Random forest forecasting of time to failure for Granite hydraulic fracturing using acoustic emission signals

Geothermal energy is a key resource to support carbon-neutral targets for its high energy baseload production. However, its utilization often involves hydraulic fracturing that can induce earthquakes. Accurate forecasting of the timing of these fractures and associated seismicity can inform hazard mitigation strategies where traditional methods often fall in short. Here, we utilize acoustic emission (AE) signals obtained from a series of hydraulic fracturing experiments on Barre Granite under a variety of stress regime to develop a random forest model to forecast the time to failure. The failure time is defined by a large pressure drop denoting the propagation of unstable macro-scale hydraulic fractures to the edges of the specimen. We achieve a coefficient of determination R ² of 0.97 on test data using 50% of the data as the training set. We find that the first 20 statistical features constitute 100% of the contribution to the forecast, where the top 4 features are mean, minimum and kurtosis of the first finite difference and skewness of the signal voltage. At our given injection rate, our model can accurately forecast 1000 to 1800 s before failure compared to the 18 to 69 s in advance when using a benchmark inverse AE rate model. Our results suggest that it is possible to forecast failure of a rock specimen prior to the onset of accelerated seismic release, with implications for managing induced seismicity hazards.