Data clustering for model-prediction discrepancy reduction – A case study of solids transport in oil/gas pipelines

• Solid particles are not fully avoidable and pose danger for oil and gas pipelines.
• There are 62 models that predict the minimum velocity to transport solids.
• The predictions of these models differ several orders of magnitude for same input.
• Data clustering and model selection approach is outlined to lower this discrepancy.

The minimum fluid-flow velocity to ensure particle transport in pipelines is an essential design and operation consideration for oil and gas production. This flow velocity is difficult to estimate due to complex nature of the physical processes. It has been shown that the predictions of different, alternative models may vary several orders of magnitude for the same inputs. This paper introduces a systematic approach to reduce this discrepancy using data clustering, model selection, and cluster identification techniques. The approach is tested using 772 experimental data points (published in open literature), and the results show that the average of the error percentages between the predictions and experimental velocities are reduced from several orders of magnitude to 37%.