CN-121995534-A - Lithology recognition method and system based on neural network

CN121995534ACN 121995534 ACN121995534 ACN 121995534ACN-121995534-A

Abstract

The invention discloses a lithology recognition method based on a neural network, which comprises the steps of obtaining a logging curve, recognizing missing values existing in the logging curve, inputting parameter data in the logging curve as sequence data into an established neural network model, predicting the neural network model to obtain complement data corresponding to the missing values, filling the missing values in the logging curve by using the complement data to obtain a complete logging data set, namely a first data set, carrying out equalization treatment on lithology classification samples in the complete logging data set based on the first data set to obtain an equalized logging data set, namely a second data set, optimizing super-parameters of a classification model by adopting a whale optimization algorithm to obtain target super-parameter configuration, configuring the classification model, inputting the second data set into the classification model, and recognizing and outputting lithology categories. The invention has the advantages of high filling precision of the missing value, excellent sample equalization effect, obvious model performance optimization and system practicability.

Inventors

WANG PAN
XU LIN
ZHOU CAILONG
HAN JING
WANG YIZHAN
ZHU KUNPENG

Assignees

东华理工大学

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. The lithology recognition method based on the neural network is characterized by comprising the following steps of: acquiring a logging curve, identifying missing values in the logging curve, inputting parameter data in the logging curve as sequence data into an established neural network model, predicting by the neural network model to obtain complement data corresponding to the missing values, and filling the missing values in the logging curve by using the complement data to obtain a complete logging data set, namely a first data set; performing equalization processing on lithology classification samples in the first data set to obtain an equalized logging data set, namely a second data set; optimizing the super parameters of the classification model by adopting a whale optimization algorithm to obtain target super parameter configuration, and configuring the classification model; Inputting the second data set into the classification model, and identifying and outputting lithology categories.
2. The lithology recognition method based on the neural network of claim 1, wherein the neural network model comprises a time domain convolutional network, a bi-directional gating cycle unit and an attention mechanism; The time domain convolution network is composed of a causal convolution layer, an expansion convolution layer and residual connection, wherein the causal convolution layer is used for extracting time sequence features in the logging curve, the expansion convolution layer is used for acquiring long sequence dependency relations in the logging curve, and the residual connection is used for superposing and transmitting original input information of the causal convolution layer and processed output features of the expansion convolution layer; The bidirectional gating circulation unit is used for comprehensively extracting sequence characteristics of the log curve samples; An attention mechanism is used to assign weights to different parameters in the log.
3. The neural network-based lithology recognition method of claim 1, wherein the equalization of lithology classification samples based on the first data set comprises the following steps: Step1 from the sample set Randomly selecting a sample point as an initial clustering center point; step2, calculating the shortest distance D (x) between each sample and the initial clustering center point, and calculating the probability of each sample being selected as the next clustering center based on the shortest distance D (x), wherein the probability is the ratio of the square of D (x) of the sample to the sum of squares of all samples D (x): ; Step3, selecting the next clustering center by adopting a wheel disc method; step4, distributing all samples to the class of the cluster center closest to the sample according to the Euclidean distance minimum principle; step5, calculating the average value of all sample points in each class, taking the average value as a clustering center of the next iteration, and taking an error square sum formula as an objective function: ; Wherein Ci represents the cluster center of the ith cluster, k represents k cluster clusters, and x represents the Euclidean distance of the corresponding cluster clusters; step6, repeating Step4 and Step5 until the clustering result converges.
4. The neural network-based lithology recognition method of claim 1, wherein the equalization of lithology classification samples based on the first data set comprises the steps of performing equalization on lithology classification samples in the data set by using a Kmeans++ SMOTE oversampling algorithm formed by combining a Kmeans++ clustering algorithm with an SMOTE algorithm: clustering the first data set by adopting a Kmeans++ clustering algorithm, distributing each sample to the nearest cluster center, and carrying out iterative updating to obtain k locally optimal cluster clusters; Screening cluster clusters with the proportion of the minority samples meeting a preset threshold as oversampling target clusters, and distributing new sample generation quantity according to sparsity of the minority samples in the target clusters, wherein if the minority samples are sparse, the distributed new sample generation quantity is more; And generating a new minority class sample in each target cluster by adopting an SMOTE algorithm, and realizing the balance between the majority class and the minority class sample.
5. The neural network-based lithology recognition method of claim 1, wherein optimizing the super-parameters of the classification model using a whale optimization algorithm comprises: Taking the super-parameter combination obtained by current optimization as a target solution vector X The surrounding approximation of the target solution vector is realized by updating the position vector of each searching individual, and the position updating formula is as follows: X(n+1)=X(n)-(2a・r-a)・D 1 ; wherein D 1 = |2r seed X (N) -X (t) |, which is the distance between whale and target, X For the position vector of the current optimal solution, a is a control parameter, r is a random vector between [0,1], n is the iteration number, X (n) is the position vector of the searching individual in the nth iteration, X (n+1) is the updated position vector of the searching individual in the (n+1) th iteration, and t represents the current iteration number; synchronously executing a contraction surrounding strategy and a spiral position updating strategy, and updating the position vector of each search individual through a spiral function model to realize the approximation of the target solution vector: ; Wherein D2 is the distance of the whale individual from the target, b is a constant for controlling the spiral shape, l is a random number between [ -1,1 ]; and selecting one search individual contained by the search unit as a reference individual, and updating the position vectors of other search individuals based on the position vectors of the reference individual so as to enlarge the search range of super-parameter optimization and realize the global search of the target super-parameters.
6. The neural network-based lithology recognition method of claim 1, wherein the super parameters comprise a learning rate, a tree depth, a leaf number, and a number of iterations; The super-parameter range obtained after the whale optimization algorithm is optimized comprises learning rate of 0.1-0.2, tree depth of 8-12, leaf number of 80-120 and iteration number of 250-300.
7. The method of claim 1, wherein the log parameters include natural gamma, borehole diameter, shallow resistivity, middle resistivity, deep resistivity, lithologic density, neutron porosity, mechanical rotational speed, longitudinal wave time difference, and density correction.
8. The lithology recognition method based on the neural network of claim 1, further comprising the steps of preprocessing data, processing original logging data in a logging curve by adopting a standard deviation normalization method, wherein a normalization formula is as follows: x'=(xi-μ)/σ; Where xi is the sample value before normalization, μ is the sample mean, σ is the sample standard deviation.
9. The neural network-based lithology recognition method of claim 1, wherein the step of recognizing the missing values in the log is characterized by further comprising the step of evaluating the missing value filling effect by means of root mean Square error, average absolute error and determinable coefficient, wherein the root mean Square error RMSE is less than or equal to 0.12, the average absolute error MAE is less than or equal to 0.09 and the determinable coefficient R-Square is more than or equal to 0.82.
10. A neural network-based lithology recognition system, comprising: The missing value filling module acquires a logging curve, identifies missing values existing in the logging curve, inputs parameter data in the logging curve as sequence data into an established neural network model, obtaining complement data of a corresponding missing value through the neural network model prediction, and filling the missing value in the logging curve by using the complement data to obtain a complete logging data set, namely a first data set; the sample equalization module is used for carrying out equalization processing on lithology classification samples in the first data set to obtain an equalized well logging data set, namely a second data set; The super-parameter optimization module is used for optimizing super-parameters of the classification model by adopting a whale optimization algorithm to obtain optimal super-parameter configuration, and configuring the classification model; And the lithology recognition module is used for loading the classification model of the target super-parameter configuration, receiving the second data set, and recognizing and outputting lithology categories.

