The data accumulated in the production of a drinking water plant in East China were selected to establish a dosing prediction model for intelligent coagulant dosing in the plant. During data cleaning, according to the operation conditions of the plant, the random error was processed by combining the local outlier factor (LOF) algorithm, K-nearest neighbor imputation (KNN) algorithm and data smoothing algorithm. The association of each raw water indicator with coagulant dosage was also evaluated using gray correlation analysis before modeling, and the closely related raw water indicators were analyzed for coagulation mechanism. The model uses BP neural network, and the Bayesian optimization algorithm was used to optimize the parameters of the model. Several models were built for evaluation, and the mean absolute error of the optimal model on sixty thousand samples in the test set was 3.66 L/h. The results show that the optimal model can accurately predict the dosage of coagulant. On this basis, the established intelligent dosing model was prospected to be applied to the water plant dosing system to help the implementation of the intelligent modification of the water plant dosing system.