摘要
为了克服手机信令数据定位信息的时空不确定性对出行识别的影响,分析手机信令数据的时空特性,在时空阈值识别停留点方法的基础上,引入兴趣面(AOI)信息和基站位置数据,提出可变参数滑动窗口的出行停留点识别方法.建立出行链模型,采用贝叶斯多目标优化获得模型的最佳参数,实现时空阈值的动态调整,提高识别的准确性.为了验证方法的有效性,组织志愿者采集真实出行GPS数据和出行信息标签作为验证数据,与应用模型于对应手机信令数据后的结果进行对比.研究结果表明,手机信令数据的采样在移动和静止状态下存在特性差异;相较于对比方法,提出的出行链识别方法在泛化性能和最优性能方面都表现出较小的误差和较高的识别率,尤其在识别率指标上,相比其他的最新算法改进了3％～26％.
Abstract
The spatio-temporal characteristic of mobile signaling data was analyzed to mitigate the impact of spatio-temporal uncertainty in the location information of mobile signaling data on trip identification. Area of interest (AOI) and base station locations were incorporated based on the spatio-temporal threshold-based method for identifying stay points. A method for identifying stay points using a variable-parameter sliding window was proposed. A trip chain model was established,and Bayesian multi-objective optimization was employed to determine the best parameters. The dynamic adjustment of spatio-temporal thresholds was realized to enhance recognition accuracy. Volunteers were organized to collect real travel GPS data and travel information labels serving as validation data and compared with the results after applying the model to the corresponding mobile phone signaling data in order to validate the effectiveness of above-mentioned method. The research results indicate that there are characteristic differences in the sampling of mobile signaling data between mobile and stationary states. The proposed method show reduced errors and improved recognition rates in terms of both generalization and optimal performance compared with the benchmark methods. There is an improvement ranging from 3％ to 26％ especially in recognition rate compared to other state-of-the-art algorithms.
万方数据