| CPC G06N 7/01 (2023.01) [E21B 47/07 (2020.05); E21B 47/10 (2013.01); E21B 2200/20 (2020.05); E21B 2200/22 (2020.05)] | 11 Claims |
|
1. A method of managing a well system, comprising:
obtaining, by a digital twin manager and based on field well dynamics behavior data of a well site of the well system, emulated well dynamics behavior data using a physics-based model,
wherein the physics-based model includes a valve module that corresponds to and emulates a valve control mechanism of the well site, and
wherein the field well dynamics behavior data include:
normal operational data; and
abnormal operational data corresponding to at least one of system faults, system failures, system shut-downs, safety conditions, weather conditions, and abnormal well conditions;
obtaining, by the digital twin manager and based on a predetermined monitoring criterion, predicted well dynamics behavior data of the well site using a physics constrained machine learning model that is based on the emulated well dynamics behavior data and the field well dynamics behavior data;
determining, by the digital twin manager and using a second machine learning model, an impact level that associates the predicted well dynamics behavior data with a well system abnormality in which a well mass flow of the well site is beyond a predetermined range,
wherein the second machine learning model includes a Naïve Bayes algorithm that uses the predicted well dynamics behavior data and the abnormal operational data as inputs to a learned Naïve Bayes model,
wherein the learned Naïve Bayes model comprises attributes, class probabilities, and conditional probabilities based on the predicted well dynamics behavior data and the abnormal operational data,
wherein the attributes comprise an ordered list of a well mass flow attribute, a well temperature attribute, and a well pressure attribute,
wherein the determining of the impact level includes, for each attribute, determining a corresponding normalized impact value that is a percentage contribution of the predicted well dynamics behavior data, and
wherein the normalized impact values are based on the abnormal operational data;
determining, by the digital twin manager and using the second machine learning model, a likelihood level that associates the predicted well dynamics behavior data with the well system abnormality,
wherein the determining of the likelihood level includes determining likelihood values that correspond to the normalized impact values, and
wherein the likelihood values are scaled based on the percentage contributions of the corresponding normalized impact values;
determining, by the digital twin manager and using the second machine learning model, a probability and a risk level of the well system abnormality based on the impact level and the likelihood level,
wherein determining the probability of the well system abnormality includes summing the likelihood values;
determining, by the digital twin manager, a command for adjusting the valve control mechanism based on the risk level and the probability of the well system abnormality; and
based on the command, opening or closing the valve control mechanism to regulate the well mass flow of the well site to remediate the well system abnormality and restore the well mass flow of the well site to the predetermined range,
wherein the second machine learning model is configured to determine, for each attribute in the ordered list, a posterior probability that is input for a next iteration of the second machine learning model for the next attribute of the ordered list.
|