深水平台施工过程自然风险定量评估

Quantitative risk evaluation of natural disasters in construction of deepwater platform

在分析现有风险理论的基础上,结合海洋施工实际,提出了一种对海洋施工自然风险进行定量评估的模型。根据目标系统各风险事件与影响因素之间存在高度非线性复杂映射的特点,提出了利用BP神经网络拟合方法,对主要风险事件发生概率进行量化的模型。针对传统的梯度下降优化算法收敛速度慢的特点,采用可避免计算Hesse矩阵的Levenberg-Marquardt算法来训练神经网络。这一模型可以模拟专家评价,并准确地按照专家的评定法则进行估算,具有一定的通用性。

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 backpropagation. 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