Advanced Deep Learning for microseismic events prediction for hydraulic fracture treatment via Continuous Wavelet Transform

Abstract

Understanding hydraulic fracture propagation events would enable operators to estimate the fracture geometry, understand formation/fracture interactions, and estimate the total microseismic events cloud. This paper introduces a new method for understanding the dynamic fracture propagation events using a signal processing technique. Detection of the dynamic fracture event using continuous wavelet transform (CWT) is used to train a deep learning model with microseismic events to predict this cloud for every hydraulic fracture job. The deep learning model is used for the same formation. The CWT for treatment pressure is a convolution process that involves the application of a wavelet signal to the acquired pressure signal in a continuous manner. The wavelet is stretched, or compressed, and is moved along the acquired pressure signal, acting as a generic microscope to highlight small changes in the treating pressure signal. Using CWT as a mathematical microscope to discern diverse signals has been adopted by numerous engineering and medical applications since the 1990s.A normalized CWT scalogram for fracture-treating pressure is thus created as a unique representation of each fracture propagation mode in the hydraulic fracturing job. The new technique is calibrated with several fracture simulation runs and real field data using the previous techniques like moving reference point (MRP). That previous technique which has been previously validated through fracture simulation, is used as a calibration for the new normalized CWT scalogram technique. The technique was finally calibrated by comparing it with micro-seismic events recorded by the Marcellus Shale Energy and Environment Laboratory (MSEEL).The results of this comparison were exceptionally accurate. It can be used to train a deep learning model that can be used to estimate the cloud of microseismic events for hydraulic fracture jobs in Marcellus shale. The new deep learning framework uses CWT to convert the treating pressure during hydraulic fracture propagation to dimensionless representation (normalized CWT scalogram) which acts as a mathematical microscope for the fracture treating pressure on different wavelet stretch or compressions, enabling the detection of the major changes in the treating pressure then train the deep learning model to estimate microseismic events cloud.