摘要
为了减少计算量,提高故障检测准确性,提出一种基于改进K-means聚类的kNN故障检测方法.首先通过改进K-means聚类将原始建模数据分成C个类,然后利用kNN分别对每个类建立模型.在每类的训练数据集合中找到每个样本的k个最近邻,计算k个最近邻距离的平方和.基于训练数据确定进行故障检测的阈值.对新的待检测样本,先判断属于哪一类,然后应用对应类的kNN模型进行故障检测.仿真结果表明:该方法不但可以提高过程故障检测的可靠性,而且大大缩短了故障检测时间.
In order to reduce computation and improve the accuracy of fault detection, a new kNN fault detection method based on the improved K-means clustering is proposed. First the original modeling data is divided into C clusters according to the batch through improved K-means. And then the model for each cluster is builded by kNN. In each training data cluster, the k-nearest neighbors for each sample are found,and the squared distances are calculated. The threshold for fault detection based on the training data is determined. For an incoming new batch, we first determine which cluster it belongs to, and then apply the kNN model of the corresponding clusters for fault detection. Simulation results show that the proposed method not only improves the reliability of fault detection, but also greatly reduces the time.
出处
《沈阳化工大学学报》
CAS
2013年第1期69-73,共5页
Journal of Shenyang University of Chemical Technology
基金
国家自然科学基金资助项目(61034006
61174119)
