To improve the accuracy and efficiency of short-term load forecasting

we proposed a load forecasting method based on an improved manifold regularization extreme learning machine. Firstly

to improve the generalization performance and efficiency

and to solve the potential problem of ELM caused by random initialization parameters

Manifold Regularization Extreme Learning Machine（MRELM） was put forward. Secondly

in view of the problems about the selection of parameters in MRELM and the selection on hidden layer nodes of MRELM

Bayesian optimization algorithm（BOA） was introduced to the MRELM to optimize the MRELM. According to analysis of experimental data based on the BOA-MRELM

the mean relative error（MRE） is decreased by 1.903%. After 30 experiments

the variance of the MRE is decreased by 1.9‰. What’s more

the average single running time can be reduced to 6.113 s.