Thermal Management System 16
• 3.2 Thermodynamic Modeling of Battery Pack 19
• 4. AI Algorithms for Battery Temperature Prediction 20
• 4.1 Methods of Machine Learning 20
• 4.2 LSTM Method Prediction Algorithm 24
• 4.3 Random Forest Algorithm 27
• 5. Experiment Methodolgy 29
• 5.1 Setup for the Battery temperature estimation 29
• 5.2 LSTM training model for temperature prediction 32
• 5.3 Random Forest training model for temperature prediction 38
• 6. Results 42
• 6.1 Experiment Result for Battery Temperature 42
• 6.1.1 Result of battery temperature at 25°C 42
• 6.1.2 Result of battery temperature at 50°C 43
• 6.1.3 Result of battery temperature at -10°C 45
• 6.2 Machine Learning Training Results 48
• 6.2.1 LSTM Training result at 25°C 48
• 6.2.2 RF Training result at 25°C 51
• 6.2.3 LSTM Training result at 50°C 54
• 6.2.4 RF Training result at 50°C 57
• 6.2.5 LSTM Training result at -10°C 61
• 6.2.6 RF Training result at -10°C 64
• 6.3 Comparison of Experimental and ML prediction results 67
• 6.4 Training Results with Different Epochs in LSTM Model 71
• 6.5 Training Results with Different Epochs in RF Model 72
• 7. Conclusion 75
• References 77