基于GA-LSSVR模型的路网短时交通流预测研究

发布时间：2018-03-29 01:29

本文选题：智能交通　切入点：变量选择　出处：《交通运输系统工程与信息》2017年01期

【摘要】：目前,很多短时交通流预测方法仅利用某一路段历史数据的时间相关性或者道路上下游路段的时空相关性进行交通流预测,未充分考虑路网所有路段之间的时空相关性.提出了一种基于稀疏混合遗传算法优化的最小二乘支持向量回归(LSSVR)模型,并应用于路网短时交通流预测.该预测模型不仅可以自动优化LSSVR模型参数,而且可以从高维路网交通流数据中选择有助于交通流预测的变量子集.实验结果表明,与LSSVR模型相比,所提方法具有更好的预测能力;而且,少量时空变量被选择出来构建预测模型,极大减少了信息冗余,改进了模型可解释性.
[Abstract]:At present, many short-term traffic flow forecasting methods only use the temporal correlation of historical data of a section or the temporal and spatial correlation of upstream and downstream sections of the road to carry out traffic flow prediction. A least squares support vector regression (LSSVR) model based on sparse hybrid genetic algorithm is proposed, which does not fully consider the temporal and spatial correlation among all sections of the road network. The model can not only optimize the parameters of LSSVR model automatically, but also select a subset of variables from the traffic flow data of high-dimensional road network. The experimental results show that the proposed model can be used to predict traffic flow in short time. Compared with the LSSVR model, the proposed method has better prediction capability, and a small number of spatio-temporal variables are selected to construct the prediction model, which greatly reduces the information redundancy and improves the interpretability of the model.
【作者单位】：江苏大学汽车工程研究院;江苏大学汽车与交通工程学院;
【基金】：国家自然科学基金(51108209,61203244,61403172,61573171) 交通运输部信息化项目(2013-364-836-900) 中国博士后科学基金(2015T80511,2014M561592) 江苏大学高级人才科研启动基金(14JDG066) 福建省信息处理与智能控制重点实验室(闽江学院)(MJUKF201724)~~
【分类号】：U491.14

【相似文献】