棉 花 学 报 Cotton Science 2007，19(3)：220-226
Prediction of Time Series of Main Yield Characters for Cotton Varieties in the Yellow River Valley Region Using ARIMA Models Abstract：According to Box-Jenkins theory and ARIMA(p，d，q) dynamic model，the time series of main yield characters of cotton varieties in Yellow River Valley region was identified，simulated and predicted. The results showed：(1) After stationarity identification，the time series of boll weight and lint percentage appeared significantly increasing trend，which indicated that the effort of cotton breeding in the Yellow River Valley Region for improving boll weight and lint percentage achieved great success. But the time series of lint yield kept stationary and random without increasing trend ，and that was correlate to that the time series of boll numbers per plant had no increasing trend. According to most researchers' reports，boll numbers per plant contributed most to cotton yield increasing.(2)After model identification，the deviation of ARIMA(p，d，q) model were all the white noise series，and the AIC varied from-133.894 to 274.425，the correlation coefficient was 0.9478-0.9767 ，and the goodness of fittest was 89.83%-95.39%；The relative accuracy of prediction test for 2004 and 2005 was 95.60%-99.75%，The lint yield，boll numbers per plant，boll weight and lint percentage of varieties that joined Yellow River Valley Regional test in 2006 were predicted as 1398.05 kg·hm-2，16.61，5.96 g and 40.16%，respectively，which should to be identified in the future. This method was perfect in maths，had a high applicable value，and offered a new way for breeders to grasp the development trend and prospect of cotton varieties. (3) This method was mainly applied to make short-term forecasting. What's more，like other prediction models，ARIMA(p，d，q) model was a dynamic model ，too，and it would change with data addition of time series {yt}. So in order to achieve optimum prediction result ， the model should be adjusted in application.
|