基于组合模型的主元分析预测监控方法

传统的基于主元分析（PCA）的过程监控技术可以对工业过程当前的状况进行监控，但难以预测系统未来的运行情况。本文在PCA监控方法的基础上建立预测模型，首先根据历史数据建立PCA的综合监控统计量模型，其次结合k邻近（k-NN）的最小二乘支持向量机（LSSVM）和灰色理论（GM（1,1））技术建立在线组合预测模型，实现对工业过程运行状态的预测。利用组合模型对一个多变量动态过程实例进行仿真，将预测效果与k-NN的LSSVM以及GM(1,1)各自单独的预测效果作比较，验证了所提方法在长期预测方面的准确性，表明基于组合模型的PCA预测监控方法特别适用于对缓慢漂移故障的长期预测。

Long-term prediction monitoring by principal component analysis（PCA） based on hybrid models

Traditional principal component analysis (PCA)-based process monitoring technologies focus on the current states of industrial processes rather than their future behavior. Motivated by this observation, this paper introduces a PCA-based predictive model which implements prediction of the operating states of industrial processes. Monitoring models of a PCA combined statistical index are first established based on normal historical data for the process. A k-NN least squares support vector machine (LSSVM) combined with GM (1,1) is then adopted to establish a on-line combinatorial prediction model, which allows forecasting of future states of the industrial process. Simulations on a multi-variables dynamic process demonstrated that the proposed method is more accurate than either k-NN based LSSVM or GM(1,1) in long-term prediction monitoring. The results of this work prove that PCA based on hybrid models is especially suitable for the long-term prediction of slowly drifting faults.