多重核学习非线性时间序列故障预报

Nonlinear time series fault prediction by multiple kernel-learning

针对非线性时间序列故障预报问题,提出了多重核学习故障预报方法.利用多重核学习可以减少支持向量的个数,提高预测性能.而且在多重核学习定义的混合核空间中运用减聚类能够提取正常原型.最后,将本文提出的方法应用于连续搅拌釜式反应器的故障预报,仿真结果表明该方法能够提高故障预报的准确性与实时性.

A novel fault prediction method based on multiple kernel-learning is proposed for fault prediction in nonlinear time series. In the support-vector regression, the multiple kernel-learning will reduce the number of support vectors, and improve the performance of the prediction model. Furthermore, the normal prototypes could be extracted by conducting subtractive clustering on the mixed kernel space defined by multiple kernel-learning. The proposed method is applied to a continuous stirred-tank reactor（CSTR） for fault prediction. Simulation results indicate that this method predicts faults quickly and accurately.