基于改进极限学习机的软测量建模方法张东娟.pdf

Computer Engineering and Applications 计算机工程与应用 2012 ，48 （20 ） 51 基于改进极限学习机的软测量建模方法 张东娟，丁煜函，刘国海，梅从立 ZHANG Dongjuan, DING Yuhan, LIU Guohai, MEI Congli 江苏大学 电气信息工程学院 自动化系，江苏 镇江 212013 Department of Automation, School of Electrical and Information Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China ZHANG Dongjuan, DING Yuhan, LIU Guohai, et al. Soft sensor modeling based on improved extreme learn- ing machine algorithm. Computer Engineering and Applications, 2012, 48 （20 ）：51-54. Abstract ：To solve the problem that biomass concentration is difficult to measure directly in the fermentation pro- cess, a soft sensor modeling method based on Improved Extreme Learning Machine （IELM ）is proposed. The least squares method is combined with the ELM algorithm to calculate the optimal learning parameters. And the training error is used as feedback input to improve the stability and prediction of ELM. In order to further improve the stabili- ty of the model, the Lanczos Bidiagonalization （LBD ）is used to calculate the output weights. The proposed modeling method is used to construct a novel soft sensor model for the erythromycin fermentation process. Compared with ELM 、IRLS-ELM and PL-ELM model, IELM model has higher prediction accuracy and stronger generalization ca- pability. Key words ：extreme learning machine; soft sensor; Lanczos bidiagonalization; fermentation process 摘 要：针对生物发酵过程中一些生物参量难以测量的问题，提出一种基于改进极限学习机（IELM ）的软测量 建模方法。该方法通过最小二乘方法和误差反馈原理计算出最优的网络输入到隐含层的学习参数，以提高模 型的稳定性和预测精度。通过双对角化方法计算出最优的输出权值，解决输出矩阵的病态问题，进一步提高 模型的稳定性。将所提方法应用于红霉素发酵过程生物量浓度的软测量。结果表明，与ELM 、PL-ELM 、 IRLS-ELM 软测量建模方法相比，IELM 在线软测量建模方法具有更高的预测精度和更强的泛化能力。 关键词：极限学习机；软测量；双对角化；发酵过程 文章编号：1002-8331（2012 ）20-0051-04 文献标识码：A 中图分类号：TP273 1 引言 的基本思想是根据某种最优准则，选择一组既与主 生物发酵过程是一个具有高度非线性和不确定 导变量有密切关系又容易测量的变量，通过构造某 性的复杂生化反应过程，生物量浓度是微生物发酵 种数学关系，来估计主导变量。 过程中的关键过程参数。由于受生物传感技术发展