摘要
结合神经元模型,提出了一种新的BP算法权值、阈值同步学习的BP算法,该方法将神经元权值、阈值均看作自适应的学习变量,在学习过程中同步修改,从而提高传统BP算法的性能。应用于结构损伤检测的数值模拟算例表明,该方法收敛速度较快、检测精度较高,可以改善传统算法收敛速度慢、易出现过拟合的缺陷。
Based on the nerve cell model, a new BP algorithm with weight and threshold synchro-learning ability is developed. The weight and threshold of the nerve cell model is considered as the adaptive learning variables in this new BP algorithm and are adjusted synchronously in the learning process, which improves the behavior of the traditional BP algorithm. The numerical analysis of the structural damage detection shows that this new BP algorithm provides a faster convergence to the solution and a higher accuracy of the detection, which is an improvement to the traditional BP algorithm with slower convergence and tendency to over fitting.
出处
《世界地震工程》
CSCD
北大核心
2005年第2期52-56,共5页
World Earthquake Engineering
基金
北京市自然科学基金(8031001)重点资助项目(8031001)
