Short-term load forecasting is a very important task of power system.Accurate short-term load forecasting is meaningful for the economical

safe and credible operation of power system.With the development of power system

especially the development of power market

the forecasting method with high accuracy must be researched.A new method of pre-disposing history data is proposed based on the daily periodicity of load

which lays the foundation for using history data by the model.Least square support vector regression（LSSVR） algorithm is a new generation of machine learning algorithms.So

based on daily periodicity and the same moment similarity of load

a LSSVR-based short-term load forecasting model is put forward.The optimal load linear regression function is obtained by use of the samples in the same time but in the different days to train LSSVR.While the minimum load sample error is achieved

the model generalization error on the sector is reduced

so a better load forecasting performance is obtained.