陳瑄易Chen, Syuan-Yi吳治廷Wu, Chih-Ting2025-12-092030-01-072025https://etds.lib.ntnu.edu.tw/thesis/detail/7e2512e891e0007a313d8394ee98021b/http://rportal.lib.ntnu.edu.tw/handle/20.500.12235/125047本論文針對一具有鋰電池與超級電容之複合電力電動巴士(Hybrid Electric Bus, HEB)設計一套能量管理系統(Energy Management System, EMS)，期望利用所研究之控制策略，有效調度雙電力之間的功率流向，同時平滑鋰電池輸出電流以達到保護鋰電池作用。為評估所提方法之能源使用效益，本研究建立一套HEB能源模擬軟體，並設計與比較不同控制策略之成本。首先以基本規則庫控制策略與全域搜索控制策略在不同需求功率和超級電容殘電量的情況下，計算出最佳功率分配比以達到最小化電能消耗；傳統頻率解耦控制策略利用低通濾波器將需求電能解耦為高頻與低頻兩頻段，使雙電力源各自依據其特性滿足需求功率。為使低通濾波參數依據當前環境自適應調整，本研究進一步發展強化式頻率解耦控制策略，使 EMS 可於系統運作過程中，不斷學習以找到最佳解。而在學習方法上本論文採用Q-Learning 策略並將其與頻率解耦控制策略結合，利用Q-Learning 動態調整低通濾波器之濾波參數以平滑化鋰電池輸出電流，達到保護電池之目的。結果比較中，以全球輕型車輛測試規範(Worldwide Harmonized Light Vehicles Test Procedures, WLTP)行車形態為例，基本規則庫控制策略為比較基準，使用全域搜索控制策略的總能耗改善率達20.62%，標準差改善率達10.90%；傳統頻率解耦控制策略的總能耗改善率達5.62%，標準差改善率達13.19%；強化式頻率解耦控制策略的總能耗改善率達8.88%，鋰電池輸出標準差改善率達最高的16.71%。This thesis focuses on designing an Energy Management System (EMS) for a Hybrid Electric Bus (HEB) equipped with lithium battery and supercapacitor. The objective is to utilize the proposed control strategies to efficiently manage the power flow between the two energy sources while smoothing the output current of the lithium battery, thereby protecting them. To evaluate the energy efficiency of the proposed method, this research developed an HEB energy simulation software and studied different control strategies. These include a Rule-Based Control (RBC) strategy and a Global Search Control (GSA) strategy, which calculate the optimal power distribution ratio under varying power demands and supercapacitor State of Charge (SOC) to minimize energy consumption. The Frequency Decoupling Control (FDC) strategy is used to decouple the power demand, enabling more efficient energy release or storage between the two energy sources. Furthermore, this study develops a Reinforcement Learning Frequency Decoupling Control (RLFDC) strategy, which allows the system to continuously learn and find optimal solutions during operation, offering greater flexibility compared to conventional algorithms. RLFDC adopts the Q-Learning strategy as its learning method. Q-Learning is used to dynamically adjust the filter parameters of the low-pass filter to smooth the output current of the lithium battery, thereby achieving the goal of battery protection.In the comparison of results using the WLTP driving cycle as an example, with the RBC serving as the benchmark, the GSA control strategy achieved a total energy consumption improvement rate of 23.18% and a standard deviation improvement rate of 14.41%. The traditional FDC strategy resulted in a total energy consumption improvement rate of 4.36% and a standard deviation improvement rate of 12.84%. The RLFDC strategy achieved a total energy consumption improvement rate of 8.61% and the highest improvement in lithium battery output standard deviation at 15.45%.強化式頻率解耦控制策略頻率解耦控制策略全域搜索法控制策略基本規則庫控制策略複合式能量管理系統複合電力電動巴士Reinforcement Learning Frequency Decoupling Control StrategyFrequency Decoupling Control StrategyGlobal Search AlgorithmRule-Based Control StrategyHybrid Energy Management SystemHybrid Electric Bus學術論文