Quantitative risk evaluation of natural disasters in construction of deepwater platform
Tan Zhen-dong · Ziran zaihai xuebao · 2007
Based on the existing risk theory and combined with the practice of oceanic construction,this paper introduced a new model for quantitative asessment of natural risk in oceanic construction.To deal with the complex nonlinear mapping between the risk events and their influencing factors,the BP neural network simulation was applied to quantifying the risk probability of each risk event.Because of slow speed of the traditional arithmetic for training the network,a new numerical optimization technique was used to accelerate the convergence of the back-propagation.And the Levenberg-Marquardt iteration,which can avoid to compute Hessian matrix,is adopted as the learning rule to train the samples of this feedforward network.This model,practically universal in this field,can simulate experts' estimates and calculate the risk probability according to standard rules of assessment.