摘要
AOSVR(Accurate Online Support Vector Regression)具有在线学习和模型在线更新的优点,可应用于交通流量的实时预测,其中算法的核函数的选择对模型的学习、推广和泛化能力起着重要的作用,但是至今有关核函数的选择缺乏科学的理论依据。为了进一步提高模型的学习和推广能力等,提出一种WT-AOSVR(Weight Table And Accurate Online Support Vector Regression)模型。对交通流进行数据挖掘,分类处理,构造支路AOSVR模型和权值表,在交通流预测时,通过搜索权值表就可以得到多条支路模型的一种加权组合模型。仿真实验表明该方法既提高了模型学习精度又保证了模型的泛化和推广能力,具有一定的应用价值。
AOSVR takes the advantage of online learning and online updating its model and can be used in real-time prediction of traffic flow. In it the selection of kernel function of the algorithm plays an important role in model＇ s learning, promotion and generalisation ability. However, until now there lacks the scientific and theoretical basis in regard to the choice of kernel functions. In order to further improve the learning and promotion ability of the model, we bring forward a new method, WT-AOSVR （weight table and accurate online support vector regression） model. It makes data mining and classified processing on the traffic flow, constructs branch AOSVR models and the weight table. When forecasting the traffic flow, by searching the weight table it is able to obtain a kind of weighted composition model for multiple branch models. Simulation experiments show that this method is effective： it improves the learning accuracy of the model as well as ensures the generalisation and promotion ability of the model, and has certain applied value.
出处
《计算机应用与软件》
CSCD
北大核心
2013年第1期277-280,共4页
Computer Applications and Software
关键词
交通流
预测
权值表
AOSVR分类树
Traffic flow Prediction Weight table AOSVR Classification tree
