基于融合先验知识SVM的航班延误预警模型

Early Warning Model of Flight Delay Based on SVM with Incorporated Prior Knowledge

针对日益严重的航班延误问题,有必要通过预警机制减少其负面影响,对机场各时段延误程度的预警其实质是一个多类分类问题。利用先验知识确定了一种新的基于相对紧密度的方法来计算各样本的权值,并将其融合到支持向量机模型中,进而结合一对一法(OAO)形成一种一对一加权间隔支持向量机(OAO-WM SVM),用以实现一个多级航班延误预警模型。实验证明,用OAO-WM SVM建立的预警模型能有效判定机场各时段的延误等级;同时还证明,融合了先验知识的SVM比标准SVM具有更好的分类性能。

Aimed at the problem of severe flight delays, early warning can be used to reduce the negative effects. Judging the delay level of each interval is a multi-class classification problem. Weighted margin support vector machine （WMSVM） is adopted with the weights of samples determined by a new relative affinity function based on prior knowledge. Referring to one against one-support vector machine（OAO- SVM）, an OAO-WMSVM method is proposed to implement a multi-level early warning model of the flight delay. Experiments show that the early warning model based on OAO-WMSVM can effectively give a delay level to each interval. Experiments show that SVM with the incorporated prior knowledge has better classication performance than standard SVM.