基于人工神经网络的结构代理模型性能分析

Performance Analysis of Structural Agent Model Based on Artificial Neural Network

在实际结构施工过程中,由于复杂结构的高次超静定以及强非线性等特点,使得应用于施工过程即时决策的计算模型非常复杂。针对在施工过程中特定的结构响应无法直接获得显式关系以及应用于即时决策时的有限元模型计算缓慢的问题,通过使用代理模型对结构响应进行替代,能够很大程度上提升计算效率。本文通过对现阶段应用最为广泛的人工神经网络代理模型的构建原理进行论述,并通过一刚架算例对BP神经网络(BPNN)以及广义径向基神经网络(GRNN)的拟合性能进行了对比,旨在为结构代理模型的选用提供参考。

In the actual structure construction process,due to the characteristics of high-order static stability and strong nonlinearity,the calculation model applied to the immediate decision-making of the construction process is very complicated.Aiming at the problem that the explicit relationship of the specific structural response cannot be obtained during the construction process and the slow calculation of the finite element model applied to the instant decision,the use of the surrogate model to replace the structural response can greatly improve the computational efficiency.In this paper,the theory of the most widely used artificial neural network discussed.The fitting performance of BP neural network(BPNN)and general regression neural network(GRNN)is studied by a frame example.The comparison aims to provide a reference for the selection of structural surrogate models.