Improving Energy Governance in the Cement Industry: OPTUNA-BILSTM Framework for Long-Range Electrical Load Forecasting

Abstract

Precise long-range forecasting of industrial electrical demand constitutes a strategic imperative for energy governance, particularly within energy-intensive, continuous-process industries such as cement manufacturing. This investigation conduct a systematic evaluation of two advanced recurrent neural networks, Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM), fine-tuned via the Optuna optimization. Using realistic data from an operational cement plant, we quantify the trade-offs between predictive accuracy, computational overhead, and operational viability. Empirical results demonstrate that the Optuna-optimized BiLSTM framework achieves superior predictive performance, yielding an R² score of 0.890 and a Mean Absolute Error (MAE) of 0.220. These findings delineate a modeling framework that reconciles high-fidelity forecasting with practical implementation, providing foundational guidelines for designing energy management systems that optimize maintenance, reduce costs, and advance sustainable operations.