Modeling method and method for diagnosing lost circulation

Patent metadata

Field	Value
Publication number	US-11360235-B2
Application number	US-201916530387-A
Country	US
Kind code	B2
Filing date	Aug 2, 2019
Priority date	Dec 17, 2018
Publication date	Jun 14, 2022
Grant date	Jun 14, 2022

Abstract

A modeling method is provided. The modeling method includes: extracting multiple pieces of logging data corresponding to a plurality of logging characteristic parameters at different drilling depths of a sample well, based on a logging information of the sample well; marking the different drilling depths based on a lost circulation information of the sample well to distinguish lost circulation points and non-lost circulation points; and classifying the lost circulation points and the non-lost circulation points by adopting a random forest algorithm, based on the plurality of logging characteristic parameters and the multiple pieces of marked logging data at the different drilling depths, to establish a plurality of corresponding relations between the logging characteristic parameters and a lost circulation or non-lost circulation result, so as to obtain a diagnosis model.

First claim

We claim:

1. A modeling method for diagnosing lost circulation, comprising:
extracting multiple pieces of logging data corresponding to a plurality of logging characteristic parameters at different drilling depths of a sample well, based on logging information of the sample well;
marking the different drilling depths based on a lost circulation information of the sample well, to distinguish lost circulation points and non-lost circulation points; and
classifying the lost circulation points and the non-lost circulation points by adopting a random forest algorithm, based on the plurality of logging characteristic parameters and the multiple pieces of marked logging data at the different drilling depths, to establish a plurality of corresponding relations between the logging characteristic parameters and a lost circulation or non-lost circulation result by employing a plurality of decision-making trees, so as to obtain a diagnosis model for diagnosing lost circulation,
wherein the classifying the lost circulation points and the non-lost circulation points by adopting a random forest algorithm, based on the plurality of logging characteristic parameters and the multiple pieces of marked logging data at the different drilling depths comprises the following steps:
randomly selecting multiple pieces of corresponding logging data at the same or different drilling depths, based on the multiple pieces of corresponding logging data at different drilling depths to form training sets, wherein a random selection frequency is same as a number of samples at different drilling depths;
gradually segmenting the training sets by adopting the plurality of logging characteristic parameters, according to a preset order; and
stopping executing a segmentation action under the condition that a segmentation stopping condition is met, to obtain the decision-making trees corresponding to the training sets and furthermore realize a classification of the lost circulation points and the non-lost circulation points.

2. The modeling method for diagnosing lost circulation according to claim 1 , wherein the segmentation stopping condition is a preset segmentation frequency.

3. The modeling method for diagnosing lost circulation according to claim 1 , wherein the preset order is decided by a size of a Gini index of the training set under the logging data corresponding to each of the logging characteristic parameters at the different drilling depths.

4. The modeling method for diagnosing lost circulation according to claim 3 , wherein the Gini index of the training set under the logging data corresponding to each of the logging characteristic parameters at the different drilling depths is

Gini
⁡

(

D
,
A

)

=



D
1




D


⁢

Gini
⁡

(

D
1

)

+



D
2




D


⁢

Gini
⁡

(

D
2

)

,

wherein

Gini
⁡

(
D
)

=

1
-

∑

k
=
1

K

⁢

⁢

(



C
k




D


)

2

,

K is equal to 2, C 1 is a sample subset belonging to the lost circulation points in a training set D, C 2 is a sample subset belonging to the non-lost circulation points in the training set D, and if a logging characteristic parameter is A and a logging data corresponding to the logging characteristic parameter A at a certain drilling depth is a, the sample set D is divided into D 1 and D 2 : D 1 ={(x, y)ϵD|A(x)>a}, D 2 =D−D 1 by adopting the logging characteristic parameter A, wherein x is the logging characteristic parameter, and y is a mark for distinguishing the lost circulation points and the non-lost circulation points.

5. The modeling method for diagnosing lost circulation according to claim 3 , further comprising:
performing segmentation on the training set by adopting a logging characteristic parameter and a logging data corresponding to the logging characteristic parameter, under the condition that the Gini index of the training set under the logging data corresponding to the logging characteristic parameter at a certain drilling depth is minimum.

6. The modeling method for diagnosing lost circulation according to claim 1 , further comprising:
ranking the importance of a plurality of logging characteristic parameters used for segmentation during establishment of the decision-making trees, based on all the training sets and the decision-making trees corresponding to the training sets.

7. The modeling method for diagnosing lost circulation according to claim 6 , wherein the ranking the importance of a plurality of logging characteristic parameters used for segmentation during esta

Assignees

• Univ China Petroleum Beijing

Inventors

• Lu Yunhu
• Jin Yan
• Chen Mian
• Li Zejun
• Wang Hanqing
• Xia Yang
• Zhou Zhou
• Hou Bing
• Lin Botao
• Guo Xuyang

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N20/00

  Machine learning · CPC title

• E21B21/003

  Means for stopping loss of drilling fluid (plastering the borehole wall E21B33/138) · CPC title

• G06N20/20 Primary

  Ensemble learning · CPC title

• G01V99/00

  Subject matter not provided for in other groups of this subclass · CPC title