基于多机器学习竞争策略的短时交通流预测

Prediction of Short-term Traffic Flow Based on Ensemble Learning Mechanism

交通流量预测是当前交通大数据应用的重要议题之一.经典的交通流量预测方法通常只根据被预测道路自身的数据进行分析和决策,而往往较少考虑由于同一区域不同道路之间的交通流量关联性.本研究根据城市核心区交通流量数据的特点,构建同区域内多条相关道路的交通流量多维度数据模型.并且,基于该数据模型提出了一种基于多机器学习竞争策略的交通流量预报算法.该算法的主要核心思想是利用时间序列聚类的方式将多维交通流量数据进行降维,然后通过引入多种多机器学习方法进行并发训练,其训练结果通过竞争获得最优分类器群.最后设计了多最优机器学习集成预测方法进行交通流量预测.本模型通过在南昌市中心道路进行的实验显示,其预测结果优于传统单时间序列机器学习方法.

Short- term traffic flow prediction is an important issue of traffic- oriented big data application.Traditionally, traffic flow prediction is determined by the local recordings, whereas the natural relations of adjacent roads are neglected. Based on the characteristic of downtown traffic flow, a multi-dimensional data model concerning relevant roads is proposed. Meanwhile, a multi- machine learning competing strategy is developed. The essence of proposed prediction strategy is training commonly used machine learning algorithms independently with the clustered temporal information as input. By selecting high prediction performance as dominating weak classifier, the weight is further assigned to corresponding algorithm so that a strong classifier is eventually developed. As a result, the traffic flow prediction problem is converted to a weighted multi- machine learning problem. Experiments on downtown streets of Nanchang show that the superiority of proposed method as compared to traditional single time series based solutions.