Abstract: The disclosed embodiments include reservoir simulation systems and methods to dynamically improve performance of reservoir simulations. The method includes obtaining input variables for generating a reservoir simulation of a reservoir, and generating the reservoir simulation based on the input variables. The method also includes determining a variance of computation time for processing the reservoir simulation. In response to a determination that the variance of computation time is less than or equal to a threshold, the method includes performing a first sequence of Bayesian Optimizations of at least one of internal and external parameters that control the reservoir simulation to improve performance of the reservoir simulation. In response to a determination that the variance of computation time is greater than the threshold, the method includes performing a second sequence of Bayesian Optimizations of at least one of the internal and external parameters.

Abstract: A drilling device may use a concurrent path planning process to create a path from a starting location to a destination location within a subterranean environment. The drilling device can receive sensor data. A probability distribution can be generated from the sensor data indicating one or more likely materials compositions that make up each portion of the subterranean environment. The probability distribution can be sampled, and for each sample, a drill path trajectory and drill parameters for the trajectory can be generated. A trained neural network may evaluate each trajectory and drill parameters to identify the most ideal trajectory based on the sensor data. The drilling device may then initiate drilling operations for a predetermined distance along the ideal trajectory.

Abstract: Aspects and features of a system for real-time drilling using automated data quality control can include a computing device, a drilling tool, sensors, and a message bus. The message bus can receive current data from a wellbore. The computing device can generate and use a feature-extraction model to provide revised data values that include those for missing data, statistical outliers, or both. The model can be used to produce controllable drilling parameters using highly accurate data to provide optimal control of the drilling tool. The real-time message bus can be used to apply the controllable drilling parameters to the drilling tool.

Abstract: System and methods for simulating fluid flow during downhole operations are provided. Measurements of an operating variable at one or more locations within a formation are obtained from a downhole tool disposed in a wellbore within the formation during a current stage of a downhole operation being performed along the wellbore. The obtained measurements are applied as inputs to a hybrid model of the formation. The hybrid model includes physics-based and machine learning models that are coupled together within a simulation grid. Fluid flow within the formation is simulated, based on the inputs applied to the hybrid model. A response of the operating variable is estimated for a subsequent stage of the downhole operation along the wellbore, based on the simulation. Flow control parameters for the subsequent stage are determined based on the estimated response. The subsequent stage of the operation is performed according to the determined flow control parameters.

Abstract: Embodiments of the subject technology for deep learning based reservoir modelling provides for receiving input data comprising information associated with one or more well logs in a region of interest. The subject technology determines, based at least in part on the input data, an input feature associated with a first deep neural network (DNN) for predicting a value of a property at a location within the region of interest. Further, the subject technology trains, using the input data and based at least in part on the input feature, the first DNN. The subject technology predicts, using the first DNN, the value of the property at the location in the region of interest. The subject technology utilizes a second DNN that classifies facies based on the predicted property in the region of interest.

Abstract: A system and method for controlling a drilling tool inside a wellbore makes use of simulated annealing and Bayesian optimization to determine optimum controllable drilling parameters. In some aspects, a computing device generates sampled exploration points using simulated annealing and runs a Bayesian optimization using a loss function and the exploration points to optimize at least one controllable drilling parameter to achieve a predicted value for a selected drilling parameter. In some examples, the selected drilling parameter is rate-of-penetration (ROP) and in some examples, the controllable drilling parameters include such parameters as rotational speed (RPM) and weight-on-bit (WOB). In some examples, the computing device applies the controllable drilling parameter(s) to the drilling tool to achieve the predicted value for the selected drilling parameter and provide real-time, closed-loop control and automation in drilling.

Abstract: A system and method for controlling a drilling tool inside a wellbore makes use of Bayesian optimization with range constraints. A computing device samples observed values for controllable drilling parameters such as weight-on-bit (WOB) and drill bit rotational speed in RPM and evaluates a selected drilling parameter such a rate-of-penetration (ROP) for the observed values using an objective function. Range constraints can be continuously learned by the computing device as the range constraints change. A Bayesian optimization, subject to the range constraints and the observed values, can produce an optimized value for the controllable drilling parameter to achieve a predicted value for the selected drilling parameter. The system can then control the drilling tool using the optimized value to achieve the predicted value for the selected drilling parameter.

Abstract: A system for real-time steering of a drill bit includes a drilling arrangement and a computing device in communication with the drilling arrangement. The system iteratively, or repeatedly, receives new data associated the wellbore. At each iteration, a model, for example an engineering model, is applied to the new data to produce an objective function defining the selected drilling parameter. The objective function is modified at each iteration to provide an updated value for the selected drilling parameter and an updated value for at least one controllable parameter. In one example, the function is modified using Bayesian optimization The system iteratively steers the drill bit to obtain the updated value for the selected drilling parameter by applying the updated value for at least one controllable parameter over the period of time that the wellbore is being formed.

Abstract: A system for determining completion parameters for a wellbore includes a sensor and a computing device. The sensor can be positioned at a surface of a wellbore to detect data prior to finishing a completion stage for the wellbore. The computing device can receive the data, perform a history match for simulation and production using the sensor data and historical data, generate inferred data for completion parameters using the historical data identified during the history match, predict stimulated area and production by inputting the inferred data into a neural network model, determine completion parameters for the wellbore using Bayesian optimization on the stimulated area and production from the neural network model, profit maximization, and output the completion parameters for determining completion decisions for the wellbore.

