Chinese Journal of Ship Research, Volume. 19, Issue 6, 219(2024)

Attitude control of catamaran based on deep reinforcement learning

Leihong QIN, Songtao ZHANG, Xiaofeng NAN, and Qiming ZHONG
Author Affiliations
  • College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin 150001, China
  • show less
    Figures & Tables(18)
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    [in Chinese]
    • Table 1. Hull parameters of catamarans

      View table
      View in Article

      Table 1. Hull parameters of catamarans

      参数数值参数数值
      船长/m22.000吃水/m1.700
      船宽/m9.500横稳心高/m10.229
      船高/m4.750排水量/t104.800
    • Table 2. Parameters of wave and fin model

      View table
      View in Article

      Table 2. Parameters of wave and fin model

      鳍参数数值海浪参数数值
      $ \rho /({\text{t}} \cdot {{\text{m}}^{ - 3}}) $1.025$ g/({\text{m}} \cdot {{\text{s}}^{ - 2}}) $9.81
      $ {S_{{\text{f}}1}}/{{\text{m}}^2} $1.0$ {h_{1/3}}/{\text{m}} $1.0
      $ {S_{{\text{f2}}}}/{{\text{m}}^2} $1.2$ \beta/(^\circ) $180
      $ {L_{{\text{P1}}}}/{\text{m}} $6
      $ {L_{{\text{P2}}}}/{\text{m}} $6
      $ {\text{d}}{C_{\text{L}}}/{\text{d}}\alpha $0.03
    • Table 3. Parameters of catamaran model

      View table
      View in Article

      Table 3. Parameters of catamaran model

      参数$ {A_{33} }/{\text{t} } $$ {A_{35}}/({\text{t}} \cdot {\text{m)}} $$ {A_{53}}/({\text{t}} \cdot {\text{m)}} $$ {A_{55} }/({\text{t} } \cdot { {\text{m} }^2}) $$ {B_{33}}/({\text{t}} \cdot {{\text{s}}^{ - 1}}) $$ {B_{35}}/({\text{t}} \cdot {\text{m}} \cdot {{\text{s}}^{ - 1}}) $$ {B_{53}}/({\text{t}} \cdot {\text{m}} \cdot {{\text{s}}^{ - 1}}) $$ {B_{55}}/({\text{t}} \cdot {{\text{m}}^2} \cdot {{\text{s}}^{ - 1}}) $
      工况1138.111−241.452260.5555228.66789.505318.409−298.5084947.048
      工况2137.127−275.305296.7285510.86488.597374.072−357.6665839.849
      工况3137.127−327.308348.0005979.41388.597440.594−424.1887123.976
      参数$ {C_{33}}/({\text{t}} \cdot {{\text{s}}^2}) $$ {C_{35}}/({\text{t}} \cdot {\text{m}} \cdot {{\text{s}}^{ - 2}}) $$ {C_{53}}/({\text{t}} \cdot {\text{m}} \cdot {{\text{s}}^{ - 2}}) $$ {C_{55}}/({\text{t}} \cdot {{\text{m}}^2} \cdot {{\text{s}}^{ - 2}}) $
      工况1514.374−195.159−195.15916586.51
      工况2514.374−195.159−195.15916586.51
      工况3514.374−195.159−195.15916586.51
    • Table 4. DDPG algorithm parameters

      View table
      View in Article

      Table 4. DDPG algorithm parameters

      参数折扣因子$ \gamma $平滑因子$ \tau $Actor隐层数Critic隐层数隐层神经元个数每回合步数T
      只控制前鳍0.990.00523646 000
      只控制后鳍0.990.00523646 000
      前后鳍协同控制0.990.005231286 000
      参数回合数NActor学习率Critic学习率学习率衰减因子回放池容量B每批训练量M
      只控制前鳍5000.000 10.0010.9951 000 000256
      只控制后鳍5000.000 10.0010.9951 000 000256
      前后鳍协同控制5000.000 10.0010.9951 000 000256
    • Table 5. Parameters for different operating conditions

      View table
      View in Article

      Table 5. Parameters for different operating conditions

      工况航速/kn有义波高/m
      工况1181
      工况2222
      工况3264
    • Table 6. Reward function parameters

      View table
      View in Article

      Table 6. Reward function parameters

      参数数值参数数值
      $ {\lambda _1} $0.6$ {\lambda _4} $0.03
      $ {\lambda _2} $0.06$ {\lambda _5} $0.003(方式1和3)0(方式2)
      $ {\lambda _3} $0.3$ {\lambda _6} $0.003(方式2和3)0(方式1)
    • Table 7. Pitch control effect under different control modes

      View table
      View in Article

      Table 7. Pitch control effect under different control modes

      控制方式MAV/(°)RMS/(°)Max/(°)Min/(°)$ {\eta _{\text{p}}} $/%
      无控制4.9035.91413.775−13.983
      方式1(DDPG)3.0313.62311.131−11.54838.751
      方式1(GA-LQR)3.0903.69011.133−11.78037.613
      方式2(DDPG)2.5733.14510.560−10.81146.828
      方式2(GA-LQR)2.8413.39910.383−11.11142.828
      方式3(DDPG)1.2551.6227.266−7.13072.581
      方式3(GA-LQR)1.6802.0776.716−7.85164.881
    • Table 8. Heave control effect under different control modes

      View table
      View in Article

      Table 8. Heave control effect under different control modes

      控制方式MAV/mRMS/mMax/mMin/m$ {\eta _{\text{h}}} $/%
      无控制0.5130.6161.608−1.516
      方式1(DDPG)0.3110.3851.150−1.16437.437
      方式1(GA-LQR)0.2990.3681.075−1.16940.296
      方式2(DDPG)0.2870.3611.100−1.14641.368
      方式2(GA-LQR)0.2810.3430.953−1.13044.307
      方式3(DDPG)0.1860.2380.732−0.90261.280
      方式3(GA-LQR)0.2030.2480.704−0.94559.656
    Tools

    Get Citation

    Copy Citation Text

    Leihong QIN, Songtao ZHANG, Xiaofeng NAN, Qiming ZHONG. Attitude control of catamaran based on deep reinforcement learning[J]. Chinese Journal of Ship Research, 2024, 19(6): 219

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Ship Design and Performance

    Received: Aug. 3, 2023

    Accepted: --

    Published Online: Mar. 14, 2025

    The Author Email:

    DOI:10.19693/j.issn.1673-3185.03492

    Topics