基於教與學最佳化演算法之模糊MIT-規則控制應用於Zeta直流-直流轉換器

Abstract

本論文針對Zeta直流-直流轉換器平台提出了一種最佳化模糊MIT規則的自適應控制策略，用於在負載變動下控制DC-DC Zeta直流-直流轉換器的輸出電壓。本論文首先說明了Zeta直流-直流轉換器的系統特性與操作原理，接著根據Zeta直流-直流轉換器充放電特性推導出系統動態模型，並依據本文所需之輸出規格設計了系統所需元件值，藉由電壓的回授訊號與理想的參考值比較，並透過控制器使得系統輸出更加精確及快速。接著，利用模糊理論設計一個模糊MIT規則(FMIT)控制器，透過動態調整學習率來提升系統動態響應，之後比較MIT規則控制器以及FMIT規則控制器，並經由MATLAB SIMULINK軟體模擬以驗證FMIT規則控制器之優越性。而為了進一步改善轉換器輸出在負載變動下之強健性，本論文設計一個基於教與學最佳化模糊歸屬函數之適應性MIT規則控制器(TFMIT)，使得控制器能夠在輸入誤差及其變化量的變動下以調整模糊歸屬函數之區間，使模糊系統在不同誤差下能反應出更精準之歸屬度。最後經由推論引擎計算和解模糊化，得到MIT規則控制器之學習率變化量，其變化量將進一步提高系統之響應速度及強健性，使其具有更快的收斂時間。本論文透過數位訊號處理器(TMS320F28335)實現上述控制策略，並以上升時間和安定時間作為性能比較之指標以驗證上述三種控制策略。透過實驗結果驗證，相比傳統MIT規則控制器之性能，本論文所提出之控制器有顯著的改善效益，並使系統在負載變動下仍能保持其穩定性。

This paper proposes an optimized fuzzy MIT rule-based adaptive control strategy for the Zeta DC-DC converter platform to regulate its output voltage under load variations. First, the paper explains the system characteristics and operating principles of the Zeta DC-DC converter. Based on the charging and discharging characteristics of the Zeta converter, a dynamic model of the system is derived. The necessary component values are designed according to the required output specifications. By comparing the feedback voltage signal with the ideal reference value, the controller ensures the system output is more accurate and faster.Next, a fuzzy MIT rule (FMIT) controller is designed using fuzzy theory to dynamically adjust the learning rate, thereby improving the system's dynamic response. The performances of the MIT rule controller and the FMIT controller are compared, and MATLAB SIMULINK simulations are used to verify the superiority of the FMIT controller. To further enhance the converter's robustness under load variations, an adaptive MIT rule controller based on teaching-learning optimization of fuzzy membership functions (TFMIT) is developed. This controller adjusts the intervals of the fuzzy membership functions based on input errors and their changes,enabling the fuzzy system to achieve more precise membership values under different error conditions.Finally, inference engine calculations and defuzzification are used to determine the learning rate variation of the MIT rule controller. This variation further improves the system's response speed and robustness, achieving faster convergence times. The proposed control strategies are implemented using a digital signal processor (TMS320F28335), with rise time and settling time used as performance comparison indicators to validate the three control strategies. Experimental results demonstrate significant performance improvements of the proposed controller compared to the conventional MIT rule controller, ensuring system stability even under load variations.