身為先進的儲能系統，鋰離子電池組存在著老化壽命方面的複雜性問題。其中，內部電阻（Internal Resistance, IR）對鋰離子電池組的特性產生莫大影響。 IR是鋰離子電池組的一個重要參數，它與能量效率、功率性能、物理壽命以及鋰離子電池組的老化有關。為了獲得IR，本論文使用混合脈衝功率特性測試(Hybrid Pulse Power Characterization)在不同的充電狀態（State of Charge,SOC）和循環下從電池管理系統（Battery Management System）獲得數據。 本論文提出了一種使用帶有徑向基底函數（Radial Basis Function）內核的支援向量迴歸（Support Vector Regression, SVR）來預測鋰離子電池組退化的方法，以及一種使用1至500循環週期之間IR 和 20-90％ SOC的關係進行建模的方法。在此，使用資料驅動(data-driven)方法來實現電池壽命預測，該方法透過使用監督式機器學習（SVR）於每個循環週期的IR來實現。實驗結果表明，直到增加到500個循環週期時IR才發生非線性增長約0.24％。透過對量測到的數據進行分析，發現IR在循環老化過程中呈非線性增長，並下降到80%放電深度(Depth-of-Discharge)。電池電量在第500循環週期預測值為8746 mAh。此外，使用SVR算法，透過R2係數評估擬合質量，得分為0.96，在電池組建模中，RMSE的值為0.55x10-2。;Li-ion battery packs as one of the pioneer energy storage systems have complexity issues about degradation life. It gives some impact on the characterization of Li-ion battery pack, especially internal resistance (IR). The IR is an essential parameter of Li-ion battery packs. It relates to energy efficiency, power performance, physical life, and degradation of Li-ion battery pack. To obtain the values of IR, an IR evaluation test was applied to obtain the data from a battery management system (BMS) using Hybrid Pulse Power Characterization test in a different state of charge (SOC) and cycles. We propose an approach to predict the degradation of Li-ion battery pack using support vector regression (SVR) with Radial Basis Function (RBF) kernel and modeling approach using the relationship between IR, different SOC levels 20%-90%, and cycle in the beginning of life 1 cycle until cycle 500. The data-driven method is used here to achieve battery life prediction.based on IR behavior in every cycle using supervised machine learning, SVR. Our experiment result shows that the IR increasing non-linear approximately 0.24%, and it happened if the cycle increased until 500 cycles. By analyzing the data collected from measurement, it was found out that the IR is increasing non-linearly during cycle aging and degraded to 80% Depth-of-Discharge (DOD). The battery capacity in cycle 500 prediction capacity is 8746 mAh. Besides, using SVR algorithm the quality of the fitting was evaluated using coefficient determination R2, and the score is 0.96. In the proposed modeling process of the battery pack, the value of RMSE is 0.55x10-2.
