Mid-Term Energy Consumption Prediction Method for Large-Scale Natural Gas Pipelines Using Multi-Timescale Information

As the most widely used approach for long-distance natural gas transport, large-scale natural gas pipelines consume substantial amounts of energy. Effective scheduling and management of these pipelines necessitate accurate energy consumption predictions. However, mature mid-term prediction techniques for natural gas pipelines are currently lacking in engineering practice. This study aims to apply knowledge distillation techniques to develop a mid-term energy consumption prediction model for natural gas pipelines. By training with the assistance of a complex teacher model, the knowledge distillation model leverages small-scale data to enhance its accuracy. Unlike the teacher model, which requires contemporary flow rate inputs, the knowledge distillation model can achieve similar accuracy using only historical data, thereby increasing its practical applicability. The proposed method was evaluated using field data from two pipelines, and the performance as well as the impact of various parameters were thoroughly discussed.