CN-121997741-A - SSA-LSSVM reservoir lithology fine prediction method based on lithology granularity factor construction

CN121997741ACN 121997741 ACN121997741 ACN 121997741ACN-121997741-A

Abstract

The invention provides an SSA-LSSVM reservoir lithology fine prediction method based on lithology granularity factors, which comprises the steps of obtaining multi-attribute logging curves with known lithology and granularity grading in a research area, calculating texture features and gradient features to construct a high-dimensional engineering feature pool, screening optimal attribute feature combinations sensitive to lithology granularity level identification from the high-dimensional engineering feature pool by adopting a filtering method-wrapping method mixing strategy based on a binary sparrow search algorithm, constructing a lithology granularity factor model, optimizing weights of the lithology granularity factor models by adopting a sparrow search algorithm to obtain an optimal lithology granularity factor model, constructing an SSA-LSSVM lithology granularity identification model, training the model based on the sparrow search algorithm to obtain a trained SSA-LSSVM lithology granularity identification model, and effectively capturing nonlinear features and redundant information in logging data to improve classification accuracy.

Inventors

WU WEI
WANG GUANGXU
MOU CHENYANG
LIN CHANGSONG
LIU WEIHONG
ZHAO XIAOMING

Assignees

河南理工大学

Dates

Publication Date: 20260508
Application Date: 20260126

Claims (10)

1. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factors is characterized by comprising the following steps: S101, acquiring a multi-attribute logging curve with known lithology and granularity grading in a research area, and preprocessing; S102, calculating texture features and gradient features by utilizing the preprocessed multi-attribute logging curve to construct a high-dimensional engineering feature pool, wherein the texture features are used for representing lithology granularity heterogeneity, and the gradient features are used for representing lithology granularity mutation; S103, screening an optimal attribute feature combination sensitive to lithology granularity level identification from a high-dimensional engineering feature pool by adopting a filtering method-wrapping method mixing strategy based on a binary sparrow searching algorithm; s104, constructing a lithology granularity factor model based on the optimal attribute feature combination, optimizing the weight of the lithology granularity factor model by adopting a sparrow search algorithm with the aim of maximizing the differentiation degree among lithology of different granularity levels, acquiring the optimal lithology granularity factor model and calculating lithology granularity factors corresponding to the optimal attribute feature combination; S105, constructing an SSA-LSSVM lithology granularity recognition model, fusing the optimal attribute feature combination with a corresponding lithology granularity factor to serve as an input feature, taking known lithology and granularity grading as labels, performing super-parameter optimization on a least square support vector machine by utilizing a sparrow search algorithm, training the LSSVM lithology granularity recognition model based on the optimal super-parameter, and obtaining a trained SSA-LSSVM lithology granularity recognition model; S106, predicting the whole research area based on the trained SSA-LSSVM lithology granularity recognition model.
2. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factor according to claim 1, wherein the properties of the logging curve at least include natural gamma, deep lateral resistivity, sonic time difference, compensating neutrons, density and well diameter; the pretreatment comprises the following steps: Removing invalid values and abnormal zero values in the original data, and carrying out depth alignment and equidistant resampling on logging curves of different sources; smoothing the curve by adopting a median filtering algorithm; Fitting and filling are carried out by adopting a cubic spline interpolation algorithm, and the continuous variation trend of the logging response is recovered; and eliminating dimensional differences and magnitude influences among logging attributes by adopting a Z-score standardization method, and converting the data into standard normal distribution.
3. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factor according to claim 2, wherein calculating texture features and gradient features using the preprocessed multi-attribute log to construct a high-dimensional engineering feature pool comprises: Based on a sliding window technology, taking standard deviation of original logging curve depth points in a window as a central depth point of the window, and moving along the depth point by point to form a texture characteristic curve for representing lithology granularity heterogeneity; calculating a first derivative of the original logging curve in depth by adopting a central difference method as a gradient characteristic to form a gradient characteristic curve for representing lithology granularity mutation; and constructing a high-dimensional engineering characteristic pool comprising an original logging curve, a texture characteristic curve and a gradient characteristic curve.
4. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factors according to claim 3, wherein filtering-wrapping method mixing strategy based on binary sparrow search algorithm is adopted to screen preferred feature combinations sensitive to lithology granularity level identification from a high-dimensional engineering feature pool, comprising: The first filtering method comprises the steps of obtaining a preliminary filtering characteristic set from a high-dimensional engineering characteristic pool based on a maximum information coefficient, and obtaining a final filtering method filtering characteristic set from the preliminary filtering characteristic based on a variance expansion factor test; The packing method is carefully selected, filtering method screening features are encoded into binary feature vectors, LSSVM model classification accuracy is used as a first fitness function, and a binary sparrow search algorithm is utilized to screen the best attribute feature combination from the filtering method screening feature set.
5. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factor according to claim 4, wherein obtaining the preliminary screening feature set from the high-dimensional engineering feature pool based on the maximum information coefficient comprises: calculating nonlinear correlation strength between each feature in the high-dimensional engineering feature pool and rock granularity category by adopting a maximum information coefficient algorithm, and screening the nonlinear correlation strength based on the nonlinear correlation strength MIC screening threshold characteristics, removing, and retaining nonlinear correlation strength The feature of the MIC screening threshold is used as a primary screening feature to obtain a primary screening feature set; Obtaining a final filter screening feature set from the preliminary screening features based on a variance expansion factor test, comprising: For the primary screening feature set, taking the kth primary screening feature as a dependent variable, taking the rest primary screening features as independent variables, performing multiple linear regression, and calculating the decision coefficient of the multiple linear regression Based on determining coefficients Calculating a variance expansion factor of the kth preliminary screening feature: Iteratively calculating variance expansion factors of each preliminary screening feature, and eliminating the variance expansion factors And (3) obtaining a filtering method screening feature set by the features of the VIF screening threshold until the variance expansion factors of all the preliminary screening features are smaller than the VIF screening threshold.
6. The method for fine rock property prediction of the SSA-LSSVM reservoir based on the rock property granularity factor according to claim 5, wherein the method for screening the best attribute feature combination from the filtering method screening feature set by utilizing a binary sparrow search algorithm comprises the steps of dividing the filtering method screening feature set into a training set and a verification set, training an LSSVM model by adopting a K-fold cross verification strategy, taking the average classification accuracy of the LSSVM model on the verification set as a first fitness function, taking binary feature vectors of filtering method screening features as optimization variables, carrying out iterative search by adopting the binary sparrow search algorithm to obtain a filtering method screening feature representing the selected binary feature vector, and taking the filtering method screening feature represented by the selected binary feature vector after the iteration is ended as the best attribute combination.
7. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factors according to any one of claims 1 to 6, wherein the lithology granularity factor model constructed based on the optimal feature combination is: Wherein, the Is depth point Is characterized by a lithology granularity factor of (1), Is the first The best attribute is characterized by depth Processing the normalized numerical value; Is the first The weight of the individual features is determined, Is the best attribute feature quantity.
8. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factors according to claim 7 is characterized in that a sparrow search algorithm is adopted to optimize the weight of a lithology granularity factor model to obtain an optimal lithology granularity factor model, wherein the method comprises the steps of taking the weight vector and the bias term of the lithology granularity factor model as optimization variables, taking the degree of distinction among the maximum coarse sand, medium sand, fine sand, silt, mudstone and pebble sandstone as optimization targets, introducing a generalized Fisher criterion as a second fitness function, and adopting the sparrow search algorithm to optimize the weight vector and the bias term of the lithology granularity factor model to obtain the optimal lithology granularity factor model.
9. The SSA-LSSVM reservoir lithology fine prediction method constructed based on lithology granularity factors according to claim 8, wherein optimizing weight vectors and bias terms of the lithology granularity factor model by adopting a sparrow search algorithm comprises: Step 4.1, calculating an intra-class divergence matrix by taking the optimal attribute feature combination as input Inter-class divergence matrix According to the internal divergence matrix And an inter-class divergence matrix Defining a second fitness function based on Fisher discriminant criteria: , For the weight vector to be optimized, At the level of the minimum value of the total number of the components, Representing a transpose; step 4.2, mapping the weight vector and bias of the lithology granularity factor model to be optimized into the spatial position of a sparrow individual, setting sparrow search algorithm parameters, and initializing a population; Step 4.3, calculating the fitness value of each individual in the population by using the second fitness function, selecting discoverers and joiners according to the sequence of the fitness values, and recording the global optimal position; Step 4.3, updating the positions of the discoverers and the joiners according to the position updating formulas of the discoverers and the joiners; step 4.4, randomly selecting an alerter from the discoverer and the joiner, and updating the alerter position by using an alerter position updating formula; And 4.5, repeating the steps 4.3-4.4, and outputting the global optimal position when the maximum iteration times or the fitness value are not lifted for a plurality of continuous generations.
10. The method for finely predicting the lithology of the SSA-LSSVM reservoir based on the lithology granularity factor according to any one of claims 3-6, 8 or 9, wherein the method for predicting the lithology of the reservoir based on the trained SSA-LSSVM lithology granularity recognition model comprises preprocessing a target logging curve to be predicted in the research area, performing texture feature and gradient feature calculation by adopting a method of S102 to obtain a texture feature curve and a gradient feature curve, calculating lithology granularity factors of each depth point by utilizing the optimal lithology granularity factor model of S104, calculating to generate a continuous lithology granularity factor curve, fusing the lithology granularity factor curve with the target logging curve, the texture feature curve and the gradient feature curve as input, and recognizing the specific lithology granularity level corresponding to each depth point by utilizing the trained SSA-LSSVM lithology granularity recognition model.

