基于概率选择的主动学习智能软测量建模

Active Learning Intelligent Soft Sensor based on Probability Selection

针对复杂工业过程中有标签样本少,样本标记成本高等问题,提出一种基于概率选择的主动学习软测量建模方法。该方法首先采用主成分分析对无标签样本进行子空间集成;然后基于所有子学习器的输出,计算无标签样本的不确定度,从而对无标签样本进行信息评估,选取最有价值的样本进行人工标记;最后通过分析无标签样本和有标签样本的作用,引入训练集的性能指标完成对终止条件的设计。通过实际工业过程数据的应用仿真,验证了所提方法在减小标记代价的同时,还能够实现模型精度的提高。

Aiming at lack of tag samples and high cost of sampling tags in complex industrial processes, an active learning algorithm based on probability selection is proposed. Firstly, unlabeled samples are performed subspace integration by using the principal component analysis. Then, the information of unlabeled samples is evaluated by the uncertainty, which is calculated based on the out put of all sub learners. And the most valuable samples are selected to mark manually. Finally, the function of unlabeled samples and labeled samples are analyzed, and the termination conditions are designed by introducing the performance index of training set. Through simulations of industrial processes data, it is verified that the proposed method can improve the accuracy of the model while reducing the cost of marking.