Abstract: Methods for forecasting chaotic time series based on maximum Lyapunov exponent are analyzed,and an improved method is proposed,in which several neighboring reference vectors are selected in reconstructed phase space to increase forecasting precision.Time series of chaotic Lorenz system are forecasted,and the results show that the improved method has a higher accuracy than the original method,meanwhile the influence from noise and number of nearly neighbor vector on the forecasting results is discussed.The average relative error which is forecasted with real traffic volume time series by the improved method is less than 8%,which verifies the effectiveness of the improved method.