【作者】 陈建文；

【导师】 张东站；

【作者基本信息】 厦门大学 ， 计算机技术， 2019， 硕士

【摘要】 航运业是一个重资产的行业,船舶资产价值重大,造船业是航运业的基础。中日韩三个国家的造船业的订单量占据了世界绝大部分的市场份额,是世界造船业研究的重点。然而目前在这一领域的量化研究,多采用传统经济学研究方法。近年来,计算机科学领域数据挖掘、机器学习相关技术快速发展,应用领域也在不断拓宽。因此,本文将结合机器学习算法来对中日韩三国的造船业进行分析研究,探索中日韩三国造船业的订单量分析预测模型。本文选取了中日韩这三个世界造船业的主要产能国家为研究对象,首先阐述了该领域当前研究背景和现状,提出本文的研究目的和意义。本文选取了手持订单量作为预测目标,并根据经济理论和市场实际背景选取了对中日韩三国造船业手持订单量有较显著影响的8个影响指标进行预测建模。针对传统采用时序预测方法如灰色预测、ARMA预测仅对单一变量进行时序规律分析,忽略了其他指标对预测目标的影响的问题和缺陷,本文提出了使用回归进行造船手持订单量预测的改进思路。首先构建了基于多元线性回归的预测模型;针对指标间可能存在的非线性问题,提出了基于多项式回归改进的预测模型;又为了解决指标间可能存在的多重共线性问题,提出了基于岭估计的多元线性回归模型。针对预测目标的影响指标,构建了灰色时间序列预测,对影响造船手持订单量的特征指标进行预测。再次,为探索能更好的预测造船手持订单量的模型,构建了基于神经网络的预测模型。接着,针对灰色模型对指标数据集的利用不足的问题,同时为了实现对未来手持订单量的预测和进一步提高预测精度,提出了灰色-神经网络组合的FGBP-SOB与GBP-SOB模型改进思路,充分综合了GM(1,1)和BP神经网络模型的优势。实验证明GBP-SOB模型较单一模型具有更好预测效果。最后,为挖掘中日韩三国之间造船业存在的竞争关系,提出了基于Apriori算法的中日韩三国新接造船订单量涨跌联动情况分析模型进行研究分析。最后,本文将研究结果应用到中国-东盟海洋大数据平台的航运模块,实现中日韩造船订单量及指标检索、预测等功能,为平台提供研究和预测参考。更多还原

【Abstract】 Shipping industry is a heavy asset industry.The value of ship assets is significant.Shipbuilding industry is the foundation of shipping industry.China,Japan and Korea occupy the maj ority of the world shipbuilding market share and China,Japan and Korea are the focus of the research of world shipbuilding industry.However,at present,the quantitative research in this field mostly adopts the traditional economic research methods.In recent years,in computer science field,data mining and machine learning technology have developed rapidly and their application fields are also expanding.Therefore,this paper will use machine learning algorithm to analyze the shipbuilding industry of China,Japan and South Korea,and explore the shipbuilding new contracts and orderbook of China,Japan and South Korea analysis and prediction model.This paper chooses China,Japan and South Korea as the research objects.Firstly,it expounds the current shipbuilding background and current research situation in shipbuilding field,and puts forward the purpose and significance of this paper.In this paper,shipbuilding orderbook are selected as the forecasting target.According to the economic theory and the actual market background,eight influencing indicators were choosen for forecasting and modeling.They have a significant impact on orderbook in shipbuilding industry of China,Japan and South Korea.To deal with the imperfection of traditional time series forecasting methods such as grey forecasting and ARMA forecasting,which only analyze the time series rule of a single variable and neglect the impact of other indicators on the forecasting target,this paper puts forward an improved idea of using regression to predict shipbuilding orderbook.Firstly,a forecasting model based on multiple linear regression is constructed.In view of the possible non-linear problems among the indicators,an improved prediction model based on polynomial regression is proposed.As well,a multi-linear regression model based on ridge estimation is proposed to solve the possible multi-collinearity problems among the indicators.Aiming at the impact indicators of the forecasting target,the time series forecasting grey model is constructed to forecast the indicators related to shipbuilding orderbook.Then,in order to find a better prediction model for shipbuilding orderbook,a prediction model based on neural network is constructed.Then,aiming at the problem of insufficient use of grey model for indicators data set,and in order to realize the prediction of future orderbook and to improve the forecast accuracy,this paper put forward an improvement idea that combined the grey and neural network model into FGBP-SOB and GBP-SOB.In this way,the FGBP-SOB and GBP-SOB models can fully integrate the advantage of GM(1,1)and BP neural network model.And experiments prove that GBP-SOB model has better forecast accuracy than single model.After that,in order to find the competitive relationship between China,Japan and Korea,an analysis model based on Apriori algorithm is proposed to analyze the rising and falling of shipbuilding new contracts of China,Japan and Korea.Finally,this paper applies the research results to the shipping module of the China-ASEAN Ocean Big Data Platform,realizes the retrieval and prediction functions of orderbook and indicators of shipbuilding between China,Japan and Korea,and provides reference of prediction and research for-the platform.更多还原