基于BP神经网络的港口吞吐量预测模型

港口吞吐量预测是港口决策和规划的基础。为了合理预测港口吞吐量,本文利用外贸进出口总量、第一产业总产值和第三产业总产值作为BP神经网络的输入变量,港口吞吐量为输出变量,建立了港口吞吐量预测的BP神经网络预测模型。然后根据2000年-2010年广西北部湾港口吞吐量、外贸进出口总量、第一产业总产值和第三产业总产值,利用Matlab 6.5软件的神经网络工具箱,通过对BP神经网络模型的反复训练,发现当隐含层节点数为6,学习率为0.05,训练次数为500次,训练精度为0.001,动力因子为0.9时得到的效果最好。并对BP神经网络模型与多元回归模型的预测结果进行比较分析,认为BP神经网络模型预测的总体效果更优。最后利用所确定的BP神经网络模型,对2011年和2012年两年的港口吞吐量进行了预测。

The Port Throughput Forecast Model Based on the BP Neural Network

Port throughput forecasting is the premise of port decision-making and planning.In order to forecast the port throughput reasonably,this paper takes foreign trade volume,the first industrial output and the third industry output as input variables and port throughput as output variables of the BP neural network to build the its forecast model of the port throughput.Then,according to the 2000-2010’s date of the North Gulf in Guangxi,the author uses Matlab 6.5 software neural network toolbox to train the BP neural network model repeat,and finds that the forecast effect is best when hidden node number is 6,adopted is 0.05,train time is 500,train accuracy is 0.001 and momentum factors is 0.9.By comparing the results with that from multivariate linear regression prediction model,the paper concludes that the BP neural network model is more superior.Finally,the author uses the determined BP neural network model to forecast the 2011 and 2012 two years of port throughput in the future.