Promoted predictions of screen-out probability by model inputs optimization based on (a) Well A, (b) Well B, (c) Well C and (d) Well D. The inputs of the model are consist of the original measurements and three optimized indicators (Vs/Vf + DPP + ΔP/ΔVf). The blue dashed circle marks the false alarm of the screen-out in the Wells C and D cases. The solid orange line is the labelled screen-out probability curve. The dashed line is the predicted probability curve.

Contexts in source publication

Context 1

... successfully reports all of the screen-out events and produces the lowest RMSE in all testing rounds. Therefore, the indicators in the first three rounds are all selected. The σmin/Ps in the last round is abandoned due to the increasing RMSE and false-negative error, as shown in Table 5. The predictions in the third-round test are shown in Fig. 3. Three false-positives are observed for Wells C and D that are located distant from the training cases (wells), as marked by dashed circles ...

Context 2

... curve before the screen-out also provides clues for the inducement analysis. In Fig. 3 (a), the long proppant injection slug between 4000s and 5000s increases the screen-out probability near to the value of 0.8, indicating that the continuous-proppant-injection length may be one of the key factors causing the screen-out in this case. In Fig. 3 (b), a vertical ascent of probability is observed around 3000s, when the 40/70 mesh proppant is injected right after the 100 mesh one. The fracture may not be well cracked with a relatively narrow width that is sensitive to larger proppant grain. The underdeveloped fracture may be an important inducement for Well B. The pressure ...

Context 3

... Wells C and D, the sign showing in pressure is inconspicuous because it deviates only slightly from the norm, as shown in Figs. 3 (c) and (d). The probability can still define distinct indications precursory to screen-out. A long proppant slug may be one of the main reasons causing the screen-out in Well C, according to the rapid climb in probability between 6000s and 8000s in Fig. 3 ...

Context 4

... in pressure is inconspicuous because it deviates only slightly from the norm, as shown in Figs. 3 (c) and (d). The probability can still define distinct indications precursory to screen-out. A long proppant slug may be one of the main reasons causing the screen-out in Well C, according to the rapid climb in probability between 6000s and 8000s in Fig. 3 ...

Context 5

... predicted probability of screen-out varies in a similar period to the wellhead pressure variations (Fig. 3), indicating that the GRU model interprets the relationship between wellhead pressure and screen-out events when the data are manually trimmed. The predicted curves fluctuate around the labelled curves due to the finite and substantial variation in evolving fracture volume, which supports the linear hypothesis of screen-out probability. ...