基于互信息的多块k近邻故障监测及诊断

Multiblock k-nearest neighbor fault monitoring and diagnosis based on mutual information

由于传统的k近邻故障监测不考虑过程的局部信息,只建立一个全局模型,因此提出一种基于互信息的多块k近邻故障监测方法。首先,考虑建模数据的非线性和非高斯等特性,基于变量间的互信息进行子块构建;然后,利用k近邻方法对每个子块进行建模与监测,子块中的k近邻模型反映了更多的过程局部特征;最后,将所有子块的监测结果通过贝叶斯推断方法进行融合,并采用基于马氏距离的故障诊断方法辨识故障源。通过对田纳西−伊斯曼过程和高炉炼铁过程中的应用仿真,监测结果表明所提方法的可行性和有效性。

The traditional k-nearest neighbor(kNN)fault monitoring does not take into account the process of local information and only builds a global model.Thus,a multi-block kNN fault monitoring algorithm based on mutual information is proposed.First,with the nonlinear and non-Gaussian characteristics of the modeled data taken into consideration,subblocks are constructed based on mutual information between variables.Then,the kNN algorithm is used to model and monitor each subblock,in which the kNN model reflects more local characteristics of the process.Lastly,the monitoring results of all subblocks are fused by the Bayesian inference method,and a fault diagnosis method based on Mahalanobis distance is used to identify the source of faults.Through the application simulation in the Tennessee Eastman process and the blast furnace ironmaking process,the monitoring results show the feasibility and effectiveness of the proposed method.