2025/07/09 by Tavakoli Garmaserh, Ehsan, Emadi, Mehran
#Deep Learning #Electricity Consumption Forecasting #Long Short-Term Memory #Petrochemical Industrial Units
paper · doi:10.57647/j.mjee.2025.1902.41
Accurate forecasting of electricity consumption in petrochemical industrial units is essential for optimizing energy management and ensuring operational efficiency. This study presents a novel deep learning framework that integrates advanced feature engineering and Long Short-Term Memory (LSTM) networks to address the challenges posed by irregular seasonal trends and dynamic consumption patterns. Key innovations include the use of Fourier Transform-based feature extraction for enhanced data representation and a hybrid genetic-sparse matrix optimization technique for feature selection, ensuring high predictive performance. The proposed method effectively mitigates issues related to data irregularities through preprocessing techniques, resulting in improved accuracy and stability in both univariate and multivariate time series forecasting scenarios. Experimental evaluations using benchmark datasets demonstrate significant improvements, achieving a Root Mean Square Error (RMSE) of 0.0693 and a Mean Absolute Percentage Error (MAPE) reduction of over 15% compared to state-of-the-art methods. These results highlight the robustness and practical applicability of the proposed framework for industrial energy consumption forecasting and sustainable energy management.