CAJ | 학술논문

针对BP神经网络对于海量数据训练及多维数据训练收敛困难的问题，在使用增加动力项、自适应学习速率等方法的基础上，引入均值影响度算法（MIV）构造了贝叶斯正则化反向传播（BRBP）神经网络，以此提高电子线路板红外故障诊断算法的效率。利用红外测温方式，获取了不同室温及运行状态下电路板中21个元器件温度数据。将此21个参数作为故障诊断模型的初始输入变量，经过MIV算法简约为12个参数输入至BRBP神经网络，进行故障评估和诊断。结果表明：相对于传统的BRBP神经网络，本文设计的基于MIV和BRBP神经网络模型诊断方法极大简化了数据训练的数据量并解决了数据收敛的困难，因此效率更高，用时更省。
침대BP신경망락대우해량수거훈련급다유수거훈련수렴곤난적문제，재사용증가동력항、자괄응학습속솔등방법적기출상，인입균치영향도산법（MIV）구조료패협사정칙화반향전파（BRBP）신경망락，이차제고전자선로판홍외고장진단산법적효솔。이용홍외측온방식，획취료불동실온급운행상태하전로판중21개원기건온도수거。장차21개삼수작위고장진단모형적초시수입변량，경과MIV산법간약위12개삼수수입지BRBP신경망락，진행고장평고화진단。결과표명：상대우전통적BRBP신경망락，본문설계적기우MIV화BRBP신경망락모형진단방법겁대간화료수거훈련적수거량병해결료수거수렴적곤난，인차효솔경고，용시경성。
The training algorithm for BP network is hard to converge when the input data is large and has high dimen-sion. Aiming at this problem,a novel fault diagnosis method based on MIV and BRBP neural networks by infrared temperature measuring is put forward. Sample data about 21 variables of circuit board under different room tempera-ture and operating conditions are measured,and these 21 parameters are used as the initial input variables of fault di-agnosis model. After MIV optimization,the reduced 12 variables will be input into BRBP neural networks to predict faults and classify the circuit board running conditions. Experiments show that the proposed neural networks model is more efficiently and more rapidly compared with the traditional BRBP neural network. The neural network model pres-ented in the paper can effectively diagnose the circuit board faults.