| DC 欄位 |
值 |
語言 |
| DC.contributor | 化學學系 | zh_TW |
| DC.creator | 袁文緯 | zh_TW |
| DC.creator | Wen-Wei Yuan | en_US |
| dc.date.accessioned | 2023-1-17T07:39:07Z | |
| dc.date.available | 2023-1-17T07:39:07Z | |
| dc.date.issued | 2023 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=109223048 | |
| dc.contributor.department | 化學學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 染料敏化太陽能電池 (DSSCs) 是一種新興的光伏技術，由於其具有低成本、靈活性和良好的穩定性等優點而得到了廣泛的研究。染料敏化劑在DSSCs中起著關鍵的作用，然而，由於DSSCs的複雜結構以及缺乏可靠和可理解的定量（分子和電子）結構-性質關係，設計出最佳效率的敏化劑的仍然是一個挑戰。而機器學習 (ML) 已成為一種很有前途的方法，可以通過識別大數據中的相關性進行自動學習，從而加速新材料的開發。在本研究中，我們開發了預測 N3 系列釕染料 PCE 值的 ML 模型，旨在幫助和加速N3系列釕染料的優化。因此為ML訓練集收集了118個N3系列釕染料，利用密度泛函理論(DFT)和時間相關 DFT 計算基態和激發態的 46 個分子和電子性質，作為開發預測 ML 模型的分子描述 子， 採用卷積神經網絡（CNN）、輕量化梯度提升機（LightGBM）和人工神經網絡（ANN）三種 ML 演算法進行模型訓練， 優化的 ML 模型用於預測 93 種不同框架（包括非N3染料） 的釕染料，這些染料未包含在訓練數據集中。預測結果 表明，我們的 ML 模型可以區分具有不同效率（從低到高）的染料。這種 ML 模 型的可用於導出設計出具潛力的釕染料的化學規則，並且還可以用於快速且有效 地虛擬篩選大量新設計的染料。 | zh_TW |
| dc.description.abstract | Dye-sensitized solar cells (DSSCs) are emerging photovoltaic technologies that have been extensively studied due to their low cost, flexibility, and stability. Sensitizers play a key role in DSSCs; design of an optimal sensitizer is challenge due to the complex architecture of DSSCs and the lack of a reliable and understandable quantitative (molecular- and electronic-) structure–property relationship. Herein, we develop a predictive and accurate ML model for PCE of Ru-dyes aimed to assist and accelerate the optimization of Ru-dyes for DSSC applications. 118 N3-series Ru dyes were collected for the ML training set. DFT and time-dependent DFT were employed to calculate 46 molecular and electronic properties at the ground and excited states as the molecular descriptors for developing predictive ML models. Three ML algorithms, light gradient boosting machine (LightGBM), artificial neural network (ANN), and convolutional neural network (CNN) were employed for model learners. The trained and optimized ML models were used to predict 93 Ru-dyes (unseen molecules) with different frameworks (including non-N3 dyes). The prediction results showed that our ML models can distinguish dyes with low-to-high efficiencies. The interpretability of such ML model is used to derive chemical rules for designing the potential Ru-dyes and further can be used to rapidly and effectively virtual screen a large number of newly designed dyes to discover promising dyes for practical DSSCs applications. | en_US |
| DC.subject | 染料敏化太陽能電池 | zh_TW |
| DC.subject | Dye-Sensitized Solar Cells | en_US |
| DC.title | Enhanced Development of Potential Ruthenium-based Dyes for Dye-Sensitized Solar Cells by Machine Learning | en_US |
| dc.language.iso | en_US | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |