改进的FCM-LSSVM青霉素发酵过程预测建模

发布时间：2018-01-05 00:34

本文关键词：改进的FCM-LSSVM青霉素发酵过程预测建模　出处：《控制工程》2017年11期 　论文类型：期刊论文

【摘要】：针对青霉素发酵过程周期长,每个阶段表现出不同的特性,最小二乘支持向量机(least squares support vector machines,LSSVM)的全局模型预测精度难以保证的问题,提出了改进的基于模糊C均值聚类(fuzzy C-means clustering algorithm,FCM)和LSSVM的青霉素发酵过程分段建模方法。首先,在分析影响青霉素产物浓度相关因素的基础上选取输入变量,对样本数据采用FCM算法聚类,按照最大隶属度将样本归类为稳定过程或过渡过程;然后,分别为稳定过程的4个阶段和过渡过程的3个阶段分别建立LSSVM子模型,最后通过子模型切换策略得到系统输出。利用Pensim仿真平台数据,将提出的方法与FCM-LSSVM和LSSVM方法进行比较,平均绝对误差分别为0.013 2、0.014 3、0.014 9,均方根误差分别为0.017 8、0.019 2、0.021 6,实验结果表明,所提出的方法具有良好的精度和泛化能力。
[Abstract]:In view of the penicillin fermentation process cycle is long, each stage shows different characteristics. The least squares support vector machine (LS-SVM) is least squares support vector machines. LSSVM) global model prediction accuracy is difficult to ensure. An improved fuzzy C-means clustering algorithm based on fuzzy C-means clustering is proposed. Firstly, based on the analysis of the factors related to the concentration of penicillin products, the input variables were selected, and the sample data were clustered by FCM algorithm. According to the maximum degree of membership, the sample is classified as a stable process or a transition process; Then, the LSSVM submodel is established for the four stages of the stabilization process and the three stages of the transition process respectively. Finally, the output of the system is obtained by submodel switching strategy. Using the data of Pensim simulation platform, the proposed method is compared with FCM-LSSVM and LSSVM methods. The mean absolute error and root mean square error were 0.013 2 0. 014 3 0. 014 9 and 0. 017 8 0. 019 2 卤0. 021 6, respectively. The experimental results show that the proposed method has good accuracy and generalization ability.
【作者单位】：宜春学院物理科学与工程技术学院;
【基金】：国家自然科学基金(51366013)
【分类号】：TP18;TQ927
【正文快照】： 青霉素发酵过程具有不确定性、非线性和严重的时变性等特点,且发酵过程机理复杂,数学模型难以建立,因此,一般采用人工神经网络[1,2]、支持向量机[3](support vector machines,SVM)等数据驱动方法建立经验模型。神经网络建模存在训练样本量大,网络结构难确定,易陷入局部极值,收

【相似文献】