Abstract: In order to improve the accuracy of port throughput prediction, an improved Elman neural network prediction model based on Adaboost algorithm was established for port throughput prediction. Firstly, Elman neural network was used for many times of training and iteration, and then each Elman neural network was used as a weak predictor. Based on Adaboost algorithm, multiple weak predictors were weighted and combined to form an Elman-Adaboost strong predictor model. The strong predictor optimized by Adaboost algorithm had a stronger ability to identify data samples with large errors and could realize dynamic reinforcement learning of data. The port throughput data of Ningbo-Zhoushan port from 2011 to 2017 were taken as samples for simulation. BP neural network, Elman neural network, BP-Adaboost neural network and Elman-Adaboost neural network were used for prediction respectively, and the prediction accuracy of the four models was compared. The data results show that: the Elman-Adaboost strong predictor model can be used for port throughput prediction, the maximum relative error of its prediction result is 1.91% and the minimum relative error is 0.06%. When the prediction error can be controlled below 2%, the data fitting effect is better and the prediction accuracy is higher. It can be used as a method of port throughput prediction.

Key words: traffic and transportation engineering, port throughput, Adaboost algorithm, Elman neural network, dynamic prediction

摘要： 为提高港口吞吐量的预测精度，建立基于Adaboost算法改进的Elman神经网络预测模型，进行吞吐量的预测。首先对Elman神经网络进行多次训练和迭代，然后将每个Elman神经网络作为弱预测器，基于Adaboost算法将多个弱预测器加权组合，形成Elman-Adaboost强预测器模型。经过Adaboost算法优化的强预测器对误差较大的数据样本有更强的识别能力，可以实现对数据的动态增强学习。以宁波-舟山港2011—2017年的港口吞吐量数据为样本进行仿真，分别使用BP神经网络、Elman神经网络、BP-Adaboost神经网络以及Elman-Adaboost神经网络进行预测，比较四种模型的预测精度。研究结果表明：Elman-Adaboost强预测器模型用于港口吞吐量的预测，预测结果的相对误差最大值1.91%，最小值0.06%，可以将预测误差控制在2%以下，数据拟合效果更好预测精度更高，可以作为港口吞吐量预测的一种方法。

关键词: 交通运输工程, 港口吞吐量, Adaboost算法, Elman神经网络, 动态预测