Search for a command to run...
Abstract Hydraulic fracturing is performed to intensify hydrocarbon production by applying high pressure to the productive formation in order to initiate the creation of a fracture and its further development in the formation with subsequent placement of a propping agent (proppant) to ensure inflow to the production well. The job is performed as per preliminary design according to the model, which is necessary for achieving the optimal parameters of the fracture. At the modeling stage, the expected parameters of the hydraulic fracture and production rate assessment is made on the basis of information on the physical properties of the formation according to core studies and well logging data, data from previously performed fracturing jobs on adjacent wells, a job plan, information on the types and properties of fracturing fluids and proppants. After conducting hydraulic fracturing and calibrating the fracture model based on the obtained data, it becomes possible to form a more accurate representation of the actual fracture parameters for productivity evaluation. However, this representation comes with a significant degree of uncertainty regarding the actual fracture geometry and fracture planes propagation direction, the uniformity of change in the fracture half-length within the reservoir in both directions from the wellbore (symmetry), and the distribution of proppant volume between the fracture planes. Under these circumstances, making decisions regarding the optimization of the hydraulic fracturing program for a field remains a highly challenging task. For the studied waterhammer signals, it is difficult to obtain the true values of the fracture parameters; the available data is a result of related inverse problems` solving. Recording the pressure signal with a high sampling rate (in practice, up to 200 Hz) makes possible a detailed analysis of the oscillations observed after pumps shutdown (Ma et al., 2019; Qiu et al., 2023). One of the problems that can be solved by analyzing the high-frequency pressure signal is the localization of the hydraulic fracturing zone (Oparin et al., 2024). Knowing the fracturing zone and assuming the influence of the fracture geometry on the shape of the recorded signal, the task of estimating the fracture parameters is set. The waterhammer signal recorded at the wellhead can have a complex shape due to the well design, fluid properties, fracture geometry, and reservoir properties. The variable velocity of acoustic wave propagation along the wellbore further complicates the solution of the inverse problem. Currently, there is an increasing number of signal analysis techniques based on the utilization of various novel neural network algorithms. The development and implementation of these solutions in production processes allows to explore new approaches to data management and significantly reduce production costs. On the other hand, it is of the highest priority for the oil industry to ensure the reliable operation and safety of any forecasts, and therefore, it is necessary to monitor the physical plausibility of neural network algorithms or apply it in combination with robust mathematical modeling. This paper proposes a method for estimating fracture parameters based on the analysis of waterhammer signals. For this purpose, the use of the latent space parameters of a variational autoencoder trained on a set of simulated waterhammer data is proposed. Based on the field data analysis results, the possibility of additional estimation of fracture parameters is demonstrated.