粗糙分类器的多模型软测量建模方法

Multiple models soft-sensing technique based on rough classifier

根据多模型可以改善模型估计精度,提高泛化性的思想,提出了1种粗糙分类器的多模型软测量建模方法。该方法采用聚类、分类相结合的方式对数据进行分组训练,在一定程度上消除了矛盾样本点可能对模型精度造成的影响。对各组样本利用支持向量回归机建立回归子模型,得到多模型软测量系统。同时,通过向粗糙集引入相似度作为评价样本间相似性的指标,解决了传统粗糙集无法识别训练样本集中未出现过的模式的问题。通过引入概率测度,利用概率公式作为粗糙集分类的决策规则,简化了算法。基于上述理论构造的粗糙分类器,有效地提高了分类器的分类精度,确保了各子模型的估计精度。将该方法应用于双酚A生产过程的质量指标软测量建模,仿真结果表明了该算法的有效性。

According to the idea that multi-models can improve the estimated accuracy and generalization,a soft-sensing method with multiple models based on rough classifier is presented.The training data set is separated into several classes according to clustering and classification so that the impact of conflicting samples to model can be removed.Support vector machine is used for building regression model in subclass,and finally obtain the soft-sensing multiple models.Meanwhile,similarity degree introduced to rough set for evaluation index of sample similarity,it solves the problem of conventional rough set can hat recognize patterns who never appear in training set before.By introducing probability measure into rough set,and subsequently probability formula is used as decision rule of rough classifier,therefore,simplified algorithm.The constructed rough classifier based on the theory above,precision of classifier has a remarkably improve.It ensures the estimated precision of sub models.The proposed algorithm is used in a soft-sensor model for the bisphenol -A productive process,and the result of simulation shows the effectiveness of the algorithm.