摘 要

近几年，作为我国重要运输方式之一的内河航运发展迅速，其2015年总货运量为21.8亿吨，单位运输成本却仅为铁路的1/2，公路的 1/4^[1]，为我国每年带来巨大经济利润。但是，我国内河航运依然存在船舶航行安全、物流优化、港口调度等问题，直接影响内河航运的发展。船舶自动识别系统AIS(Automatic Identification System) 可采集丰富的数据信息，进行多种应用。因此，可以对内河AIS数据运用数据挖掘理论方法进行挖掘及分析，实现船舶航行时间的预测，以提高航务管理部门的管理水平，促进内河航运的迅猛发展。

本文首先梳理了国内外AIS数据挖掘及行程时间预测的研究，在已有的研究中，虽然行程时间预测算法种类较多，但应用AIS对内河船舶航行时间预测的研究较少。论文以AIS数据为基础，依照缺失数据的排查、上下行船舶的划分以及冗余数据的剔除三大步骤进行数据的预处理工作，在对预处理后的AIS数据分析研究基础之上，综合考虑各类行程时间预测算法的特点，选取适用于内河船舶的支持向量机（Support Vector Machine, SVM）算法作为模型的核心算法。在模型的构建过程中，论文基于SVM算法自学习能力，寻找过去时段航行时间及多种随机影响因素与当前时段航行时间之间的复杂函数关系，实现了基于AIS内河船舶航行时间预测方法的研究。在最终的预测模型的实现与分析中，论文采用R统计软件，以长江武汉段的AIS数据为例，基于预处理后的AIS数据构建训练集，采用网格搜索法获取SVM最优参数依次为C=0.5、=0.1、=0.0625，同时，在最优参数预测下的结果散点图趋近观测值，预测结果的平均相对误差绝对值为10.53%，模型具有较高的可靠性和鲁棒性。

本文的创新之处首先是对AIS数据的预处理，为算法的选择提供理论依据的同时，有效改善了算法实现中输入数据的质量；其次在模型的构建中，选取内河航运应用较少、预测效果较好的SVM算法，利用SVM的自我学习能力，构建思路简单创新；最后是预测模型的实现，采用基于网格搜索的SVM参数寻优法获取参数最优值，预测精度大幅提高。

综上，本文所建立的基于网格搜索法的SVM参数寻优的内河船舶航行时间预测模型实现了对内河船舶航行时间的预测，且预测效果较为精确、稳定，应用领域广泛。

关键词：内河船舶；航行时间；船舶自动识别系统；支持向量机

Abstract

In recent years, as an important mode of transport, the inland shipping has developed rapidly. In 2015, it completed a total of 2.18 billion tons of cargo, and the unit transportation cost is only half of the railway, quarter of the highway ^[1], so that it brings huge economic profits annually for our country. However, China's inland shipping still have some problems, such as the safety of navigation, optimization of logistics, scheduling of port, and it will have direct impact on the development of the inland shipping. Automatic Identification System can collect a wealth of data and perform a variety of applications. Therefore, we can use data mining theoretical methods on the AIS data of inland to analyze and predict ship sailing time accurately, so that it can improve the management level of flight operations department and promote the rapid development of inland navigation.

This paper reviews the researches on the AIS data dig and the travel time prediction at home and abroad firstly, and in previous studies, although the travel time prediction algorithm has more categories, but it is short of the research on the time prediction in inland navigation vessels depending on AIS data. The paper does pretreatment on AIS data follow the three steps of troubleshoot missing data, dividing up and down ship and eliminate redundant data, and considers the features of various types of travel time prediction algorithm according to analyzing AIS data to select Support Vector Machine, SVM algorithm as the core algorithm which is suitable for inland vessels. During the construction of the model, the paper uses SVM algorithm’s self-learning ability to look for the complex function between the sailing time in the past periods and a variety of random factors with the current sailing time, so that it realized the study of inland vessels sailing time prediction method finally. In the final implementation and analysis of prediction model, with R statistical software, depending on the AIS data in Wuhan section of the Yangtze River, the paper uses grid search method and training set of the AIS to get the optimal parameters, and the optimal parameters are C=0.5, =0.1, =0.0625. Also, with optimal parameters, the results scatterplot approaching observations and the average prediction error is 10.53%, the predicting model has a high reliability and robustness.

The first innovation of this paper is pretreatment of the AIS data, which provide a theoretical basis for the selection algorithm and effectively improve the quality of the input data of the algorithm realization. Secondly, in the process of building predictive model, the paper selects the SVM algorithm, which less applications in inland shipping and has better forecast result, and uses SVM self-learning ability, whose model idea is simple and innovative. Finally, in the achievement part of the forecast model, the paper obtains optimum value based on grid search of SVM parameter optimization method to improve the accuracy of prediction model.

In a conclusion, the inland vessels sailing time prediction model established in this paper based on the grid search of SVM parameter optimization method achieves the inland ship sailing time forecast, and the prediction results of the mode is accurate, stable, wide range of applications, and has good prospects.

Key Words：inland vessels; sailing time; Automatic Identification System; Support Vector Machine

目 录

第1章 绪论 1

1.1 研究目的及国内外现状 1

1.1.1 研究背景 1

1.1.2 研究目的及意义 1

1.1.3 国内外研究现状 2

1.2 研究内容与方法 3

1.2.1 研究内容 3

1.2.2 研究方法 4

1.3 论文结构 4

1.4 本章小结 6

第2章 AIS数据的预处理与分析 7

2.1 AIS数据信息的介绍 7

2.1.1 AIS系统的定义及原理 7

2.1.2 AIS数据信息 7

2.2 内河船舶航行特点 8

2.3 AIS数据的预处理方法 8

2.3.1 预处理步骤 8

2.3.2 实例分析 8

2.4 行程时间预测算法的研究 12

2.4.1 基于线性理论的预测算法 13

2.4.2 基于非线性理论的预测算法 13

2.4.3 内河船舶航行时间预测算法的选择 14

2.5 本章小结 15

第3章 内河船舶航行时间预测模型的建立 16

3.1 支持向量机理论 16

3.1.1 支持向量机原理 16

3.1.2 核函数的选择 16

3.2 内河船舶航行时间预测模型的建立 17

3.2.1 支持向量机算法 17

3.2.2 预测模型构建方法 18

3.3 模型参数的确定 20

3.4 本章小结 20

第4章 内河船舶航行时间预测实现与分析 22

4.1 基于网格搜索法的SVM参数优化 22

4.1.1 R语言介绍 22

4.1.2 训练数据集的构建 22

4.1.3 基于网格搜索法的参数寻优 24

4.2 预测模型的实现 27

4.2.1 svm函数介绍 27

4.2.2 模型的验证 27

4.3 预测模型的评价 32

4.4 本章小结 32

第5章 工作总结与展望 33

5.1 工作总结 33

5.2 经济效益分析 34

5.3 展望 34

参考文献 35

附 录 37

致 谢 43

第1章 绪论

1.1 研究目的及国内外现状

1.1.1 研究背景