基于主组分分析的概率神经网络损伤定位研究

概率神经网络(PNN)以贝叶斯概率的方法描述测量数据,因而PNN在有噪声条件下的结构损伤检测方面具有巨大潜力。与此同时,PNN的网络规模随着训练样本的增加而增大,这极大地降低了网络运行速度。基于此,本文提出了基于主组分分析(PCA)的PNN损伤定位方法,分别用传统PNN(TPNN)、主组分分析PNN(PCAPNN)和自适应PNN(APNN)三种模型进行了悬索桥的损伤定位研究。研究发现,APNN的识别精度最好,PCAPNN次之,TPNN最差。但APNN需要很长的训练时间,网络规模较大;其他两个网络几乎不需要训练时间,且PCAPNN网络规模较其他两个网络减少了1／3～1／4。在低噪声情况下,PCAPNN的识别效果基本上等同于APNN。

Research on damage localization of principal component analysis-based probabilistic neural network

As the probabilistic neural network (PNN) describes measurement data in a Bayesian probabilistic approach, it shows great potential for structural damage detection in noisy conditions. Meanwhile, the size of PNN increases as the learning samples increase. This reduces the running velocity. Based on this, a damage localization method called PNN is proposed based on principal component analysis in this paper. Three PNN models, namely, the traditional PNN (TPNN),the principal component analysis PNN (PCAPNN) and the adaptive PNN (APNN) models are utilized to detect the damage location of a suspension bridge respectively. This study shows that the (identification) accuracy (IA) of damage localization using the APNN is the best, the IA using the TPNN is the worst , and the IA using the PCAPNN is between the former two models.But the training time using the APNN is very long,and the size of the model is relatively large. Meanwhile, the others hardly need time to train the PNN models, and the size of PCAPNN reduces to that of other two models from 1/3 to 1/4. Furthermore, in low noise level, the IA using PCAPNN is almost the same as the APNN.