The new methods of calculating initial value and selecting smoothing coefficient in exponential smoothing
HE Ai-lin · Journal of Guangzhou University · 2011
This paper proves that after times of iterations,the results from exponential smoothing method could rapidly approach those results form exponent weighted average method.It further proposes the use of different methods to calculate exponential smoothing value according to data amount: when t=t*(t* refers to the minimum data amount for recurrence algorithm),iterations or full exponent weighted average method is used,and when tt*,recurrence algorithm method is used.When addressing the stability of time series,the paper puts forward two new methods to select smoothing coefficient α: one is based on the calculation of accumulative contribution rate of the most recent k data,the other is the use of half-life for the most recent data's weighting coefficient.