基于负二项模型区域日度事故预测及影响因素分析

A Prediction of Daily Road Accidents in an Area of Beijing and Factors Analysis Based on Negative Binomial Model

分析区域日度交通事故规模的影响因素,是做好区域日度交通事故短时预测与防控的基础。搜集北京某区域2012-2015年道路交通事故、气象和日期性质等数据,采用负二项回归技术,建立了区域交通事故预测模型;以2012-2014年的数据作为训练集,以2015年的数据为测试集,拟合模型伪R2为0.645,预测期内绝对百分误差的中位数为17.04%,模型预测效果较好,达到了精度要求。模型还表明:①节假日期间事故减少,节假日前1 d事故增加,节假日后1 d天事故平稳;②1周内,周一和周日事故规模相对较小;③1年内,2月、3月事故规模稍小,7月、9月、10月、11月、12月事故规模稍高;④尾号限行对事故规模影响大,但针对尾号为4和9的限行几乎没有影响;⑤相较于晴天,多数非晴朗天气情况下事故规模反而下降;⑥日平均气温提高会小幅降低事故规模,但日最高气温和最低气温之差增大会增加事故规模。

An analysis of the factors affecting regional daily traffic accidents is the basis for accurate short-term accident prediction and prevention. A model to predict regional traffic accidents is developed by using negative binomial regression technique. Data is collected on road traffic accidents,weather conditions,and date characteristics of a district in Beijing from 2012 to 2015. The data from 2012 to 2014 is used as a training set,and the data from 2015 as a test set,the pseudo-R2 of the fitted model is 0.645,and the median absolute percentage error during the forecast period is 17.04%. The model has good performances and meets the accuracy requirement. The results show that: ①accidents decrease during holidays,increases on 1 day before holidays,and no significant change on 1 day after holidays;②accidents are relatively fewer on Mondays and Sundays during a week;③ accidents are relatively fewer in February and March,and slightly more in July,September,October,November,and December during a year;④traffic restrictions based on the last number on license plate has large impacts on accidents,but it has little effects when the number is 4 and 9;⑤accidents actually decrease in most non-sunny weather conditions compared to sunny days;⑥an increase in the average daily temperature will slightly reduce accidents,but an increase in the difference between the daily maximum and minimum temperatures will increase accidents.