Abstract: It is necessary to monitor multiple load parameters during the operation of wet ball mill. However, the change of working condition will lead to the fact that the real time data and the modeling data do not satisfy the assumption of independent and identically distributed. In order to solve the deterioration of model performance caused by the change of probability distribution of historical data and real time data under multiple conditions, and the correlation between the load parameters cannot be fully considered by the traditional soft sensing model. The transfer learning strategy and multi-task learning mechanism is introduced to establish a soft sensing model of wet ball mill load parameters based on multi condition transfer learning. Firstly, the joint distribution adaptation is used to fit the marginal and conditional distribution in the dimension reduction process. Then, the mill load parameters are predicted by multi-task least squares supports vector machines. Experimental result indicates that the performance of the proposed method is superior or at least comparable with existing benchmarking methods, can solve the problem of soft sensor under multiple loading condition.

Key words: Multi-mode, transfer learning, wet ball mill load parameters, joint distribution adaptation, multi-task, least-squares support vector machine

摘要： 湿式球磨机运行过程中需要对多个负荷参数进行监测，然而运行工况改变会导致实时数据和建模数据的同分布假设不再成立。针对传统软测量方法不能考虑负荷参数之间的关联性，以及多工况情况下建模数据和实时数据概率分布变化引起的模型性能恶化问题，有针对性的引入迁移学习策略与多任务学习机制，建立一种基于多工况迁移学习的湿式球磨机负荷参数软测量模型。首先采用联合分布适配在降维过程中共同适配不同工况的边缘和条件分布，然后利用多任务最小二乘支持向量机方法对磨机负荷参数进行回归预测。实验结果表明，该软测量方法显著优于现有的方法，适用于多工况情况下的软测量建模。

关键词: 多工况, 迁移学习, 湿式球磨机负荷参数, 联合分布适配, 多任务, 最小二乘支持向量机