In recent years, the automotive industry has informed the development situation. Under the problems related to automobile exhaust emissions and serious air pollution and energy shortages, new electric vehicles have sufficient advantages due to their low emissions, high energy efficiency, and low noise. It has been recognized by the people and governments of all countries. The research purpose of this paper is to meet the electricity demand of users, adopt the boundary model of charging and discharging energy and fully adopt deep learning to optimize the real-time charging optimization scheduling strategy of electric vehicles. In order to meet the electricity demand of users, the charge-discharge energy boundary model is used to characterize the charge-discharge behavior of electric vehicles. After the day-ahead training and parameter saving of the proposed model, according to the real-time state of system operation at each moment of the day, the charge-discharge scheduling strategy at that moment is generated. It is verified that the proposed charging scheduling method based on deep reinforcement learning can effectively reduce the power fluctuations in the microgrid and reduce the daily charge and discharge costs on the premise of meeting the charging needs of users; during the development of electric vehicles, different electronic components, especially the power consumption of electric motors, must be faced. A deep learning algorithm based on an improved recurrent neural network (MRNN) is proposed. The system is modeled according to different data and parameters inside the vehicle, and the network is modeled by the MRNN deep learning algorithm. Carry on training, predict the power demand and provide the best power, so as to expand the mileage, better optimize the power distribution of the motor, and compare the improved models. Experimental research results show that the efficiency of the related scheduling strategy model is increased by about 37.2% compared with the traditional model. The proposed method is fast in calculation and does not require iterative calculation, which fully meets the needs of real-time scheduling.