In recent years, the concept of microgrid has gained attention as part of the energy infrastructure of smart grids. To improve reliability, stability margins, and flexibility, microgrids can interconnect to each other and form a multiple microgrid system. The integration of distributed energy resources and the formation of a multiple microgrid system create various technical challenges and issues, including uncertainty and variable power from wind and solar-based distributed generation, reduction of system inertia, voltage and frequency stability control in islanded operation, and the need for coordination among microgrids. Studying the effects of these problems and providing suitable analytical solutions requires access to appropriate and detailed models that accurately represent the dynamic behavior of the microgrid and its components. In this thesis, a detailed model of distributed generation resources such as DFIG wind turbines and battery storage systems, along with load models and network lines, is presented, and by integrating these models, a multi-microgrid model is obtained. With the expansion of microgrid dimensions, the use of detailed and high-order models leads to an increase in computational burden and the time required for simulations, making it difficult to eva‎luate the dynamic characteristics of these systems. When studying a specific microgrid, the external microgrids and its elements can be replaced with simpler models. For this purpose, this thesis presents a reduced-order model for external microgrids titled "Black Box External Grid". In this method, external microgrids are replaced by a black box model that has been trained using a nonlinear exogenous self-organizing neural network method (NARX). For each microgrid, two neural networks related to the injected active and reactive power, named NN-P and NN-Q, are trained using data obtained from the detailed model simulation, and their estimated output values are injected as a controllable current source at the boundary bus of the microgrid. The effectiveness of the proposed reduced-order model is eva‎luated by comparing it with the results obtained from the detailed model.

محمداسماعيل همداني گلشن

حميدرضا كارشناس , محمد سعيد مهدوي