A novel predictive model of mixed oil length of products pipeline driven by traditional model and data

Mixed oil will inevitably form during the batch transportation in products pipeline, and accurate calculation of mixed oil length is of great significance for the economic benefit of the oil pipeline network. Traditional models (White Box/WB) are based on underlying engineering principles, however, there are some problems including low accuracy or high computing complexity. Data model (Black Box/BB), developed by historical data and machine learning algorithms, could act as a problem-solving approach while it takes the risk of overfitting due to the lack of understanding of the physical process. In this article, the Austin-Palfrey equation is used as the WB model to convert original features and input them into the BB model built on the Gradient Boosting Decision Tree algorithm (GBDT), and finally a novel predictive model (Grey Box/GB) incorporated prior knowledge is proposed. Experimental results of industrial examples reveal that the new model has higher predictive accuracy and better robustness, which means the dual-driven modeling method is more effective and can even become a new and valuable research direction.