【摘要】 提出了一种基于多分支BP神经网络建立气动肌肉位移/气压迟滞模型的新方法。首先,搭建气动肌肉位移/气压迟滞特性测试系统,得到气动肌肉位移/气压迟滞曲线;然后分别采用传统BP神经网络、多分支BP神经网络和Prandtl-Ishlinskii模型对气动肌肉的位移/气压迟滞开展建模研究;最后通过比较分析发现,采用多分支BP神经网络方法能有效避免传统BP神经网络训练过程中的过拟合现象,且建模精度明显优于传统的PrandtlIshlinskii模型;多分支BP神经网络的平均误差、均方差与最大误差相较于Prandtl-Ishlinskii模型减小了87.45%,86.68%与74.73%。更多还原

【Abstract】 A novel modeling method based on multi-branch BP neural network is proposed to describe the displacement/pressure hysteresis of pneumatic muscle. Firstly,the displacement/pressure hysteresis characteristic test system was built to obtain the displacement/pressure hysteresis loops of pneumatic muscle. Then,the classical BP neural network,multi-branch BP neural network and Prandtl-Ishlinskii model are used to fit the hysteresis loop of pneumatic muscle,respectively. Finally,the comparative study found that the multi-branch BP neural network can effectively avoid the over-fitting phenomenon in the fitting process of classical BP neural network,and the modeling capacity is obviously better than the traditional Prandtl-Ishlinskii model. Compared with Prandtl-Ishlinskii model,the mean average,mean square and maximum errors of multi-branch BP neural network are reduced by 87. 45%,86. 68% and 74. 73%.更多还原