摘要
改进一种基于瞬时最优控制的神经网络训练算法。本方法以瞬间最优控制价值函数最小化为训练目标,考虑了地震输入的能量,利用最速下降梯度法计算权值的改变量,并对敏感度矩阵进行近似处理,可解决神经网络控制中神经网络控制器难以获得的训练输入/输出样本对的难题。该方法适合多输入/多输出结构体系,整个推导过程都是针对此体系进行的。文中通过对一个三层框架结构体系进行有效的仿真计算,说明了算法的有效性。
The neural network training algorithm is improved, which is based on instantaneous optimal control method, taking into consideration of the energy of earthquake. The training object of the algorithm is minimization of the cost function in optimal control, variations of the weight values are obtained by the steepest descend gradient method and moreover an approximate treatment is made on the sensitivity matrix evaluation. The method can overcome the difficulties of obtaining the training input/output samples for neuro-controller in neural network control. The algorithm can be used in multiple input multiple output (MIMO) system, and the simulation of active control of a three-story frame structure subjected to ground excitation shows that the proposed method is effective.
出处
《振动与冲击》
EI
CSCD
北大核心
2005年第2期5-8,共4页
Journal of Vibration and Shock
基金
国家自然科学基金(10372084)
陕西省自然科学基金(2002A17)
大连理工大学工业装备结构分析国家重点实验室开发基金资助项目
关键词
人工神经网络
瞬时最优控制
主动控制
Algorithms
Construction
Control equipment
Earthquakes
Gradient methods
Neural networks
Optimal control systems
Structural frames
