CAJ | 학술논문

将脊波与神经网络进行有机的结合,构造了一种新型的用于模拟电路故障诊断的脊波网络系统,并给出了基于最速梯度下降法与动量法相结合的训练算法和电路故障诊断的方法及步骤。由于脊波能有效地处理高维函数直线状和超平面状的奇异性,因此将脊波作为神经网络隐层的激活函数从而构成的脊波网络能有效地处理复杂的电路故障信息从而对故障做出正确的分类。实验结果证实了这种新型方法的有效性。
장척파여신경망락진행유궤적결합,구조료일충신형적용우모의전로고장진단적척파망락계통,병급출료기우최속제도하강법여동량법상결합적훈련산법화전로고장진단적방법급보취。유우척파능유효지처리고유함수직선상화초평면상적기이성,인차장척파작위신경망락은층적격활함수종이구성적척파망락능유효지처리복잡적전로고장신식종이대고장주출정학적분류。실험결과증실료저충신형방법적유효성。
A novel ridgelet network combing ridgelet and neural network appropriately used for diagnosing the faults of analog circuit is constructed. Then a training algorithm based on the steepest gradient descent method and momentum method and the approach and procedure of fault diagnosis of analog circuit are presented. Due to the ridgelet handling effectively the line-like and hyperplane-like singularities in high dimensions, the ridgelet network adopts the ridgelet as the activation functions in the hidden layer so that the ridgelet network can deal with effectively the complicated fault information of circuit under test to classify the faults correctly. The experimental results demonstrate the effectiveness of the novel approach.