Methods and systems for predicting lithology and formation boundary ahead of the bit
A method includes drilling a wellbore in a current well. An interval of the wellbore comprises a first portion of the wellbore and a second portion of the wellbore. The method also includes obtaining an offset drilling log and an offset lithology log for a geologically similar interval in an offset well and training a first machine learning model, using the offset drilling log, to produce a first trained machine learning model. The method further includes producing, using the first trained machine learning model, a forecasted drilling log for the second portion of the wellbore in the current well, training a second machine learning model, using a gradient boosting machine learning technique, the forecasted drilling log, and the offset lithology log, to produce a second trained machine learning model, and producing, using the second trained machine learning model, a forecasted lithology log for the second portion of the current well.
1 . A method comprising:
drilling, using a bottom hole assembly (“BHA”), a wellbore through a formation in a current well using a drill bit, wherein the formation comprises an interval having a first portion and a second portion, wherein the first portion of the formation includes a segment of the formation through which the wellbore has already been drilled and the second portion of the formation includes a segment of the formation through which the wellbore will be drilled;
recording, using one or more sensors mounted on the BHA, a measured drilling log in the wellbore penetrating the first portion of the formation;
obtaining a real-time lithology log of the first portion of the formation, wherein the real-time lithology log comprises a rock type of the first portion;
obtaining an offset drilling log and an offset lithology log for a geologically similar interval in an offset well, the geologically similar interval being geologically similar to the interval of the current well;
training a first machine learning model, using a computer processor and the offset drilling log, to produce a first trained machine learning model, wherein the first machine learning model is trained using a long short-term memory (LSTM) architecture to process the offset drilling log and produce a forecasted drilling log for the second portion of the formation, wherein training the first machine learning model includes:
assuming the offset drilling log as sequential to leverage an ability of the LSTM architecture to capture time-dependent patterns;
preparing the offset drilling log for supervised learning by normalizing input variables and framing a problem as predicting drilling log curves based on prior depth step measurements;
configuring LSTM cells with input, forget, and output gates to regulate information flow and maintain long-term dependencies in the offset drilling log;
generating training data slices from the offset drilling log, where input features correspond to specific depth intervals and response variables predict the forecasted drilling log for the second portion of the formation;
optimizing weights of the first machine learning model, biases, and activation functions to enhance accuracy in sequential data forecasting;
producing, using the computer processor and the first trained machine learning model, the forecasted drilling log for the second portion of the formation in the current well;
training a second machine learning model, using the computer processor, a gradient boosting machine learning technique, the forecasted drilling log, and the offset lithology log, to produce a second trained machine learning model;
producing, using the computer processor and the second trained machine learning model, a forecasted lithology log for the second portion of the formation of the current well from the measured drilling log in the wellbore penetrating the first portion of the formation,
wherein the forecasted lithology log comprises a rock types classification for the second portion of the formation;
adjusting a density of drilling mud using the rock types classification;
drilling, using the BHA, the wellbore through the second portion of the formation using the adjusted drilling mud.
2 . The method of claim 1 , further comprising updating, using the computer processor, the first trained machine learning model with a real drilling log for at least a section of the second portion of the interval of the current well to produce an updated first trained machine learning model.
3 . The method of claim 2 , further comprising producing, using the computer processor and the updated first trained machine learning model, an updated forecasted drilling log for the second portion of the interval of the current well.
4 . The method of claim 3 , further comprising updating, using the computer processor, the second trained machine learning model with the updated forecasted drilling log and a real lithology log for at least the section of the second portion of the interval of the current well to produce an updated second trained machine learning model.
5 . The method of claim 4 , further comprising producing, using the computer processor and the second trained machine learning model, an updated forecasted lithology log for the second portion of the interval of the current well.
6 . The method of claim 1 , further comprising correlating, using the computer processor, the offset drilling log and the offset lithology log from the offset well by depth.
7 . The method of claim 1 , wherein a size of the interval determines a parameter in the LSTM architecture.
