In recent years, numerous international scholars are paying a good deal attention to what is going on in plasma ignition and assisted combustion area. The traditional spark discharge reactor is designed based on repetitive experiment. But the heating effect is unable to cope with bad environment gradually, and the bad environment restricts the safe working boundary of the combustion chamber along with the increase of height, the decrease of pressure, and the increase of flow velocity in the combustion chamber. In order to maintain the spark discharge, the optimization of power supply is needed. The plasma discharge reaction is under more rationally and efficiently control to generate the required active particles and energy. The effect of discharge frequency and electron density growth method on spark discharge is analyzed by using two-dimensional model, zero-dimensional model and deep learning model. The results show that the growth rate of voltage amplitude becomes a negative linearity related to the growth rate of initial electron density by building neural network model. The slow-fast and segmented growth in electron density is more energy-saving in air discharge system, and the number of O atoms and O (1D) particles produced by the segmented growth of electron density is higher and more favorable for ignition. In consideration of the energy consumption and the particles of assisted combustion, the voltage waveform corresponding to the segmented growth of electron density is the best one in choices for spark discharge.