Research On Operational Prediction And Performance Evaluation Of Container Terminals Based On Deep Learning

The continuous development of China’s economy and society brings great demand for logistics transportation.The long and tortuous coastline of the mainland has brought many excellent harbors.Combined with the current multimodal transport scheme,it provides a solid foundation for the development of waterway transportation.Waterway transportation has considerable advantages in logistics transportation because of its large volume and low price.Container has the characteristics of standardization,integration,loading and unloading mechanization and transportation convenience.As a transportation hub integrating waterway transportation and container handling and storage,container terminal plays an important role in economy and trade.With the growth of container transportation volume and the increasing competition between container terminals,container terminals improve their operation level and service efficiency of operation system.Terminal operation condition is the key of container terminal handling system(CTHS).Liner berthing time(LBT),liner handling time(LHT),total working time of quay crane farm(TWT-QCF)and liner handling volumes(LHV)are not only the central indexes to evaluate the service efficiency of container terminals,but also the important basis and guidance for task scheduling and resource allocation of container terminals.On the basis of summarizing the research background of container terminal production operation,combined with the current continuous development of deep learning theory,this paper proposes the combination of computational logistics and deep learning,summarizes,unifies,and integrates the essence and connotation of computing,and discusses the prediction of LBT,LHT,TWT-QCF and LHV.This paper combines the learning computation of artificial neural network with the logistics generalized computation for container terminals(LGC-CT)to connect the information space and the physical world.Based on the operation data of a typical container terminal in China in recent five years,three deep learning prediction models are designed and implemented according to the index characteristics of LBT,LHT,TWT-QCF and LHV,and the prediction results are evaluated to provide support for the allocation,deployment,and implementation of container terminal operation resources.Based on the deep learning framework of tensorflow 2.3,three deep learning models including recurrent neural network(including long short-term memory network,gated recurrent unit,etc.),convolutional neural network and attention mechanism are designed.Through the preprocessing of container terminal production operation data set sequence and the design and debugging of model architecture,good prediction results are obtained,the feasibility and credibility of the proposed composite computing system and computing paradigm are preliminarily verified.It also provides some help and insights for the scheduling decision of container terminal operation system,improves the operation efficiency of container terminal,and finally enhances the competitiveness of the terminal.