城市道路短时交通流量预测
|
摘要
结合BP神经网络模型和自回归求和滑动平均(ARIMA)模型对城市道路交通短时区间流量进行预测.影响交通流的因素有很多,难以一一量化,但这些因素都可以由线性自相关结构和非线性结构结合线性组合得到.而BP神经网络对非线性关系有很好的拟合效果,ARIMA模型则具有良好的线性拟合能力.在训练模型时,先用ARIMA模型拟合训练集,与原始数据作差得到一组残差;用BP神经网络模型拟合残差;将两个模型结合得到组合模型.将2017年7月1日7:00到2017年7月1日18:00期间,贵阳市某个路口断面所采集的过车数据作为训练集,建立ARIMA模型和BP神经网络模型以及组合模型,预测2017年7月1日18:00到2017年7月1日19:00的短时交通流.过车数据统计时间间隔为5min,则训练集共有有效数据132组,测试集的有效数据为12组.分别用三类误差分析指标比较三个模型的拟合、预测效果,结果显示组合模型的预测效果比两个模型单独使用的预测效果更准确.
This paper combines the BP neural network model and the Autoregressive Integrated Moving Average(ARIMA) Model to predict the short-term traffic volume of urban roads. There are many factors that affect the traffic volume which is very difficult to quantize.But these factors can be expressed as the linear combination of linear and nonlinear structures. As we know, the BP neural network is good at fitting the nonlinear relationship and the ARIMA model is good at fitting the linear fitting relationship. During training model,first, we use ARIMA model to fit the training set which gets a set of residuals. Second the BP network was used to fitting the residuals. In the last we combine the two results and got a hybrid model. The data of traffic volume, from July 1, 2017 7:00 to July 1, 2017 18:00, was from an urban road in GuiYang as a training set. And we established the ARIMA model, BP neural network and the hybrid model to predict the short-term traffic volume during July 1,2017 18:00 and July 1, 2017 19:00. We count the number of cars every 5 minutes, which got 132 groups data of training set and 12 groups data of testing set. In the end, it analyzes the fitting and the prediction efficiency by using three kinds of index about error analysis. The result show that the hybrid model is more accurate than the two models alone in prediction.
This paper combines the BP neural network model and the Autoregressive Integrated Moving Average(ARIMA) Model to predict the short-term traffic volume of urban roads. There are many factors that affect the traffic volume which is very difficult to quantize.But these factors can be expressed as the linear combination of linear and nonlinear structures. As we know, the BP neural network is good at fitting the nonlinear relationship and the ARIMA model is good at fitting the linear fitting relationship. During training model,first, we use ARIMA model to fit the training set which gets a set of residuals. Second the BP network was used to fitting the residuals. In the last we combine the two results and got a hybrid model. The data of traffic volume, from July 1, 2017 7:00 to July 1, 2017 18:00, was from an urban road in GuiYang as a training set. And we established the ARIMA model, BP neural network and the hybrid model to predict the short-term traffic volume during July 1,2017 18:00 and July 1, 2017 19:00. We count the number of cars every 5 minutes, which got 132 groups data of training set and 12 groups data of testing set. In the end, it analyzes the fitting and the prediction efficiency by using three kinds of index about error analysis. The result show that the hybrid model is more accurate than the two models alone in prediction.
引文
[1]谭满春,冯荦斌,徐建闽.基于ARIMA与人工神经网络组合模型的交通流预测[J].中国公路学报,2007,20(4):118-121.
[2]杨显立,许伦辉等,基于小波神经网络的道路交通流量实时预测模型研究[J].公路交通技术,2013, 10(5):111-114.
[3]Box G E P. Time series analysis forecasting and control[M]. San Francisco:Holden-day, 1976.
[4]王燕.应用时间序列分析[M].北京:中国人民大学出版社,2012:135-183.
[5]吴岸城.神经网络与深度学习[M].北京:电子工业出版社,2016:66-86.
[2]杨显立,许伦辉等,基于小波神经网络的道路交通流量实时预测模型研究[J].公路交通技术,2013, 10(5):111-114.
[3]Box G E P. Time series analysis forecasting and control[M]. San Francisco:Holden-day, 1976.
[4]王燕.应用时间序列分析[M].北京:中国人民大学出版社,2012:135-183.
[5]吴岸城.神经网络与深度学习[M].北京:电子工业出版社,2016:66-86.