Abstract: In accordance with the hardship to get real-time and on-line biology parameters in fermentation process，a soft sensor model based on Support Vector Machines（SVM） is established for estimating the biology parameters.It is well known that the model complexity and generalization performance of this Support Vector Regression（SVR） model depend on a good setting of the three parameters（?着，C，?酌）.In this article，an algorithm called Particle Swarm Optimization（PSO） is applied to optimize the parameters（?着，C，?酌） at the same time.Based on the proposed method，a PSO-SVR model is developed to estimate the products concentration of L-Asparaginase II.The results from fermenter control show that the PSO-SVR state estimation model has good learning accuracy and generalization performance so as to acquire the real-time and on-line estimation value of L-Asparaginase II products concentration.

Key words: Support Vector Regression（SVR）, state estimation, Particle Swarm Optimization（PSO） algorithm, L-Asparaginase II

摘要： 针对发酵过程中生物参数难以实时在线测量的问题，建立了用于生物参数状态预估的支持向量机软测量模型。考虑到该支持向量回归模型的复杂性和推广能力的好坏很大程度上取决于其3个参数（?着，C，?酌）能否取到最优值，采用粒子群算法实现对参数（?着，C，?酌）的同时寻优。在此基础上，以L-天冬酰胺酶II为对象，建立其基于PSO-SVR的发酵过程产物浓度状态预估模型。发酵罐控制结果表明：该模型具有很好的学习精度和泛化能力，可实现对L-天冬酰胺酶II产物浓度的实时在线预估。

关键词: 支持向量回归（SVR）, 状态预估, 粒子群优化（PSO）算法, L-天冬酰胺酶II