基于GPS定位数据的出行端点推断

Inference method of trip ends based on GPS track data

为了从大规模GPS定位数据中提取出出行行为建模所必需的出行信息,针对临界逗留时间、轨迹点聚集的临界距离和临界重复路段长度各设定了5个离散值,并使用全部125个参数组合推断了出行端点。以查全率和错误率作为预测精度评价指标,以基于智能手机的出行调查采集的出行信息为依据,比较了推断得到的出行端点与真实的出行端点。在最优参数组合的条件下,推断方法的查全率达到了96.02%,且错误率仅为4.74%。研究结果表明:基于智能手机的出行调查能够对传统出行调查起到有效的补充作用,且可为进一步开展大规模GPS出行调查提供理论和方法指导。

To extract travel characteristics essential for modeling travel behavior from large-scale GPS track data streams,each of the critical dwell time,the critical distance for GPS point clustering and the critical overlapping length of roads is tested with 5discrete values.The resulting 125 parameter combinations are applied to infer the trip ends.The derived trip ends are compared with the actual ones,which are collected in a smartphone-based survey,by taking the identification rate and the error rate as the measures.Under the condition of the optimal parameter combination,the identification rate is up to 96.2%,while the error rate remains only 4.74%.Results of the current study indicate that travel survey based on smartphones can not only make up for conventional travel surveys effectively to a certain extent,but also provide theoretical and practical guidance for further largescale GPS travel surveys.