摘要
从神经网络的基本方程出发，在理论、数值计算方面对弹性力学神经计算的稳定时间和计算精度进行了研究。理论分析表明：偏置电容Ｃ、信号网络神经元的放大倍数β影响网络的稳定时间；偏置电阻Ｒ、约束网络神经元的惩罚系数α影响求解精度，但在工程上没有必要对Ｒ，Ｃ，β，α等进行严格的限制。数值模拟结果也证实了理论分析的结论。
On the basic of dynamic equations of neural networks, the stabilizing time and the computing precision in the neurocomputation of elastic problem are analyzed bath theoretically and numerically. Clearly, there are effects of capacity C, amplifying gain β on the stabilizing time, and the resistance R, penalty coefficient on the precision. However, it is no necessary to constrain the R, C, β, strictly when it is applied to engineering. The conclusions of numerical simulation are well with that of theoretical analysis.
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2000年第3期22-26,共5页
Journal of Mechanical Engineering
基金
国家自然科学基金
四川省跨世纪杰出青年学科带头人培养基金
关键词
有限元
神经网络
稳定时间
误差特性
结构分析
Finite element Neural networks Stabilizing time Error characteristic Numerical simulation
