【作者】 李俊；

【导师】 高岚；

【作者基本信息】 武汉理工大学 ， 轮机工程， 2008， 硕士

【摘要】 由于水上运输运量大、能耗低、运输适应性强以及可国际通航等特点,水运在国民经济建设中占有重要地位。目前我国水运虽迅猛发展,但随着船舶数量增多,船舶交通流量增加,航行密度增大,船舶交通事故频发,致使人命安全、财产损失及环境污染等问题时有发生。因此,对船舶交通事故预测研究势在必行,具有重要的实际意义。船舶交通事故预测是一个具有影响因素繁多、不规则、随机性、非线性预测并且数据不全的预测难题。支持向量机是由V.Vapnik等提出的一种学习技术,是借助于最优化方法解决机器学习问题的新工具。支持向量机是机器学习领域若干标准技术的集大成者,它集成了最大间隔超平面、Mercer核、凸二次规划、稀疏解和松弛变量等多项技术,在工业预测应用中,获得了目前为止最好的效果。支持向量机(SVM)作为一种全新的学习机器,具有拓扑结构简单、提供全局唯一最优解、推广性能好、能从未知分布的小样本中抽取大量的有用信息,解决了样本空间中的高度非线性分类和回归等问题,利用回归作预测能取得很好的预测效果的特点。因此本文选择支持向量机对船舶交通事故进行预测。本论文的主要研究工作如下:(1)全面研究了统计学习理论和支持向量机理论,阐述了支持向量机的基本原理。重点论述了统计学习理论的三个核心概念:VC维、推广能力的界、结构风险最小化,支持向量机的最优分类面、核函数及支持向量回归机,并研究支持向量回归机算法。(2)研究了船舶交通事故的特点、种类和等级,并根据船舶交通事故的特点,从人为因素、船舶因素及环境因素等对船舶交通事故的成因进行全面地分析和研究。(3)研究了指数平滑模型(采取二次、三次平滑预测)、回归模型(采取从一元至五元回归预测)、灰色系统模型和支持向量回归机模型的算法,深入研究了这四种预测方法在船舶交通事故预测中的应用。(4)在对支持向量回归机方法进行船舶事故预测研究中,探讨了事故数据的预处理、支持向量回归机预测模型参数的选取及事故数据的特征选取原理和方法,并进行最优核函数选取研究。另外用高斯径向基为核函数的支持向量机以时间、事故种类和事故等级为特征对长江某流域的事故进行预测,比较得出以时间为特征的预测效果较好。(5)集成基于VC++平台的船舶交通事故预测系统,利用船舶交通事故预测系统对船舶事故数据进行试验并比较,证实支持向量回归机预测效果最佳,并总结了GM(1,1)模型、指数平滑模型及SVM模型的特点。更多还原

【Abstract】 Due to large, low energy consumption, strong adaptability and international navigation of transportation in water, water transportation occupies very important position in national economic construction. At present, water transportation keeps fast development in our country.With the development of maritime trade , maritime accidents often take place because of more and more wartercraft, increasing flow of marine traffic, increasing density of voyage, these maritime accidents casued by serious loss of people’s lives, property and environmental pollution.. So the research in forecast of maritime accidents has actual significance. A forecast of maritime accidents has many characteristics, such as various influent-factors, irregular, randomness, higher-dimension nonlinear, data missing and so on. So the forecast of maritime accidents is difficult problem.SVM(Support vector machine) is a new machine learning technique developed from the middle of 1990s by Vladimir Vapnik. Support vector machines are a very specific class of algorithms, characterized by the use of a maximal margin hyper-plane the theory of kernels, the absence of local minima, convex optimization the sparseness of the solution, Mercer’s theorem and the capacity control obtained by acting on the margin. A large number of experiments have shown that support vector machine has not only simpler structure, but also better performance, especially its better generalization ability. As a new machine learning technique, SVM has simple topology, providing the global only optimal solution, good of the generalization ability, useful information extrction in unknown distribution of small sample, solution higher-dimension and nonlinear convex optimization of sample space.Using regression forecast exacts good effective. This paper applies SVM method to the forecast of maritime accidents.Base on above all, this paper did the following works:1. Research STL (Statistical Learning Theory) and the theory of SVM. It mainly introduces three core concepts of STL, which are VC dimension, minimizing the bound by minimizing hand structural risk minimization. It also elaborates the ideas, counting steps and optimize algorithm of support vector classification and regression.2. Research maritime accident’s characteristics. According to its characteristics , the paper overall thorough analysis and research three factors about human factors, environmental factors and ship factors.3. Research exponential smoothing model, regression model, grey system model and SVM regression model. Further study applications the forecast of maritime accidents.4. Researching the forecast of maritime accidents by SVM regression model, investigate maritime accidents data’s pretreatment, selection of SVM regression model’s parameters, principle and method of maritime accidents data’s characteristics. Optimize kernel function, slecet gaussian kernel function forecast of maritime accidents in Yangtze river’s certain valley by time characteristic, accdent’s species charateristic and accident levels. The forecast result is better performance.5.Using VC++ program to get the forecast system of maritime accidents, comparing experimental results proves that SVM regression model is optimum. Summary the GM(1,1) model, exponential smoothing model and SVM regression model characteristics.更多还原