A blending model for predicting flow rate of crude oil pipeline by integrating Leapienzon friction formula with random forest algorithm

Abstract

Flow parameters are crucial in pipeline operations. To address the challenge of accurately measuring the flow rate in crude oil pipelines, this paper proposes a hybrid model, i.e., the L-RF Blending model, which integrates the Leapienzon friction formula with the Random Forest (RF) model using a Blending approach. Compared to the Leapienzon friction formula and the existing predictive models that rely solely on RF, Support Vector Regression (SVR), Gaussian Process Regression (GPR), or Backpropagation (BP) algorithms, the L-RF Blending model better captures the complex relationships between the input and output variables. This is achieved by combining critical information from field sensor data with existing mechanistic knowledge. To assess the generalization of the Blending model, its performance is compared with other models using new datasets collected from an actual crude oil pipeline, demonstrating its potential for enhancing safety monitoring in pipeline management.