Search

CN-116877068-B - Element sensitivity analysis method for intelligently distinguishing shale small layers by using element logging

CN116877068BCN 116877068 BCN116877068 BCN 116877068BCN-116877068-B

Abstract

The invention discloses an element sensitivity analysis method for intelligently distinguishing shale small layers by using element logging, which comprises the following steps of S1, mining identification degree characteristics of shale small layers with multiple elements and single and double compound intersection, S2, screening homogeneous elements based on a twin neural network, and S3, performing generalization heterogeneous element sensitivity analysis based on a residual neural network. By adopting the method designed by the invention, the system, objectivity and overall extraction of the shale small-layer sensitive elements can be realized, and the automation of shale small-layer sensitive element analysis can be realized.

Inventors

  • OU CHENGHUA
  • Quan Haosen
  • LI CHAOCHUN

Assignees

  • 西南石油大学

Dates

Publication Date
20260505
Application Date
20230803

Claims (5)

  1. 1. An element sensitivity analysis method for intelligently distinguishing shale small layers by using element logging is characterized by comprising the following steps: step S1, multi-element single-double composite intersection shale small layer identification feature mining, further comprising the steps of constructing a shale small layer element basic data set, and comprising the following steps: Step S11, all unit element curves acquired through element logging are intersected in pairs to generate a double-element curve; s12, intersecting the unit element curve with the double element curve to generate a three-element curve; Step S13, generating multi-element curves obtained by any number of intersection; step S14, synthesizing a single element curve, a double element curve and a multi element curve to generate a shale small layer element basic data set; Step S2, based on homogeneous element screening of the twin neural network, optimizing a shale small layer element basic data set into a heterogeneous element data set, wherein the method comprises the following substeps: step S21, constructing a homomorphic element sample with the same information, which comprises the following substeps: normalizing the curves, wherein only one curve with similar curve shape is reserved; Only one curve with the same information provided when identifying shale small layers is reserved; s22, designing and optimizing a twin neural network structure; step S23, calculating homogeneity parameters of all curves in the shale small-layer element basic data set, screening out all homogeneous single-element curves, and intersecting the left heterogeneous single-element curves to generate a double-element curve set, wherein the step S23 further comprises the step of continuously generating multiple element curve sets; and step S3, heterogeneous element sensitivity analysis based on residual neural network generalization.
  2. 2. The method for intelligently identifying the element sensitivity analysis of shale fractions by using element logging according to claim 1, wherein the twin neural network structure comprises a multi-layer perceptron or a convolutional neural network.
  3. 3. The method for intelligently identifying shale small layers by using element logging according to claim 1, wherein the step S3 designs and optimizes a residual multi-layer perceptron structure by means of a heterogeneous element data set, trains and predicts the shale small layers of heterogeneous elements in a segmented way, further optimizes the heterogeneous element data set into a sensitive element data set as a heterogeneous element sensitivity evaluation index according to the generalization performance difference of different heterogeneous element curves to the residual multi-layer perceptron during prediction, and finally forms the shale small layer element identification sensitivity analysis method, and comprises the following substeps: S31, structural design and optimization of a residual multi-layer perceptron; Step S32, residual error multi-layer perceptron generalization contrast and sensitive element screening of a heterogeneous element curve; step S33, forming a heterogeneous element sensitivity analysis method; step S31 is to introduce a residual error module, and establish a residual error multi-layer perceptron model as a model for evaluating sensitivity.
  4. 4. The method for intelligently identifying the element sensitivity analysis of the shale small layer by using the element logging according to claim 3, wherein the step S32 is characterized in that a heterogeneous element data set is divided into a training segment and a verification segment at the same depth, after the residual multi-layer perceptron is used for learning the training segment characteristics, the generalization performance of the verification segment is used as an evaluation element sensitivity index, insensitive elements in the heterogeneous element data set are removed, and the heterogeneous element data set is further optimized into a sensitive element data set.
  5. 5. The method for intelligently identifying shale fractions by using element logging according to claim 4, wherein the step S33 is characterized in that multi-element single-double composite intersection curve generation, twin neural network element homogeneity evaluation and residual multi-layer perceptron generalization element sensitivity evaluation are combined to form a heterogeneous element sensitivity analysis method, and the shale fractions at the bottom of the well are identified.

