EP-4737679-A1 - SYSTEMS AND METHODS FOR ENHANCING WIRELINE CONVEYANCE MODELING IN WIRELINE CONVEYANCE USING PHYSICS-INFORMED NEURAL NETWORK

EP4737679A1EP 4737679 A1EP4737679 A1EP 4737679A1EP-4737679-A1

Abstract

Systems and methods for enhancing wireline conveyance modeling in wireline conveyance using a physics-informed neural network (PINN) are provided. A method includes: receiving historical job data from at least one past wireline conveyance operation, the historical job data including: static input data, and timeseries input data, processing the historical job data through a neural network (NN) configured to learn the relationships between input features and target variables including: an input layer, a plurality of hidden layers, and an output layer, integrating the neural network with a numerical physics-based wireline conveyance model to form a physics-informed neural network (PINN), training the PINN using the historical job data and estimation results of the physical model to enhance prediction accuracy, utilizing the trained PINN to predict at least one physical parameter for a current wireline conveyance operation, and outputting the predicted at least one physical parameter for use in the wireline conveyance operation.

Inventors

WANG, JIACHENG
BAKLANOV, NIKOLAY

Assignees

Services Pétroliers Schlumberger
Schlumberger Technology B.V.

Dates

Publication Date: 20260506
Application Date: 20251029

Claims (15)

A method, comprising: receiving historical job data from at least one past wireline conveyance operation, the historical job data comprising: static input data; and timeseries input data; processing the historical job data through a neural network (NN) configured to learn the relationships between input features and target variables comprising: an input layer, a plurality of hidden layers, and an output layer; integrating the neural network with a numerical physical tension model to form a physics-informed neural network (PINN); training the PINN using the historical job data and estimation results of the physical model to enhance prediction accuracy; utilizing the trained PINN to predict at least one physical parameter for a current wireline conveyance operation; and outputting the predicted at least one physical parameter for use in the wireline conveyance operation.
The method of claim 1, further comprising planning the current wireline conveyance operation using the outputted predicted at least one physical parameter before the current wireline conveyance operation is started.
The method of claim 1 or 2, wherein the historical job data further comprises timeseries input data from the current wireline conveyance operation.
The method of claim 3, further comprising: re-training the PINN using the historical job data including the timeseries input data from the current wireline conveyance operation and the estimation results of the physical model to further enhance prediction accuracy; utilizing the re-trained PINN to update a prediction for the at least one physical parameter for the current wireline conveyance operation; outputting the updated predicted at least one physical parameter; and adjusting the current wireline conveyance operation using the outputted updated predicted at least one physical parameter.
The method of any one of the preceding claims, wherein the PINN comprises a customizable weight ranging between a weight using only the neural network and a weight using only the physical model estimation results, and any weighted combination of the neural network and the physical model estimation results.
The method of any one of the preceding claims, wherein the static input comprises one or more of: a tool string property, a cable property, a well profile, a well geometry, a wellbore fluid property, a viscous mud property, or a restriction.
The method of any one of the preceding claims, wherein the timeseries input comprises one or more of: cable speed, cable length, pumping rate, or tractor force.
The method of any one of the preceding claims, further comprising performing the current wireline conveyance operation based on the outputted predicted at least one physical parameter.
A system, comprising: a processor; and a non-transitory computer-readable medium storing instructions that, when executed, cause the processor to: receive historical job data from at least one past wireline conveyance operation, the historical job data comprising: static input data; and timeseries input data; process the historical job data through a neural network (NN) configured to learn the relationships between input features and target variables comprising: an input layer, a plurality of hidden layers, and an output layer; integrate the neural network with a numerical physical tension model to form a physics-informed neural network (PINN); train the PINN using the historical job data and estimation results of the physical model to enhance prediction accuracy; utilize the trained PINN to predict at least one physical parameter for a current wireline conveyance operation; and output the predicted at least one physical parameter for use in the wireline conveyance operation.
The system of claim 9, wherein the instructions further cause the processor to plan the current wireline conveyance operation using the outputted predicted at least one physical parameter before the current wireline conveyance operation is started.
The system of claim 9 or 10, wherein the historical job data further comprises timeseries input data from the current wireline conveyance operation.
The system of claim 11, wherein the instructions further cause the processor to: re-train the PINN using the historical job data including the timeseries input data from the current wireline conveyance operation and the estimation results of the physical model to further enhance prediction accuracy; utilize the re-trained PINN to update a prediction for the at least one physical parameter for the current wireline conveyance operation; output the updated predicted at least one physical parameter; and adjust the current wireline conveyance operation using the outputted updated predicted at least one physical parameter.
The system of any one of the claims 9 - 12, wherein the PINN comprises a customizable weight ranging between a weight using only the neural network and a weight using only the physical model estimation results, and any weighted combination of the neural network and the physical model estimation results.
The system of any one of the claims 9 - 13, wherein the static input comprises one or more of: a tool string property, a cable property, a well profile, a well geometry, a wellbore fluid property, a viscous mud property, or a restriction; and wherein the timeseries input comprises one or more of: cable speed, cable length, pumping rate, or tractor force.
The system of any one of the claims 9 - 14, wherein the instructions further cause the processor to perform the current wireline conveyance operation based on the outputted predicted at least one physical parameter.

