大跨平屋盖结构风压分布特性的神经网络模型

影响大跨平屋盖结构风荷载分布特性的因素很复杂，且具有不确定性，仅靠风洞试验难以完整的描述其风荷载分布的整体特性，针对这种情况，本文中提出用改进的BP神经网络和模糊神经网络两种方法来建立反映大跨平屋盖结构风压分布特性的模型，并用试验数据进行了验证．结果表明，这两种模型都能很好的逼近和预测大跨平屋盖结构风压分布的特性，相比之下，改进的BP神经网络稳定性较好，但逼近速度慢，精度也不高；而模糊神经网络由于结合了模糊系统和神经网络的优点，其稳定性好，逼近速度快，且精度高，这表明模糊神经网络方法是预测大跨平屋盖结构风压分布特性的有效途径，在结构风工程中具有广阔的应用前景。

Neural Network Models for Describing Characteristics of Wind Pressure Distribution on Large Span Flat Roof

The factors which control and affect the distribution of wind load on large span flat roof are complicated and full of umcertainties.Therfore, it is difficult to describe them completely depending only on wind tunnel tests. In view of at this situation, the reformed back propagation (BP) neural network and the fuzzy neural network (FNN) are used to establish new models reflecting the distribution of wind load on large span flat roof. The new models are verified by wind tunnel test data. The results show that these two methods can also approach and predict the distribution of wind load on large span flat roof correctly. Compared with these two methods, the BP has good stability but the speed is slow and the precision is not high. Due to the combination of fuzzy system with neural network, the FNN possesses not only has good stability but also rapid speed and the precision is much higher than the BP. The study shows that using FNN is an effective way to predict the distribution of wind load on large span flat roof, and it can be widely used in structural wind engineering.