Electrolytic capacitor-less converters have the advantages of low cost, long life, high power density and high power factor. But the bus voltage is bound to fluctuate significantly with the greatly reduced electrolytic capacitor, causing damage to capacitor, switches, and motor. It’s necessary to suppress the fluctuation of bus voltage for a better performance. This paper presents a neural network-based bus voltage fluctuation suppression of electrolytic capacitor-less PMSM control system. The neural network is trained with the current speed status and margin of voltage to generate a dynamic reference speed that can reduce the power fluctuation on bus capacitor. It is verified that the bus voltage can be effectively suppressed compared to the non-optimized electrolytic capacitor-less model with only 4% of large electrolytic capacitor.