摘要
软测量技术广泛用于流程工业中实时估计难以测量的关键变量,但其性能常常受限于标记样本缺乏、特征提取不当、单一模型性能不佳等问题。为此,提出了一种新的半监督集成学习软测量方法。该方法将隐特征提取、半监督学习、集成学习融合到同一建模框架下,实现了优势互补。首先,通过极限学习机自编码器(extreme learning machine auto-encoder,ELMAE)对过程数据进行多样性隐特征提取,进而建立多样性高斯过程回归(Gaussian process regression,GPR)基模型;然后,通过多学习器伪标记生成策略为每个基模型生成伪标记样本,进而扩充标记样本集;最后,利用扩充的标记样本集重新训练基模型后,对基模型进行集成,从而构建最终的软测量模型。将所提方法应用在金霉素发酵过程的基质浓度预测中,实验结果验证了所提方法的有效性和优越性。
Soft sensor technology has been widely used to estimate the key difficult-to-measure variables in the process industry.However,its performance is often limited by problems such as lack of labeled samples,improper feature extraction,and poor performance of the single model.Therefore,a new semi-supervised ensemble soft sensor is proposed,which integrates latent feature extraction,semi-supervised learning,and ensemble learning into the same modeling framework to achieve complementary advantages.Firstly,diverse latent features are extracted from process data by the extreme learning machine auto-encoder(ELMAE),and a set of diverse Gaussian process regression(GPR)base models are established.Then,to augment the limited labeled sample set,pseudo-labeled samples are generated for each base model by a multi-learner pseudo-label generation strategy.Finally,the base models are retrained based on the augmented labeled sample set,and the base models are integrated to build the final soft sensor model.The proposed method is applied to the prediction of substrate concentration in the process of chloromycin fermentation,and the experimental results verified the effectiveness and superiority of the proposed method.
作者
李友维
金怀平
杨彪
陈祥光
LI Youwei;JIN Huaiping;YANG Biao;CHEN Xiangguang(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Key Laboratory of Artificial Intelligence in Yunnan Province,Kunming University of Science and Technology,Kunming 650500,China;School of Chemistry and Chemical Engineering,Beijing Institute of Technology,Beijing 100081,China)
出处
《控制工程》
北大核心
2025年第4期653-663,共11页
Control Engineering of China
基金
国家自然科学基金资助项目(61863020,62163019)
云南省应用基础研究计划项目(202101AT070096)。
关键词
软测量方法
半监督学习
集成学习
极限学习机自编码器
伪标记
协同训练
Soft sensor
semi-supervised learning
ensemble learning
extreme learning machine auto-encoder
pseudo labeling
co-training
