Probabilistic prediction of electric vehicle load in service area incorporating highway spatiotemporal traffic information

The stochastic high-rate charging behavior observed on the highway poses substantial pressure on both highway service area (HSA) energy system and the upstreaming network. Therefore, the prediction of electric vehicle (EV) charging demand can greatly contribute to the proactive development of scheduling strategy in the HSA energy system, enabling optimized coordination between transportation and energy system. In this context, this paper proposes a probabilistic prediction method of EV load that can fully utilize the spatiotemporal traffic information from highways. In the proposed solution, a highway traffic flow (TF) prediction method based on the Temporal Convolutional Network and Graph Attention Network (TCN-GAT) algorithm is developed, which enables the spatiotemporal features of TF to be effectively captured for the formulation of TF prediction quantiles. On this basis, a novel method for calculating the EV entry rate grounded in historical highway driving data is proposed. A probabilistic queuing theory model considering user behavior uncertainty is further established to determine the quantile prediction of EV load. Moreover, the effectiveness of the proposed solution is widely validated through numerical simulations, and the comparison with benchmarks comprehensively indicates the superiority of the proposed solution.