停车场泊位占有率预测方法评价

采用上海市五角场地区的停车泊位检测数据,分析了商业、办公和体育场3种不同类型停车场泊位占有率(parking occupancy rate,POR)的时变特征,并评价了ARIMA(autoregressive integrated moving average)、卡尔曼滤波和BP(back propagation)神经网络等3种常用方法在POR预测中的适用性.结果表明,ARIMA和BP神经网络的预测精度总体优于卡尔曼滤波,BP神经网络在商业和办公停车场的短时预测中有较好的精度;3种方法的预测精度均随预测时间步长的增加而逐渐降低;不同类型停车场的POR预测精度存在较大差异,工作日的预测精度一般高于非工作日,且模型具有较好的自适应性.

Evaluation of Prediction Methods for Parking Occupancy Rate

This paper analyzed temporary characteristics of parking occupancy rate (POR) at three different types of parking lots, i.e., shopping mall, office building and stadium, and evaluated the applicability of autoregressive integrated moving average method(ARIMA), Kalman filter and BP neural networks on the prediction of POR, based on parking lot detection data at Wujiaochang District, Shanghai. The results show that ARIMA and BP neural networks can achieve higher prediction accuracies as compared with the Kalman filter method, and the BP neural networks performs best for the short term prediction of shopping mall and office building. The prediction accuracy of the three methods decreases as the forecasting time step increases. Different prediction accuracies exist for different types of parking lots, and the prediction accuracy for weekdays is higher than that for weekends. And the model has good adaptability. This paper can provide reference for the selection of prediction methods for different types of parking lots.