Aiming at the shortcomings of teaching-learning-based optimization ( TLBO ) algorithm, which is easy to fall into local optimization and has low accuracy in solving complex optimization problems, an efficient TLBO ( ETLBO ) is proposed to improve the global optimization performance of standard TLBO. In ETLBO, the population is divided into two groups by the two-swarm shuffling strategy, and the worst students are taught separately by the teacher to speed up the algorithm to converge to the global optimum quickly. By solving four typical numerical functions, the simulation results verify the effectiveness of the ETLBO algorithm. Finally, the parameters of extreme learning machine (ELM) model are optimized and selected through the ETLBO algorithm, and the ETLBO-ELM model is constructed and applied to urban water demand prediction. The simulation results show that the ELM model optimized by ETLBO has good prediction accuracy and generalization ability.