摘要
为了进一步提高短时交通参数多步预测的效果,以自适应指数平滑法、BP神经网络法和小波分析理论作为基础模型,利用前一时刻预测误差确定基础模型在组合模型中所占权重,提出了一种交通参数一步预测组合模型;通过分析交通参数合成和分解机理,在分别提出多时间尺度交通参数合成方法和交通参数分解方法的基础上,设计了一种基于多时间尺度一步外推的短时交通参数多步预测方法,采用某大城市感应线圈1 min时间尺度的交通参数数据进行了验证和对比分析。验证结果表明,交通参数一步预测组合模型的预测效果明显优于任一基础模型,且该方法的多步预测效果明显优于循环一步外推短时交通参数多步预测方法。
In order to improve the multi-step prediction effect of short-term traffic parameters, based on the adaptive exponential smoothing method, the BP neural network method and the wavelet analysis theory, a traffic parameter one-step forecast combination model is proposed by using the weight of the basic model in the combination model determined by the last time prediction error. By analyzing the composition and decomposition mechanism of traffic parameters, and based on the proposed multi-time scale traffic parameter composition and traffic parameter decomposition method, a multi-step prediction method of short-term traffic parameters is designed based on one-step extrapolation of multi time scale. The method is verified and compared by 1 min scale traffic parameter data collected by loop detectors in a city. The results show that the forecast effect of traffic parameter one-step combination model is better than that of any basic model, and the forecast effect of the proposed multi-step prediction method is better than that of the cycle step extrapolation based short-term multi-step prediction method.
出处
《公路交通科技》
CAS
CSCD
北大核心
2013年第2期90-98,共9页
Journal of Highway and Transportation Research and Development
基金
国家自然科学基金项目(51278257)
高等学校博士学科点专项科研基金项目(20110061110034)
浙江省自然科学基金项目(LY12F01013)
关键词
交通工程
多步预测
多时间尺度
短时交通参数
合成与分解
traffic engineering
multi-step prediction
multiple time scale
short-term traffic parameter
composition and decomposition
