本研究針對配製皮帶式馬達發電機(Belt-driven Starter Generator, BSG)之輕度混合並聯式油電混合車(Parallel Hybrid Electric Vehicle)之架構，提出一能量管理策略(Energy Management Strategy, EMS)，其使用以遺傳基因演算法(Genetic Algorithm, GA)為即時最佳化演算法之等效油耗最小策略(Equivalent Consumption Minimization Strategy, ECMS)為基礎，利用油電車輸出之電池電量(State Of Charge, SOC)偏差、引擎油耗及排汙(Emission)組成多目標問題，並使用多目標基因演算法(Multi-Objective Genetic Algo-rithm)求出控制策略之柏拉圖最佳解集合(Pareto-Optimal Solution)，此解集合為車輛行駛時，控制策略決定操控關鍵參數的資料庫，藉由徑向基底函數網路(Radial Basis Function Network, RBFN)及遞迴最小平方(Recursive Least Squares, RLS)，決定操控關鍵參數即可即時的決定出引擎之操作點及CVT(Continuously Variable Transmission)齒比(Gear Ratio)。 此外，油電車之油耗及排汙會受到不同的行車型態(Driving Cycle)類型(市區、郊區及高速公路)很大的影響，因此本研究使用基於多層感知機(Multilayer Percep-tron, MLP)之行車型態辨識(Driving Pattern Recognition, DPR)演算法，即時的辨識車輛所處的行車型態，並利用此辨識結果挑選相對應之資料庫，使電池電量可以維持，並同時改善油耗及排汙。 為了驗證本研究提出之多目標等效油耗最小策略(Multi-Objective ECMS, MOECMS)，初步驗證使用由車輛模擬軟體ADVISOR(Advanced Vehicle SimulatOR)與MATLAB/Simulink建立之基於後視法(Backward-Facing Method)之油電車模型，將MOECMS整合至其中進行模擬以分析及驗證其有效性。 為了評估MOECMS在嵌入式控制器(Embedded Controller)之可行性，本研究搭建了硬體迴路(Hardware-in-the-Loop, HiL)實驗平台，將MOECMS寫入此嵌入式控制器並使用控制器區域網路(Controller Area Network, CAN)做訊號交換，同時導入真實駕駛藉此增加實驗之真實性。
由後視模型之電腦模擬結果得知，相較於傳統汽油車，在油耗方面，MOECMS最高有42.31 %的改善。 排汙方面，MOECMS在HC、CO及NOx最高分別有39.83 %、47.85 %及32.77 %的改善。 在電池電量方面，MOECMS可將電量維持在[0.3, 0.7]的範圍內。 由硬體迴路實驗可知，MOECMS可以完整的寫入嵌入式控制器並且與CAN總線配合且即時的運行，而其實驗結果與電腦模擬相當一致，驗證了本研究提出之MOECMS在理論及實務中均有很好的成效。

A novel Energy Management Strategy (EMS) named as Multi-Objective Equivalent Consumption Minimization Strategy (MOECMS) is proposed by this thesis. On the basis of ECMS, the control parameters are determined by Multi-Objective Genetic Algorithm (MOGA) to take fuel consumption, emissions and State Of Charge (SOC) of battery into account. Recursive Least Squares (RLS), Radial Basis Function Network (RBFN) and Multilayer Perceptron (MLP) are employed to generate the most suitable control parameters. In the proposed MOECMS, Genetic Algorithm (GA) is applied to real-time optimize the operation points of ICE. Different from conventional ECMS, MOECMS takes both fuel consumption and emissions, namely HC, CO and NOx, into consideration. The key manipulation parameters, including two equivalent factors and three emission weights, can be quantified and optimized by a Multi-Objective Genetic Algorithm: Non-dominated Sorting Genetic Algorithm (NSGA-II). Through NSGA-II, a group of Pareto-Optimal Solutions can be generated. In order to decide the suitable key manipulation parameters out of the database established by Pareto-Optimal Solutions, RLS and RBFN are introduced. To be more adaptive, the parameters embedded in RBFN can be fine-tuned via Reinforcement Learning (RL) by using the real-time data of battery SOC and TSSF. The battery SOC, therefore, can be stabilized and retained within a narrow range. The database of the key manipulation parameters, however, is dependent of driving cycles. It implies different driving cycles significantly affect the performance of HEV with respect to fuel consumption and emissions. Therefore, a Driving Pattern Recognition (DPR) algorithm based on MLP is included so that MOECMS can be more adaptive to adopt the appropriate database for various road conditions.
