Affiliations: [a] Department of Management Science and Technology, Guanghua School of Management, Peking University, Beijing, P.R. China | [b] State Grid Corporation of China, Xicheng District, Beijing, P.R. China
Correspondence:
[*]
Corresponding author. Ming Lei, Department of Management Science and Technology, Guanghua School of Management,
Peking University, No. 5 Yiheyuan Road Haidian District, Beijing, P.R. China. Tel./Fax: +86 10 62756243;
E-mail: leiming@gsm.pku.edu.cn.
Abstract: Demand forecasting is very important, both in academic and business practice. Intermittent demand forecasting has bothered managers and scholars for a long time. The zero demand point and extreme high values in the demand time series make it difficult to forecast. Under this condition, the fuzzy time series model performs well in eliminating the side effects of extreme data. To make our model more stable, we review the latest intermittent forecasting literature and combine the Multi-Aggregation Prediction Algorithm (MAPA) and the fuzzy Markov chain model to generate a new forecasting algorithm. In this paper, we use the material demand data of STATE GRID Corporation of China to test the forecasting accuracy of the new forecasting algorithm, as well as compare the results to the exponential smoothing (ES) model and the new forecasting algorithm. The results show that FMC-MAPA with an equal weight method in the time disaggregating process is the best forecasting method in this case. The forecasting ability of this method is more stable and robust in different fuzzy partition numbers and data adjustment than the ES and FMC model. We also study the impact of data adjustment on forecasting error. The results indicate that the unadjusted data have lower forecasting errors when compared to the linear trend adjustment, additive seasonal adjustment and the combination of these two adjustment methods.
