基于不同预测模型的海运危险品安全评价

Safety Evaluation of Hazardous Goods Shipping Using Different Predicting Models

危险品大多具有易燃易爆、易腐蚀和放射性等性质,对该类货物的水路运输进行定量风险评估非常复杂。GRNN和BP神经网络作为智能算法的典型理论,有着传统建模方法不具有的很多优点,如对建模对象的结构、参数及运动特性可以不作要求,只需知道对象的输入和输出即可通过自身的学习达到预期目的,因此在不确定因素的评价方面有着广泛的应用。针对水运危险品本身的特性和相关影响因素,在对GRNN和BP两种不同评价方法进行研究的基础上,根据近年来发生在我国内地的危险品事故案例,按照各影响因素对事故风险等级的相关性分析,建立相应的预测模型,并进行对比。运行结果表明,GRNN在小样本安全评价方面有着比BP网络更高的准确度及稳定性。

Goods transported by sea are mostly inflammable, explosive, corrosive or radioactive. The quantitative assessment of the risks involved is very complex. Typical intelligent algorithm theories, GRNN（Generalized Regres sion Neutral Networks）and BP neutral network are able to solve the problem more effectively than traditional model- ing. GRNN and BP neural networks are widely used for complex problems because they are able to model a process without precise system parameters and dynamic characteristics relying on training. The different risk prediction models are created on the basis of GRNN and BP neural network and the hazardous goods accident cases in recent years are analysed. The significances of various influencing factors are identified according to the correlation analysis. The results of analysis indicate that GRNN has higher accuracy and stability than BP network in the safety eval uation.