Description

Element sensitivity analysis method for intelligently distinguishing shale small layers by using element logging Technical Field The invention belongs to the technical field of shale oil gas horizontal well drilling, and particularly relates to an element sensitivity analysis method for intelligently distinguishing shale small layers by using element logging. Background Shale oil gas is an important unconventional oil gas resource, and a high-quality reservoir is relatively thin, so that in the shale oil gas horizontal well drilling process, the relative position of the current well bottom and the high-quality reservoir needs to be judged in real time, so that the well track can be adjusted in time, and the drilling meeting rate of the high-quality reservoir of shale oil gas is improved. At present, the data for judging the relative positions of the bottom of a well and a high-quality reservoir are mainly data such as element logging. And on-site engineers establish element characteristic modes of the high-quality reservoir through element logging data of the high-quality reservoir adjacent to the well drilling horizontal well and the vertical well, and use the element characteristic modes as identification basis for the current horizontal well drilling. The key point of accurately judging the relative position of the bottom hole and the high-quality reservoir is to establish an accurate, objective and reliable high-quality reservoir element characteristic mode, and the first step of establishing the element characteristic mode is to analyze the sensitivity of each element to the high-quality reservoir, screen out a relatively sensitive element logging curve and reduce the interference of insensitive elements in the subsequent analysis. In the horizontal well drilling process, the element curves obtained through logging are usually up to thirty, the change characteristics of all the element curves are difficult to grasp in a manual analysis mode, in addition, due to measurement errors, irregular operation of field workers and other reasons, certain noise exists in the element content curves, so that the sensitivity of each element to shale small-layer changes is analyzed and screened, the sensitivity element curves are necessary and necessary, the interference of insensitive curve noise can be removed, the data volume of manual analysis is reduced, and the reliability of manual analysis results of the relative positions of the bottom of a well and a high-quality reservoir is improved. The main mode of judging the sensitivity of the element curve to the high-quality reservoir is that a field engineer determines the relevant characteristics of the high-quality reservoir according to the past drilling construction experience of the horizontal well and by combining the corresponding evaluation well (vertical well) of the horizontal well to be drilled. A few curves are typically combined empirically to produce a new characteristic curve. In addition, the intersection curves of a few element curves are generated empirically, so that the element intersection characteristics of the well cannot be completely mined, and meanwhile, due to the problem of homogeneity among certain elements, a plurality of nonsensical calculation easily occurs in a manual analysis mode. Meanwhile, the analysis process is not systematic, so that a lot of unnecessary calculation is easy to generate in the analysis process, and the analysis result is not comprehensive. In addition to the systematic lack, the whole analysis process has lower automation degree, and most of data operation, processing, visualization and the like need to be carried out manually, so that the whole analysis process has lower efficiency. In summary, the shale small-layer element discrimination sensitivity analysis at the present stage has the problems of systematic lack of analysis process, lower automation degree, subjectivity and unilateral analysis result and the like. Disclosure of Invention In order to solve the defects in the background technology, the invention provides an element sensitivity analysis method for intelligently distinguishing shale small layers by using element logging. The aim of the invention is realized by the following technical scheme: An element sensitivity analysis method for intelligently distinguishing shale small layers by using element logging comprises the following steps: Step S1, multi-element single-double composite intersection shale small layer identification feature mining; step S2, screening homogeneous elements based on a twin neural network; and step S3, heterogeneous element sensitivity analysis based on residual neural network generalization. Specifically, the step S1 further includes a step of constructing a shale minor layer element basic dataset, including the following sub-steps: Step S11, all unit element curves acquired through element logging are intersected in pairs to generate a double-element curve; s12, inters