For an actual water diversion project, a mechanistic model based on hydraulic principles was established, and hydraulic parameters were calibrated. Based on data such as system layout parameters, upstream reservoir’s water level, measured pressure and flow rate along the pipeline, and measured flow rate and opening of the regulating valve, a BP neural network data model was established, and GA was used to optimize the initial weights and biases to improve the accuracy of the data model. The target opening data of the regulating valve calculated by the two models were comparatively analyzed, and the applicable conditions of the two models were proposed. The results indicate that when the cosine similarity between the input data and the training data is high (≥ 0.95), the average error of the data model is 0.58%, representing a reduction of 0.11 percentage points compared with that of the mechanistic model, and the maximum error is reduced by 0.46 percentage points; however, when the cosine similarity is less than 0.95, the maximum error of the data model can reach 9.02%, while the average error of the mechanistic model does not exceed 1.13%. The data model depends on data and has high calculation speed and accuracy under conditions of high cosine similarity.