Neural Network Based Microgrid Voltage Control

Abstract

The primary purpose of this study is to improve the voltage profile of Microgrid using the neural network algorithm. Neural networks have been successfully used for character recognition, image compression, and stock market prediction, but there is no directly application related to controlling distributed generations of Microgrid. For this reason the author decided to investigate further applications, with the aim of controlling diesel generator outputs. Firstly, this thesis examines the neural network algorithm that can be utilized for alleviating voltage issues of Microgrid and presents the results. MATLAT and PSCAD are used for training neural network and simulating the Microgrid model respectively. The Feedforward Back-propagation algorithm is used in this study and the Microgrid consists of wind, solar, and diesel power generations, and battery storage. Neural network will indicate how much real and reactive power is needed from each generator. In the second stage, several scenarios are proposed to verify that the monitoring points are very important for training neural networks. Finally, the comparison of results is shown for further discussion of critical points. In conclusion, the results of the study show that neural network algorithm is well suited for the application, and it was effectively certified for the purpose of improving the voltage profile of Microgrid.