采用 KPCA-BP 神经网络的并联机构全局综合性能评价方法研究

摘要

基于并联机构单一性能指标的多样性和各单一性能指标之间的非线性关系，将核主成分分析（ Kernel Principal Component Analysis，KPCA）方法与误差反传播（ Back Propagation，BP）神经网络技术相结合，建立一个对可完成不同工作任务的并联机构进行综合性能评价的KPCA-BP神经网络模型。通过合理的系统抽样，利用KPCA对BP神经网络的输入数据进行预处理，既能处理各单一性能指标间的非线性关系，又能简化BP神经网络结构，加快网络学习速度，提高预测精度，进而提出一种基于多种单一性能指标的并联机构全局综合性能评价新方法，为并联机构工作任务优序关系研究提供科学的参考依据。%Due to the diversity and nonlinear correlation of the parallel mechanism’ s single performance indexes,Kernel Principal Component Analysis ( KPCA) and Back Propagation( BP) neural network technology could be applied to construct KPCA-BP neu-ral network model for comprehensive performance evaluation of parallel mechanism.Through proper sampling,the input data of BP neural network are dealt by KPCA in advance, then the nonlinear relationships among single performance indexes could be re-vealed,and BP neural network could always be simplified to accelerate learning speed and improve forecast accuracy.Further-more,a new global comprehensive performance evaluation method of parallel mechanism is proposed based on various performance indexes,providing scientific basis for optimum order of parallel mechanism’ s working tasks.