基于激活函数可调的改进型BP网络故障检测应用研究  

Research on Fault Inspection Based on TAF-BP Improved Neural Network

作　　者：王正武[1] 宁顺杰[1] 

机构地区：[1]空军第一航空学院,河南信阳464000

出　　处：《测试技术学报》2013年第2期174-177,共4页Journal of Test and Measurement Technology

摘　　要：神经网络方法被广泛应用在故障检测领域.针对某型航空装备供油系统故障率高,造成燃油压力波动较大这一复杂现象,文章对传统的BP网络传递函数和传递模式进行了改进,用变化的可调节函数代替固定的传递函数,构造了激活函数可调的改进型BP神经网络,提高了网络的储存能力和映射能力.建立了动态的网络结构和学习算法,用于该型故障问题的检测和诊断.仿真表明,改进后的网络收敛较快,精度较高,泛化能力强,检测效果比较理想.The neural network methods are widely used in the fault inspection field. This paper presented a dynamic studying algorithm and improved the network structure. It used the tunable activation function to re- place the traditional fixed passing function to enhance the memory capacity and mapping capacity. We ana- lyzed the principle about the oil feeding system fuel-pressure wave fault and used the improved method to in- spect the fault. The simulation indicates that the improved network and method can increase the convergence speed, enhance the inspection precision and the generalization capability.

关 键 词：可调函数 神经网络 燃油压力 故障检测 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]