LSTM-PF 앙상블 모델 기반 리튬 이온 배터리의 저온거동 특성예측

다국어 초록 (Multilingual Abstract)

Battery life prediction is important to maintain stability and performance in battery use. This study presents an approach that combines cycling tests under various current conditions with deep neural network algorithms to identify and predict the trend of battery capacity reduction in low-temperature conditions and constructed 18 characteristic data including test environment and conditions, as well as geometric and statistical features. The importance of these features was analyzed using the Random Forest algorithm, and the top 12 feature data were selected to improve the efficiency and accuracy of the LSTM model. Furthermore, we applied a sequential ensemble
technique that uses the output of the LSTM model as input for the Particle Filter, significantly enhancing the accuracy of the prediction model. This method enabled us to real-time predict battery capacity within a 0.9% error margin based on the Worldwide Harmonized Light Vehicles Test Cycle (WLTC) driving cycle data. Our approach demonstrates the effective prediction of battery capacity trends using real operational environment data, without relying on Electrochemical Impedance Spectroscopy (EIS).

목차 (Table of Contents)

• 1. 서론 1
• 1.1. 연구 배경 1
• 1.2. 연구 사례 4
• 1.3. 연구 목표 7
• 2. 실험 및 방법 9
• 2.1. 실험 · 9
• 2.1.1. 시험 배터리 및 사이클링 장비 · 9
• 2.1.2. 사이클링 시험 및 배터리 열화 데이터 · 10
• 2.1.3. WLTC 프로파일 기반 사이클링 시험 및 데이터 세트 12
• 2.2. 알고리즘 설계 · 15
• 2.2.1. 특성 데이터 추출 15
• 2.2.2. LSTM-PF 앙상블 모델 18
• 2.2.3. 훈련, 검증, 시험을 위한 데이터셋 분리 24
• 2.2.4. Random Forest 26
• 2.2.5. Rolling Window Cross Validation · 27
• 3. 결과 · 29
• 3.1. 특성 데이터 선별 및 모델 파라미터 설정 29
• 3.1.1. 특성 데이터 선별 29
• 3.1.2. 모델 파라미터 설정 30
• 3.2. LSTM-PF 앙상블 모델 기반 용량 예측 33
• 4. 토론 · 36
• 4.1. LSTM-PF 앙상블 모델 검증 36
• 4.2. 모델의 확장성 및 한계 · 37
• 5. 결론 · 38
• 6. 참고문헌 · 39
• ABSTRACT · 45