﻿ 基于深度学习算法的船用高频电路工作状态检测研究
 舰船科学技术  2023, Vol. 45 Issue (12): 156-159    DOI: 10.3404/j.issn.1672-7619.2023.12.031 PDF

Research on working state detection of marine high frequency circuit based on deep learning algorithm
HE Yi-jie, WANG Bo
BeiDou School, Wuhan Qingchuan University, Wuhan 430204, China
Abstract: A deep learning algorithm based working state detection method for marine high frequency circuit is studied to improve the operational reliability of marine high frequency switching power supply. The working state signals of Marine high frequency circuits are collected and used as input of the depth limited Boltzmann machine. The depth limited Boltzmann machine uses the two-layer limited Boltzmann machine to extract the working state characteristics of marine high frequency circuits through two nonlinear mapping. The extracted working state characteristics of the high-frequency circuit are set as the input of the support vector data description method, which maps the input sample to the high-dimensional inner product space, determines whether the sample exists in the optimal hypersphere of the high-dimensional inner product space, and detects whether the working state of the marine high-frequency circuit is normal or abnormal. Experimental results show that the proposed method can accurately detect the working state of marine high-frequency circuit and meet the operational reliability requirements of marine high-frequency switching power supply.
Key words: deep learning algorithm     marine high frequency circuit     working condition detection     nonlinear mapping     higher dimensional inner product space     optimal hypersphere
0 引　言

1 船用高频电路工作状态检测 1.1 船用隔离式高频开关的高频电路分析

 图 1 船用隔离式高频开关电源结构图 Fig. 1 Structure diagram of marine isolated high frequency switching power supply
1.2 深度受限玻尔兹曼机的特征提取

 $E\left( {v,h\left| \mu \right.} \right) = - \sum\limits_{i = 1}^n {{w_i}{v_i}} - \sum\limits_{j = 1}^m {{w_j}{h_j}} - \sum\limits_{i = 1}^n {\sum\limits_{j = 1}^m {{v_i}{w_{ij}}{h_j}} }，$ (1)

 $P\left( {v,h\left| \mu \right.} \right) = {e^{ - E\left( {v,h\left| \mu \right.} \right)}}/Z\left( \mu \right) ，$ (2)
 $Z\left( \mu \right) = \sum\limits_{v,h} {{e^{ - E\left( {v,h\left| \mu \right.} \right)}}} 。$ (3)

 $P\left( {{h_j} = 1\left| {v,h} \right.} \right) = \sigma \left( {{w_j} + \sum\limits_i {{v_i}} {w_{ij}}} \right)。$ (4)

 $\sigma \left( x \right) = \frac{1}{{1 + \exp \left( { - x} \right)}} 。$ (5)

 ${\mu ^ * } = \arg \max l\left( \mu \right) = \arg \max \sum\limits_{t = 1}^T {\lg P\left( {{v^t}\left| \mu \right.} \right)}。$ (6)

 $H = sigmiod\left( {V \times {w_{ij}}} \right)。$ (7)

Sigmoid函数属于非线性表达式，即输入的船用高频电路工作状态样本数据，经过Sigmoid函数映射后，转换为隐含层数据 $H = \left( {{h_1},{h_2}, \cdots ,{h_n}} \right)$ 。将原始样本数据通过非线性映射，转化为另一种状态，受限玻尔兹曼机可以挖掘船用高频电路工作状态样本数据的隐藏特征。将隐含层输出数据 $H = \left( {{h_1},{h_2}, \cdots ,{h_n}} \right)$ 作为另一个受限玻尔兹曼机的输入，合并完成训练后的2个网络，通过逐层训练方法，构建船用高频电路工作状态特征提取的深层数据的非线性表达。

1.3 支持向量数据描述的工作状态检测方法

 $\mathop {\min }\limits_{R,o{\xi _i}} F\left( {R,o{\xi _i}} \right) = {R^2} + C\sum\limits_i {{\xi _i}}，$ (8)
 ${\rm{ s.t.}}\left\{ {\begin{array}{*{20}{l}} {{{\left\| {{x_i} - o} \right\|}^2} \leqslant {R^2} + {\xi _i}} ，\\ {{\xi _i} \geqslant 0} 。\end{array}} \right.$ (9)

