Abstract: For the online precise forecasting of time series，the hybrid forecasting algorithm based on multiple algorithms is established.The Judicature De-Noising algorithm（JDN） is first presented，which realizes de-noising & new steady-state disposal of time series data and reserves their original information.The Empirical Mode Decomposition（EMD） is then utilized to decompose the de-noised data.Next，the BP Neural Network（NN） and Least Square Support Vector Machines（LSSVM） are used to predict the low frequency items and high frequency items of the decomposed sequence data respectively，and the synthesis online realization of the proposed precise forecasting method for time series is finally provided.The algorithm can conquer its divergence by dealing with high frequency items using BP and its long computing time by predicting low frequency items using LSSVM.The simulation and practical experiment results，based on the breath period data of many patients，indicate that the algorithm realized online can forecast the time series precisely.Moreover，the forecasting error is less than that of single BP and the computing time is less than that of single LSSVM.

摘要： 针对时间序列的在线精确预测问题，建立了融合预测算法。创新地提出了司法消噪算法，在保留数据的原始信息前提下，实现了对时间序列中数据噪声和新稳态的处理；利用经验模式分解方法对除噪后的数据进行平稳化分解处理；结合BP神经网络、最小二乘支持向量机分别对分解后的低频、高频项进行预测，实现对时间序列的在线精确预测。该算法克服了BP神经网络的高频易发散和最小二乘支持向量机的计算高耗时问题。基于患者呼吸周期序列预测的仿真和临床实验结果表明，该算法能实现时间序列的在线精确预测，且误差小于单一的BP算法，耗时小于单一的最小二乘支持向量机预测算法。