自回归预测模型变权组合定阶

Order determination for AR(p)model weight changeable combination

针对自回归(AR(p))预测模型定阶问题中回溯阶的不确定性和时变性,以及基于单一回溯阶预测方法的局限性和组合预测中的冗余信息等问题,该文提出了一种基于遗忘因子的变权组合定阶方法。利用冗余定理筛选基于多个回溯阶预测方法的有效信息,并利用遗忘因子实现了组合权重的时变性,克服了基于单一回溯阶预测模型稳健性欠缺、预测精度低等局限性,提高了预测能力。通过实例表明该算法具有高度的可靠性和可行性,为类似预测方法的定阶问题提供了研究思路。

Aiming at the uncertainty and time dependence of retrospective order in the AR（p）model, a weight changeable combination order determination model based on forgetting factor was proposed in this paper considering the limitation of single AR（p）model and the redundant information of combination fore- casting. The model utilized redundant theory to choose the proper AR（p）models, using forgetting factor to realize the time dependence of combination weight. The model overcomed the limitations of single AR （p）model, such as the lack of robustness and low accuracy of prediction, which improved the prediction ability. The example validations indicated that the forecast precision of the proposed model was reliable and feasible, which provided idea for similar prediction method in the way of order determination.