摘要
针对化工过程强非线性和多工况的特性,提出了一种基于BP神经网络(BPNN)有效非线性融合多关联向量机(MRVM)的建模方法.首先选择不同的核函数,采用样本数据建立单一RVM子模型;然后利用BPNN的强非线性拟合能力,对各子模型的预测信息进行非线性融合,并采用人工鱼群算法(AFSA)对BPNN的初始权重和阈值进行优化;最终建立MRVM非线性融合模型.将该建模方法应用于甲醇制烯烃生产过程(MTO)乙烯收率预测研究中,研究结果表明:与单一RVM模型和最优加权组合模型相比,基于MRVM的非线性融合模型具有更佳的预测精度.
Aiming at the nonlinearity and multiple operating modes features of chemical process, a modeling method based on MRVM nonlinearly integrated by BPNN was proposed. Firstly, several RVM sub-models with the different kernel function were developed. Then pr information of sub-models was combined by BPNN, and the initial weight and thresho optimized by AFSA. Finally, the nonlinear fusion modeling of MRVM was develope lCtlve were The modeling method was applied to develop a soft sensor of ethylene yield in the MTO production. The results indicated that the soft sensor based on the modeling method had better prediction precision than single RVM model and optimum weighted combination model.
出处
《浙江工业大学学报》
CAS
北大核心
2017年第3期294-299,共6页
Journal of Zhejiang University of Technology
基金
国家自然科学基金资助项目(21676251)