Description

SSA-LSSVM reservoir lithology fine prediction method based on lithology granularity factor construction Technical Field The invention relates to the technical field of drilling exploration analysis, in particular to a reservoir lithology fine prediction method. Background Lithology identification is a fundamental and critical technical work in oil and gas exploration and development and geological research, and particularly, the fine division of the granularity level (such as coarse sand, medium sand, fine sand, silt and the like) of reservoir rocks, and the accuracy of the lithology identification is directly related to the scientificity of sedimentary facies analysis, reservoir physical property evaluation (pore-permeability characteristics) and subsequent reservoir description. Currently, the main means for obtaining formation lithology and granularity information include well coring analysis and well logging data interpretation. The drilling and coring analysis method is to obtain the core sample of underground rock stratum via drilling and to observe and analyze the lithology and granularity of the core directly. While providing the most intuitive and accurate lithology and granularity information, it is costly, time consuming, and difficult to obtain a sufficient number of core samples in many practical projects, especially at early exploration stages, thereby limiting continuous and comprehensive lithology granularity evaluation of the entire wellbore interval. In order to make up for the deficiency of coring, well logging techniques are widely used. Logging curves (e.g., natural gamma, resistivity, sonic moveout, neutrons, density, etc.) can continuously reflect the properties of the subsurface formation. The existing logging lithology recognition method mainly comprises a plate intersection method, an experience discrimination method and a part of traditional machine learning algorithm. With the development of big data technology, the person skilled in the art gradually starts to use the data mining and machine learning methods for lithology recognition, so that the subjectivity of the traditional method is reduced to a certain extent. However, existing logging lithology recognition techniques, even when applied with machine learning algorithms, have significant limitations in facing the need for refined granularity prediction: First, the model is easy to fit and has poor generalization ability for "granularity level" tags. The prediction accuracy of machine learning algorithms is severely dependent on the number and quality of training samples. In areas of investigation where core samples are scarce, samples with well-defined "granularity grading" labels (e.g., distinguishing fine sand from silt) are extremely scarce. This makes machine learning algorithms extremely prone to overfitting, making it difficult for the model to capture subtle logging response differences between different size fractions lithology when predicting an uncancelled wellbore interval. Secondly, the feature processing has fundamental defects, and granularity confusion caused by the average effect is difficult to overcome. In the prior art, there is a fundamental cognitive blind spot in processing log data, which relies on the "absolute value" of the log (e.g., gr=80 API) as a model input. However, in the face of thin interbed or graded-granularity formations, the tool signals can produce a "vertical averaging effect" on rapidly changing lithologies, i.e., averaging different lithologies, due to the vertical resolution limitations of the logging tool. This effect results in a high overlap, or even no distinction, between "thin layer of fine sandstone" and "thick layer of siltstone or mudstone" at the log response values (e.g., average GR values). Therefore, all conventional models that rely on "absolute values" as inputs inevitably misjudge fine sandstone as siltstone or mudstone, resulting in distortion of the granularity level judgment of good quality reservoirs. In summary, existing machine learning methods rely on simple linear analysis at the feature optimization stage, and cannot effectively capture nonlinear features and redundant information in the well logging data, resulting in extremely easy over-fitting of the model under the condition of scarce coring samples and difficulty in fine division of rock granularity. Disclosure of Invention Aiming at the technical problems, the invention provides an SSA-LSSVM reservoir lithology fine prediction method based on lithology granularity factor construction, which comprises the steps of firstly adopting a sliding window standard deviation statistics and a central difference method to respectively extract texture features and gradient features of a logging curve, constructing a high-dimensional engineering feature pool, dominating the 'granularity heterogeneity' and the 'lithology abrupt change boundary' hidden in a stratum, secondly adopting a Maximum Information Coefficient