Pattern Discovery Based Preventive Control Strategy for Voltage Source Inverter Based Microgrids Against Islanding

Abstract-This article presents a statistical learning-based method for preventive control of microgrids against isolating from the main grid. Based on the stability criteria, the microgrid pre-islanding conditions are divided into secure and insecure classes. Critical variables regarding microgrid dynamic security are first selected via a feature selection procedure known as RELIEF. An unsupervised learning method, the pattern discovery method, is then applied to the space of the critical variables to extract the detailed structure knowledge called patterns. The patterns are in hyperspace and can effectively encode the microgrid security boundaries. The extracted patterns are used as preventive control laws to change the pre-islanding microgrid operating point if it is in the insecure region. Applying the proposed preventive control strategy guarantees that the microgrid can be isolated from the main grid without any instability problem. The studies are conducted in the time domain using the developed program in a MATLAB environment (The MathWorks, Natick, Massachusetts, USA). The results demonstrate the effectiveness and accuracy of the proposed preventive control method of microgrid against islanding in comparison with the decision tree based method.