助训练框架下的半监督软测量建模方法

A semi-supervised soft sensor modeling method under Help-training framework

为了充分利用工业过程中大量无标签样本信息,并减少过程的不确定因素对无标签样本质量的影响,提出一种助训练框架下的半监督孪生支持向量回归软测量建模方法。采用孪生支持向量回归机构建主学习器,对高置信度无标签样本添加伪标签;同时,基于K近邻算法构建辅学习器,最大化学习器在近邻样本集上的均方误差,经过此项指标筛选后的待处理样本集包含了更多的数据信息;主、辅学习器二者相辅相成,一定程度上提高了模型的泛化性;再利用所构建的助训练框架提高样本利用率后得到预测模型,实现对无标签样本信息的充分挖掘。通过对脱丁烷塔工业过程中的实际数据进行建模仿真,所得结果表明此模型具有良好的预测性能。

In order to make full use of a large number of unlabeled sample information in industrial processes and reduce the impact of process uncertainties on the quality of unlabeled samples,a semi-supervised soft sensor modeling method under the help-training framework is proposed.The twin support vector regression is used to build the main learner and add pseudo labels to the unlabeled samples with highest confidence;At the same time,the auxiliary learner is constructed based on the k-nearest neighbor algorithm to maximize the root mean square error of the learner on the nearest neighbor sample set.The candidate sample set screened by this index contains more data information;The main and auxiliary learners complement each other,which improves the generalization of the model to a certain extent;Then,the prediction model is obtained by using the help-training framework to improve the sample utilization,so as to fully mine the unlabeled sample information.Through the modeling and simulation of the real data in the industrial process of debutanizer,the results show that the model has good prediction performance.