Natural gas pipeline integrity management efficiency evaluation and prediction: A DEA-Malmquist-ML approach

Evaluating Natural Gas Pipeline Integrity Management (NGPIM) efficiency is essential for ensuring pipeline performance and continuous improvement. Traditional evaluation methods typically concentrate on assessing historical efficiencies at the enterprise level of operators and are unable to forecast future efficiency. This study proposes a DEA–Malmquist–ML model to evaluate and predict NGPIM efficiency across 42 U.S. states. From 2010 to 2023, 18 states recorded average annual static efficiency above 1, showing an “eastern high, western low” spatial pattern, while 37 states exhibited dynamic efficiency above 1, mainly driven by technological progress and scale efficiency change. Overall, U.S. NGPIM efficiency remains high and shows a gradual upward trend. Panel Tobit regression identifies factors influencing static and dynamic efficiency. The proposed SACBi-LSTM model—integrating Convolutional neural network, bi-directional long and short-term memory network, and sparse attention—achieves 90 % average accuracy across 11 indicators, with prediction errors for 33 states within 0.1, confirming strong model adaptability. This regional efficiency evaluation and forecasting framework supports evidence-based policymaking and rational resource allocation.