两种溢油事故损害赔偿方法的应用对比  

Two contrasting methods in compensation assessment of the oil-shipping spill

作　　者：俞徐波[1] 朱鸣鹤[1] 牛程飞[1] 

机构地区：[1]宁波大学海运学院,浙江宁波315211

出　　处：《安全与环境学报》2013年第3期269-273,共5页Journal of Safety and Environment

基　　金：浙江省科技厅资助项目(2012C35053)

摘　　要：针对船舶溢油事故的赔偿问题,基于径向基神经网络(RBF)和GA-SVM两种方法,以国际油污基金公约所承认的著名船舶油污事故损害赔偿案例作为学习样本和检验样本,通过建立基准模型和3个对比方案,比较两种方法在赔偿评估中的优劣。其中基准模型为GA-SVM模型,以前16个赔偿案例作为学习样本,后3个赔偿案例作为检验样本,以模糊定量溢油量、油种比重、海况、环境敏感度、清污情况5个主要影响因素作为SVM的输入特征向量,赔偿额作为输出量。3个对比模型为径向基神经网络模型,其中方案1与基准模型基本设定相同;方案2在方案1的基础上,以溢油量、油种比重、风速、波高、涌高、视距(能见度)、环境敏感度、油膜扩散面积、污染海岸线长度、清污情况10个影响因素作为RBF的输入特征向量;方案3以方案2为基准,使用前12个数据为学习样本,后7个为检验样本进行建模分析。结果表明:GA-SVM模型在预测中具有较高的准确率;而RBF在理论上虽占优势,但预测结果不理想,并对其可能的原因作了相应的分析。This paper is focused on the study of the compensation methods for the oil-shipping spill accidents by establishing a benchmark model with a three contrasting schemes,respectively based on the neural network of the Radial Basis Function（RBF） and the Genetic Algorithm with the Supporting Vector Machine（GA-SVM）so as to compare the two methods for quality evaluation of compensation.Since all the models can use the same data as learning samples and test the samples from the famous compensation cases of oil shipping contaminants well admitted by the International oil-spill contamination compensation fund convention,the benchmark model can serve as the GA-SVM model,whose basic settings are supposed to be as follows： the 16 compensation cases can be taken as the learning samples,with the other 3 groups of data taken as testing samples.Whereas the five basic influential factors,including the oil quantity,oil specific gravity,the ocean climate,the environmental sensitivity and the oil cleansing situations,can be taken as SVM input feature vectors,the compensation amounts can be taken as the output vectors.As to the three contrasting models,they can be taken as the RBF models,of which Scheme 1 is supposed to have the same set with that of the benchmark model.Scheme 2 can be thought of as set up on the basis of Model 1,though it may have more influential factors,such as the oil quantity,the oil specific gravity,the speed of the wind in situ,the wave height,surge height,the wave visibility,the environmental sensitivity,the oil-spilling diffusion area,the contamination coastline length,the oil cleansing situations,which are to be taken as the input of RBF characteristic vectors.At the same time,Scheme 3 is to be set up on the basis of Scheme 2,though only 12 compensation cases can be taken as the learning samples,with the other 7 sets of data being taken as test samples.Thus,all the models can be regarded as being in contrast to each other,with the GA-SVM model displaying a higher accuracy in forecasting.Apart from

关 键 词：安全工程 船舶油污 环境损害 RBF 径向基神经网络 GA-SVM 赔偿评估 

分 类 号：X55[环境科学与工程—环境工程]