Motion Control of Robotic Fish Based on <scp>BP</scp> Neural Network‐Fuzzy Fusion Optimization Algorithm

Abstract To address the complex dynamic control of multi‐source coupled drive systems in bionic robotic fish, this paper proposes a BP neural network–fuzzy fusion optimization algorithm. The method employs a parallel architecture that takes depth error, depth change rate, and pitch angle as inputs. Fuzzy control ensures rapid response to nonlinear disturbances, while an offline‐trained BP neural network enhances the modeling accuracy of complex dynamics. The core innovation involves dynamically adjusting the fusion weights of both controllers using sliding window error trends, enabling coordinated control of propeller speed, tail fin oscillation, and slider displacement. Experimental results indicate that under disturbance‐free conditions, the fused algorithm achieves an average integral absolute error (IAE) of 18.6 ms, representing reductions of 16.2% and 40.5% compared to individual fuzzy and BP neural network control, respectively. Response time is reduced to 23 s with overshoot limited to 45%. In anti‐interference and mass perturbation tests, the algorithm exhibits significantly better convergence speed and stability than conventional methods. The dynamic weight allocation mechanism successfully combines the rapidity of fuzzy logic with the precision of neural networks, offering an effective solution for multimodal motion control of underwater bionic robotic fish. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.