本論文旨在改善微電網在併網模式時因負載變化而造成的實虛 功控制暫態響應，以及當市電端發生異常轉為孤島模式時，而造成的 電壓擾動的暫態響應，因此使用一具有模糊推理機制和線上學習能力 的 遞迴式小波 模糊類神經網路 (Recurrent Wavelet Fuzzy Neural Network, RWFNN)及來取代傳統比例積分(Proportional Integral, PI)控 制器，來改善暫態響應，達到更好的控制效果，該網路經常被應用在 處理具有非線性和不確定性的控制系統。 本文使用MATLAB R2017a/Simulink來建置一微電網之架構且分 別操作於併網模式與孤島模式來模擬，驗證 PI 與 FNN 及本文所使用 之 RWFNN 的演算法之可行性。實驗方面，將各演算法寫入德州儀器(Texas Instruments, TI)公司的 DSP TMS320F28335 微控制器中，再使 用 Opal-RT 所建立的硬體迴圈(Hardware-in-the-loop, HIL)作為架構， 驗證本文所提演算法及跟其他演算法相比之差異。;This paper dedicated to improve the transient response of the real and reactive power control at grid-connected mode in the microgrid and the voltage disturbance caused by the load change at islanding mode . Therefore, a Recurrent Wavelet Fuzzy Neural Network (RWFNN) with fuzzy inference mechanism and online learning capabilities is used to replace the traditional Proportional Integral (PI) controller to improve transient response . However, to achieve better control effect, the network is often used to deal with non-linear and uncertain control systems. This article uses MATLAB R2017a/Simulink to build a microgrid architecture and operates in grid-connected mode and island mode respectively to simulate the feasibility of PI and FNN and the RWFNN algorithm used in this article. In terms of experiments, each algorithm is implented on DSP TMS320F28335 microcontroller of Texas Instruments (Texas Instruments, TI), and then the hardware-in-the-loop (HIL) established by Opal-RT is used as architecture, verify the algorithm proposed in this article and its differences compared with others.
