摘要
在径向基函数(RBF)神经网络实现无人机复合材料超声检测脱粘缺陷识别时,针对最小均方(LMS)算法在确定网络输出权值时存在稳态失调误差和收敛速度相矛盾的问题,提出一种改进的自适应的变步长LMS算法.该算法根据反馈误差自适应确定步长,通过引进动量项加快收敛速度.将改进LMS算法应用到RBF网络缺陷识别中,结果表明该方法在稳态失调误差较小的情况下,能快速确定RBF网络的权值.改进的RBF网络能够较好地识别超声检测脱粘缺陷.
To solve the problem of contradiction between steady state error and convergence speed when choosing the weight within the least mean square(LMS)algorithm, an improved variable step--size adaptive LMS algorithm is proposed,recognizing the flaw in the composite materials for unmanned aerial systems using radial basis function(RBF)neural network with ultrasonic tes- ting. The algorithm first decides the step size adaptively depending on the error, then adds the mo-mentum project to have a better convergence speed. The experimental results show that the new algorithm can decide the RBF network weight more quickly within low steady state error and the improved RBF network can classify the flaws better.
出处
《军械工程学院学报》
2013年第5期66-69,共4页
Journal of Ordnance Engineering College
基金
军队武器装备预研项目