| China's contribution to the world's economic growth has maintained around 30%since the financial crisis in 2008.Early in 2010,China became the world's second largest economy and in 2013 the world's biggest cargo trading nation.The total volume of imports and exports of China in 2017 reached RMB27.79 trillion.International shipping serves international trade,of which 90%transportation are achieved by sea.Technologies relate to containers popularized since 1970s greatly reduced intercontinental transportation cost,which further facilitate global trade.In recent years,international container freight rates fluctuated fiercely,adding more unpredictability to it.As international container shipping industry is a capital-intensive industry that requires huge investment,fluctuate container freight rates bring big risk to shipping companies,traders,and the industry as a whole.Container freight rates trend study and forecast have always been the hotspot of international container shipping industry.Studies on international container shipping rates forecast play a positive role in enhancing industrial cost management,reducing industrial default rate,improving international maritime transportation organizations' efficiency and providing reference for government decision-making quantification.Taking into consideration that intelligence prediction is an important application of intelligence science,aiming at studying both theory and method on international container freight rates forecast,this article proposes the theme "Research on Forecasting International Container freight rates",under the guidance of Know Discovery framework in Information Science,Following the concept of"source","communication",and "application" in intelligence science,considering today's "big data era",this article briefly elaborates the research significance,overseas and domestic research status,and establishes the research framework.The research content includes the general framework of forecast,information integration model and methods,data characteristics processing method,forecast model and empirical analysis of international container freight rates under the big data environment.In the section of general framework of forecast,this article discusses the overall process and framework of international container freight rates forecast based on information integration and data mining.The overall process mainly includes six parts such as information collection of freight rates,freight rates information processing and integration,exploratory data analysis and freight rates data characteristics processing,establishment,operation,assessment,and application of freight rates forecast model,among which the establishment,operation and assessment of the forecast model are challenges and key points of the whole research,while freight rates information processing,integration and data characteristics processing require a lot of labor work in real practice.The section of international container freight rates information integration model and methods mainly deal with the integration of heterogenous and isomerous freight rates.This article found that with the development of internet,e-commerce,and information technologies,international freight rates information are becoming more and more digital,integrated,and real-time,which provides basis for using data mining technologies to forecast freight rates,and requires solving isomerism of freight rates information with the study of information integration methods.After discussed the requirements of forecasting freight rates and current status of freight rates information,this article proposes a model of data warehouse based freight rates integration,explains how to get and integrate Web freight rates and incremental information,designs knowledge base and rule base,in order to integrate heterogenous and isomerous freight rates information by using information integration modelThe section of freight rates data characteristics processing includes data processing and dimensional derivation,and discusses techniques and methods to ensure that the freight rates data conform to the requirements of data mining algorithm and to improve the accuracy of freight rates forecast.Data processing mainly includes the processes and methods of processing freight rates data that is abnormal and invalid,and historical zipper chain scission;integration method of key information and original freight rates data,and transformation method of special features and attributes.The dimensional derivative strategies centering on the basic freight rates data mainly include the methods of dimensional derivative,such as the perspective of horizontal,vertical,historical change,and index date,as well as the derivative strategies for predicting the target dimension.The section of data mining based freight rates forecast model probes into the overall framework of international container shipping rates forecast;discusses forecast model and result evaluation index system targeting freight rates spot trends forecast(classification problems)and the ups and downs(regression problems).This article tries to optimize the traditional data mining algorithm.It discusses adaptive grid search strategy,optimization algorithm with hyperparameter adjustment and optimization;explores the time series based THO method-in view of international freight rates data has obvious characteristics of time series,so as to optimize the forecast result evaluation strategy and to reduce the generalization error.It discusses GBDT algorithm based parallel computation and iterative operation optimization strategy of the loss function after pre-ordering,to improve the operation efficiency of the GBDT algorithm in a big data environment.In the section of empirical research on freight rates forecast,considering the future big data application environment for freight rates forecast model,the author adopted real international container freight rates data to forecast the spot trend and ups and downs of the freight rates,by using the processes,methods,models and optimization strategy to design and establish the big data technology based information platform.There are three channels for the data:firstly,the backend freight rates database of some city's international shipping space booking e-platform(ITEU);secondly,the freight rates data from some large international freight forwarder's business system;thirdly,the Web freight rates data from a domestic well-known international shipping space booking business website(365wuliu etc.).The total data volume is about 9.6 million.The empirical study shows that the processes,models and methods of forecasting international container freight rates explored in this article successfully present the way of realizing freight rates forecast from freight rates collecting,analysis,integration and processing,and are obviously superior to the forecast results of the traditional time series approach.This article summarizes the researches and its deficiency at the end,and envisions future research. |
PDF Full Text Request