论文部分内容阅读
首先利用粗糙集理论对原始数据进行约简，并按一定的原则选取多个的简。然后在每一个约简的基础之上构建一个前馈子网络，并将多个子网综合成统一的容错前馈神经网络，达到对冗余信息的综合利用。通过对相应权值的训练调节，使网络的输出更精确合理。即，当某些量测信号丢失或难以获得时，可以通过其它不包含该量测信号的的简所构成的网络来进行正确的诊断，从而在信息不完备、不精确的情况下，仍保持较好的诊断性能。最后，通过对某液体火箭发动机泄漏故障检测的仿真，表明该容错网络可以满足高可靠性诊断场所的需要。
First of all, the rough set theory is used to reduce the original data and select more than one according to a certain principle. Then, a feedforward sub-network is constructed on the basis of each reduction, and a plurality of sub-networks are integrated into a unified fault-tolerant feedforward neural network to achieve the comprehensive utilization of redundant information. Through the training of the corresponding weight adjustment, the network output more accurate and reasonable. That is, when some of the measurement signals are lost or difficult to obtain, the correct diagnosis can be performed through other networks that do not contain the measurement signals, so that the information remains incomplete and inaccurate Better diagnostic performance. Finally, the simulation of leakage detection of a liquid rocket engine shows that the fault-tolerant network can meet the needs of high-reliability diagnosis sites.