Abstract
The manufacturing sector is a major source of global greenhouse gas emissions, yet existing forecasting approaches often rely on sectoral or national data, offering limited operational insights for individual plants. This lack of plant-level forecasting tools forces industries into reactive emission management, risking inefficiencies and compliance penalties. This study addresses this critical gap by developing a BiLSTM-based forecasting framework tailored to daily industrial operations, validated on 1,704 records. The proposed model achieved an RMSE of 0.140, representing a 15-20 % improvement over conventional models (ARIMA, Linear Regression, and XGBoost). Feature analysis using SHAP identified electricity consumption as the dominant driver of emissions, translating directly into actionable strategies for decarbonization. The framework is deployed through an interactive dashboard supporting real-time forecasting and "what-if" scenario analysis. By bridging advanced deep learning with plant-level applicability, this study provides a scalable pathway for industries to transition from reactive compliance to proactive emission reduction.
