摘要
智能交通是现代交通发展的前沿领域,交通事件预测是其中的一个研究热点。传统BP神经网络模型是交通事件分析中常用的模型分析方法之一,但易陷入局部极值,不适合处理长期且连续的交通事件数据。为解决上述问题,提出使用循环神经网络处理交通事件数据,利用循环神经网络模型的有限时间长度记忆优势,构建序列数据分类模型来训练数据,采用随机失活方法去除模型的过拟合问题,通过达拉斯地区的实际流量数据验证,将预测结果与传统BP神经网络模型方法作对比。实验对比结果表明,该综合算法在预测精度和损失值方面均有较明显提升,验证了方法的有效性。
Intelligent transportation is the frontier field of modern traffic development, and traffic event prediction is one of the research focal points. The traditional BP neural network model is one of the commonly used model analysis methods in traffic event analysis, but it is easy to fall into local extreme value and is not suitable for dealing with long-term and continuous traffic event data. In this paper, a cyclic neural network was proposed to deal with traffic event data, and a sequential data classification model was constructed to train the data by taking advantage of the finite time length memory advantage of the cyclic neural network model. The random inactivation method was used to remove the over-fitting problem of the model, and the prediction results were verified with the traditional BP neural network through the verification of the actual flow data in the Dallas area. The model method was compared. The results of the experiment show that the comprehensive algorithm is improved in the prediction precision and the loss value, and the effectiveness of the method is verified.
作者
刘伟
张晓蕾
孙士保
赵鹏程
LIU Wei;ZHANG Xiao-lei;SUN Shi-bao;ZHAO Peng-cheng(College of Information Engineering,Henan University of Science and Technology,Luoyang Henan 471023,China;Department of Software Engineering,Beijing Information Technology College,Beijing 100018,China)
出处
《计算机仿真》
北大核心
2021年第6期78-82,87,共6页
Computer Simulation
基金
国家自然科学基金项目(51474095)
河南省重点攻关项目(152102210277)
赛尔网络下一代互联网技术创新项目(NGII20180313)(NGII20160517)
河南省高校科技创新团队支持计划项目(17IRTSTHN010)
河南科技大学科技创新团队项目(2015XTD011)
河南科技大学重大产学研合作培育基金项目(2015ZDCXY03)。
关键词
交通事件预测
循环神经网络
序列数据
随机失活
Traffic event prediction
Recurrent neural network
Sequence data
Random inactivation
