摘要
在分析Levenberg-Marquardt(L-M)算法和Nguyen-Widrow(N-W)方法原理的基础上,提出了一种多层前馈神经网络训练算法,该算法在使用N—W方法初始化神经网络可变参数的基础上使用L-M算法训练多层前馈神经网络。构造了适合于变压器油中溶解气体分析故障诊断的神经网络,使用了标准BP算法、加动量项BP算法和结合N-W方法的L-M算法训练该网络,结果表明算法收敛速度快、不容易陷入局部极小点。将训练所得网络用于变压器油中溶解气体分析故障诊断,诊断结果验证了该方法的有效性。
By analyzing the principles of Levenberg-Marquardt (L-M) algorithm and Nguyen-Widrow (N-W) ANN variable parameters initialization method, the network-training algorithm of multi-layer feedforward is proposed. The N-W method is used to initialize ANN variable parameters. The L-M algorithm is used to train Artificial Neural Network (ANN). The ANN for transformer fault diagnosis based on dissolved gas-in-oil analysis (DGA) is constructed. Training results of BP algorithm, BP algorithm with momentum and L-M algorithm with N-W method are used. The effectiveness of the proposed algorithm is verified by fault diagnosis result of transformers.
出处
《华北电力大学学报（自然科学版）》
CAS
北大核心
2005年第4期1-4,共4页
Journal of North China Electric Power University：Natural Science Edition
