Rolling Time Domain Charging Allocation of Electric Vehicles Under Time Varying Demand

Abstract

To study the impact of traffic conditions of urban road networks and the distribution of potential demand of charging users on the charging distribution in the region. In this paper, the long short-time memory (LSTM) neural network is used to learn the historical user charging behavior data and predict the distribution of user charging behavior in real-time. An integer programming model for charging allocation of electric vehicles in the rolling time domain under time-varying demand is established by taking multiple charging stations and road networks in the region as research objects. Aiming at the highest comprehensive return, the model rejects some users with a long charging residence time before charging is allocated at the peak time and carries out the rolling time domain allocation of charging. When the demand for charging is high during peak charging hours, early rejection of long-staying users during the real-time allocation of charging appointments can increase the revenue of charging stations and the utilization rate of charging piles. In the case that the charging pile can be saturated at peak times, the allocation based on LSTM neural can increase the revenue of charging stations by 20.45% on average and the utilization rate of charging piles by 2.35% on average throughout the day. During peak hours, the revenue of the charging station can be increased by 36.63% on average, and the utilization rate of the charging pile can be increased by 13.54% on average.

Keywords

• Demand prediction
• deep learning
• dynamic charging allocation
• integer programming