A new short-term load forecasting method based on empirical mode decomposition(EMD)and feature correlation analysis is proposed. The method begins with the decomposition load sequence and uses the EMD to decompose the original load time series into different frequency intrinsic mode function(IMF)components and residual components to weaken the volatility of the original sequence under the environment of complex influence factors and obtain more regular components. Then

the minimal redundancy maximal relevance(mRMR)criterion is used to analyze the correlation between each IMF component and feature information(such as the type of day

weather and electricity price)to obtain the best feature set. Finally

the least squares support vector machine(LSSVM)load forecasting model based on intelligent algorithm is adopted to predict each component and superpose each component prediction result to get the final load forecasting. Taking actual data of a power grid as an example

the results show that the proposed composition model can predict the short-term load which is sensitive to external factors more accurately.