Machine learning application in batch scheduling for multi-product pipelines: A review

Abstract

Batch scheduling is a crucial part of pipeline enterprise operation management, especially in the context of market-oriented operation. It involves 3 main tasks: quickly preparing batch plans, accurately tracking interface movement, and operation condition in real time. Normally, the completion of multi-product pipeline batch scheduling depends on simulation models or optimization models and corresponding conventional solving algorithm. However, this approach becomes inefficient when applied to large-scale systems. The rapid development of machine learning has brought new ideas to batch scheduling research. This paper first reviews the current state of batch scheduling technology, and suggests that applying machine learning to it is a promising development direction. Then, we summarize the progress of machine learning applications in batch planning, interface movement tracking, and operational condition monitoring, and point out their limitations. Finally, considering the separation of refined oil production, transportation, and sales processes, 5 recommendations are put forward: oil supply and demand prediction and pipeline capacity prediction, batch planning, batch interface movement tracking, mixed oil development monitoring, and pipeline operation condition identification.