This study proposes two neural network-based approaches for computing the degrees required to generate pulse-width modulation (PWM) signals. The conventional method of using a lookup table to determine the orientations is replaced by these ANN models, which demonstrate superior performance across a range of modification indices. The first ANN is trained to estimate the phase angle based on input from the Empire Bird Optimizer, eliminating the need for a lookup table. A separate ANN is then trained using these configurations to generate the PWM signals. Experimental data on voltage level, inverter output current, and power flow are collected using a scope and power quality tester. The recommended hybrid ANN-based selective harmonic elimination PWM effectively removes low-order vibrations. The research is implemented in the MATLAB/Simulink platform, and the results indicate that the proposed system is efficient, accurately determining firing angles with minimal repetitions, and effectively addressing local minima values. The practical implementation of an eleven-level modular multilevel converter serves as the basis for this proposal, exemplified by an eleven-level H-bridge inverter.