基于径向基函数神经网络预测器的多传感器故障在线检测和信号恢复的研究

APPLICATION OF RADIAL BASIS FUNCTION NEURAL NETWORK PREDICTOR IN SENSOR FAULT ON-LINE DIAGNOSIS AND SIGNAL RECOVERY

理论研究及大量实践表明:径向基函数神经网络具有较强的函数逼近能力,学习速度优于常用的BP网络。本文利用径向基神经网络构成传感器输出预测器实现了多传感器故障在线检测和信号恢复。文中阐述了预测器的构成及其在线学习算法。通过仿真研究证明:该预测器对传感器输出具有很好的在线预测、跟踪能力。当某传感器发生故障时,在及时准确地发出报警信号的同时,对瞬时故障,能很好地恢复故障期间传感器正常的输出,即消除瞬时故障对系统正常运行的影响;对长期故障,能在故障发生后一定的时间范围内,正确估计出传感器正常输出,以保证系统的正常运行。

In comparison with the BP network, the radial basis function neural network is superior in capability of function approximation, classification and convergence rate. In this paper a predictor based on radial basis function neural network for multisensor's fault detection and signal recovery is proposed. The principle of the predictor and its on-line algorithm are addressed. The simulation results demonstrate that the proposed predictor has nice ability of sensor's on-line output prediction. When a fault occurs in a sensor, the predictor can detect it quickly and give the alarm signal. Even more, for instant faults it can recover the sensor's normal outputs accurately so that the faults have no influences on the system, while for permanent faults the predictor can do the same thing in a beginning period of the faults, as result the system may work normally in this period.