| Title |
Demand and Load Forecasting Methods Based on Operational Information in Factory Energy Management Systems |
| Authors |
김선혁(Seonhyeog Kim) ; 이일우(Il-woo LEE) ; 허태욱(Taewook Heo) |
| DOI |
https://doi.org/10.5370/KIEE.2025.74.9.1440 |
| Keywords |
Factory Energy Management System; Load Forecasting; Long Short-term Memory; Manufacturing Execution System; Quality Control Charts |
| Abstract |
This paper proposes a method for load prediction and analysis in a factory energy management system (FEMS). Since the electricity demand of a factory is largely dependent on the type of products produced, the temperature and humidity inside the factory, seasonality, and the state of the process, the variability of the process data and the deviation of the electricity demand appear significantly, so a load prediction method that considers the manufacturing execution system (MES) that reflects and manages the characteristics of the process is necessary. In addition, since the process data includes process control charts and abnormal data outside the normal range, it is expected to improve the performance of the learning model by applying control chart analysis techniques to remove data that hinder the performance of load prediction. Understanding each process, utility, and equipment within the factory is also essential for effective energy analysis. The Shewhart individuals control chart can be utilized to eliminate anomalous data, thereby improving the performance of the training data-set. The successful implementation and operation of FEMS require data synchronization, missing value treatment, and periodicity analysis in the data refinement process. Additionally, an AI-based load forecasting model allows for more precise energy demand predictions, enabling factory operators to derive optimal energy-saving strategies.
This study makes the following contributions: (1) it proposes a data preprocessing framework tailored to different industrial sectors such as biotechnology and paper manufacturing; (2) it integrates MES-based production schedules into load forecasting to reflect real-world operational variability; and (3) it demonstrates the practical applicability of LSTM-based prediction models with validated performance in actual factory datasets. This approach significantly contributes to reducing power consumption and enhancing energy efficiency in South Korea's industrial sector. |