何梓亮 李扬 周博俊 徐鑫哲

摘要：在我国交通事故统计中，人为失误导致的交通事故占90%，超速行为的危险性已引起各界重视。尤其大型货物运输车辆，其重量大，制动性能較差，该类车辆超速导致的交通事故通常比较严重。对货物运输平台提供的车辆行驶相关数据（包括 GPS定位数据、时间数据、速度数据）进行处理和分析。通过OSMnx在 Open Street Map下载广东地图数据并处理，运用隐马尔科夫模型对原始 GPS定位数据进行地图匹配，对车辆行驶数据进行 LSTM时间序列建模和学习训练，将训练好的网络用于各个路段车辆超速的预测。结果表明，该方法能够很好地拟合超速率的变化趋势，物流平台可以根据预测结果对司机进行提醒，对减少驾驶员的超速行为和保障安全行驶具有重要意义。

关键词：交通安全;超速;地图匹配;LSTM神经网络;时间序列预测

中图分类号：U279文献标志码：A文章编号：1009-9492（2021）11-0044-04

Calculation and Analysis of Road Over-speed Ratio Based on Freight Vehicle

GPS Big Data

He Ziliang，Li Yang ※，Zhou Bojun，Xu Xinzhe

（School of Mechanical and Electrical Engineering， Guangdong University of Technology， Guangzhou 510006， China）

Abstract： In the statistics of traffic accidents in our country， the traffic accidents caused by human error account for 90%. The danger of speeding behavior has attracted attention from all walks of life. Freight vehicle have high weight and poor braking performance. The traffic accidents caused by such vehicles speeding are usually serious. The vehicle driving data included GPS information， speed and record time which provided by the transport platform were processed and analyzed. The Guangdong street map data was downloaded from OpenStreetMap and OSMnx tool was used to process it. Then， the Hidden Markov Model was used to match the record location data to the map street. After that， LSTM model was used to fit the vehicle driving data and predict the over-speed ratio of the street. The result show that the model can fit the trend of change of over-speed ratio well， which is helpful for the transport platform. It is of great significance to reduce the over-speed behavior of drivers and ensure safe driving.

Key words： traffic safety; over-speed; map matching; LSTM neural network; time series prediction

0 引言

2020年8月6日，最高人民法院发布《交通肇事罪特点和趋势司法大数据专题报告》指出，在交通肇事案件中，货车肇事率占32.04%，远高于货车保有率（10.71%），其超速、超载等问题尤为突出。因此，研究大货车的超速行为，预测各路段的超速率具有重要的理论意义和实际意义。

目前，国内外关于超速行为在时间维度的研究较少。2020年，Yufang LI 等[1]选用路程、限速、平均速度、车道数等特征，对车辆在城市道路、郊区道路和高速公路的速度采用 GA-BP 算法和传统 BP 算法进行建模对比，结果表明 GA-BP 具有更好的准确性和动态跟踪性。……