CN-121997694-A - Artificial intelligence interpretation method for multi-arm borehole logging data

CN121997694ACN 121997694 ACN121997694 ACN 121997694ACN-121997694-A

Abstract

The invention provides an artificial intelligence interpretation method of multi-arm well diameter logging data, which comprises the steps of 1, carrying out graphic processing on logging interpretation data, 2, building a working condition interpretation model based on a convolutional neural network, 3, training the built working condition interpretation model, and 4, judging the working condition of multi-arm well diameter logging according to the acquired actual multi-arm well diameter logging data by utilizing the trained working condition interpretation model. According to the artificial intelligence interpretation method for the multi-arm borehole logging data, the interpretation efficiency and the interpretation precision of the multi-arm borehole logging interpretation work are effectively improved by combining the high-precision data acquisition capability of the multi-arm borehole logging with the high-efficiency intelligent analysis advantages of the artificial intelligence, and the artificial labor intensity is reduced.

Inventors

XU WEI
LI RONGQIANG
YANG PING
YI JUN
LI GUANGMING
WEI LINNA
GUO HAIYAN
MENG FANYU
GONG JIAN

Assignees

中国石油化工股份有限公司
中国石油化工股份有限公司胜利油田分公司

Dates

Publication Date: 20260508
Application Date: 20241108

Claims (11)

1. The artificial intelligence interpretation method of the multi-arm borehole logging data is characterized by comprising the following steps of: step 1, carrying out graphic processing on logging interpretation data; step 2, building a working condition interpretation model based on a convolutional neural network; Step 3, training the constructed working condition interpretation model; and 4, judging the multi-arm well diameter logging working condition according to the acquired actual multi-arm well diameter logging data by using the trained working condition interpretation model.
2. The artificial intelligence interpretation method of multi-arm borehole log data as recited in claim 1, characterized in that in step 1, the method is performed by Converting the grid data signal into a two-dimensional gray picture with pixel values between 0 and 255, wherein P (i, j) and P' (i, j) respectively represent values before and after normalization of the grid data of the ith row and the jth column, R i represents the ith row data, and a round function rounds the values according to a specified number of bits and defaults to reserve 0 bit fraction.
3. The artificial intelligence interpretation method of multi-arm borehole log data as recited in claim 1, wherein step 2 comprises: step 21, constructing three convolution feature extraction layers with different sizes, which are used for processing graphic features with different scales; Step 22, each convolution characteristic extraction layer is composed of three layers of convolution-pooling-activation networks; Step 23, repeating the step 22, sequentially introducing convolution kernels with different sizes, and extracting to obtain a final feature map; step 24, using a Softmax classifier to classify the interpretation conditions of the multi-arm borehole log.
4. The artificial intelligence interpretation method of multi-arm borehole log data as claimed in claim 3, characterized in that in step 21, the first layer uses three sizes of 3*3, 5*5, 7*7 convolution kernels to extract features of the current layer, the second layer uses three sizes of 7*7, 9*9, 11 x 11 convolution kernels to extract features of the current layer, and the third layer uses three sizes of 9*9, 11 x 11, 13 x 13 convolution kernels to extract features of the current layer.
5. The artificial intelligence interpretation method of multi-arm borehole log data as recited in claim 3, wherein step 22 comprises: a. Each layer of convolution features is extracted by S(i,j)=(K*I)(i,j)=∑ m ∑ n I(i+m,j+n)K(m,n) The method comprises the steps of (1) inputting grid data, wherein I represents the input grid data, K represents convolution kernel grid data, S represents feature mapped data, namely extracted feature grid data, S (I, j) represents feature mapped grid data of an ith row and a jth column after processing, (K x I) represents matrix multiplication operation of a convolution kernel and the input grid data, and the expansion of the matrix multiplication operation is sigma m sigma n I (i+m, j+n) K (m, n), wherein m and n respectively represent the width and the height of the convolution kernel; b. after extracting the grid characteristics, carrying out batch standardization operation to solve the problem that training is difficult due to deepening of the layer number of the neural network; c. The Relu activation function is used in the convolution layer, and is a linear rectification function, and the formula is as follows: σ(x)=max(0,x) x is the input value into the activation function and the max function returns the maximum value of a set of values or specified parameters.
6. The artificial intelligence interpretation method of multi-arm borehole logging data as claimed in claim 5, characterized in that in step 23, the above three steps a, b and c are repeated, convolution kernels of different sizes are sequentially introduced, a final feature map is extracted, and the obtained feature map is converted into a one-dimensional output vector y= [ y 1 y 2 ,,,y l ] of length L through a full connection layer.
7. The artificial intelligence interpretation method of multi-arm borehole log data as claimed in claim 6, wherein at step 24, the output vector is Softmax operated using the formula And obtaining a probability distribution, and outputting the working condition category corresponding to the maximum probability value.
8. The artificial intelligence interpretation method of multi-arm borehole log data as claimed in claim 1, characterized in that in step 3, the working condition interpretation model is trained by using a large number of marked multi-arm borehole log interpretation sample data, so that the model can learn characteristic representation and classification rules of borehole log data graphs of different working condition types, wherein the working condition types comprise normal, diameter reduction, diameter expansion, corrosion, perforation, dislocation and bending.
9. The artificial intelligence interpretation method of multi-arm borehole logging data as claimed in claim 8, characterized in that in step 3, the judgment standard of normal working condition is that all measured arm values are reduced compared with standard size values, the average value reduction is within the allowable error, and the judgment is normal; the judgment standard of the diameter reduction working condition is that compared with the standard size value, all the measurement arm values are reduced, the average value reduction exceeds the allowable error, and the diameter reduction is judged; the judgment standard of the expanding working condition is that all the measurement arm values are increased compared with the standard size value, and the average value increase exceeds the allowable error, and the expanding working condition is judged; the judgment standard of the corrosion working condition is that compared with the standard size value, the increase of the partial measurement arm value is close to or exceeds the wall thickness of the sleeve, and the average value increase exceeds the allowable error, so that the corrosion is judged; The judgment standard of the perforation working condition is that compared with the standard size value, the increase of the partial measurement arm value is close to or exceeds the wall thickness of the sleeve, and perforation is judged; The judgment standard of the fault-breaking working condition is that all the measurement arm values are suddenly changed at the same depth, all the measurement arm values are increased, and all the arm value increase amounts are close to or exceed the wall thickness of the sleeve, and the fault-breaking working condition is judged; the judgment standard of the bending working condition is that the standard size value is compared with the standard size value, the part of the measurement arm values are increased, the measurement arm values of the relative orientation are reduced, the deformation is continuous, and the bending is judged.
10. The artificial intelligence interpretation method of multi-arm borehole logging data according to claim 1, wherein in step 4, the actual data is preprocessed to be used as test data, and the test data is input into a trained working condition interpretation model, and the working condition category corresponding to the maximum probability value is output.
11. An artificial intelligence interpretation system of multi-arm borehole log data, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the computer program to perform the steps corresponding to the method of any of claims 1-10.

