基于时间窗口选择和SVR的船舶交通事故率预测

为提高对船舶交通事故率的预测精度,建立基于时间窗口选择和支持向量回归的船舶交通事故率预测模型,该预测模型考虑船舶交通事故的发生具有随机不确定性,以及海事规则生效等导致的船舶交通事故率的变动趋势。在构造船舶交通事故率不确定时间序列的基础上,采用滑动窗口技术对置信区间比较集中的不确定时间序列点进行聚类并分段,选择最接近当前时间的区段作为统计时间窗口,并利用支持向量回归模型对船舶交通事故率进行预测。以英国籍船舶交通事故数据为例,对预测模型进行实例分析。通过与指数平滑法和自回归移动平均模型预测结果的对比验证预测模型的有效性,为船舶交通安全管理者的决策提供指导。

Prediction of Vessel Traffic Accident Rate Based on Time Window Selection and SVR

The prediction model based on time window selection and support vector regression is proposed for improving the accuracy of the vessel traffic accident rate prediction. The uncertainty of vessel traffic accident occurrence and the change of vessel traffic accident rate caused by new maritime rules coming into effect are reflected in the prediction model. The random time sequences of the vessel traffic accident rate are built and sliding window is used to implement the segmentation of the time sequences, and the segments of each time stamp are clustered according to the confidence interval. The segment in the time window closest to the interested time is used to predict the vessel traffic accident rate by support vector regression model. A case study with the vessel traffic accident data from the United Kingdom is conducted. The effectiveness of the method is checked comparing the results with predictions made by exponential smoothing method and by autoregressive integrated moving average model.