Informer-Based Power Load Forecasting Model for Electrolytic Aluminium Smelters

Maomao Ding¹Shiyao Cheng¹Tianpeng Xia¹Zhongwei Cai¹Boyang Chen¹Huixian Zhu^2*

• ¹Customer Service Center, State Grid Corporation of China, Tianjin, China
• ²Beijing Tsingsoft Technology Co., Ltd, Beijing, China

As the most energy-intensive stage in aluminium industries, electrolytic aluminium smelters account for 40% of global industrial load. This significant percentage signifies a crucial source of flexibility on the industrial demand side. Consequently, accurately forecasting its power load is fundamental to unlocking and utilizing its significant regulation potential. This study addresses the complex task of mid-to-long term load forecasting for electrolytic aluminium smelters, which requires analyzing yearly patterns, growth trends, and unpredictable fluctuations. By leveraging the advanced features of the informer mechanism, the proposed approach introduces a multifaceted ensemble strategy. It is characterized by: 1) utilizing a hierarchical decomposition approach to meticulously uncover and emphasize the intrinsic characteristics present in mid to long-term power load for electrolytic aluminium smelters; 2) employing a dedicated long sequence time series data forecasting mechanism to precisely capture and model the underlying trends in the data; 3) integrating an Adversarial Autoencoder and Long Short-Term Memory ensemble model to creatively assimilate and predict the residual components of power load by effectively considering random fluctuations. The effectiveness and accuracy of the proposed approach are rigorously validated using historical power load data from some electrolytic aluminium suppliers in China. This validation process involves a comparative analysis with various traditional algorithms, thereby establishing the superior performance and reliability of the proposed strategy in capturing nuances of electrolytic aluminium load forecasting.