Abstract: During the casting process of the tundish, the removal rate of inclusions is a key index used to measure the optimization effect of the flow field of the tundish. Water model and numerical simulation methods respectively have significant operational errors, complex particle flow models and high computing power costs when simulating the migration and removal of inclusions, which are the bottlenecks restricting the determination of the optimized structure of the tundish flow field. To predict the flow field optimization effect of different structures of tundishes quickly and accurately, this paper proposes a prediction model for the removal rate of inclusions in tundishes based on BP neural network. By learning the characteristic rules among the flow characteristic parameters of different tundishes, the removal rate of inclusions of different particle sizes can be predicted. Through the grid search algorithm optimization, it is found that when the number of hidden layer nodes n=13 and the learning rate η=0.08, the root mean square loss function value of the model test set reaches the minimum. Three algorithms, genetic algorithm, particle swarm optimization, and simulated annealing, are used to optimize the weights and biases, and it is found that the PSO-BP model has the best optimization effect, with an average hit rate of 92.29%. To further verify the production practicability of the model, it is combined with the water model and mathematical model, and the prediction results of the model are in good agreement with the experimental results with 5% Gaussian noise.