Description

Artificial intelligence interpretation method for multi-arm borehole logging data Technical Field The invention relates to the technical field of petroleum and natural gas exploration and development, in particular to an artificial intelligence interpretation method of multi-arm borehole logging data. Background At present, the explanation of multi-arm borehole diameter logging needs to be manually explained by depending on the experience of service personnel, and mainly has the following two problems: (1) Manual interpretation of multi-arm borehole logging results, large workload and low efficiency In practical applications, the multi-arm borehole logging results often need to be interpreted and interpreted manually, which is not only labor-intensive, but also inefficient. As one of the most commonly used casing quality evaluation logging projects in the oilfield, the average interpretation time of a multi-arm borehole log is about 3 hours, accounting for 43.3% of the total duration of all log interpretations. Manually interpreting multi-arm borehole log results requires an expert or professional to review, analyze and judge a large number of multi-arm borehole log data one by one. This not only requires a lot of time and effort, but also is susceptible to human factors such as fatigue, subjective judgment, etc., resulting in inaccuracy or inconsistency of interpretation results. With the increasing and complicating of multi-arm borehole logging data, the difficulty and effort of manual interpretation is also increasing. This not only makes the whole multi-arm borehole log interpretation process more cumbersome and time consuming, but may also result in some important information being missed or misinterpreted, thereby affecting the final analysis results and application results. (2) The explanation of multi-arm borehole logging depends on manual experience, and the recognition efficiency is needed to be further improved Interpretation of multi-arm borehole log results often relies on human experience. Relying on human experience for multi-arm borehole log interpretation means that each time a new multi-arm borehole log data or situation is faced, a practitioner is required to make a decision by virtue of his or her rich knowledge and experience. However, this approach is not only inefficient, but also prone to missing data or situations where changes are not obvious. Because experience and cognition of each person are limited, it is difficult to ensure that accurate judgment can be made every time. The multi-arm borehole logging data consists of forty curves, and is observed by naked eyes according to manual experience, so that the multi-arm borehole logging data is large in workload, easy to fatigue and extremely easy to cause missing of problems. At present, the interpretation coincidence rate of the multi-arm borehole diameter logging is 80 percent. In the Chinese patent application with application number of CN202311118038.4, a horizontal well sleeve change working condition prediction method, system, equipment and storage medium are related, wherein engineering parameters and geological parameters are taken as characteristic data, sleeve change working condition information is taken as labeling data, a sleeve change data large table is constructed, correlation matrix analysis of the relation between continuous characteristic parameters is carried out based on the sleeve change data large table to form an analysis result, characteristic dimension reduction processing is carried out on the analysis result, characteristic parameters with autocorrelation lower than a set threshold value are calculated, quantitative analysis of the relation between the sleeve change working condition parameters and the continuous characteristic parameters is carried out to obtain sensitive characteristic parameters, similarity relation analysis between the sleeve change working condition parameters and the discrete characteristic parameters is carried out to obtain main control factors affecting the sleeve change working condition. With the disclosed exemplary embodiments, uncertainty of set-variant prediction can be reduced to the maximum extent through a machine learning intelligent classification decision system, and precision and generalization capability of a set-variant prediction model can be improved. The method for establishing and applying the paleo-water depth prediction model comprises the steps of obtaining a well-logging curve and paleo-water depth data of a known well along the well depth direction, correcting the paleo-water depth data by a data correction method to obtain corrected paleo-water depth data along the well depth direction, obtaining a paleo-water depth sensitive well-logging curve with the correlation degree between the corrected paleo-water depth data and the well-logging curve exceeding a set degree according to the correlation degree between the corrected paleo-water depth data and the wel