| 研究生: |
蔡齊恩 Tsai, Chi-En |
|---|---|
| 論文名稱: |
機器學習輔助鋰離子電池電解液分子之氧化還原電位預測與新型分子設計 Machine Learning-Assisted Redox Potential Prediction and Molecular Design of Lithium-Ion Battery Electrolytes |
| 指導教授: |
許文東
Hsu, Wen-Dung |
| 學位類別: |
碩士 Master |
| 系所名稱: |
工學院 - 材料科學及工程學系 Department of Materials Science and Engineering |
| 論文出版年: | 2026 |
| 畢業學年度: | 114 |
| 語文別: | 中文 |
| 論文頁數: | 117 |
| 中文關鍵詞: | 機器學習 、密度泛函理論 、鋰離子電池 、電解液溶劑 、分子設計 |
| 外文關鍵詞: | Machine Learning, Density Functional Theory, Lithium-Ion Battery, Electrolyte Solvent, Molecular Design |
| 相關次數: | 點閱:8 下載:0 |
| 分享至: |
| 查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
鋰離子電池電解液的電化學穩定窗口(Electrochemical Stability Window, ESW)是決定電池操作電壓上限的關鍵熱力學指標,然而傳統密度泛函理論(Density Functional Theory, DFT)計算成本高昂,難以應用於大規模電解液分子的虛擬篩選。本研究建立一套以機器學習為核心的計算輔助分子設計框架:首先比較兩種氧化還原電位預測方法並擇優採用,再依序執行遺傳演算法分子生成(第一階段)與密度泛函理論驗證(第二階段),系統性地探索高電壓鋰離子電池電解液分子的氧化還原電位預測與新型分子設計。
在預測方法的選擇上,方法一以減碰撞指紋(Reduced Collision Fingerprint, RCFP)取代傳統 ECFP 作為分子指紋輸入,在 Materials Project(MP)數據集上以決策樹模型比較,結果顯示 RCFP(r = 1,1,566 位元)在還原電位預測上的 R² 相較最佳 ECFP 提升約 0.17,在游離能預測上亦保持領先;進一步以 LightGBM 在 PubChemQC 約 2,300 萬筆分子上評估特徵組合,確立 RCFP 結合 RDKit 分子描述符的複合輸入為最佳配置。SHAP 值分析揭示 IE 預測主要仰賴 VSA 類表面積描述符與含氮電拓撲狀態指數,EA 預測則以 BCUT2D_MRHI為最強單一信號。然而 RCFP 的生成依賴 RDKit 的子結構識別,約 800 萬筆 PubChemQC 分子無法轉換為有效指紋而必須捨棄,構成方法一在分子覆蓋率上的固有限制。
方法二改以分子 SMILES 字串為直接輸入的 D-MPNN,不受指紋可表示性限制而能涵蓋全部分子。以 PubChemQC 大規模氣相量子化學數據預訓練後,遷移至 MP 溶液相氧化還原電位數據集進行微調,系統性比較三種微調策略與一組從零訓練基準後,確立以 200 萬筆預訓練配合策略 B(凍結 D-MPNN 編碼器、僅微調前饋神經網路)為最佳組合,在測試集上取得 IE R² = 0.9688與 EA R² = 0.9449。兩種方法的比較結果顯示,方法二在分子覆蓋率與預測精度上均優於方法一,故後續分子生成與篩選均採用方法二所建立之模型。
在第一階段,以方法二之最佳 D-MPNN 模型作為適應度評估器,結合遺傳演算法對電解液候選分子進行大規模結構演化,最終篩選出 25 個候選分子,預測 ESW 分布於 7.09 至 8.39 eV 之間,76% 的分子合成可及性指數(SA Score)低於 3.5。25 個候選分子中有 11 個對應現有商用電解液(包含 DMC、DEC、EMC、EC 等),這些分子原先包含於初始族群中,其再現應理解為保留而非框架之從頭發現,說明模型預測結果與已知優良電解液特性一致,為模型可靠性提供間接支撐。此外,另以官能基片段取代已知電解液分子作為初始族群進行對照實驗,結果顯示初始族群所攜帶的化學先驗知識對演化收斂具決定性影響:不依賴已知電解液骨架的起步雖能觸及與訓練集重疊度較低的化學空間,惟受限於突變算子僅能擴增分子結構、且計算預算與適應度函數設定不同,其收斂表現顯著遜於以已知分子起步的版本。
在第二階段,以 B3LYP/6-31+G(d)/IEF-PCM 對 25 個分子進行 DFT 驗證,計算偶極矩、極化性及 Li⁺ 溶劑化複合物的前線軌域能量。結果顯示 Li⁺ 配位後所有分子的 LUMO 和 HOMO 能量亦同步下降,碳酸乙烯酯的 HOMO-LUMO 能帶隙縮小幅度最大,與其在負極形成 SEI 膜的已知行為吻合,驗證了計算方法的物理合理性。此研究同時揭示,以自由分子 ESW 評估電化學穩定性存在固有局限,Li⁺ 配位態的穩定性與自由分子預測值並不完全對應,DFT 配位態驗證是機器學習篩選不可取代的互補步驟。
The electrochemical stability window (ESW) of lithium-ion battery (LIB) electrolytes is the key thermodynamic property limiting cell operating voltage, yet conventional density functional theory (DFT) calculations remain too costly for large-scale virtual screening. We establish a computational framework for high-voltage LIB electrolytes that compares two redox-potential prediction methods — an improved molecular fingerprint with conventional machine learning (Method 1) and a directed message-passing neural network (D-MPNN) with transfer learning (Method 2) — then applies the better one in two design stages: genetic-algorithm (GA) molecular generation (Stage 1) and DFT validation (Stage 2).
