题目：Large scale business forecasting: a hierarchical approach    

报告人：李丰副教授 （北京大学）  

摘要：Hierarchical forecasting with noisy and intermittent time series is a challenge in both research and empirical studies. Extensive research focuses on improving the accuracy of each hierarchy. This talk focus on advancing forecasting methodologies for hierarchical and ultra-long time series. We will start with a hierarchical forecasting with alignment approach that treats bottom-level forecasts as mutable, improving overall accuracy, especially in intermittent time series, using N-BEATS and LightGBM. Then we develops a reconciliation method that maintains immutability for a subset of forecasts in hierarchical structures, ensuring coherence and accuracy while preserving operational constraints, demonstrated with online retail sales data. In the end, we introduce a distributed framework for ultra-long time series forecasting using MapReduce, which enhances accuracy and computational efficiency by splitting the series into subseries and applying ARIMA models, proving especially effective for long forecast horizons.  

报告人简介：李丰博士就职于北京大学光华管理学院，任商务统计与经济计量系副教授。本科毕业于中国人民大学，博士毕业于瑞典斯德哥尔摩大学，研究领域包括贝叶斯统计学，大规模时间序列预测方法，大数据分布式学习等。李丰博士最新研究成果发表在统计期刊JCGS, 管理期刊EJOR (ABS4)，会计期刊CAR (FT50) 等。他同时著有 Bayesian Modeling of Conditional Densities，《大数据分布式计算与案例》和《统计计算》。  

报告时间：2024年10月31日（周四）上午10:30-12:30  

报告地点：教二楼727  

联系人：胡涛