Passive learning to address nonstationarity in virtual flow metering applications

Steady-state process models are common in virtual flow meter applications due to low computational complexity, and model development maintenance cost. Nevertheless, the prediction performance of steady-state typically degrades with time inherent nonstationarity underlying being modeled. Few studies have investigated how learning methods can be applied sustain accuracy meters. This paper explores passive learning, where is frequently calibrated new data, as a way address improve long-term performance. An advantage that it compatible used industry. Two methods, periodic batch online varying calibration frequency train Six different types, ranging from data-driven first-principles, trained on historical production data 10 petroleum wells. The results two-fold: first, presence arriving measurements, frequent updating sustains an excellent over time; second, intermittent infrequently addition utilization expert knowledge essential increase accuracy. investigation may interest experts developing soft-sensors for nonstationary processes, such • Passive VFMs. VFM types Frequent key VFMs physical considerations advantageous little data.