Unveiling Key factors governing seismogenic potential and seismogenic productivity of hydraulic fracturing pads: Insights from machine learning in the Southern Montney Play

•Machine learning reveals key new geolgoical/operational factors in seismogenic pattern.•Targeted formation in HF stimulations determines seismogenic productivity.•Prior HF pads' cumulative injected volumes are crucial for seismogenic potential.•Different mechanisms drive seismogenic productivi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Earth and planetary science letters 2024-01, Vol.626, p.118511, Article 118511
Hauptverfasser: Wang, Bei, Kao, Honn, Yu, Hongyu, Li, Ge, Dokht, Ramin M.H., Visser, Ryan
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•Machine learning reveals key new geolgoical/operational factors in seismogenic pattern.•Targeted formation in HF stimulations determines seismogenic productivity.•Prior HF pads' cumulative injected volumes are crucial for seismogenic potential.•Different mechanisms drive seismogenic productivity and seismogenic potential.•Insights on key seismic controlling factors enable seismogenic pattern forecasts. One of the most challenging works in mitigating seismic hazard related to injection-induced earthquakes (IIE) is to adequately estimate the activity level of IIE caused by individual hydraulic fracturing (HF) pad. Reliable estimation is largely dependent on comprehensively deciphering the relationships between IIE and the controlling factors. To deal with this conundrum, in this study, we first build an enhanced earthquake catalog with more than 30,000 earthquakes for the southern Montney Play (SMP), in western Canada between 2014 and 2022. We then associate these detected earthquakes to their corresponding HF pads (357 in total). We train the supervised machine-learning algorithm “eXtreme Gradient Boosting” (XGboost) to systematically evaluate how seven geological and eight operational factors contribute to the seismogenic potential (occurrence of IIE) and seismogenic productivity (number of IIE) of these HF pads. Our XGBoost classification model first suggests that cumulative injected volume and the location of the HF pads are the key factors deciding whether IIE would occur or not. Next, using the Shapley Additive Explanations (SHAP) values to quantitatively interpret the outputs from XGBoost, we reveal that contributions from operational and geological factors could be comparable to the seismogenic productivity. The top 3 most influential features are the number of HF stages targeting the Lower Middle Montney formation, cumulative volume from preceding injections, and whether the HF pads are located within the Fort St. John graben. In the interim, the seismogenic potential model reveals that the ranking of determinants governing the seismogenic productivity of IIE is somewhat different, signifying distinct fundamental mechanisms that influence the two phenomena (occurrence vs. number). Our results provide comprehensive understanding on the key operational and geological factors controlling seismogenic pattern and pave the way for forecasting the activity level of HF-related IIE for individual HF pad in the SMP area.
ISSN:0012-821X
1385-013X
DOI:10.1016/j.epsl.2023.118511