At the preliminary design stage, the computer simulations based on backward-facing HEV mathematical model are undertaken to verify the performance of MOECMS through the softwares: ADVISOR (ADvanced VehIcle SimulatOR) and MATLAB/Simulink. To ensure the feasiblity in real-world circumstances, the proposed MOECMS is physically coded into the embedded DSP (Digital Signal Processor) chip and verified by Hardware-in-the-Loop (HiL) experiments.
According to the simulation results, the fuel consumption can be reduced by 42.31% in terms of NYCC driving cycle in comparison to pure ICE vehicles. As to emissions, the computer simulations show that HC, CO and NOx can be reduced by 39.83%, 47.85% and 32.77% respectively. The battery SOC can be retained within [0.3, 0.7] for any unpredictable driving cycles. From the HiL experiments, the performance of MOECMS is superior, compared with traditional EMS. The experimental results by HiL are, in fact, pretty close to the computer simulations by MATLAB/Simulink. Therefore, it implies that the proposed MOECMS can be potentially applied to the real-world BSG HEVs in the future.

[1] 張佑生（2016）。 巴黎氣候協定今生效 台灣2030年將減碳23%。 聯合新聞網。 取自https://udn.com/news/story/10543/2084080
[2] Opportunities to reduce light vehicle emissions in Australia. (2014). from http://climatechangeauthority.gov.au/reviews/light-vehicle-emissions-standards-australia/opportunities-reduce-light-vehicle-emissions
[3] 李淑蓮（2012）。 電動車市場規模在2020年將逾500萬輛、市場規模可望達1兆5千億美金。 北美智權報。 取自http://www.naipo.com/Portals/1/web_tw/Knowledge_Center/Industry_Economy/publish-8.htm
[4] P. Harrop. (2013). The electric vehicle business: new developments. from http://www.idtechex.com/research/articles/the-electric-vehicle-business-new-developments-00005542.asp
[5] N. Murgovski, J. Sjöberg, J. Fredriksson. (2011). A methodology and a tool for evaluating hybrid electric powertrain configurations. International Journal of Electric and Hybrid Vehicles, 3, 219-245.
[6] L. Thibault, O. Grondin, C. Querel, G. Corde. (2012). Energy Management Strategy and Optimal Hybridization Level for a Diesel HEV. SAE International, 260-271.
[7] A. Panday, H. O. Bansal. (2014). A Review of Optimal Energy Management Strategies for Hybrid Electric Vehicle. International Journal of Vehicular Technology, 1-19.
[8] S. Liu, C. Du, F. Yan, J. Wang, Z. Li, Y. Luo. (2012). A Rule-based Energy Management Strategy for a New BSG Hybrid Electric Vehicle. Paper presented at the Third Global Congress on Intelligent Systems, Wuhan, China.
[9] I. Arsie, C. Pianese, G. Rizzo, M. Santoro. (2001). A model for the energy management in a parallel hybrid vehicle. Paper presented at the 3rd International Conference on Control and Diagnostics in Automotive Applications, Genova, Italy.
[10] B. Huang, Z. Wang, Y. Xu. (2006). Multi-Objective Genetic Algorithm for Hybrid Electric Vehicle Parameter Optimization. Paper presented at the International Conference on Intelligent Robots and Systems, Beijing, China.
[11] X. Hu, Z. Wang, L. Liao. (2004). Multi-Objective Optimization of HEV Fuel Economy and Emissions Using Evolutionary Computation. SAE TECHNICAL PAPER SERIES.
[12] D. Kum, H. Peng, N. K. Bucknor. (2011). Supervisory Control of Parallel Hybrid Electric Vehicles for Fuel and Emission Reduction. Journal of Dynamic Systems, Measurement, and Control, 133.
