Abstract: In order to help the construction of smart port, aiming at the problem of insufficient accuracy of port container throughput prediction, a port container throughput prediction method was constructed by using random forest algorithm to deal with high-dimensional variables. Firstly, considering that the port container throughput was affected by complex environment, the characteristic variable training set was established. Then, the random forest model was trained by generalized error analysis, and the importance of characteristic variables was analyzed according to MDA analysis to screen the set of important influence characteristic variables. Finally, the decision tree of random forest prediction was constructed, and the prediction model based on random forest algorithm was established. Dalian Port was taken as a case to verify, and compared with three kinds of prediction methods such as cubic exponential smoothing, multiple regression analysis and BP neural network. The results show that the proposed random forest algorithm has higher prediction accuracy.

Key words: traffic and transportation engineering; container throughput; random forest algorithm; port; prediction

摘要： 为助力智慧港口建设，针对港口集装箱吞吐量预测准确性不足的问题，利用随机森林算法处理高维变量，构建一种港口集装箱吞吐量预测方法。首先考虑港口集装箱吞吐量受复杂环境影响，建立特征变量训练集；再通过泛化误差分析训练随机森林模型，根据MDA分析对特征变量重要性进行分析，筛选重要影响特征变量集合；最后构建随机森林预测决策树，建立基于随机森林算法的预测模型。以大连港为案例进行验证，并与三次指数平滑、多元回归分析和BP神经网络3种方法预测进行对比，结果表明：随机森林算法预测准确性更高。

关键词: 交通运输工程；集装箱吞吐量；随机森林算法；港口；预测