Description

TECHNICAL FIELD This disclosure generally relates to systems and methods for enhancing wireline conveyance modeling, and more particularly, systems and methods for enhancing wireline conveyance modeling in wireline conveyance using a physics-informed neural network (PINN). BACKGROUND Wireline conveyance involves delivering a tool string to its target depth within a well for operation and retrieving it afterward using wireline system, e.g., in well-logging and perforation operations. This process generates real-time surface sensor readings of cable speed, cable length, surface tension, head tension, and/or sticking force. The cable speed refers to the rate of releasing or retrieving the wireline cable, while cable length denotes the total length of the cable deployed into the well. Surface tension refers to the force on the wireline cable measured at the ground level generated by the toolstring and cable as the tool string is being lowered into or pulled out of the well (or wellbore). Surface tension may be influenced by various factors, including cable speed and length, tool and cable weight and shape, buoyancy, friction, hydrodynamic force, well geometry, etc. Head tension refers to the tension measured at the uppermost section of the downhole tool inside the wellbore, and indicates the load experienced by the wireline. Sticking force refers to a product of the differential pressure between the wellbore and the reservoir and the area upon which the differential pressure is acting. Understanding of the interaction between wireline components and wellbore elements is crucial for accurately interpreting tool string dynamics during wireline conveyance. Traditional physics-based tension models, which have been widely used for conveyance job planning, often lack prediction accuracy due to the simplification of physics and inaccessibility of downhole parameters, particularly in deeper, deviated wells with viscous fluids. Accordingly, there is a need and desire for improvement in wireline conveyance modeling in wireline conveyance. SUMMARY This disclosure pertains to systems and methods for enhancing wireline conveyance modeling in wireline conveyance using a physics-informed neural network (PINN). Example embodiments of the present disclosure may include a computational model that combines a neural network and a numerical physical model to accurately simulate wireline conveyance jobs (e.g. surface tension responses during wireline conveyance). Embodiments may also update or re-train the model during job operation using data gathered from the job operation to improve the model. A first aspect of this disclosure pertains to a method, including: receiving historical job data from at least one past wireline conveyance operation, the historical job data including: static input data, and timeseries input data, processing the historical job data through a neural network (NN) configured to learn the relationships between input features and target variables including: an input layer, a plurality of hidden layers, and an output layer, integrating the neural network with a numerical physical tension model to form a physics-informed neural network (PINN), training the PINN using the historical job data and estimation results of the physical model to enhance prediction accuracy, utilizing the trained PINN to predict at least one physical parameter for a current wireline conveyance operation, and outputting the predicted at least one physical parameter for use in the wireline conveyance operation. A second aspect of this disclosure pertains to the method of the first aspect, and further includes planning the current wireline conveyance operation using the outputted predicted at least one physical parameter before the current wireline conveyance operation is started. A third aspect of this disclosure pertains to the method of the first aspect, wherein the historical job data further includes timeseries input data from the current wireline conveyance operation. A fourth aspect of this disclosure pertains to the method of the third aspect, and further includes: re-training the PINN using the historical job data including the timeseries input data from the current wireline conveyance operation and the estimation results of the physical model to further enhance prediction accuracy, utilizing the re-trained PINN to update a prediction for the at least one physical parameter for the current wireline conveyance operation, outputting the updated predicted at least one physical parameter, and adjusting the current wireline conveyance operation using the outputted updated predicted at least one physical parameter. A fifth aspect of this disclosure pertains to the method of the first aspect, wherein the PINN includes a customizable weight ranging between a weight using only the neural network and a weight using only the physical model estimation results, and any weighted combination of the neural network and the physical model estimation