大跨悬索桥损伤定位的自适应概率神经网络研究

由于概率神经网络(PNN)以贝叶斯概率方法描述测量数据,因而PNN在有噪声条件下的结构损伤检测方面,具有巨大的潜力。而PNN中高斯核函数的宽度,严重影响网络的泛化能力,本文提出了一种运用自适应PNN进行复杂结构的损伤定位研究方法,并与传统PNN对大跨悬索桥的损伤定位进行了仿真性能比较;同时讨论了噪声程度、特征向量简化对损伤识别精度的影响。研究发现,运用自适应PNN进行损伤定位,不仅性能优于传统PNN,而且进行特征向量简化时,可以提高损伤定位的识别精度。

RESEARCH ON ADAPTIVE PROBABILISTIC NEURAL NETWORK FOR DAMAGE LOCALIZATION OF A LONG-SPAN SUSPENSION BRIDGE

The probabilistic neural network (PNN) can describe the measured data by the Bayesian probabilistic approach. It shows great potential for the structural damage detection in noisy conditions. The width of Gaussian kernel function effects on the generality of network. An adaptive PNN is proposed for the structural damage detection in the paper. It is compared with the traditional PNN for a long-span suspension bridge. At the same time , the effect of the noise level and eigenvector reduction on the identification accuracy (LA) is discussed. The study shows that the LA of the damage localization using the adaptive PNN is better than that of the traditional one. The IA will be increased after adopted the eigenvector reduction.