本文对多层递阶预报方法进行了深入的研究,提出了一种初值选取方法,并选择AR模型作为预报模型,采用高效的Houscholder变换来确定模型阶次,从而使得这种预报方法简便而实用,预报精度亦有进一步提高。本文还提出一种季节性时间序列预报的新方法,它具有简单、易行,且精度高的特点。本文将多层递阶预报方法成功地应用于冷库自动控制系统,实现了对温度的实时预报,取得了满意的效果。 In this paper, we study deeply the multi-level forecasting method, propose a method of initial value selection, select AR model as forecasting model and use efficient Houscholder transform to determine model order, which makes this forecasting method easy and convenient Practical, forecasting accuracy has also been further improved. This paper also proposes a new method of seasonal time series forecasting, which is simple, easy and accurate. In this paper, the multilevel forecasting method is successfully applied to the automatic control system of cold storage, which realizes the real-time forecast of temperature and achieves satisfactory results.