基于RBF-SVM的拼宽预应力混凝土T梁桥可靠度分析

Study on the Flexural Reliability for the Old T Beams of Widening PC Bridges Based on RBF-SVM

结合有限元法进行结构可靠度计算时,常采用响应面法和BP神经网络法拟合功能函数的隐式表达式,但这些方法对样本点数量的要求较高,在小样本下的拟合中效果并不理想.因此,结合SVM理论,验证了RBF-SVM法可用于小样本下的功能函数隐式拟合.研究结果表明:采用SVM法拟合的隐式函数,对规范设计值的预测偏小,但其预测误差符合工程精度要求,可以用于小样本下的拼宽桥隐式函数拟合及可靠度计算;拼宽可有效提高旧梁的结构可靠指标并对拼接附近的旧梁可靠指标增加最为有利.

For the calculation of structural reliability with the finite element method,response surface and BP neural network method are usually used for the implicit expression of functional functions.But these methods have high requirements for the numbers of sample points,which may be not good for the fitting of little samples.RBF-SVM is proved to be used for the fitting of functional functions under little samples combined SVM theory.The results show that the predicted values by SVM are less than current standard,which prediction error meet the requirements of computational accuracy.It can be applied to the fitting of implicit function and reliability calculation under little samples.The project of widening can effectively increase the reliability index of old beams,and is good for the increase of reliability index for old beams which are near the split joint.