Abstract: In accordance with the characteristics of a non-stationary material requirement time series of the equipment manufacturing industry, we build a combination forecasting model based on empirical mode decomposition (EMD) and least square support vector regression (LSSVR). We divide the non-stationary time series into a series of intrinsic mode functions (IMF) and a residual by using EMD. We then analyze the business in a real situation and combine every IMF into high frequency and low frequency, which represent short-term fluctuations and long-term trends, respectively. After these steps, we mine more information. Then, we make a combination forecast by using LSSVR. An empirical study shows that the combination forecast of the EMD-LSSVR can forecast the non-stationary time series of material demand efficiently, and its prediction accuracy is higher than that of traditional methods.