[13] G. PAGANELLI, S. DELPRAT, T. M. GUERRA, J. RIMAUX, J. J. SANTIN. (2002). Equivalent Consumption Minimization Strategy For Parallel Hybrid Powertrains. Paper presented at the IEEE 55th Vehicular Technology Conference, Birmingham, AL, USA.
[14] C. Musardo, G. Rizzoni, Y. Guezennec, B. Staccia. (2005). A-ECMS: An Adaptive Algorithm for Hybrid Electric Vehicle Energy Management. European Journal of Control, 11, 509-524.
[15] S. Onori, L. Serrao, G. Rizzoni. (2015). Hybrid Electric Vehicles: Energy Management Strategies: Springer London.
[16] A. Brooker, K. Haraldsson, T. Hendricks, V. Johnson, K. Kelly, B. Kramer, T. Markel, M. O’Keefe, S. Sprik, K. Wipke, M. Zolot, D. Bharathan, S. Burch, M. Cuddy, D. Rausen. (2003). ADVISOR Documentation. from http://adv-vehicle-sim.sourceforge.net/advisor_doc.html
[17] L. Guzzella, A. Amstutz. (2005). Vehicle Propulsion Systems: Introduction to Modeling and Optimization. Germany: Springer Berlin Heidelberg.
[18] 曾小華, 宮維鈞（2014）。 ADVISOR 2002電動汽車仿真與再開發應用。 北京，中國：機械工業出版社。
[19] J.-S. Won, R. Langari, M. Ehsani. (2005). An energy management and charge sustaining strategy for a parallel hybrid vehicle with CVT. IEEE Transactions on Control Systems Technology, 13(2), 313-320.
[20] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan. (2002). A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, 6(2), 182-197.
[21] M. Beza, M. Bongiorno. (2014). Application of Recursive Least Squares Algorithm With Variable Forgetting Factor for Frequency Component Estimation in a Generic Input Signal. IEEE Transactions on Industry Applications, 50(2), 1168-1176.
[22] K. J. Åström, B. Wittenmark. (2008). Adaptive Control: Courier Corporation.
[23] D. Zhou, W. Liu, W. Zhou, S. Dong. (2013). Research on Network Traffic Identification Based on Multi Layer Perceptron. Telecommunication Computing Electronics and Control, 12, 201-208.
[24] H. He, C. Sun, X. Zhang. (2012). A Method for Identification of Driving Patterns in Hybrid Electric Vehicles Based on a LVQ Neural Network. Energies, 5, 3363-3380.
[25] J. Wang, Q. N. Wang, X. H. Zeng, P. Y. Wang, J. N. Wang. (2015). Driving cycle recognition neural network algorithm based on the sliding time window for hybrid electric vehicles. International Journal of Automotive Technology, 16(4), 685-695.
[26] L. Fang, S. Qin, T. Li, K. Zhu. (2011). Simultaneous Optimization for Hybrid Electric Vehicle Parameters Based on Multi-Objective Genetic Algorithms. Energies, 4, 532-544.
[27] What Is Hardware-In-The-Loop Simulation? from https://www.mathworks.com/help/physmod/simscape/ug/what-is-hardware-in-the-loop-simulation.html
[28] L. Wang, A. H. C. Ng, K. Deb. (2011). Multi-objective Evolutionary Optimisation for Product Design and Manufacturing: Springer London.
[29] H. Yu, T. Xie, S. Paszczynski, B. M. Wilamowski. (2011). Advantages of Radial Basis Function Networks for Dynamic System Design. IEEE Transactions on Industrial Electronics, 58(12), 5438-5450.
[30] S. Chen, C. F. N. Cowan, P. M. Grant. (1991). Orthogonal least squares learning algorithm for radial basis function networks. IEEE Transactions on Neural Networks, 2(2), 302-309.
[31] 蘇木春, 張孝德（2004）。 機器學習：類神經網路、模糊系統以及基因演算法則。 台灣：全華科技圖書股份有限公司。