摘要
为实现城市交通异常管理的主动式响应,给交通异常处置争取更多的时间,减少交通异常对城市路网的影响,提出了一种基于神经网络及关联性修正的交通异常预测方法。基于历史异常数据构建交通异常数据库,并定义了预测模型中的主要参数;构建了基于改进神经网络算法的交通异常预测模型,在此基础上,创新性地提出了结合不同单元区域背景概率及交通异常相关关系挖掘的预测修正算法,对预测结果进行关联性修正以得到最终更加准确的预测结果,大幅提升了模型的预测精度。应用哈尔滨市30 d实例数据训练了所提出的交通异常预测模型,用15 d数据进行了验证,结果表明经过关联性修正的预测模型成功预测次数明显增加,相较于传统方法,预测成功率提升了31.46%,皮尔逊检验值均大于1.642,预测的结果的可信度大于80%的置信水平。
In order to achieve active response and strive for more disposal time of abnormal traffic in urban management,and reduce impacts of abnormal traffic on urban road networks,a model of forecasting abnormal traffic based on neural network and relevant correction.A database of abnormal traffic is established based on historical abnormal data,and defines main parameters of the model.Then,a method based on improved neural network algorithm is developed.This paper innovatively proposes a correction algorithm for forecasting with a combination of background probability of different unit regions and the mining results of abnormal traffic correlation.Accuracy of forecast is greatly improved after relevant correction.A case study with 30 day training data and 15 day test data collected in Harbin is applied to the model.The results show that compared with traditional methods,success rate of forecast improves by 31.46%when using the new method.The values of Pearson test are all above than 1.642,and confidence level of reliability is more than 80%.
作者
安实
王雷
周超
AN Shi;WANG Lei;ZHOU Chao(School of Transportation Science and Engineering,Harbin Institute of Technology,Harbin 150090,China;Logistics Group of Harbin Institute of Technology,Harbin 150001,China)
出处
《交通信息与安全》
CSCD
北大核心
2019年第2期10-17,共8页
Journal of Transport Information and Safety
基金
国家自然科学基金项目(51578199)资助
