Achieving accurate and balanced regional electric vehicle charging load forecasting with a dynamic road network: a case study of Lanzhou City

Spatial and temporal predictions of electric vehicle (EV) charging loads provide a basis for further research on synergistic operation of road-vehicle-electricity networks with different attributes, which is important for siting and capacity building of urban road networks and charging stations, as well as for long-term planning and operation of power systems. However, some of the variables in EV charging load prediction are often given as random variables or constants, and to enhance the modeling and prediction accuracy for EV charging loads, data- and model-driven concepts were combined, and a novel model was proposed to achieve accurate balanced regional EV charging load forecasting with dynamic road networks: a case study of Lanzhou city. First, a road-vehicle-grid integration model was constructed, and an EV was used as an intermediate medium to initially integrate the flexible domains of the two layers of the road network and grid for establishing the actual road network topology in the city and for comparing static and dynamic road network models. Second, particle swarm optimization (PSO) was employed to optimize the backpropagation (BP) neural network for predicting future regional EV ownership. In addition, the PSO-BP-Monte Carlo (MC) model was refined for obtaining accurate spatial and temporal predictions of regional short-term charging loads through the introduction of an EV real-time unit mileage power consumption model with the use of M/M/c queue theory for determining charging waiting times. Finally, a simulation was conducted in the urban area of Lanzhou city as an example of actual traffic roads, and the results showed that the proposed model could be applied to reasonably effectively and more accurately predict the regional load conditions. The average growth in the peak load within the region over the next five years will reach 21.07%, and accurately balances the occupancy rate of each charging pile of the charging station, and the dynamic road network will experience a decrease in the peak-valley difference between the loads of the static network of 5.92%. Compared to those obtained with other methods, the single-day load distribution was more balanced, and the road access time better conformed with the actual road access conditions, which verifies the effectiveness and feasibility of the proposed method and lays a foundation for the next step of research on synergistic operation of road-vehicle-grid networks with different attributes.