In Method 1, Reduced Collision Fingerprint (RCFP, r = 1, 1,566 bits) outperforms all tested Extended-Connectivity Fingerprints, and a composite RCFP + RDKit-descriptor input gives the best LightGBM performance on ~23 million PubChemQC molecules (ionization energy, IE, R² ≈ 0.760; electron affinity, EA, R² ≈ 0.905). In Method 2, a D-MPNN pretrained on PubChemQC and fine-tuned on the Materials Project dataset with a frozen encoder achieves IE RMSE = 0.1748 eV and EA RMSE = 0.2486 eV. In Stage 1, this model guides a GA generating 25 candidates with predicted ESW of 7.09–8.39 eV, eleven matching known commercial electrolytes. In Stage 2, B3LYP/6-31+G(d)/IEF-PCM calculations show that Li⁺ coordination lowers the LUMO energy of every candidate; ethylene carbonate shows the largest HOMO–LUMO gap narrowing (−0.28 eV), consistent with its known SEI-forming behavior.
Free-molecule ESW alone is therefore an incomplete stability descriptor: DFT validation of the Li⁺-coordinated state is an indispensable complement to machine-learning screening.
1. MacDougall, T., Reduced Collision Fingerprints and Pairwise Molecular Comparisons for Explainable Property Prediction Using Deep Learning, in Maître ès sciences. Université de Montréal, 2021.
2. 陳祐頎、吳昱賢、張家欽, 鋰離子電池電解質─鋰離子傳遞的橋梁, in 科學發展月刊. p. 16–20, 2019.
3. Shakourian-Fard, M., G. Kamath, and S.K. Sankaranarayanan, Evaluating the Free Energies of Solvation and Electronic Structures of Lithium-Ion Battery Electrolytes. Chemphyschem, 17(18): p. 2916–2930, 2016.
4. Xu, K., Electrolytes and interphases in Li-ion batteries and beyond. Chemical reviews, 114(23): p. 11503–11618, 2014.
5. Shao, N., et al., Electrochemical windows of sulfone-based electrolytes for high-voltage Li-ion batteries. J Phys Chem B, 115(42): p. 12120–12125, 2011.
6. Chen, X., et al., Ion–solvent chemistry in lithium battery electrolytes: From mono-solvent to multi-solvent complexes. Fundamental Research, 1(4): p. 393–398, 2021.
7. Xu, K., Nonaqueous liquid electrolytes for lithium-based rechargeable batteries. Chemical reviews, 104(10): p. 4303–4418, 2004.
8. Goodenough, J.B. and Y. Kim, Challenges for rechargeable Li batteries. Chemistry of materials, 22(3): p. 587–603, 2010.
9. Chen, M., et al., Progress on predicting the electrochemical stability window of electrolytes. Current Opinion in Electrochemistry, Article no. 34, 2022.
10. Gibson, L.D., J. Pfaendtner, and C.J. Mundy, Probing the thermodynamics and kinetics of ethylene carbonate reduction at the electrode–electrolyte interface with molecular simulations. The Journal of Chemical Physics, 155(20), 2021.
11. Kazemiabnavi, S., et al., Electrochemical Stability Window of Imidazolium-Based Ionic Liquids as Electrolytes for Lithium Batteries. J Phys Chem B, 120(25): p. 5691–702, 2016.
12. Ong, S.P., et al., Electrochemical windows of room-temperature ionic liquids from molecular dynamics and density functional theory calculations. Chemistry of Materials, 23(11): p. 2979–2986, 2011.
13. Weininger, D., SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules. Journal of chemical information and computer sciences, 28(1): p. 31–36, 1988.
14. Landrum, G., RDKit documentation. Release, 1(1-79): p. 4, 2013.
15. Rogers, D. and M. Hahn, Extended-Connectivity Fingerprints. Journal of Chemical Information and Modeling, 50(5): p. 742–754, 2010.
