Multiphase flow rate prediction using chained multi-output regression models

Virtual flow meters (VFM) are emerging as an attractive and cost-effective alternative to traditional multiphase flow meters to meet monitoring demands, reduce operational costs, and improve oil recovery efficiency. However, no previous studies have accounted for the correlations between the oil, water, and gas flow rates when developing machine learning models. This study proposes a chained regression model for multiphase flow rate prediction to account for such relationships. Real-field data consists of 375 data points for sensory measurements, including pressure, temperature, and choke opening levels, and 42 data points for oil, water, and gas flow rates that were measured downstream of the wellhead, which was acquired over one month. Two robust algorithms, Support Vector Machine (SVM) and Gaussian Process (GP), were employed to develop the chained regression model. The evaluation metrics such as mean absolute percentage error (MAPE) for all the models were estimated using a repeated hold-out approach of cross-validation. The response variables, i.e., the three flow rates, were moderate to strongly correlated. The results showed that the GP-based chain regression model was significantly better than the direct model using the GP algorithm for oil (MAPE: 2.07% vs. 2.27%) and gas (MAPE: 2.5% vs. 2.65%) flow rate prediction (p < 0.01). Overall, the chained model is generally superior to the direct model for flow rate prediction, which was supported by the ranking scores, consistently outperforming the latter in both SVM (79 vs. 87) and GP (64 vs. 70) based approaches. The sensitivity analysis showed that the GP-based chained model accurately predicted oil, water, and gas flow rates within 39.45 m3/day, 14.69 m3/day, and 5.63 m3/day, respectively, of the actual values for approximately 92% of the data points. This study’s findings can be instrumental in designing and developing practical and accurate VFM for multiphase flow rate prediction.