In order to improve the accuracy of short-term forecasting of wind power, a short-term forecasting method based on improved TLBO optimization LSSVM is proposed. Firstly, to improve TLBO algorithm of "teach" stage, the adaptive teaching factor is used, at the same time, changing average of all search individuals, which can enhance the performance of TLBO in the whole search space; Then, the 'learning' stage of TLBO algorithm is improved to maintain the diversity of the population and avoid the premature convergence and local optimization of TLBO algorithm, and the gaussian mutation operator is introduced in the learning stage. Finally, the improved LSSVM prediction model is optimized with improved TLBO. Take the measured data of beiyan wind farm in Shanghai as an example, the simulation results show that with the PSO and TLBO compared to optimizing LSSVM, improved TLBO optimizing LSSVM method for short-term wind power prediction has better stability and higher accuracy