As chaotic time series are large

least squares support vector machines(LSSVM) have disadvantages of bigger memory spending and slower training speed on prediction. According to data characteristic of large chaotic time series

it adopts ideas of data sets partition and data correlation coefficient to propose a LSSVM prediction model for large chaotic time series based on new reduction strategy. The model partitions large chaotic time series is split up into several different subsets based on the mean cycle of chaotic time series.Some nonsupport vectors from all subsets is reduced except the end based on the values of Lagrange multipliers.The reduced data sets combines with the end subset based on the correlation coefficients

and is used to regress and predict by LSSVM method. The proposed model is applied to the forecast of large chaotic time series on correlative experiments

and the results show it hardly loses prediction precision and takes quicker training speed.