Efficient privacy-preserving federated learning method for Internet of Ships

标签	孔径故障尺寸/mm	状态	故障程度	样本矩阵维度	数量/个
N0		正常		28×28	500
IR1	18	内圈故障	微小	28×28	500
IR2	36	内圈故障	中度	28×28	500
IR3	53	内圈故障	显著	28×28	500
OR1	18	外圈故障	微小	28×28	500
OR2	36	外圈故障	中度	28×28	500
OR3	53	外圈故障	显著	28×28	500
B1	18	滚球故障	微小	28×28	500
B2	36	滚球故障	中度	28×28	500
B3	53	滚球故障	显著	28×28	500

Table 2. Execution time of the encryption/decryption operations
View table
View in Article
Table 2. Execution time of the encryption/decryption operations
操作方法执行时间/s
2 000个参数 10 000个参数 65 162个参数
Paillier加密 14.61 71.58 594.67
Paillier解密 4.22 20.62 134.94
Paillier密文加法 0.03 0.15 1.04

Table 3. Classification results of different FLs
View table
View in Article
Table 3. Classification results of different FLs
算法准确率/% 精准率/% 召回率/% 上传参数量(|w|∙N) 压缩比
FL 93.60 94.13 93.60 2 000 1
MFL 99.10 99.13 99.10 2 000 1
EPFL 98.80 98.85 98.80 199.80 0.1

Table 4. Classification results of the EPFL with different momentum rates
View table
View in Article
Table 4. Classification results of the EPFL with different momentum rates
算法 γ 准确率/% 精准率/% 召回率/%
EPFL 0 96.90 97.08 96.90
EPFL 0.3 98.40 98.50 98.40
EPFL 0.5 98.70 98.79 98.70
EPFL 0.7 98.80 98.85 98.80

Table 5. Classification results of the EFL with different node numbers
View table
View in Article
Table 5. Classification results of the EFL with different node numbers
算法准确率/% 精准率/% 召回率/%
EFL(N=5) 99.10 99.10 99.10
EFL(N=10) 98.80 98.85 98.80
EFL(N=15) 98.20 98.25 98.20
EFL(N=20) 97.10 97.24 97.10

Tools

Get Citation

Copy Citation Text

Zehui ZHANG, Cong GUAN, Hang GAO, Tiegang GAO, Hui CHEN. Efficient privacy-preserving federated learning method for Internet of Ships[J]. Chinese Journal of Ship Research, 2022, 17(6): 48

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites