Noise-robust fusion power supply fault diagnosis based on wavelet integrated one-dimension convolutional neural network

Fig. 2. Transformation of feature to low frequency component

Fig. 3. Process diagram of model structure and training

Fig. 4. The representation of converter units of power supply using Matlab $/$ Simulink

Fig. 5. Schematic diagram of short-circuited thyristor VT1

Fig. 6. Original thyristor short-circuit signal at Id1, the noise (6 dB), and the noisy signal

Fig. 7. Normalized confusion matrix constructed from the results of three branches in proposed approach (left column is branch 1, 2; right column is branch 3)

Fig. 8. Accuracy comparison results of WDCNN, TICNN, ConvLSTM, FaultNet and our method under different intensities of noise (color online)

Table 1. Main simulation parameters of converter units

Table 1. Main simulation parameters of converter units

仿真参数 Simulation entity	参数值 Value
额定电压Rated voltage / kV	440
短路容量Short circuit capacity / GVA	10
降压变容量Step-down transformer capacity / MVA	300
降压变电压Step-down transformer voltage / kV	400 $/$ 66
环流电抗器Circulation reactor	200 μH,300 μ $Ω$
短路阻抗Short circuit impedance / %	11 $/$ 15.8
整流变容量Rectifier transformer capacity / MVA	78
整流变电压Rectifier transformer voltage / kV	66 $/$ 1.05

Table 2. Description of classes and labels for the simulated power supply dataset

Table 2. Description of classes and labels for the simulated power supply dataset

第一级 Branch 1	第二级 Branch 2	第三级 Branch 3	工况 Operating condition	故障详情 Fault details and source
0	0	0	单桥/并联Single bridge/parallel	整流 $/$ 逆变无故障Rec&Inv (normal)
—	—	1	整流Rectification	整流直流环抗前短路Before (Rec)
—	—	2	整流Rectification	整流桥晶闸管VT1短路Arm 1 (Rec)
—	1	3~5	整流Rectification	整流桥晶闸管VT2-5短路Arm 2-5 (Rec)
—	—	6	整流Rectification	整流桥晶闸管VT6短路Arm 6 (Rec)
1	—	7	整流Rectification	整流直流环抗后短路After (Rec)
—	—	8	逆变Inversion	逆变直流环抗前短路Before (Inv)
—	—	9	逆变Inversion	逆变桥晶闸管VT1短路Arm 1 (Inv)
—	2	10~14	逆变Inversion	逆变桥晶闸管VT2-5短路Arm 2-5 (Inv)
—	—	15	逆变Inversion	逆变桥晶闸管VT6短路Arm 6 (Inv)
—	—	16	逆变Inversion	逆变直流环抗后短路After (Inv)

Table 3. Accuracy of the proposed method under different noise environments

Table 3. Accuracy of the proposed method under different noise environments

层级 Branch levels	SNR / dB
层级 Branch levels	0	2	4	6	8	10
Branch 1	97.98 $\pm$ 1.74	99.12 $\pm$ 0.41	99.71 $\pm$ 0.22	99.93 $\pm$ 0.07	100 $\pm$ 0.01	100 $\pm$ 0.01
Branch 2	97.32 $\pm$ 1.33	98.65 $\pm$ 1.14	98.97 $\pm$ 0.54	99.39 $\pm$ 0.23	99.56 $\pm$ 0.11	99.73 $\pm$ 0.03
Branch 3	88.39 $\pm$ 4.50	92.13 $\pm$ 3.12	95.26 $\pm$ 2.57	96.97 $\pm$ 1.87	97.48 $\pm$ 0.71	98.31 $\pm$ 0.19

Table 4. The diagnostic results of the proposed model and the compared methods under different intensities of noise