<p>One of the most critical challenges in operating microgrids is the control and power flow. Considering the momentary changes of the system conditions, especially the load, and the time-consuming nature of solving the power flow to obtain the optimal share of generators. In this paper, a two-layer scheme to resolve this issue is presented. A multi-bus microgrid with two upper and lower levels is considered. At the upper level, a droop-based method is used to control the voltage and frequency and send active and reactive power to the lower level. In the second layer of control, which is related to the lower level of the microgrid, an objective function is solved offline to obtain optimal operating points in different load conditions, and it is used to train artificial neural networks (ANN). Then, in real-time operating at different load conditions, the output range of each inverter is obtained by ANN and is defined as a constraint in the objective function. This is done to increase the speed and accuracy of real-time operation. Finally, the system is tested under different load changes. The outcomes displayed that the suggested method has brought favorable results regarding speed and accuracy of system performance. Simulations are performed by MATLAB software.</p>