| 摘要: | 隨著能源科技與行動裝置的快速發展，鋰電池已成為現代電力儲存的關鍵技術，廣泛應用於電動車、可攜式電子產品、無人機、醫療設備、智慧家庭裝置及儲能系統等多元領域。鋰電池具備高能量密度、低自放電率與長使用壽命等優勢，使其在各類產品中扮演不可或缺的角色。面對全球碳中和與淨零排放目標的推動，鋰電池更被視為實現能源轉型的核心主軸。隨著應用需求不斷提升，如何掌握鋰電池之健康狀態(State of Health, SOH)已成為產業與學界重視的重要課題。
SOH 為衡量鋰電池目前可用容量相較於出廠初始容量之比值，常用以評估電池效能衰退情形與使用壽命。傳統SOH估測多仰賴阻抗分析、循環測試與開路電壓監測等方式，雖具準確性，但在實務應用上常面臨高成本、耗時長與無法即時診斷等限制。
本研究運用資料探勘技術結合機器學習方法，建立鋰電池SOH之預測模型，以提升狀態評估的即時性與準確性。研究核心採用監督式學習演算法，藉由分析電池歷史資料中的變化趨勢，建構可應用於SOH預測的模型。本研究使用四種監督式學習模型進行分析與比較，分別為人工類神經網路(Artificial Neural Network, ANN)、線性迴歸(Linear Regression, LR)、梯度提升(Gradient Boosting)與支持向量機(Support Vector Machine, SVM)。訓練資料來自21700圓柱型鋰電池之老化實驗數據，涵蓋不同廠牌、型號與操作條件。資料前處理步驟包括缺漏值處理、特徵選擇、容量正規化與數據整併，以提升模型穩定性與泛用性。實驗亦模擬資料不連續及歷史不足之情境，以驗證模型在實際應用條件下的預測能力與彈性。研究建立一套具應用性與擴充性的SOH預測模型架構，適用於不同鋰電池型號與操作環境，可協助企業進行電池效能監控、壽命管理與預測性維護。未來亦可延伸應用於智慧手機、筆記型電腦、電動車與儲能設備，提升能源使用效率並降低營運風險，為智慧能源管理系統之發展奠定基礎。
;Lithium-ion batteries have become a core technology in modern energy storage systems, widely used in electric vehicles, portable electronics, drones, medical devices, and smart energy infrastructure. Their high energy density, low self-discharge rate, and long cycle life make them essential across industries. With the global emphasis on carbon neutrality and sustainable development, accurate prediction of a battery’s State of Health (SOH) has become increasingly important for efficient energy management.
SOH refers to the ratio of a battery’s current usable capacity to its original capacity and is a key indicator of battery degradation and remaining useful life. While traditional estimation methods—such as impedance analysis and cycle testing—offer high accuracy, they are often time-consuming, expensive, and unsuitable for real-time applications.
This study applies data mining techniques and supervised learning algorithms to develop a predictive model for lithium battery SOH. Four algorithms are utilized: Artificial Neural Network (ANN), Linear Regression (LR), Gradient Boosting, and Support Vector Machine (SVM). The model is trained using aging data collected from 21700 cylindrical lithium-ion cells under various operational conditions. Data preprocessing includes handling missing values, normalization, and feature selection. To assess real-world applicability, the model is tested on datasets with incomplete historical records.
The proposed model supports battery performance monitoring, lifecycle management, and predictive maintenance. It can be applied to diverse battery-powered products, contributing to improved energy efficiency and system reliability. |
