基于Attention-ConvLSTM的出租车需求预测研究

Taxi Demand Prediction Based on AttentionConvLSTM

准确合理的出租车需求预测能平衡乘客出行需要且缓解城市交通拥堵,但出租车需求是动态的,它会随着时间的周期性和空间的相关性变化而变化。使用神经网络的出租车需求预测模型广受关注,但目前大部分模型对需求数据的时空间特征提取有效性不足,缺乏筛选数据特征的能力,使得预测结果不够贴合现实情况。对此,本文提出AttConvLSTM预测模型。首先通过k-means聚类和Granger因果关系检验寻找与检验时空相关性,再结合影响需求的气候和节假日等外部因素,运用卷积长短期记忆网络(ConvLSTM)融合注意力机制捕获及评估时空间特征,对特征进行选择性注意,从而提高预测结果的精度与可靠性。最终选取纽约市出租车需求数据进行实验,结果表明该模型相比多种知名基线模型表现出更高的准确度和稳定性。

Accurate and reasonable taxi demand forecasting can balance passenger travel needs and reduce urban traffic congestion,but taxi demand is dynamic and changes with time cycles and spatial correlations.Taxi demand forecasting models using neural networks are widely popular,but most of the current models are not effective enough in extracting the spatio-temporal features of the demand data and lack the ability to filter the data features,making the forecasting results not close to the real situation.In response,this paper proposes the Att-ConvLSTM prediction model.First,find and test spatio-temporal correlation through kmeans clustering and Granger causality test,then combine external factors such as climate and holidays that affect demand,use convolutional long and short-term memory network(ConvLSTM)to fuse attention mechanism to capture and evaluate spatiotemporal features,and selectively pay attention to features,so as to improve the accuracy and reliability of prediction results.New York City taxi demand data were selected for the experiment,and the results show that the model exhibits higher accuracy and stability compared to a variety of well-known baseline models.