2024

Mirasci Seray; Uygyr Sara; Aksoy Asli

The increasing energy demand and costs in the industrial sector necessitate effective energy management strategies. This study investigates a food processing company with an on-site cogeneration system, which faces challenges of high energy costs and fluctuating energy demand due to its seasonal production. During off-peak seasons, surplus energy is generated and frequently sold at reduced rates, thereby increasing operational inefficiencies. Conversely, during on-peak seasons, the company faces heightened energy demands and increased costs, further complicating energy management and impacting overall operational effectiveness. To address these challenges, an energy consumption forecasting model (ECFM) has been developed which employs Quantile Regression (QR) as a statistical method and different machine learning (ML) algorithms, including Decision Trees (DT), Boosted Trees, Bagged Trees, and Artificial Neural Networks (ANN). Although QR is an effective method for handling non-normally distributed data, it is inadequate for capturing the high volatility of energy consumption in this study. Among the ML models, the bi-layered ANN demonstrated the most effective performance achieving the lowest forecasting errors and demonstrating a 52.42% reduction in CO2 emissions. This reduction is consistent with the company's decarbonization strategies and regulatory compliance goals. The findings highlight the potential of advanced ML models, particularly the bi-layered ANN, to enhance the accuracy of energy forecasting, reduce greenhouse gas emissions, and create competitive advantages in industrial settings. This study contributes to the growing body of knowledge on the integration of operational efficiency with environmental sustainability in energy management practices. It demonstrates the potential of advanced forecasting models to support the development of robust and sustainable energy solutions across a range of industrial contexts.