论文部分内容阅读
提出一种新的基于神经网络多步时序预测的非线性系统故障诊断方法 .该方法先利用回归神经网络对多个传感器检测序列并行进行多步预测 ,再由多步预测序列和传感器检测序列生成历史残差序列和预测残差序列 .最后 ,根据统计定义的几个决策指标进行故障检测与诊断 .与其它方法相比 ,本文方法所需信息较少、可诊断的故障较多 .仿真表明该方法是有效的 ,可有效地增强故障信息、抑制非故障信息 .
A new fault diagnosis method for nonlinear system based on neural network multi-step sequence prediction is proposed. The method firstly uses multi-step predictive sequence and sensor detection sequence to generate multi-step prediction by using regression neural network. Historical residual sequence and prediction residual sequence.Finally, according to the statistical definition of several decision indicators for fault detection and diagnosis.Compared with other methods, this method requires less information and more faults can be diagnosed.Simulation shows that the The method is effective, which can effectively enhance the fault information and restrain the non-fault information.