基于FTSVM的固态发酵预测模型的研究  

Study on Prediction Model of Solid State Fermentation Based on FTSVM

作　　者：陈树[1] 任召金 CHEN Shu;REN Zhaojin(School of IoT Engineering,Jiangnan University,Wuxi 214122)

机构地区：[1]江南大学物联网工程学院,无锡214122

出　　处：《计算机与数字工程》2018年第4期772-778,共7页Computer & Digital Engineering

基　　金：江苏省六大人才高峰基金项目(编号:2012-WLW-006)资助

摘　　要：固态发酵过程是一个相对封闭的过程,无法进行人工干预,工艺优化只能通过对历史发酵过程中的数据进行分析建模,进而指导后续的发酵工艺。论文根据发酵过程中的温度时间序列数据建立产量预测模型,整个过程包括特征预处理、特征选择和模糊孪生支持向量机训练三个部分。首先对单个样本数据集减去平均值并乘以权值以提高样本内数据的差异性,然后通过最小冗余最大相关算法选取最优特征子集,最后将特征向量输入到模糊孪生支持向量机进行训练,得到固态发酵产量的预测模型。实验结果表明,该模型能够提高预测的正确率,提高模糊孪生支持向量机对高维数据分类的性能,为后续的工艺优化提供了一定的决策依据。Solid-state fermentation is a relatively closed process,it can not be manual intervention,process optimization can only through historical fermentation process data analysis and modeling,and then guide the follow-up fermentation process.In this paper,the prediction model of yield is established according to the temperature and time series data in the fermentation process.The whole process includes three parts:feature preprocessing,feature selection and fuzzy twin support vector machine training.Firstly,the average value is subtracted from the single sample data set and multiplied by the weight to improve the difference of the data in the sample.Then,the optimal feature subset is selected by the minimum redundancy maximum correlation algorithm.Finally,the eigenvector is input to the fuzzy twin support vector machine to obtain the prediction model of solid state fermentation yield.Experimental results show that the proposed model can improve the prediction accuracy and improve the performance of fuzzy twin support vector machines(FTSVM)for high-dimensional data classification,and provide a basis for subsequent optimization.

关 键 词：特征选择 最小冗余最大相关 模糊孪生支持向量机 预测 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]