多维时空因果关系学习的停车泊位占用率预测技术

Prediction Technology for Parking Occupancy Rate Based on Multidimensional Spatial-Temporal Causality Learning

针对停车泊位占用率预测方法中对空间因素和多维度影响因素考虑不足的问题,扩展了格兰杰因果关系模型,提出一种综合考虑时空相关性和多维度影响因素的停车泊位分析模型,并在此基础上设计了基于神经网络的停车泊位占用率预测算法.采用欧盟FP7项目提供的City Pulse数据集进行仿真实验,实验结果表明:基于多维时空因果关系的神经网络学习预测方法较其他基于时空相关性的停车泊位预测方法的预测精度都有提高;在容量为56辆车的停车场样本中,对0. 5 h后和1. 0 h后停车泊位占用率进行预测的平均绝对误差低至2. 488、3. 418,绝对误差小于20%、10%的预测结果占比更大.

To solve the problem of less attention to spatial factor and multidimensional influence factor in parking occupancy rate prediction,the Granger causality model was expanded and a parking analysis model that involves the spatial-temporal correlation and multi-dimensional influencing factors was proposed.Based on the model,a prediction algorithm for parking occupancy rate was designed by using neural network.The simulation experiments were conducted on the CityPulse dataset provided by the European Union FP7 project.Results show that neural network learning prediction methods based on multidimensional spatial-temporal causality have improved the prediction accuracy compared to other prediction methods based on spatial-temporal correlation.In a sample of parking lot with a capacity of 56 cars,the mean absolute error of parking occupancy rate prediction after 30 minutes and after 1 hour are 2.488 and 3.418,respectively,and prediction results with absolute error less than 20%and 10%account for a larger proportion of all predictions.