智能系统学报
智能繫統學報
지능계통학보
CAAI TRANSACTIONS ON INTELLIGENT SYSTEMS
2015年
1期
156-162
,共7页
高斯过程回归(GPR)%模糊C均值聚类(FCM)%Adaboost算法%L-赖氨酸%软测量%欧氏距离%隶属度%加权求和
高斯過程迴歸(GPR)%模糊C均值聚類(FCM)%Adaboost算法%L-賴氨痠%軟測量%歐氏距離%隸屬度%加權求和
고사과정회귀(GPR)%모호C균치취류(FCM)%Adaboost산법%L-뢰안산%연측량%구씨거리%대속도%가권구화
Gaussian process regression(GPR)%fuzzy C-mean clustering (FCM)%Adaboost algorithm%L-lysine%soft measurement%Euclidean distance%membership%weighted sum
为解决赖氨酸发酵过程中菌体浓度难以在线检测的难题，提出一种基于模糊C均值聚类（ FCM）与集成高斯过程回归（ GPR）的软测量建模方法。针对典型生物发酵过程可分为延滞期、指数生长期、稳定期、死亡期4个反应周期的特点，采用模糊C均值聚类算法对样本集进行聚类分析以形成若干子样本集；对每个子样本集分别采用高斯过程回归训练时，为提高GPR模型的泛化能力，利用Adaboost 算法提升GPR模型，分别在各子集建立集成GPR软测量子模型；采用欧氏距离计算新样本点对应于每一子模型的隶属度；加权求和获得最终的软测量模型的预测输出。基于氨基酸类典型菌种L－赖氨酸反应过程菌体浓度参数预测的试验研究表明：与全局单一GPR模型、集成GPR模型和基于FCM与多GPR模型相比，所建立的基于FCM与集成GPR软测量模型拟合精度高，泛化能力强，较好地满足了赖氨酸发酵过程的控制要求。
為解決賴氨痠髮酵過程中菌體濃度難以在線檢測的難題，提齣一種基于模糊C均值聚類（ FCM）與集成高斯過程迴歸（ GPR）的軟測量建模方法。針對典型生物髮酵過程可分為延滯期、指數生長期、穩定期、死亡期4箇反應週期的特點，採用模糊C均值聚類算法對樣本集進行聚類分析以形成若榦子樣本集；對每箇子樣本集分彆採用高斯過程迴歸訓練時，為提高GPR模型的汎化能力，利用Adaboost 算法提升GPR模型，分彆在各子集建立集成GPR軟測量子模型；採用歐氏距離計算新樣本點對應于每一子模型的隸屬度；加權求和穫得最終的軟測量模型的預測輸齣。基于氨基痠類典型菌種L－賴氨痠反應過程菌體濃度參數預測的試驗研究錶明：與全跼單一GPR模型、集成GPR模型和基于FCM與多GPR模型相比，所建立的基于FCM與集成GPR軟測量模型擬閤精度高，汎化能力彊，較好地滿足瞭賴氨痠髮酵過程的控製要求。
위해결뢰안산발효과정중균체농도난이재선검측적난제，제출일충기우모호C균치취류（ FCM）여집성고사과정회귀（ GPR）적연측량건모방법。침대전형생물발효과정가분위연체기、지수생장기、은정기、사망기4개반응주기적특점，채용모호C균치취류산법대양본집진행취류분석이형성약간자양본집；대매개자양본집분별채용고사과정회귀훈련시，위제고GPR모형적범화능력，이용Adaboost 산법제승GPR모형，분별재각자집건립집성GPR연측양자모형；채용구씨거리계산신양본점대응우매일자모형적대속도；가권구화획득최종적연측량모형적예측수출。기우안기산류전형균충L－뢰안산반응과정균체농도삼수예측적시험연구표명：여전국단일GPR모형、집성GPR모형화기우FCM여다GPR모형상비，소건립적기우FCM여집성GPR연측량모형의합정도고，범화능력강，교호지만족료뢰안산발효과정적공제요구。
In order to solve the problem that cell concentration is difficult to directly measure in the lysine fermenta-tion process, a kind of soft measurement modeling method is proposed on the basis of fuzzy C-mean clustering (FCM) and integrated Gaussian process regression (GPR) .The characteristics of typical biological fermentation process can be divided into 4 reaction cycles , including lag phase , exponential growth phase , stable phase , and dead phase .The cluster analysis is conducted for a sample set by applying fuzzy C-mean clustering algorithm , so as to form several sub-sample sets.In order to improve the generalization performance of the GPR , each group is trained through Gaussian Process Regression based on Adaboost and the corresponding integrated sub -models are established.The memberships between each new sample and each group are set as the weights through Euclidean distance and the predicted result is obtained by weighted sum by using typical bacterium of amino acid —L-lysine fermentation as an example .The simulation results showed that compared with the global single GPR model , inte-grated GPR model and the model based on FCM and multiple GPR , the soft measurement model based on integrat-ed GPR and FCM has high fitting precision .It also had strong generalization ability , which meets the control re-quirements of lysine fermentation process.
