基于ODCG的网约车需求预测模型

An online car-hailing demand forecasting model based on ODCG

为提高网约车需求预测的准确率,提出结合卷积神经网络(convolutional neural network,CNN)和卷积门控循环单元(convolutional gate recurrent unit,ConvGRU)的出发地-目的地需求预测分析(origin-destination demand prediction with CNN and ConvGRU,ODCG)模型。ODCG模型的网络结构分为局部空间特征(local spatial feature,LSF)提取模块、时间演化特征(time evolution feature,TEF)提取模块、全局关联模块(global association module,GCM)和输出层。LSF提取模块利用CNN分别处理出发地视图和目的地视图,得到网约车需求的局部空间依赖性;TEF提取模块将网约车需求的局部空间信息、天气信息和订单序列关联度信息整合到ConvGRU中,分析网约车的需求;GCM模块整合所有区域之间的相关性,通过将所有区域特征加权求和得到全局相关性,并将相应区域之间的相似度定义为权重。试验结果表明,ODCG模型在网约车需求预测中优于其他基线模型,同时提高了网约车需求预测的准确率。

In order to improve the prediction accuracy of online car-hailing demand,origin-destination demand prediction with CNN and ConvGRU(origin-destination demand prediction with CNN and ConvGRU,ODCG)was proposed.The network structure of ODCG model was divided into local spatial feature(LSF)extraction module,time evolution feature(TEF)extraction module,global association module(GCM)and output layer.The LSF extraction module used CNN to process the origin view and destination view respectively to obtain the local spatial dependence of online car-hailing demand.TEF extraction module integrated local spatial information,weather information and order sequence correlation information into ConvGRU(ConvGRU)to analyze the demand of online car-hailing.The GCM module integrated the correlation between all regions,obtained the global correlation by weighted sum of all regional features,and defined the similarity between corresponding regional pairs as weights.The experimental results showed that ODCG model was superior to other baseline models in the prediction of online car-hailing demand,and improved the accuracy of online car-hailing demand prediction.