8 . The method of claim 1 , wherein the offset drilling log and the forecasted drilling log comprise one or more curves selected from a list consisting of: weight on bit, rotations per minute, bit size, torque, flow pump rate, rate of penetration, mechanical specific energy, and mud-log lithology.
9 . A system comprising:
a current well having a wellbore being drilled, by a bottom hole assembly (“BHA”) comprising a drill bit, through a formation, wherein the formation comprises an interval having a first portion and a second portion, wherein the first portion of the formation includes a segment of the formation through which the wellbore has already been drilled and the second portion of the formation includes a segment of the formation through which the wellbore will be drilled, wherein the BHA further comprises at least one sensor configured to record a measured drilling log in the wellbore penetrating the first portion of the formation, wherein a real-time lithology log of the first portion of the formation is obtained, and wherein the real-time lithology log comprises a rock type of the first portion; and
a non-transitory computer readable medium in electronic communication with the current well and storing a set of instructions, executable by a computer processor, the set of instructions comprising functionality for:
obtaining an offset drilling log and an offset lithology log for a geologically similar interval in an offset well, the geologically similar interval being geologically similar to the interval of the current well,
training a first machine learning model, using the offset drilling log, to produce a first trained machine learning model, wherein the first machine learning model is trained using a long short-term memory (LSTM) architecture to process the offset drilling log and produce a forecasted drilling log for the second portion of the formation, wherein training the first machine learning model includes:
assuming the offset drilling log as sequential to leverage an ability of the LSTM architecture to capture time-dependent patterns;
preparing the offset drilling log for supervised learning by normalizing input variables and framing a problem as predicting drilling log curves based on prior depth step measurements;
configuring LSTM cells with input, forget, and output gates to regulate information flow and maintain long-term dependencies in the offset drilling log;
generating training data slices from the offset drilling log, where input features correspond to specific depth intervals and response variables predict the forecasted drilling log for the second portion of the formation; and
optimizing weights of the first machine learning model, biases, and activation functions to enhance accuracy in sequential data forecasting,
producing, using the first trained machine learning model, the forecasted drilling log for the second portion of the wellbore in the current well,
training a second machine learning model, using a gradient boosting machine learning technique, the forecasted drilling log, and the offset lithology log, to produce a second trained machine learning model,
producing, using the second trained machine learning model, a forecasted lithology log for the second portion of the current well, from the measured drilling log,
wherein the forecasted lithology log comprises a rock types classification for the second portion of the formation;
adjusting a density of drilling mud using the rock types classification;
drilling, using the BHA and drill bit, the wellbore through the second portion of the formation using the adjusted drilling mud.
10 . The system of claim 9 , wherein the set of instructions further comprising functionality for updating the first trained machine learning model with a real drilling log for at least a section of the second portion of the interval of the current well to produce an updated first trained machine learning model.
11 . The system of claim 10 , wherein the set of instructions further comprising functionality for producing, using the updated first trained machine learning model, an updated forecasted drilling log for the second portion of the interval of the current well.
12 . The system of claim 11 , wherein the set of instructions further comprising functionality for updating the second trained machine learning model with the updated forecasted drilling log and a real lithology log for at least the section of the second portion of the interval of the current well to produce an updated second trained machine learning model.
13 . The system of claim 12 , wherein the set of instructions further comprising functionality for producing, using the second trained machine learning model, an updated forecasted lithology log for the second portion of the interval of the current well.
14 . The system of claim 9 , wherein the set of instructions further comprising functionality for correlating the offset drilling log and the offset lithology log from the offset well by depth.
15 . The system of claim 9 , wherein a size of the interval determines parameter in the LSTM architecture.
16 . The system of claim 9 , wherein the offset drilling log and the forecasted drilling log are one or more curves selected from a list consisting of: weight on bit, rotations per minute, bit size, torque, flow pump rate, rate of penetration, mechanical specific energy, and mud-log lithology.