中文摘要：

提出了采用贝叶斯推理模型BIM（Bayesianinferringmodel）对时变非线性系统的输出进行在线监测的实现思路和方法．首先描述了时变非线性系统的在线输出监测问题．然后介绍了BIM结构和训练方法，BIM的特点在于训练样本完全采自于在线闭环系统，采用改进的觅食优化算法IEFOA（ImprovedE．ColiForagingOptimizationAlgorithm）离线训练门槛矩阵参数D．而在线预测应用时，采用滑动窗口数据实时更新BIM结构，从而实时跟踪系统的输出变化．最后，利用时变非线性对象对BIM的在线观测能力进行了验证，仿真结果表明BIM适合于系统的输出监测，并且具有设计简单、跟踪性能好等优点，为非线性系统的性能评估提供了一种新的底层数据预测方法．

英文摘要：

The implementation idea and solution are proposed in this article for the output on-line monitoring of the time- variant nonlinear system by using bayesian inferring model （BIM）. Firstly, the on-line monitoring problem of nonlinear system is described. Then the BIM structure and training methods are introduced. The characteristics of the BIM include that the sample data for off-line training are from the closed loop system and the optimization algorithm for the threshold matrix D is selected as the improved foraging optimization algorithm （ IEFOA ）. While in the on-line applications, the sliding window data are used to update the structure of the BIM for the on-line tracing of the system output. The time-va- riant nonlinear object is employed to validate the on-line monitoring ability of the BIM. The simulation results indicate that the BIM is adapted to the system on-line output monitoring and owns the characteristics of easy design, high accuracy tracing ability and etc, which provide a kind of data prediction method for the lowest system.