摘要
针对软测量建模过程中模型存在失效问题,提出了一种基于KFCM和AMDE-LSSVM多模型的软测量建模方法;首先,采用核模糊C均值聚类(Kernel-based fuzzy c-means algorithm,KFCM)对训练样本数据进行划分,然后利用最小二乘支持向量机(least squares vector machina,LS-SVM)对每个聚类建立子模型,并使用自适应变异差分进化算法(Adaptive Mutation different evolution,AMDE)对最小二乘向量机中的径向基宽度和惩罚系数进行寻优;将提出的算法用于秸秆发酵关键参数乙醇浓度、基质浓度(总糖浓度)、菌体浓度检测中,通过软测量建模得到的预测值与离线化验值进行对比,证明方法的有效性;实验结果表明,改进后的算法克服了差分进化算法中容易陷入局部最优,早熟收敛的现象;建立的新模型相比单一模型,乙醇浓度、基质浓度(总糖浓度)、菌体浓度测量误差分别为0.64%,1.85%和0.75%,具有更好地适应秸秆发酵过程、提高测量精度的优势。
Aiming at the problem of failure in soft-sensing model,a multiple-model soft-sensing modeling method was proposed.Separating a whole training data several clusters with different centers by KFCM,each subset was trained by LS-SVM.In the training process,AMDE algorithm was used to optimize the parameters of the LS-SVM.The proposed algorithm is applied to the key parameters of straw fermentation,such as ethanol concentration,matrix concentration（total sugar concentration）and cell concentration detection.The predicted values obtained by soft sensor modeling are compared with off-line test values,which proves the effectiveness of the method.The experimental results show that the improved algorithm overcame the phenomenon that DE algorithm is easy to fall into the local optimum and premature convergence.Compared with the single model,the measurement errors of the ethanol concentration,the matrix concentration（total sugar concentration）and the cell density in the new model were respectively 1.54%,1.05% and 0.85%,indicating the new model can better adapt to the straw fermentation process and improve the detection accuracy.
作者
姜哲宇
刘元清
朱湘临
王博
Jiang Zheyu;Liu Yuanqing;Zhu Xianglin;Wang Bo(School of Electrical and Information Engineering,Jiangsu University,Zhenjiang 212013,China)
出处
《计算机测量与控制》
2018年第8期46-50,92,共6页
Computer Measurement &Control
关键词
自适应变异差分进化算法
核模糊C均值聚类
最小二乘向量机
秸秆发酵
adaptive mutation differential evolution algorithm
fuzzy kernel C clustering
least squares vector machine
straw fermentation
