Data-Driven Predictive Model for Mixed Oil Length Prediction in Long-Distance Transportation Pipeline

During the sequential transportation of refined oil pipelines, mixed oil in the intersection area of two different types of oil needs to be identified and cut. Therefore, prediction of the length of the mixed oil (LMO) is significant for scheduling and optimization. Presently, the models used for such tasks are all developed by first-principles (FP), which suffers from the limitations of strong assumptions and ignorance of important features. In this paper, we first study the application of pure data-driven predictive models for predicting the LMO, which is able to deal with the drawback of the FP-based models. Then, the performance of the FP-aided data-driven models are further investigated. Two fundamental data-driven learning models, namely the partial least squares (PLS) and the deep neural network (DNN), are employed to construct the regression models. The performance of the PLS and DNN, as well as those of the FP-aided PLS and DNN are evaluated using real-life dataset from the database of long-distance pipelines.