混合区间多尺度分解的区间时间序列组合预测

鉴于传统预测方法一直基于“点”来衡量时间序列数据,然而现实生活中在给定的时间段内许多变量是有区间限制的,点值预测会损失波动性信息。因此,本文提出了一种基于混合区间多尺度分解的组合预测方法。首先,建立区间离散小波分解方法(IDWT)、区间经验模态分解方法(IEMD)和区间奇异普分析方法(ISSA)。其次,用本文构建的IDWT、IEMD和ISSA对区间时间序列进行多尺度分解,从而得到区间趋势序列和残差序列。然后,用霍尔特指数平滑方法(Holt's)、支持向量回归(SVR)和BP神经网络对区间趋势序列和残差序列进行组合预测得到三种分解方法下的区间时间序列预测值。最后,用BP神经网络对各预测结果进行集成得到区间时间序列最终预测值。同时,为证明模型的有效性进行了AQI空气质量的实证预测分析,结果表明,本文所提出基于混合区间多尺度分解的组合预测方法具有较高的预测精度和良好的适用性。

An Interval Time Series Combination Forecasting Approach Based on Hybrid Interval Multi-scale Decomposition

Since traditional forecasting methods have always been based on “points” to measure time series data. however, in real life situations, many variables are limited in a given time period, and point value forecasting will lose volatility information. In order to solve this problem, this paper proposes a new combination time series forecasting method, which is based on hybrid interval multi-scale decomposition. First, we establish interval multi-scale decomposition methods, including interval discrete wavelet transform method (IDWT), interval empirical mode decomposition method (IEMD) and interval singular spectral analysis (ISSA), which can decompose the interval time series into interval trend and residuals. Then, the Holt's exponential smoothing method (Holt's), support vector regression (SVR) and BP neural network are chosen to forecast the interval trend and residuals. Furthermore, the forecasting results of all the obtained components are combined to generate the aggregated interval-valued output by employing back propagation neural network. Finally, the proposed combination approach is employed for real interval AQI time series forecasting. The experimental results demonstrate that the proposed forecasting method can produce much better forecasting performance than some existing benchmark models.