Abstract: Water loss has an important impact on the efficiency of water transfer projects and is a key parameter in the formulation of scheduling plans, accurate prediction of water loss is of great significance for formulating refined water transfer plans and optimizing scheduling operations. Aiming at the existing problems of water loss prediction methods, this paper proposes a water transmission loss prediction method with high calculation accuracy, high construction efficiency and wide application scope: the correlation analysis method and principal component analysis method are used to screen the influencing factors of water loss, and the indicators reflecting the duplication of information are deleted through the correlation analysis, and the indicators with high importance are screened out by the principal component analysis; based on the filtered index system, a water transmission loss prediction model based on the improved particle swarm optimization extreme learning machine is constructed. The Liangji Canal section of the eastern line of the South-to-North Water Diversion Project is taken as an example. After screening, water depth, flow rate, air temperature and wind speed are the main influencing factors, and an IPSO-ELM water loss prediction model is established by using the predictive model trained and validated by the measured data, the daily water loss prediction is made on the data in different scenarios, and the prediction results of extreme learning machine model and the multivariate nonlinear regression model are compared and analyzed, respectively. The results show that the water loss and actual water loss calculated by the IPSO-ELM water transmission loss prediction model are very close, and the determination coefficient is 0.962 5, and the average absolute percentage error is 1.322%, which is 52.89% and 51.06% lower than the MNR model and the ELM model, respectively. The prediction error is mainly distributed within -25, 30 thousand m3, and the error distribution range is smaller than the other two models. That is, the prediction accuracy and generalization ability of the IPSO-ELM model are better than the other two models, which proves that the method is reasonable and feasible, which can be used to calculate the water transmission loss in different water transfer scenarios and provide more accurate water quantity information for water resource dispatch.