鋰離子電池充放電特性與模型建立

Abstract

現今有許多與電池測試、建模相關的研究，也有許多學者嘗試以正向脈衝充電法 (PPC) 為電池進行充電最佳化，卻常常僅分析相同種類的鋰電池，缺乏針對不同種電池的綜合評估，實驗結果也大多未獲得共識，因此本研究為不同種類、不同健康度的電池進行各式測試，實驗結果顯示鋰三元 (NMC) 電池於充放電時具有較明顯的溫度變化，磷酸鋰鐵 (LFP) 、鈦酸鋰 (LTO) 電池的溫度與電壓變化較小，兩種LFP電池中動力型電池的內阻、最大溫升皆低於能量型電池。脈衝放電測試中，本研究為鈷酸鋰 (LCO) 和NMC電池建置了二階戴維寧等效電路模型，結果顯示模型參數可反映出電池性能的優劣，且可估算出電池的內阻值。於PPC實驗中，本研究以2C-PPC充電法為各式電池進行充電，結果顯示充電過程的最大溫升均高於1C-CC充電法，且0.25 Hz 的脈衝頻率優於較為低頻的0.05 Hz，而0.25 Hz 的1C-PPC充電法可顯著降低老化NMC電池的溫升。此外，本研究製作了基於Arduino微控制器的電池監控裝置，可同時記錄電壓、電流和溫度數據。本研究亦利用中央氣象署與實驗室發電數據，以線性迴歸、多項式迴歸、嶺迴歸模型進行太陽能發電量迴歸分析，結果顯示多項式迴歸模型表現最佳，MSE為0.007，MAE為0.065，R-squared為0.94。

Many studies focus on battery testing, modeling, and charging optimization. However, these studies often lack comprehensive evaluations across different battery types, and their experimental results usually don't reach a consensus. Therefore, this study conducts various tests on batteries with different types and SOH. Experimental results show that NMC batteries have notable temperature changes during charging and discharging, while LFP and LTO batteries show smaller temperature and voltage changes. LFP power cell has lower internal resistance and temperature rise compared to LFP energy cell. In pulse discharge tests, this study establishes second-order Thevenin battery models for LCO and NMC batteries. Model parameters can reflect battery performance and estimate internal resistance. In Positive Pulse Charging (PPC) charging experiments, this study uses 2C-PPC charging method for various batteries. Results show that the maximum temperature rise during 2C-PPC is higher than 1C-CC. 0.25 Hz 2C-PPC was found to be better than 0.05 Hz 2C-PPC, and 0.25 Hz 1C-PPC can significantly reduce the temperature rise in aged NMC batteries. This study also developed a battery monitoring device based on the Arduino microcontroller, which can record voltage, current, and temperature simultaneously. Another part of this study is regression analysis of solar power. This study uses solar power data and weather data from the Central Weather Bureau, employing Linear Regression, Polynomial Regression, and Ridge Regression models. Results show that Polynomial Regression model performs best, with 0.007 MSE, 0.065 MAE, and an R-squared of 0.94.