Forecasting efficiency of two-stage Petrochemical sustainable supply chains using Deep Learning and DNDEA Model

The efficiency of supply chains is essential for improving managerial decision-making and enhancing strategic planning capabilities. This research presents a novel integration of deep learning with a two-stage supply chain framework to assess the efficiency of 28 petrochemical units over a period of 90 months. Based on sustainability principles, a dynamic network data envelopment analysis (DEA) model is employed to measure and compare the relative efficiency of supply chains operating across different time horizons. To forecast future input–output relationships in the supply chain, an advanced two-layer Long Short-Term Memory (LSTM) model is proposed. This LSTM-based prediction system demonstrated exceptional accuracy, achieving a low Mean Squared Error (MSE) of 0.0004 and a Root Mean Square Error (RMSE) of 0.0208. Additionally, the trend of the loss function during training confirmed the reliability and stability of the proposed deep learning approach. The precise forecasting capability of the LSTM model enables managers to proactively identify and address inefficiencies in production facilities before they occur, rather than relying on reactive strategies. This proactive approach allows for better resource allocation and improved operational performance across petrochemical supply chains. By integrating deep learning with dynamic network DEA models, this study offers a robust framework for predictive efficiency analysis and performance evaluation in industrial applications. The suggested framework provides decision-makers with a pragmatic assessment instrument to identify efficient and underperforming supply chains and set realistic benchmarks for improvement. This methodology is designed to be scalable and adaptable, making it suitable for real-world evaluations of multi-stage supply chains and production systems. The research culminates in a two-phase case study, illustrating the practical applicability of the proposed framework.