基于径向基神经网络的谐波叠加法

RBF-neural-network-based harmony superposition method

风荷载的数值模拟在结构设计中起着非常重要的作用。在土木工程脉动风速时程的各种数值模拟方法中,谐波叠加法最为常用。而且,通过引入FFT和不同插值技术可以在不显著地影响模拟精度的情况下,大大地缩短模拟计算所花费的时间。提出使用径向基神经网络(RBF neural network)插值技术来改进传统的谐波叠加法。使用基于径向基神经网络和传统的谐波叠加法(未引入插值技术)来模拟一幢100 m高的高层建筑上10个点的脉动风速时程,通过均方根误差(Root Mean Square Error,RMSE)和相对误差系数(Error factor,E_f)两项指标来评价改进的与传统的谐波叠加法相比较的模拟计算精度,并且记录各自所耗费的时间。结果表明:基于径向基神经网络谐波叠加法的精度令人相当满意,模拟计算效率大大提高。

Numerical simulation of wind loads is critical in design of structures.The harmony superposition method (HSM) is most widely used among all simulation methods for civil engineering structures.Likewise,the time expense of HSM can be significantly shortened by means of FFT technique and various kinds of interpolation techniques without a significant loss of the accuracy.Here,the RBF neural network interpolation method was introduced into to the traditional HSM(without introducing interpolation techniques),referred to as RBF-neural-network-based HSM.RBF-neural-network -based HSM was employed to simulate wind speed time series at 10 points on a 100-metre building.Two indices, designated as the root mean square error(RMSE) and error factor(E_f),were introduced to measure the accuracy of the proposed approach with respect to the traditional HSM and likewise their time expenses were recorded.The numerical re- suits showed that the RBF-neural-network-based HSM renders the satisfactory precision and great efficiency.