Abstract: In order to predict the port container throughput more scientifically and accurately, a multivariate input LSTM model was established based on the long and short-term memory network (LSTM) model in the deep learning method. Firstly, the hierarchical clustering method was used to carry out the cluster analysis of multiple influencing factors of container throughput of Qingdao port. According to the calculated value of Pearson’s correlation coefficient, typical influencing factors were selected. Secondly, combined with the historical container throughput data, it was input into the model as a multivariate for prediction, and the prediction results was compared with those of the univariate LSTM model and the traditional prediction model ARIMA model. The results show that the prediction error of the LSTM model using influencing factors and historical throughput data as multivariate input is reduced, the mean absolute percentage error (MAPE) is reduced to 4.170%, and the root mean square error (RMSE) is reduced to 7.736. Therefore, the predicted value is more accurate. The proposed model improves the scientificity and accuracy of prediction and promotes the application of deep learning technology in port container throughput prediction, which can provide references for reasonable decision-making and planning for ports.

Key words: traffic and transportation engineering; water transportation; hierarchical clustering; multivariate; LSTM model; Qingdao port; container throughput forecast

摘要： 为了更加科学准确地对港口集装箱吞吐量进行预测，以深度学习方法中的长短时记忆网络(LSTM)模型为基础，建立一种多变量输入的LSTM模型。首先使用系统聚类法对青岛港集装箱吞吐量的多种影响因素进行聚类分析，根据普尔逊(Pearson)相关系数计算值选取典型影响因素，其次结合历史集装箱吞吐量数据作为多变量输入到模型中进行预测，并将预测结果与单变量LSTM模型和传统预测模型（ARIMA模型）的预测结果进行比较。结果表明：使用影响因素及历史吞吐量数据作为多变量输入的LSTM模型预测误差减小，平均绝对百分比误差(MAPE)降低到4.170%，均方根误差(RMSE)降低到7.736，预测值更加精确。该模型提高了预测的科学性与准确性，促进深度学习技术在港口集装箱吞吐量预测方面的应用，可为港口的合理决策与规划提供参考。

关键词: 交通运输工程；水路运输；系统聚类；多变量；LSTM模型；青岛港；集装箱吞吐量预测