基于图像与过程数据融合的回转窑产品质量参数预报建模

项目名称： 基于图像与过程数据融合的回转窑产品质量参数预报建模

项目编号： No.61273177

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 周晓杰

作者单位： 东北大学

项目金额： 81万元

中文摘要： 高耗能设备大型回转窑工艺机理复杂，产品质量参数无法在线测量。由于难以建立精确机理模型，回转窑产品质量参数预报方法还主要是基于慢采样速率输入输出数据的静态模型方法，由于太大的采样间隔，导致过程动态信息丧失、模型预报精度较低，难以满足回转窑过程的实时监控、自动控制及实时操作优化调整的需求。 本项目在现有的静态模型研究基础上，引入多视图学习机制，充分利用回转窑火焰图像和过程数据中蕴含的与产品质量参数相关的丰富的动态信息和多源信息，并引入半监督学习机制，利用大量的未标记数据来改善学习性能，研究提出图像和过程变量动态时间序列多源高维特征提取与选择方法和基于多视图半监督集成学习的产品质量参数预报模型方法，进一步提高预报模型性能。以典型的回转窑为背景开展建模实验研究，并研发模型软件开展工业应用实验。本项目对于实现回转窑过程的实时监控、推理控制及实时操作优化至关重要，具有重要的理论研究意义和实际应用价值。

中文关键词： 回转窑；产品质量参数预报；软测量；多信息融合；特征提取

英文摘要： Large-scale rotary kilns, which consume a lot of energy, have complex process mechanism and the product quality variables are difficult to measure online. Increased complexity of the process dynamics prevents one from building accurate first-principle models. The static modeling approach based on slow-rate inputs/outputs are mainly used to build the prediction models of the product quality variables. In this case, process dynamics may be lost because of the large sampling intervals and the model is not accurate. The static model is hard to meet the demands of real-time monitoring, automatic control and real-time optimization in rotary kiln processes. On the basis of the existing static models, the project will make full use of the dynamic information and multiple source information contained in rotary kiln flame images and process data by introducing multi-view learning mechanism. Moreover, the work will make use of the unlabeled data to improve the performance of the prediction model by applying semi-supervised learning mechanism. This work aims to develop a novel feature extraction and selection method for dynamic time series data from multiple sources and of high dimension and to develop a novel modeling approach for predicting product quality variables based on multi-view learning, semi-supervised learning

英文关键词： rotary kiln；product quality prediction；soft sensing；information fusion；feature extraction