摘要
最小二乘支持向量机(1east square support vector machine,LSSVM)是1种新的机器学习算法,它采用结构风险最小化准则,能有效提高模型的泛化能力,且具有运算速度快、抗噪能力强等特点。本文针对最小二乘支持向量机发酵建模中,选择重要模型参数值的问题,提出利用全局搜索能力强的量子粒子群优化算法,优化LSSVM建模过程的重要参数,并将该混合建模方法应用于L-缬氨酸发酵,建立L-缬氨酸产物浓度、菌体浓度和底物浓度等重要过程变量的预测模型,在线预估这些不能在线测量的生化状态变量。实验表明,混合算法所建立起的L-缬氨酸发酵模型在模拟菌体生长、底物消耗及发酵产酸过程的变化等方面都比BP神经网络建模方法具有较小的拟合误差和较好的推广性能,可以为L-缬氨酸发酵生产过程提供动态模拟,具有重要的实用价值。
Least square support vector machine （LSSVM） with rapidly running and strongly anti-noise performance is a new machine learning algorithm, employing the criteria of structural risk minimization, which can improve the generalization of model. To overcome the disadvantage that it＇s difficult to get better parameter values in fermentation modeling of least square support vector machine, a method was proposed to find the better parameter values by quantum-behaved particle swarm optimization which has the better search ability. Some biochemical state variables in L-valine fermentation process which can not be measured on-line are pre-estimated by this hybrid modeling method. The results compared with from which the BP neural network models demonstrate that the models constructed by the proposed QPSO-LSSVM method can accurately predict the time course of cell growth, glucose consumption, and L-valine production during the fermentation. The models can simulate the dynamics in L-valine fermentation well. They have a practical application merit for the L-valine fermentation process modeling.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2010年第9期1221-1224,共4页
Computers and Applied Chemistry
基金
安徽省高校优秀人才基金资助项目(2006jq1244)
关键词
量子粒子群优化算法
最小二乘支持向量机
L-缬氨酸
建模
quantum-behaved particle swarm optimization, least square support vector machine, L-valine, modeling
