基于马尔可夫链组合模型的交通流量长时预测

发布时间：2018-12-15 05:34

【摘要】：交通流量长时预测在城市道路规划和交通政策制定中具有十分重要的作用。但交通流量受到气候、经济、出行等多种因素的影响,其变化是一个随机非平稳的过程,表现出随机性、波动性以及非严格的周期性。这导致了交通流量长时预测的难度较大。本文在分析了城市道路交通流量的变化特性之后,引入了马尔可夫链方法,并以马尔可夫链为基础构建了多种组合模型来进行交通流量的长时预测。本文的主要研究工作如下:首先,本文通过对实际的交通流量数据的分析,总结了交通流量的三大特点:随机性、波动性以及非严格的周期性。其中,随机性是引入马尔可夫链进行交通流量预测的前提,波动性则为引入区间预测的思想提供了契机,而非严格的周期性为长时交通流量的预测提供了可能。接着,由于真实的交通流量数据存在丢失数据以及噪声数据,本文则介绍并使用了相应的数据补偿算法以及滤波算法对其进行处理。其次,本文构建了三种基于马尔可夫链的交通流量组合预测模型:均值马尔可夫链,聚类马尔可夫链以及聚类加权马尔可夫链。其中,基于均值马尔可夫链的组合预测模型通过均值法来构建交通流量的状态,依据交通流量数据的状态转移规律实现了对未来24小时交通流量的预测,并预测出一条交通流量变化曲线。但考虑到交通流量具有一定的波动性,本文提出了区间预测的思想,利用聚类算法重构交通流量的状态,并通过聚类马尔可夫链组合模型预测出了一个由一条上曲线和一条下曲线包围而成的区域,该区域很好地反映了交通流量可能的波动范围。而为改进由聚类马尔可夫链预测得到的区域波动范围过大的问题,本文又引入加权的思想来重构交通流量预测模块,并通过聚类加权马尔可夫链组合模型预测出了一个平均区间长度较短的区域。最后,本文利用来自深圳的实测交通信息数据对上述三种组合模型的预测结果进行评价分析,评价结果表明了基于聚类加权马尔可夫链的组合模型的预测效果较好。同时,相较于检索到的其他交通流量长时预测方法,本方法预测精度较高,并能够给出交通流量可能的波动范围。
[Abstract]:Long-time forecasting of traffic flow plays an important role in urban road planning and traffic policy making. However, traffic flow is affected by climate, economy, travel and other factors. The change of traffic flow is a random and non-stationary process, showing randomness, volatility and non-strict periodicity. This leads to the difficulty of long-term traffic flow forecasting. After analyzing the changing characteristics of urban road traffic flow, this paper introduces the Markov chain method, and constructs a variety of combined models to carry out long-term traffic flow prediction based on Markov chain. The main work of this paper is as follows: firstly, through the analysis of the actual traffic flow data, this paper summarizes the three characteristics of traffic flow: randomness, volatility and non-strict periodicity. Among them, randomness is the premise of introducing Markov chain to forecast traffic flow, and volatility provides an opportunity to introduce the idea of interval forecasting, rather than strict periodicity, it is possible to forecast long-term traffic flow. Then, because the real traffic flow data have lost data and noise data, this paper introduces and uses the corresponding data compensation algorithm and filtering algorithm to deal with it. Secondly, three traffic flow combination forecasting models based on Markov chain are constructed: mean Markov chain, clustering Markov chain and clustering weighted Markov chain. Among them, the combined forecasting model based on mean Markov chain constructs the state of traffic flow by means of mean method, and realizes the prediction of traffic flow in the next 24 hours according to the law of state transfer of traffic flow data. A curve of traffic flow is predicted. However, considering the volatility of traffic flow, this paper puts forward the idea of interval prediction and reconstructs the state of traffic flow by clustering algorithm. A region surrounded by an upper curve and a lower curve is predicted by the clustering Markov chain combination model, which can well reflect the possible fluctuation range of traffic flow. In order to improve the problem that the range of regional fluctuation obtained by clustering Markov chain prediction is too large, this paper introduces the weighted idea to reconstruct the traffic flow forecasting module. A region with short average interval length is predicted by clustering weighted Markov chain combination model. Finally, the prediction results of the above three combined models are evaluated and analyzed by using the measured traffic information data from Shenzhen. The evaluation results show that the combined model based on clustering weighted Markov chain has a good prediction effect. At the same time, compared with other long time prediction methods, this method has higher prediction accuracy and can give the possible fluctuation range of traffic flow.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O211.62;U491.1

【参考文献】