基于混合优化神经网络的机器人力/位置控制

Hybrid force/position control based on mixed optimal neural networks

针对机器人在参数变化和外界工作环境的刚度变化时,系统的控制质量会因常规PID控制器没有自适应能力而明显变差,甚至无法工作,提出了一种具有混合H2/H∞性能指标的CMAC控制方法,采用CMAC神经网络加强系统对参数不确定性的补偿,引入混合优化策略来优化CMAC神经网络的结构和权值,保证了系统对外界干扰在给定干扰衰减度下具有鲁棒稳定性的同时,还能使系统达到良好的动态性能,满足一定的H2最优性能指标。仿真结果表明,本文所提控制方案在大量参数不确定性及外部扰动存在的情况下,仍能满足性能要求。

The design of a neural networks control with mixed H2/H∞ performance was manipulators force/position control. The mixed H2/H∞ tracking performance ensures the robust stability under a prescribed attenuation level for external disturbance, and the H2optimal tracking can be also achieved. Neural networks were introduced to improve the system＇s performance under parameters uncertainties. Mixed optimal strategy was introduced to optimize neural networks＇ structure and weight. The simulation shows that it is an effective method. And the control method can get better performance even when system is under large modeling uncertainties and external disturbances.