摘要
实现不易测量生物参量的在线估计是对发酵过程进行优化控制的关键。针对发酵过程中构建全局单一软测量模型适应性差的问题,提出了一种基于MPCA-GP的分阶段软测量建模方法。该方法对发酵过程的批次数据进行多向主元分析,以第一主元贡献率的变化趋势为指标实现阶段划分;通过判定测试样本T2统计量的最大后验概率,实现阶段识别;利用各个阶段的训练样本集分别建立基于高斯过程的分阶段软测量模型并实现测试样本的模型预估。实验结果表明所提方法具有较好的预测精度。
Realizing the online estimation of difficult-to-measure biological parameters is the key of the optimal con- trol in fermentation process. Aiming at the problem of poor adaptability of global single soft-sensor model in fermenta- tion process, a staged soft-sensor modeling method was developed based on MPCA-GP. The method performs multi- way principle component analysis on the batch data of the fermentation process ; and on this basis, the change trend of the first principal component contribution rate is used as the index to implement the phase partitioning. A method that judges the maximum posteriori probability of T2 statistic is used to realize the phase recognition. Various staged train- ing sample sets are adopted to establih the staged soft measurement models based on Gaussian process and the model forecast of the test samples is implemented. The experimental results show that the proposed method possesses good estimation accuracy.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第12期2703-2708,共6页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61240047)资助项目
关键词
高斯过程
软测量
分阶段建模
发酵过程
Gaussian process
soft-sensor
staged modeling
fermentation process