Abstract: A physics-influenced deep neural network (PDNN) model, or a deep neural network incorporating a physics-based cost function, can be used to efficiently denoise sensor data. To generate the PDNN model, noisy sensor data is used as training data input to a deep neural network and training output is valuated with a cost function that incorporates a physics-based model. An autoencoder can be coupled to the PDNN model and trained with the reduced-noise sensor data which is output from the PDNN during training of the PDNN or with a separate set of sensor data. The autoencoder detects outliers based on the reconstructed reduced-noise sensor data which it generates. Denoising sensor data by leveraging an autoencoder which is influenced by the physics of the underlying domain based on the incorporation of the physics-based model in the PDNN facilitates accurate and efficient denoising of sensor data and detection of outliers.

Abstract: Geosteering can be used in a drilling operation to create a wellbore that is used to extract hydrocarbons from a defined zone within the subterranean formation. According to some aspects, generating paths for the wellbore may include using path-planning protocols and pure-pursuit protocols. The pure-pursuit protocol may be executed to output a plurality of candidate drilling paths. The output may also include control parameters for controlling the drill bit. A trajectory optimizer may determine a result of multi-objective functions for each candidate path. A cost function may represent a cost or loss associated with a candidate path. Additionally, the trajectory optimizer may perform an optimization protocol, such as Bayesian optimization, on the cost functions to determine which candidate path to select. The selected candidate path may correspond to new control parameters for controlling the drill bit to reach the target location.

Abstract: Certain aspects and features relate to a system for trajectory planning and control for new wellbores. Data can be received for multiple existing wells associated with a subterranean reservoir and used to train a deep neural network model to make accurate well property projections at any other location in the reservoir. A model of features for specific well locations based on seismic attributes of the well location can be automatically generated, and the model can be used in drilling trajectory optimization. In some examples, the system builds a deep neural network (DNN) model based on the statistical features, and trains the DNN model using Bayesian optimization to produce an optimized DNN model. The optimized model can be used to provide drilling parameters to produce an optimized trajectory for a new well.

Abstract: A model optimizer for predicting a drill bit variable can select a model from multiple models based on a learned preference. The preference may be updated according to preference indicator received from a user in response to an output model selection.

Abstract: Aspects of the subject technology relate to systems and methods for predicting physical characteristics of a physical environment using a physical characterization model trained based on simulated states of a modeled physical environment. A physical characterization model can be generated based on a plurality of simulated states of a modeled physical environment. Specifically, the physical characterization model can be trained by mapping simulated spatial properties of the modeled physical environment temporally across the plurality of simulated states of the modeled physical environment. Further, input state data describing one or more input states of a physical environment can be received. One or more physical characteristics of the physical environment can be predicted by applying the physical characterization model to the one or more input states of the physical environment.

Abstract: Aspects of the present disclosure relate to receiving data associated with a subterranean reservoir to be penetrated by a wellbore and training a neural network with both the data and a physics-based first principles model. The neural network is then used to make predictions regarding the properties of the subterranean reservoir, and these predictions are in turn used to determine one or more controllable parameters for equipment associated with a wellbore. The controllable parameters can then be used to control equipment for formation, stimulation, or production relative to the wellbore.

Abstract: A method for detecting anomalies in a piece of wellsite equipment. The method may include measuring data related to the piece of wellsite equipment. The method may also include encoding the measured data with a first autoencoder to produce a first set of encoded data. The method may further include performing a first Gaussian process regression (“GPR”) on the first set of encoded data to produce a first set of results that identifies a first anomaly in the measured data and that provides a first confidence interval for the first anomaly.

Abstract: A system and method for controlling multiple drilling tools inside wellbores makes use of Bayesian optimization with range constraints. A computing device samples observed values for controllable drilling parameters such as weight-on-bit, mud flow rate and drill bit rotational speed in RPM and evaluates a selected drilling parameter such a rate-of-penetration and hydraulic mechanical specific energy for the observed values using an objective function. Range constraints including the physical drilling environment and the total power available to all drilling tools within the drilling environment can be continuously learned by the computing device as the range constraints change. A Bayesian optimization, subject to the range constraints and the observed values, can produce an optimized value for the controllable drilling parameters to achieve a predicted value for the drilling parameters.

Abstract: Systems and methods for controlling drilling operations are provided. A controller for a drilling system may provide drilling parameters such as weight-on-bit and rotation rate parameters to the drilling system, based on a machine-learned reward policy and a model-based prediction. The machine-learned reward policy may be generated during drilling operations and used to modify recommended values from the model-based prediction for subsequent drilling operations to achieve a desired rate-of-penetration.

Abstract: A system and method for controlling a gas supply to provide gas lift for wellbore(s) using Bayesian optimization. A computing device controls a gas supply to inject gas into wellbore(s). The computing device receives first reservoir data associated with a first subterranean reservoir and simulates production using the first reservoir data, using a model for the first subterranean reservoir. The production simulation provides first production data. The computing device receives second reservoir data associated with a subterranean reservoir and simulates production using the second reservoir data, using a model for the second subterranean reservoir. The production simulation provides second production data. A Bayesian optimization of an objective function of the first and second production data subject to any gas injection constraints can be performed to produce gas-lift parameters. The gas-lift parameters can be applied to the gas supply to control injection of gas into the wellbore(s).

Abstract: A system and method for controlling a gas supply to provide gas lift for a production wellbore makes use of Bayesian optimization. A computing device controls a gas supply to inject gas into one or more wellbores. The computing device receives reservoir data associated with a subterranean reservoir to be penetrated by the wellbores and can simulate production using the reservoir data and using a physics-based or machine learning or hybrid physics-based machine learning model for the subterranean reservoir. The production simulation can provide production data. A Bayesian optimization of an objective function of the production data subject to any gas injection constraints can be performed to produce gas lift parameters. The gas lift parameters can be applied to the gas supply to control the injection of gas into the wellbore or wellbores.