基于DBSVDD-RVR的多模态间歇过程质量变量在线软测量

DBSVDD-RVR based online soft sensing for quality variables in multimode batch processes

现有的多模态间歇过程软测量未考虑过程数据的批次差异及过渡模态的复杂时变特性,影响了间歇过程模态识别的合理性及质量变量在线软测量的准确性。提出了一种基于双边界支持向量数据描述-相关向量回归(double boundary support vector data description-relevance vector regression,DBSVDD-RVR)的间歇过程质量变量在线软测量方法。依据间歇过程离线模态划分获得的各稳定及过渡模态历史数据,建立DBSVDD在线模态识别模型,并引入滑动窗,构建间歇过程在线模态识别策略,利用DBSVDD模型实现在线测量数据的模态识别;在此基础上,构建了基于超球体距离的数据相似度计算方法,选择过渡模态在线数据的相似建模数据集,建立过渡模态的即时学习RVR软测量模型,并依据历史数据建立各稳定模态的RVR软测量模型,实现间歇过程质量变量的在线软测量。青霉素发酵过程的实验结果表明,所提方法有效地提高了间歇过程模态识别的合理性和质量变量在线软测量的准确性。

The existing multimode batch process soft sensor does not consider the batch difference of process data and the complex time-varying characteristics of transition modes,which affects the rationality of batch process mode identification and the accuracy of online soft sensing of quality variables.This paper proposes an online soft sensing method for batch process quality variables based on double boundary support vector data description-relevance vector regression(DBSVDD-RVR).According to the historical data of stable and transition modes obtained by offline mode partition of batch processes,the online mode identification model of DBSVDD was established.Then,the sliding window was introduced to construct the online mode identification strategy,and the online mode identification of batch data was realized by using DBSVDD model.On this basis,the data similarity calculation method based on hypersphere distance was constructed,and the similarity modeling data set of online data in transition mode was selected to establish the just-in-time learning RVR soft sensing model of transition mode.The RVR soft sensing model of each stable mode was established according to the historical data,and the online soft sensing of batch process quality variables was realized.The experimental results of penicillin fermentation process show that the proposed method effectively improves the rationality of mode identification and the accuracy of online soft sensing for quality variables in batch processes.