Research on the prediction of motion trajectory and precise control method of bionic robotic fish based on LSSVR interactive network

• Published in: Ocean engineering 2024-11, Vol.311, p.118857, Article 118857
• Main Authors: Wang, Zhiping, Li, Zonggang, Xia, Guangqing, Kang, Huifeng, Li, Bin, Zheng, Lixin, Li, Qingquan
• Format: Article
• Language: English
• Subjects: Biomimetic robotic fish; Central pattern generator; Parameter optimization algorithm; Precise trajectory control; Support vector machine regression
• ISSN: 0029-8018
• DOI: 10.1016/j.oceaneng.2024.118857

The inaccuracy of the multi-fins synergy hydrodynamic model of the robotic fish and the lack of clarity between the control parameters of the locomotion gait and swimming behavior of the robotic fish were addressed. We constructed a bionic robotic fish pectoral fins and flexible body synergy motion gait model by using a Central Pattern Generator (CPG) network. We obtained the rowing phase difference, pectoral fins rotational phase difference, flexible body fluctuation bias, frequency, and the velocity and positional attitude datasets of fish through numerical simulations using Computational Fluid Dynamics (CFD). We proposed identification and control models based on the least squares support vector machine regression (LSSVR) interactive network and constructed hydrodynamic models of pectoral fins, flexible body parameters, and locomotion modes using the identification model. The identification model data were used to design the control model offline control law, and the identification and control model parameters were updated online by combining them with the experimental dataset. The experimental results indicated that the displacement REMS were all less than 0.1, and the maximum error of the spatial motion trajectory was 0.08 m. The identification and control models enabled the robotic fish to accurately track the desired trajectory. [Display omitted] •A multi-fin synergistic gait model was constructed of bionic robotic fish.•The identification and control model were constructed by LSSVR interactive network.•The control law is designed offline and optimized online by datasets.