优化极限学习机算法及其在纸张横幅定量系统解耦中的应用

Optimization of Extreme Learning Machine and Application in Decoupling of the Cross-direction Basis Weight System for Papermaking

本课题基于奇异非混沌优化(SNO)改进了极限学习机(ELM),并用于解决纸机横幅(CD)定量系统的耦合问题。首先,采用基于分段逻辑映射的SNO方法,对输入层和隐藏层之间随机生成的权重和阈值进行优化,解决了ELM优化不足的缺点。然后,设计奇异非混沌优化极限学习机(SNOELM)解耦器,对多变量系统进行解耦。最后,将其与已提出的改进ELM、鲸鱼优化极限学习机(WOELM)和粒子群优化极限学习机(PSOELM)进行了比较。仿真结果表明,SNOELM解耦方法比ELM具有更好的优化能力,比WOELM和PSOELM具有更高的解耦精度和更快的解耦速度。

In this paper,the extreme learning machine(ELM)was improved based on strange nonchaotic optimization(SNO)and used to solve the coupling problem of cross-direction(CD)basis weight system.Firstly,the SNO based on a piecewise logistic map was used to optimize the ran-domly generated weights and thresholds between the input layer and the hidden layer,which solved the disadvantage of insufficient optimization for ELM.Then,SNO extreme leaming machine(SNOELM)decouplers were designed to decouple the multivariable system.Finally,it was compared with the improved extreme learning machine,whale optimization extreme learning machine(WOELM)and particle swarm optimization extreme leaming machine(PSOELM).Simulation results demonstrated that the SNOELM decoupling method had better optimization ability than ELM and had higher decoupling accuracy and faster decoupling speed than WOELM and PSOELM.