大型工业球磨机是流程工业的重要基础设备，其关键产品质量指标磨矿粒度分布和负荷参数缺少在线检测方法，难以形成有效的闭环反馈控制，普遍存在产品质量不稳定和高能耗问题。软测量方法是测量上述参数的可行方案，但实际过程中的多模态特性会导致软测量模型失准。.本课题致力于引入迁移学习理论解决未建模未知模态下磨矿粒度分布测量和控制问题。重点研究多模态过程中建模与实时数据概率分布失配时基于域适应的软测量建模理论和方法。综合考虑概率和流形空间的模态特征，探明数据和模型高质量迁移的机制和条件，建立多源域融合迁移软测量模型在线更新机制，实现未知模态下磨机负荷参数的软测量。研究磨矿粒度分布的基函数表示及其权值与输入变量的非线性动态关系求解方法，获得磨矿粒度分布信息。在此基础上实现磨矿粒度的闭环多目标随机分布控制，提高产品质量和降低能耗。.本研究将为流程工业中广泛存在的多工况未知模态下参数的软测量和控制提供新的思路。

Large-scale ball mill is a crucial infrastructure in process industry. For the two key product quality indicators, particle size distribution and load parameters, it lacks online measurement methods. Accordingly, it is difficult to form effective closed loop feedback control, resulting in the common issues of unstable product quality and high energy consumption. Soft sensing is a feasible solution to measure these two key parameters mentioned above. In the actual industrial process, soft sensing model misalignment is likely to occur, due to multi modes..In this proposal, transfer learning theory is introduced, aiming at the measurement and control for particle size distribution under unmodeled and unknown mode. The core research is the soft sensing modeling theory and method using domain adaptation, under circumstance of mismatch of modeling and realtime data distribution in multimode processes. First, the conditions and mechanisms for high-quality transfer of data and models are studied, considering the modal features of probability and manifold space simultaneously. Second, on-line updating mechanism for multi-source domain fusion transfer soft sensing model is researched, to realize the mill load parameters soft measurement under unknown mode. Third, the basis function representation of particle size distribution and the nonlinear dynamic relation between basis function weights and input variables are researched, obtaining the particle size distribution information. Based on this, effective closed multi-objective stochastic distribution control can be performed, leading to the improvement of product quality and consumption reduction..This study will provide new insights for soft sensing and control for the parameters under multiple-condition and unknown-mode environment in process industry.