MOBAS-ALSTM: A Deep Learning and Metaheuristic Optimization Framework for Intelligent Carbon Footprint Accounting and Emission Strategy Optimization in Listed Companies

Abstract

In response to increasing environmental accountability and regulatory pressures, publicly listed companies are actively seeking intelligent solutions to reduce carbon emissions (CE) while maintaining operational efficiency. The research proposes a deep learning (DL)-based framework for carbon footprint (CF) accounting and emission reduction strategy optimization, specifically designed for publicly traded companies. The Multi-Objective Beetle Antennae Search-driven Adaptive Long Short-Term Memory (MOBAS-ALSTM) method forecasts energy consumption and optimizes emission reduction strategies. By combining MOBAS optimization with adaptive LSTM forecasting, MOBAS-ALSTM balances the requirements of social sustainability, cost, emissions, and accuracy. CEs are calculated using data collected from IoT sensors, smart meters, and enterprise resource planning (ERP) systems, supply chain logs, environmental, social, and governance (ESG) reports, and energy usage records from listed companies, incorporating real-time operational, financial, and supply chain datasets. Data preprocessing includes data cleaning and Z-score normalization, while feature extraction employs the discrete wavelet transform (DWT) to capture key frequency-time characteristics. The ALSTM component predicts future energy demands, while the MOBAS algorithm evaluates and allocates optimal emission reduction strategies, including improvements in energy efficiency, adoption of renewable energy, carbon tax modeling, and process reengineering. These strategies are evaluated based on expected emission reductions, economic costs, renewable energy contributions, and social impacts, including employment effects and public acceptance. The experimental validation of the MOBAS-ALSTM method outperformed the existing methods (CNN+RNN+RL, LSTM, and ANN,) in terms of prediction accuracy, with lower mean squared error (MSE) (0.00030), root mean squared error (RMSE) (0.0173), mean absolute error (MAE) (0.012), and a higher coefficient of determination (R²) (0.9985). The proposed DL based framework ensures transparency, scalability, and alignment with ESG reporting standards. The research establishes a robust, data-driven framework for listed companies to achieve Maximum CE and long-term carbon neutrality goals through intelligent carbon accounting and strategic emission control, thereby supporting regulatory compliance, stakeholder expectations, and sustainable growth.