 $\begin{gathered} L\left( {R,o,{\xi _i},{\alpha _i},{\gamma _i}} \right) = {R^2} + C\sum\limits_i {{\xi _i}}- \\ \sum\limits_i {{\alpha _i}} \left[ {{R^2} + {\xi _i} - {\gamma _i}{{\left( {{x_i} - o} \right)}^2}} \right] - \sum\limits_i {{\gamma _i}{\xi _i}} 。\\ \end{gathered}$ (10)

 $\left\{ {\begin{array}{*{20}{c}} {\displaystyle\sum\limits_i {{\alpha _i}} = 1}，\\ {o = \displaystyle\sum\limits_i {{\alpha _i}{x_i}} } 。\end{array}} \right.$ (11)

 $\min L = \sum\limits_i {{\alpha _i}{x_i}^2} - \sum\limits_{i,j} {{\alpha _i}{\alpha _j}{x_i} \cdot } {x_j} 。$ (12)

 ${\left\| {z - o} \right\|^2} = {z^2} + 2\sum\limits_i {{\alpha _i}z} {x_i} - \sum\limits_{i,j} {{\alpha _i}{\alpha _j}{x_i}} {x_j} \leqslant {R^2} 。$ (13)

2 实例分析

 图 2 船用高频电路状态信号 Fig. 2 Marine high frequency circuit status signal

 图 3 测试样本的幅频特性曲线 Fig. 3 Amplitude-frequency characteristic curve of test samples
3 结　语

 [1] 牟红宇, 赵世舟, 白钰, 等. 船用增压锅炉燃烧系统燃油压差控制电路设计[J]. 热能动力工程, 2020, 35(12): 154-161. MOU Hong-yu, ZHAO Shi-zhou, BAI Yu, et al. Design of fuel oil pressure difference control circuit for combustion system of marine supercharged boiler[J]. Journal of Engineering for Thermal Energy and Power, 2020, 35(12): 154-161. DOI:10.16146/j.cnki.rndlgc.2020.12.021 [2] 李涛, 乔鸣忠, 宋庆国. 船用逆变电源两级死区补偿控制策略[J]. 船舶工程, 2022, 44(S1): 398-402. LI Tao, QIAO Ming-zhong, SONG Qing-guo. Two-stage dead zone compensation control strategy for Marine inverter power supply[J]. Ship engineering, 2022, 44(S1): 398-402. [3] 吴希杰, 周方明. 船舶控制设备用微连接器激光软钎焊电路设计[J]. 船海工程, 2021, 50(1): 64-67. [4] 汪洋, 黎恒烜, 鄂士平, 等. 基于深度学习的变电站硬压板状态检测与识别算法[J]. 沈阳工业大学学报, 2020, 42(6): 676-680. WANG Yang, LI Hengxuan, E Shiping, et al. State detection and recognition algorithm for hard platens of substation based on deep learning[J]. Journal of Shenyang University of Technology, 2020, 42(6): 676-680. DOI:10.7688/j.issn.1000-1646.2020.06.11 [5] 吴健, 尹泽, 李豪, 等. 基于分数阶理论的锂离子电池高频等效电路模型[J]. 电工技术学报, 2021, 36(18): 3902-3910. DOI:10.19595/j.cnki.1000-6753.tces.200562 [6] 岳改丽, 向付伟, 李忠. Cascode GaN高电子迁移率晶体管高频驱动电路及损耗分析[J]. 电工技术学报, 2021, 36(20): 4194-4203. YUE Gai-li, XIANG Fu-wei, LI Zhong. High-frequency drive circuit and its loss analysis of cascode GaN high electron mobility transistor[J]. Transactions of China Electro technical Society, 2021, 36(20): 4194-4203.