Abstract: The paper summarizes research progress of the multi-model fusion modeling method for process industries soft senor. According to the difference of sub-models，the multi-model fusion modeling method can be divided into data driven fusion modeling method and semi-parametric modeling method. The design ideas and research status of the data driven fusion modeling method and semi-parametric modeling method are presented，their advantages and disadvantages are analyzed，and corresponding improvement directions are proposed. According to different data processing methods，the data driven fusion modeling method can be divided into ensemble learning and cluster analysis. According to different types of model structures，semi-parametric modeling method is divided into serial and parallel structure. In the end，the future research directions of multi-model fusion modeling are presented. It is expected that breakthrough can be made in improvement of data driven models fusion technology，advancement of semi-parametric models generalization ability，and solution of dual-rate sampling. Developing soft sensor models based on multi-source information fusion by using the multi-model fusion modeling method is an effective way to realize online estimation of variables which are difficult to measure in process industries.

Key words: soft sensor, multi-model fusion, data driven fusion, semi-parametric, modeling, soft sensor, multi-model fusion, data driven fusion, semi-parametric, modeling

摘要： 对多模型融合建模方法在过程工业软测量中的研究进展进行了系统总结。根据整体模型中子模型的不同，多模型融合建模方法主要可分成数据驱动融合建模方法和半参数建模方法。详细介绍了数据驱动融合建模方法和半参数建模方法的设计思想和国内外研究现状，分析了各类方法的优缺点，并提出了相应的改进方向。根据过程数据处理方法的不同，将数据驱动融合建模方法分为集成学习和聚类分析。根据模型结构形式的不同，将半参数建模方法分为串联结构和并联结构。最后对多模型融合建模方法的未来研究方向进行了展望，期望今后的研究工作能在改进数据驱动模型融合技术、提高半参数模型外推能力和解决双率数据问题等方面取得突破性进展，并指出采用多模型融合建模方法建立基于多源信息融合的软测量模型是实现过程工业中难测变量在线估计的有效方法。

关键词: 软测量, 多模型融合, 数据驱动融合, 半参数, 建模, 软测量, 多模型融合, 数据驱动融合, 半参数, 建模