科研之友
微信
新浪微博
Facebook
分享链接
摘要
The construction method of background value is the key factor, which influences the prediction accuracy of MGM(1, m) model directly. Existing optimized background value. is calculated mainly by non-homogeneous index function fitting the first-order accumulated generating sequence. However, the corresponding background value lack of efficient theoretical basis: the prerequisite ( x(j)((1))(1) = x(j)((0))(1)) is not reasonable, and results in the seriously decrease of prediction accuracy of MGM(1, m) model. Moreover, the prediction accuracy of MGM(1, m) model is unsatisfactory, when modeling sequence that shows an oscillation characteristic. Based on the problems of the existing optimized background value and MGM(1, m) model, a new optimization method and the extended MGM(1, m) model with optimized background value are proposed. Evidence of theory and experiment results show that the optimized background value has more reasonable and reliable theoretical basis, and the extended MGM(1, m) model has higher prediction accuracy for monotone sequence and oscillation sequence. Compared with MGM(1, m) model and OBMGM(1, m) model, the prediction accuracy of the proposed OMGM(1, m) model (when each data sequence is monotonous) is improved, the average relative error of simulation and predictive values was reduced by 7.96% and 6.57% respectively. Meanwhile, the proposed OMGM(1, m) model combined with multiple linear regression model can well solve the prediction problem of multiple explanatory variables influence a dependent variable, and compared with 1) MLR model, 2) GM(0, N) model and GM(1, N) model, 3) RMGM(1, m) model and ORBMGM(1, m) model, the average relative error was reduced by 7.03%, 1.365% and 0.6675% respectively. Moreover; compared with MGM(1, m) model and OMGM(1, m) model, the proposed OOMGM(1, m) model (when each data sequence is oscillatory) the relative error of predictive values is reduce by 5.1675% and 5.205% respectively.
- 出版日期2018
- 单位重庆邮电大学