16. 吳明修, 運用機器學習設計鋰電池電解液分子/Electrolyte molecule design by using machine learning, in 材料科學及工程學系. 國立成功大學. p. 1–81, 2020.
17. Morgan, H.L., The Generation of a Unique Machine Description for Chemical Structures-A Technique Developed at Chemical Abstracts Service. Journal of Chemical Documentation, 5(2): p. 107–113, 1965.
18. Chemaxon. Extended Connectivity Fingerprint ECFP. Available from: https://docs.chemaxon.com/display/docs/extended-connectivity-fingerprint-ecfp.md.
19. Apodaca, R.L. Computing Extended Connectivity Fingerprints. Available from: https://depth-first.com/articles/2019/01/11/extended-connectivity-fingerprints/, 2019.
20. 郭力維, 利用決策樹演算法及第一原理計算開發新型鋰離子電池電解液溶劑分子, in 材料科學及工程學系. 國立成功大學碩士論文, 2024.
21. Breiman, L., et al., Classification and Regression Trees. Wadsworth, 368 p., 1984.
22. Breiman, L., Random forests. Machine learning, 45(1): p. 5–32, 2001.
23. Ke, G., et al., LightGBM: A highly efficient gradient boosting decision tree. Advances in neural information processing systems, 30, 2017.
24. Gilmer, J., et al. Neural message passing for quantum chemistry. in International conference on machine learning. PMLR, 2017.
25. Yang, K., et al., Analyzing learned molecular representations for property prediction. Journal of chemical information and modeling, 59(8): p. 3370–3388, 2019.
26. Heid, E., et al., Chemprop: a machine learning package for chemical property prediction. Journal of Chemical Information and Modeling, 64(1): p. 9–17, 2024.
27. Pan, S.J. and Q. Yang, A survey on transfer learning. IEEE Transactions on knowledge and data engineering, 22(10): p. 1345–1359, 2009.
28. Holland, J.H., Adaptation in natural and artificial systems: an introductory analysis with applications to biology, control, and artificial intelligence. MIT press, 1992.
29. Jensen, J.H., A graph-based genetic algorithm and generative model/Monte Carlo tree search for the exploration of chemical space. Chemical science, 10(12): p. 3567–3572, 2019.
30. Bemis, G.W. and M.A. Murcko, The properties of known drugs. 1. Molecular frameworks. Journal of medicinal chemistry, 39(15): p. 2887–2893, 1996.
31. Lundberg, S.M. and S.-I. Lee, A unified approach to interpreting model predictions. Advances in neural information processing systems, 30, 2017.
32. Becke, A.D., Density‐functional thermochemistry. III. The role of exact exchange. The Journal of chemical physics, 98(7): p. 5648–5652, 1993.
33. Lee, C., W. Yang, and R.G. Parr, Development of the Colle-Salvetti correlation-energy formula into a functional of the electron density. Physical review B, 37(2): p. 785–789, 1988.
34. Cancès, E., B. Mennucci, and J. Tomasi, A new integral equation formalism for the polarizable continuum model: Theoretical background and applications to isotropic and anisotropic dielectrics. The Journal of chemical physics, 107(8): p. 3032–3041, 1997.
35. Tomasi, J., B. Mennucci, and R. Cammi, Quantum mechanical continuum solvation models. Chemical reviews, 105(8): p. 2999–3094, 2005.
36. Jain, A., et al., Commentary: The Materials Project: A materials genome approach to accelerating materials innovation. APL materials, 1(1), 2013.
37. Qu, X., et al., The Electrolyte Genome project: A big data approach in battery materials discovery. Computational Materials Science, 103: p. 56–67, 2015.
38. Nakata, M., et al., PubChemQC PM6: Data Sets of 221 Million Molecules with Optimized Molecular Geometries and Electronic Properties. J Chem Inf Model, 60(12): p. 5891–5899, 2020.
39. Degen, J., et al., On the art of compiling and using 'drug-like' chemical fragment spaces. ChemMedChem, 3(10): p. 1503–1507, 2008.
40. Baell, J.B. and G.A. Holloway, New substructure filters for removal of pan assay interference compounds (PAINS) from screening libraries and for their exclusion in bioassays. Journal of medicinal chemistry, 53(7): p. 2719–2740, 2010.
41. Naejus, R., et al., Excess thermodynamic properties of binary mixtures containing linear or cyclic carbonates as solvents at the temperatures 298.15 K and 315.15 K. The Journal of Chemical Thermodynamics, 29(12): p. 1503–1515, 1997.
42. 李佳容, 以第一原理計算結合類神經網路篩選鋰電池電解液添加劑, in 材料科學及工程學系. 國立成功大學碩士論文, 2017.