Microgrid Frequency Regulation Based on Precise Matching Between Power Commands and Load Consumption Using Shallow Neural Networks

Islanded microgrids commonly use droop control methods for autonomous power distribution; however, this approach causes system frequency deviation when common loads change. This deviation can be eliminated using secondary control methods, but the core of this approach is to generate compensation values equal to the offset amount to add to the controller, thereby eliminating deviations from rated values. Such a mechanism can actually achieve the same effect by setting power reference values within the droop control method. The power references within the controller need to be adjusted dynamically, and they are associated with common load variations. Therefore, establishing a fitting relationship between the adjustment of power reference and changes in common loads can achieve better frequency regulation, keeping the system frequency operating within rated frequency ranges. These two types of data are correlated, however, due to physical parameters, the fitting between them is not strictly fixed in a mathematical sense. Thus, to find their interconnected relationships, using intelligent methods becomes crucial. This paper proposes a shallow neural network-based method to achieve fitting relationships. Moreover, to address power inputs with zero values, an input enhancement method is proposed to prevent potential gradient vanishing and ineffective learning problems. Thus, through precise matching between power commands and load consumption, the system frequency can be maintained near rated values. Various simulation scenarios demonstrate the feasibility and effectiveness of the proposed method.