人工神经网络的混合算法及其工程应用

MIXED METHOD OF ARTIFICIAL NEURAL NETWORK AND ITS APPLICATION ON FAULT DIAGNOSIS FOR ROTATIONAL MACHINE

为确定多层人工神经网络的权值和阈值建立了混合求解方法,即迭代前期采用BP算法而迭代后期采用梯度优化法进行计算。通过BP算法、梯度优化法和混合算法对相同实例网络权值和阈值的计算,验证了所提出的混合算法的有效性,实现了权值和阈值的快速准确计算。通过汽轮机故障诊断的实例计算,表明所提出算法的有效性和实际应用价值。

Mixed method is built to calculate the weights and thresholds of multi-layer neural network by utilize BP algorithm and terraced optimizations. Comparing mixed method with BP algorithm and terraced method, validity of the method is testified, which results in rapid computing of weights and thresholds.