基于流形距离的高速公路短时交通流预测模型

准确的高速公路短时交通流预测是实现交通诱导和控制的重要前提和基础。为了提高预测精度,提出一种基于流形距离(MD)的K近邻-长短期记忆(K-nearest neighbor-long short-term memory,KNN-LSTM)高速公路短时交通流预测模型。该模型利用流形相似性分析高速公路交通流的时空特性,计算多站点与目标站点之间的流形距离。然后,采用改进的KNN算法筛选出空间相关站点构造交通流数据集,通过LSTM模型提取时序特征得出预测结果。实验表明,与单一预测模型相比,该方法能更好地提取交通流时空特性且预测精度更高,可为高速公路的交通管理提供必要的依据。

Short-Term Traffic Flow Prediction Model of the Highway Based on Manifold Distance

Accurate prediction of short-term traffic flow on highway is an important prerequisite and basis for achieving traffic guidance and control. In order to improve the prediction accuracy, this paper proposes a K-nearest neighbor-long short-term memory (KNN-LSTM) highway short-term traffic flow prediction model based on manifold distance. The model used the manifold similarity to analyze the spatial-temporal characteristics of highway traffic flow and calculated the manifold distance between multiple stations and target station. Then used the improved KNN algorithm to filter out spatially related stations to construct a traffic flow dataset, and extracted temporal features to forecast by used the LSTM model. The experiments show that compared with a single prediction model, this method can better extract the spatial-temporal characteristics of traffic flow with higher prediction accuracy, and can provide the necessary basis for highway traffic management.