摘要
针对神经网络逆系统方法实现复杂非线性系统解耦存在训练时间长、实时控制较差的缺陷,提出一种改进的RELM(正则极限学习机)训练算法,根据输出权值的特点,采用不带平方根的乔累斯基分解,提高了计算效率,减少了训练时间,具有较高的学习精度及更好的泛化能力;进一步将此神经网络应用到3输入3输出多变量离散系统的解耦控制,仿真实验结果表明,所提出的方法具有较快的实时控制速度,具有较高的实用价值.
An improved RELM learning algorithm is prospected to solve the long-time training and the poor real-time control in complex nonlinear decoupling control based on neural network inverse method.The RELM with higher learning accuracy and better generalization ability can improve computational efficiency and reduce training time by use of cholesky factorization without square root based on the feature of output weights.And the method is further applied in decoupling control of multivariate discrete system with 3-input and 3-output.Simulation results demonstrate that the proposed method has faster real-time control speed,and has very high practical value.
出处
《微电子学与计算机》
CSCD
北大核心
2012年第11期70-73,共4页
Microelectronics & Computer
基金
国家自然科学基金项目(51077066)
江苏省自然科学基金项目(BK2011319)
苏州市职业大学创新基金资助项目(2011SZDCC01)
