[2I5-GS-10] AI application: Industry

[2I5-GS-10-05] A Solution Approach to Energy Plant Operation Planning Problems Using Transformers

〇Tomohiro Suzuki¹, Takumi Ozeki², Naoki Mori², Hideaki Hirano³, Shoichi Kitamura³, Kazuyuki Mori³ (1. Osaka Prefecture University, 2. Osaka Metropolitan University, 3. Mitsubishi Electric Corporation)

Keywords:Energy Plant, Transformer, Positional Embedding, Time Series Analysis

Energy plants account for most energy consumption in factories and other facilities. Benchmark problems for energy plant operation planning have been proposed for energy conservation. However, to apply the benchmark problem to real energy plants, it is necessary to have a general-purpose and fast operational planning method that can not only search for solutions to the benchmark problem but also deal with different situations.
In this study, we propose a method for regression prediction of solutions to similar problems using a deep learning model Transformer with Positional Embedding, representing time series information. In numerical experiments, we compared the solutions generated by the Transformer and the comparison method in terms of energy purchase cost and constraint violation and confirmed the effectiveness of the proposed method.