Decformer: A Multi-model Fusion Method Based on Decomposition and Transformer for Industrial Energy Consumption Prediction

As the industrial Internet of Things (IIoT) advances, the need for precise and effective long-term energy consumption forecasting grows more critical. Addressing this challenge relies on the technology of Time Sequence Forecast (TSF). Some ad-vanced deep learning methods have been used for TSF. However, existing work suffers from overly complex model structures and a lack of consideration for characteristics of the specific data. Therefore, we propose a time series prediction method called Decformer based on sequence decomposition and multi-model fusion. For components exhibiting strong periodicity, we use a simplified Transformer model with channel-independence processing to efficiently extract time-dimension feature information. Experimental results on multiple datasets demonstrate that the proposed method outperforms state-of-the-art time series prediction methods.