本論文利用OPAL-RT即時模擬系統與硬體迴圈(Hardware In the Loop, HIL)功能建置一以七美島電力系統為原型之微電網。本論文的微電網系統控制方法採用主、從控制法則，本論文由一儲能系統、太陽光發電系統、風力發電系統與三條負載饋線所組成。其中以儲能系統當作微電網控制主機(Master)，而太陽光發電系統及風力發電系統則定位為從屬(Slave)部分。此外，為了改善在併網時的實虛功率控制、併網轉孤島模式時的模式切換以及孤島狀態下時因日照或負載變化所造成之暫態響應，本論文除了以模糊類神經網路(Fuzzy Neural Network, FNN)取代傳統比例積分控制器外，也提出一線上訓練的非對稱歸屬函數之小波派翠模糊類神經網路(Wavelet Petri Fuzzy Neural Network – Asymmetric Membership Function, WPFNN-AMF)將其取代並將三者的效益做對比。本文將詳細介紹WPFNN-AMF的網路架構與線上學習法則。最後，以HIL實驗結果驗證使用FNN、WPFNN-AMF等智慧型控制器在不同操作模式下之有效性與可行性。;This paper uses the OPAL-RT real-time simulation system and hardware in the loop (HIL) function to build a microgrid based on the Cimei island power system. The microgrid using master-slave control is composed of a storage system, a photovoltaic (PV) system, a wind turbine system and three load feeders. Among them, the energy storage system is regarded as a master unit, and the photovoltaic (PV) system and the wind turbine generator system are positioned as slave units. Moreover, in order to improve the control of active and the reactive power in grid-connected mode, the transient response of the switching during the grid-connected mode to islanding mode, and the transient response which caused by irradiance or load changes in the island mode, two online trained intelligent controllers are proposed to replace the conventional proportional-integral (PI) controller in the storage system, one is fuzzy neural network (FNN), and the other is wavelet petri fuzzy neural network with an asymmetric membership function (WPFNN-AMF). This paper will introduce the WPFNN-AMF network architecture and online learning rules in detail. Finally, the HIL experiment results are used to verify the effectiveness and feasibility of using FNN, WPFNN-AMF intelligent controllers in different operating modes.
