城市出租车需求预测在降低出租车空车行驶率、缓解道路交通拥堵方面发挥着重要作用。然而,由于城市路网结构的复杂性,出租车流量的准确预测一直是一项挑战。为了更好地捕捉出租车数据的空间特征,准确预测未来的需求变化,我们提出了一种...城市出租车需求预测在降低出租车空车行驶率、缓解道路交通拥堵方面发挥着重要作用。然而,由于城市路网结构的复杂性,出租车流量的准确预测一直是一项挑战。为了更好地捕捉出租车数据的空间特征,准确预测未来的需求变化,我们提出了一种新颖的时空预测模型。该模型融合了Tucker分解和深度学习的优势,不仅能够捕获出租车需求数据之间的时空相关性,还考虑到了外部因素的潜在影响。最终,通过对五个真实世界的数据集进行出租车需求预测实验,我们验证了本文提出的模型在预测性能方面的有效性。Urban taxi demand forecasting plays an important role in reducing empty cab trips and easing road traffic congestion. However, accurate prediction of cab flows has been a challenge due to the complexity of urban road network structures. To better capture the spatial characteristics of cab data and accurately predict future demand changes, we propose a novel spatial-temporal prediction model. The model incorporates the strengths of Tucker decomposition and deep learning to not only capture the spatial-temporal correlation between cab demand data, but also take into account the potential impact of external factors. Ultimately, by conducting cab demand prediction experiments on five real-world datasets, we validate the effectiveness of the model proposed in this paper in terms of prediction performance.展开更多
文摘城市出租车需求预测在降低出租车空车行驶率、缓解道路交通拥堵方面发挥着重要作用。然而,由于城市路网结构的复杂性,出租车流量的准确预测一直是一项挑战。为了更好地捕捉出租车数据的空间特征,准确预测未来的需求变化,我们提出了一种新颖的时空预测模型。该模型融合了Tucker分解和深度学习的优势,不仅能够捕获出租车需求数据之间的时空相关性,还考虑到了外部因素的潜在影响。最终,通过对五个真实世界的数据集进行出租车需求预测实验,我们验证了本文提出的模型在预测性能方面的有效性。Urban taxi demand forecasting plays an important role in reducing empty cab trips and easing road traffic congestion. However, accurate prediction of cab flows has been a challenge due to the complexity of urban road network structures. To better capture the spatial characteristics of cab data and accurately predict future demand changes, we propose a novel spatial-temporal prediction model. The model incorporates the strengths of Tucker decomposition and deep learning to not only capture the spatial-temporal correlation between cab demand data, but also take into account the potential impact of external factors. Ultimately, by conducting cab demand prediction experiments on five real-world datasets, we validate the effectiveness of the model proposed in this paper in terms of prediction performance.