Abstract

Greenhouse gas (GHG) emission and excessive fuel consumption have become a pressing issue nowadays. Particularly, CO₂ emissions from transportation account for approximately one-quarter of global emissions since 2016. Electric vehicle (EV) is considered an appealing option to address the aforementioned concerns. However, with the growing EV market, issues such as insufficient charging infrastructure to support such ever-increasing demand emerge as well. Effectively forecasting the commercial EV charging demand ensures the reliability and robustness of grid utility in the short term and helps with investment planning and resource allocation for charging infrastructures in the long run. To this end, this article presents a time-series forecasting of the monthly commercial EV charging demand using a deep learning approach-Sequence to Sequence (Seq2Seq). The proposed model is validated by real-world datasets from the State of Utah and the City of Los Angeles. Two prediction targets, namely one-step ahead prediction and multi-step ahead prediction, are tested. Further, the model is benchmarked and compared against other time series and machine learning models. Experiments show that both Seq2seq and long short-term memory (LSTM) generate satisfactory prediction performance for one-step prediction. However, when performing the multi-step prediction, Seq2Seq significantly outperforms other models in terms of various performance metrics, indicating the model’s strong capability for sequential data predictions.

摘要

温室气体（GHG）排放和过度燃料消耗已成为当今紧迫的问题。特别是CO ₂自 2016 年以来，交通排放量约占全球排放量的四分之一。电动汽车 (EV) 被认为是解决上述问题的一种有吸引力的选择。然而，随着电动汽车市场的增长，充电基础设施不足以支持这种不断增长的需求等问题也随之出现。有效预测商用电动汽车充电需求可确保电网公用事业在短期内的可靠性和稳健性，从长远来看有助于充电基础设施的投资规划和资源分配。为此，本文使用深度学习方法——序列到序列（Seq2Seq），对每月商业电动汽车充电需求进行时间序列预测。所提出的模型已通过来自犹他州和洛杉矶市的真实数据集进行验证。测试了两个预测目标，即一步超前预测和多步超前预测。此外，该模型进行了基准测试，并与其他时间序列和机器学习模型进行了比较。实验表明，Seq2seq 和长短期记忆（LSTM）都为一步预测产生了令人满意的预测性能。然而，在执行多步预测时，Seq2Seq 在各种性能指标方面明显优于其他模型，表明该模型具有强大的序列数据预测能力。实验表明，Seq2seq 和长短期记忆（LSTM）都为一步预测产生了令人满意的预测性能。然而，在进行多步预测时，Seq2Seq 在各种性能指标方面明显优于其他模型，表明该模型具有强大的序列数据预测能力。实验表明，Seq2seq 和长短期记忆（LSTM）都为一步预测产生了令人满意的预测性能。然而，在进行多步预测时，Seq2Seq 在各种性能指标方面明显优于其他模型，表明该模型具有强大的序列数据预测能力。