An efficient fish migration modeling method integrating the random forest and Eulerian–Lagrangian–agent method for vertical slot fishways

A fishway is a fish-passing facility that helps migratory fish to cross obstacles. To minimize the considerable time and economic costs associated with fishway structure-optimized experiments, this paper proposes an Active Fish Migration (AFM) model developed based on experimental data of the juvenile Schizothorax prenanti movement and fishway flow fields. The proposed AFM model is a probability-based model in which a fish surrogate randomly determines its next movement behavior according to the computed probability. The AFM model can predict any of the possible paths of fish, making it suitable for describing fish migration in fishway engineering. In addition, the AFM model integrates the random forest and Eulerian–Lagrangian–agent methods to simulate juvenile Schizothorax prenanti migration in vertical slot fishways. The experimental results show that the average precision of the swimming behavior classification is 80.4%, the recall is 79.4%, and the F1-score is 79.8%, while the R2 value of the speed regression exceeds 0.7 (MAE < 0.07). The simulated trajectories are in good agreement with the observation data of the same flow pattern. Finally, the proposed AFM model's results exhibit a high agreement with real fish movement; thus, the AFM model has great application potential in fishway design.