基于神经网络算法的大跨度双层扭网壳的设计

应用误差反向传播的人工神经网络技术对大跨度双层扭网壳进行设计,双层扭网壳的跨度在50~80 m之间变化,训练了估算双层扭网壳的最大挠度、重量和杆单元横截面面积的神经网络。为减少数据的非线性和提高训练速度,形成了特殊的数据排序方法。这种方法提供了必要的稳定性。本文研究表明应用神经网络技术对双层网壳结构进行设计是可行的。

Design of Double Layer Torsional Reticulated Shell Using Back- Propagation Neural Network

In this paper artificial neural networks are used for design of large span double layer torsional reticulated shell. The torsional reticulated shells with spans varying between 50.0 - 80.0m are considered, Back -propagation algorithm is employed for training efficient neural networks for evaluation of the maximum deflection, weight and design of double layer torsional reticulated shell. A special method is developed for data ordering to reduce the nonlinearity of data and to increase the speed of training. This approach also provides the necessary stability. It shows that it is feasible to design double layer reticulated shell with artificial neural networks.