Data-driven modeling of droop controlled parallel inverters with generalization performance
Yuzong Wang,Jiangbin Tian,Guohui Zeng,Jinbin Zhao
DOI: https://doi.org/10.1007/s43236-024-00949-4
IF: 0.913
2024-11-17
Journal of Power Electronics
Abstract:In a microgrid inverter parallel operation system, droop control requires less communication between inverters. It has the ability of system self-regulation to maintain voltage and frequency stability. When the system load suddenly becomes large, using the traditional droop control method causes a huge drop in the system output frequency. In this paper, with the assistance of a long short-term memory neural network (LSTM), a data-driven model of a three-phase inverter shunt system is established based on droop control. Experimental data under different operating conditions are used as the training set in the training process to improve the generalization performance of the data-driven model. The improved data-driven model can accurately predict the output power of a system following the moment on the basis of the current electrical quantity of the system. It can accomplish this even under brand new operating conditions. By compensating the predicted quadratic function term of active power in the active power sag control equation, when the load power suddenly increases, the system compensates the output frequency in advance, which reduces the frequency deviation during the sag control process, and keeps the output frequency stable. This in turn, improves the power quality. Experimental results show that the data-driven model established in this paper has good generalization performance and improves droop control.
engineering, electrical & electronic