Deeppipe: A hybrid intelligent framework for real-time batch tracking of multi-product pipelines

Abstract

To improve the economy and meet the demand for transporting different oil products, multi-product pipelines are utilized to transport multi-products in sequence in the same pipelines. It is fundamentally important for operators in stations to know the accurate location of the head of each batch interface, to swing the valve at a station, and deliver oil products with minimal contamination. However, it is difficult to determine the location of the batch interface accurately, due to the complex hydrothermal conditions and mixed oil segment. In this paper, a hybrid intelligent framework is proposed to track the real-time batch interface of multi-product pipelines. The batch injection judgment module is applied to determine whether there is a new product batch injected in the pipeline. Applying the upstream and downstream flowrate, the volume calculation model is proposed to track the real-time location of each batch interface. Considering the deviation between the estimated location and the actual location of the batch interface, a self-learning modified model is proposed to compensate for the tracking errors. Taking a real-world multi-product pipeline network in China as an example, the accuracy and efficiency of the proposed model are verified. The results suggested that the hybrid intelligent framework outperforms other comparative methods, with minimal tracking errors being 3.79 min