基于未知输入观测器的非线性时间序列故障预报

Fault Prediction for Nonlinear Time Series Based on Unknown Input Observer

将未知非线性系统的输出作为时间序列并进行空间重构,针对得到的离散线性时变系统,提出了基于未知输入观测器的预测新方法.以实时拟合时间序列的线性AR模型作为时变系统的已知线性部分,将拟合误差作为时变系统的未知输入,实现了对非线性时间序列的一步预测.再利用递推预测的方法,将一步预测推广到N步预测,同时证明了该方法的预测误差有界.通过未知输入的预测值和状态的预测误差的变化可以方便地判断故障的发生,实现故障预报.仿真结果证明了方法的有效性.

The nonlinear time series, which is formed by the output of the unknown system, is converted into discrete time-varying dynamic system with space reconstruction. A novel method is presented for the prediction of time series which is achieved by the adaptive observation of system states. An AR model is used to approximate the linear part of the time series; the nonlinear part and the approximate error are regarded as the unknown-input of system. A one-step-ahead prediction method is proposed. The one-step-ahead prediction is then extended to N-steps-ahead moving horizon prediction, and the prediction error is proved bounded. The fault is predicted conveniently by the prediction of unknown-input and prediction error. The simulation results show the method is efficient.