论文部分内容阅读
针对出租车空载率高、司机寻客难的问题,提出了泊松-卡尔曼组合预测模型(Poisson-Kalman Combined Prediction Model,PKCPM)。首先,采用加权非齐次泊松模型,针对出租车历史数据进行建模,得到目标时刻的估计值。其次,基于当天的实时数据,将临近时刻乘客需求的平均值作为目标时刻预测值。最后,将预测值和估计值作为卡尔曼滤波模型的输入参数,实现对目标时刻出租车乘客需求的预测,同时引入误差反向传播机制,减小下一次预测误差。基于郑州市出租车轨迹数据集,将组合模型与非齐次泊松模型(NHPM)、加权非齐次泊松模型(WNHPM)、支持向量机(SVM)等三种模型进行对比,实验结果显示PKCPM相比于WNHPM、SVM误差分别降低了降低了8.84%、14.9%.该模型对不同时段内、不同空间网格的乘客需求进行预测,为出租车寻找乘客提供了可靠的依据。
Aiming at the problem of high taxi load and driver’s difficulty in finding a taxi, a Poisson-Kalman Combined Prediction Model (PKCPM) is proposed. First, weighted historical non-homogeneous Poisson model is used to model taxi historical data to get the estimated value of the target moment. Secondly, based on the real-time data of the day, the average passenger demand at the immediate moment is taken as the target time prediction value. Finally, using the predicted value and the estimated value as the input parameters of the Kalman filter model, the prediction of the taxi demand of the target time objective is achieved, and the error backpropagation mechanism is introduced to reduce the next forecast error. Based on the taxi track dataset in Zhengzhou, the combined model is compared with three models of non-homogeneous Poisson model (NHPM), weighted non-homogeneous Poisson model (WNHPM), support vector machine (SVM) Compared with WNHPM, PKCPM reduces SVM errors by 8.84% and 14.9%, respectively. This model predicts the passenger demand of different spatial grids in different periods and provides a reliable basis for taxi search for passengers.