動量型粒子群體智慧演算法鑑定綠能系統參數

Parameter Identification on Green Energy System with Momentum-type Particle Swarm Intelligence Algorithm

劉恩睿(En-Jui Liu)

指導教授 ： 洪哲文

國立清華大學/工學院/動力機械工程學系/博士(2021年)

摘要

綠能系統逐漸成為二十一世紀發電儲電主流，為達成有效設計與控制系統的效能，參數鑑定是一項至關重要工作，然而目前商用綠能系統難以直接從製造廠商獲得詳細之內部參數數據。本研究將探討三種主流的綠色能源系統，包含光伏太陽電池、質子交換膜燃料電池及鋰離子電池等。應用強健性高之群體智慧演算法預測這三種系統之模型參數，以辨識並取得高精確度的參數值有效地提供預測各系統模組在不同工作條件下的效能。
本論文提出一種使用廣義增量規則來更新粒子速度以求得全域最佳值的動量形式粒子群體智慧演算法並應用於綠能系統之參數辨識問題上。首先本研究以拉伸/壓縮彈簧、壓力容器、焊接組合鋼樑三種基準範例工程最佳化問題執行演算法之強健度測試以確認演算法之有效性後再進行前述之三種能源模組參數辨識。分析解包含有目標函數值的收斂率、解答精確性及平均絕對誤差與標準差。本論文之光伏太陽電池使用二極體電路模型，而質子交換膜燃料電池及鋰離子電池均使用電化學模型。透過動量型式粒子群體智慧演算法最佳化計算獲得這些綠能系統內之模組參數後將可掌控系統內之電池電極、電解質與隔離膜之間的化學反應機制。
由三種工程最佳化的問題之參數最佳化結果顯示出動量型粒子群體智慧演算法具有相當好的最佳化函數值求解能力和演算法穩健性。進一步在能源系統的參數最佳化結果顯示光伏太陽電池電壓的均方根誤差為8.839E-4 V，而質子交換膜燃料電池電壓的平方誤差和為2.0656 V。在鋰離子電池動態負載模型分析上，本論文亦建置了鋰離子電池實驗平台進行實驗，以提供模組分析之效能比對驗證。由實驗與分析結果比對後，鋰離子電池的電壓平均絕對誤差為3.6350 mV，預測1 C充放電的SOC為0.0139 %。將參數套入2C、0.5C以及路況動態效能驗證中，鋰離子電池端電壓的誤差小於0.5 %，電池之SOC小於0.8 %。從計算結果顯示動量粒子群體演算法可以有效率地計算出電池模組之待定參數的最佳化值，相當適合應用在需要高精準度之綠能系統的電池動態負載模型效能預測上。

關鍵字

動量型粒子群演算法 ； 光伏太陽能電池 ； 質子交換膜燃料電池 ； 鋰離子電池 ； 強健性測試

並列摘要

Green energy systems have gradually become the mainstream of power generation and storage in the 21st century. In order to achieve effective design and control system performance, parameter identification is a vital task. However, it is difficult for commercial green energy systems to obtain detailed data of internal parameters directly from manufacturers. This thesis will investigate three mainstream green energy systems, including photovoltaic solar cells, proton exchange membrane fuel cells, and lithium-ion batteries. The use of highly robust swarm intelligence algorithm to predict the model parameters of these three systems can identify and obtain high-precision values of parameters to effectively predict the performance of each system model under different working conditions.
This thesis proposes a momentum particle swarm intelligence algorithm that uses the generalized delta rule to update the particle’s velocity to obtain the global optimal value and applies it to the parameter identification problem of the green energy system. First, in this study, three benchmark engineering optimization problems of tension/compression springs, pressure vessels, and welded composite steel beams were used to perform the robustness test of the algorithm to confirm the effectiveness of the algorithm, and then parameter identifications were performed of the aforementioned three energy modules. The analyzed solution includes the convergence rate of the objective function value, the accuracy of the solution, and the average absolute error and standard deviation. The photovoltaic solar cell in this study uses the diode circuit model, while the proton exchange membrane fuel cell and lithium ion battery both use the electrochemical model. After obtaining these module parameters in the green energy system through the optimization calculation using the momentum-type particle swarm intelligent algorithm, the chemical reaction mechanism between the battery electrodes, electrolyte and separator in the system can be controlled.
The parameter optimization results of the three engineering optimization problems show that the momentum-type particle swarm intelligence algorithm has a very good capability to achieve the optimal function value and the robustness of the algorithm. Moreover, the parameter optimization results in the energy system show that the root mean square error of photovoltaic solar cells is 8.839E-4 V, while the sum of square errors of proton exchange membrane fuel cells is 2.0656 V. In the analysis of the dynamic load model of lithium-ion battery, this thesis also built a lithium-ion battery experimental platform for experiments to provide performance comparison verification for module analysis. After comparing the experimental and analytical results, the average absolute error of the voltage of the lithium-ion battery is 3.6350 mV, and the SOC of 1C charge and discharge is predicted to be 0.0139%. Applying the parameters to 2C, 0.5C and dynamic performance verification on road conditions, the error of the lithium-ion battery terminal voltage is less than 0.5%, and the battery SOC is less than 0.8%. The calculation results show that the momentum particle swarm algorithm can efficiently calculate the optimal value of the undetermined parameters of the battery module, so it is quite suitable for application in the performance prediction of the battery dynamic load model of the green energy system that requires high accuracy.

並列關鍵字

Momentum-type Particle Swarm Optimization ； Photovoltaic Cell ； Proton Exchange Membrane Fuel Cell ； Lithium Ion Battery ； Robustness Testing

參考文獻

[1] C. Le Quéré et al., “Temporary reduction in daily global CO 2 emissions during the COVID-19 forced confinement,” Nature Climate Change, vol. 10, no. 7, pp. 647-653, 2020.

[2] J. Moor, “The Dartmouth College artificial intelligence conference: The next fifty years,” Ai Magazine, vol. 27, no. 4, pp. 87-87, 2006.

[3] M. Dorigo, M. Birattari, and T. Stutzle, “Ant colony optimization,” IEEE computational intelligence magazine, vol. 1, no. 4, pp. 28-39, 2006.

[4] D. Karaboga and B. Basturk, "On the performance of artificial bee colony (ABC) algorithm," Applied soft computing, vol. 8, no. 1, pp. 687-697, 2008.

[5] J. Y. Liu, M. Z. Guo, and C. Deng, “GeesePSO: An efficient improvement to particle swarm optimization,” Computer Science, vol. 33, no. 11, pp. 166-168, 2006.