Description

Lithology recognition method and system based on neural network Technical Field The invention relates to the technical field of petroleum and natural gas exploration and development, in particular to a lithology recognition method and system based on a neural network. Background Lithology identification is the foundation of reservoir characteristic research and mineral resource exploration, and has important significance for the development of geological work. The existing lithology recognition methods are mainly divided into two types, namely a traditional lithology recognition method, such as a cross-plot method, probability statistics and cluster analysis, the methods depend on the experience knowledge of geological workers, the time consumption is long, the recognition accuracy is low when complex lithology data are processed, and the other type is an artificial intelligence method, such as machine learning and deep learning models (such as Bayesian, support vector machines, BP neural networks, LSTM neural networks and Xgboost, GBDT, DNN, biGRU, GNN) which achieve a certain effect in lithology recognition, but have significant limitations. The core problems in the prior art mainly comprise the first aspect that logging data are influenced by objective factors such as geological environment complexity and logging instrument faults, part of parameters have missing values, characteristic information is lost due to direct discarding of missing data, and then lithology recognition accuracy is lowered, the second aspect that samples of different lithology are unevenly distributed, the ratio of most samples is too high, the ratio of few samples is extremely low, model training is biased to most types, the few lithology is difficult to accurately recognize, the third aspect that super-parameter selection of a classification model lacks scientific and effective optimization means, the traditional trial-and-error method is low in efficiency, optimal super-parameter configuration is difficult to obtain, and classification performance of the model is limited. Aiming at the problems, the prior art does not form a complete solution, and a lithology recognition technology capable of synchronously solving the problems of data deficiency, sample imbalance and super-parameter optimization is needed to improve lithology recognition accuracy in complex scenes. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a lithology recognition method and system based on a neural network, which are used for solving the problems that logging data are influenced by objective factors such as geological environment complexity, logging instrument faults and the like, partial parameters have missing values, the missing data are directly abandoned to cause characteristic information loss so as to reduce lithology recognition accuracy, samples of different lithology are unevenly distributed, the proportion of most samples is too high, the proportion of few samples is extremely low, model training is biased to most types and few lithology is difficult to accurately recognize, the super-parameter selection of a classification model lacks scientific and effective optimization means, the traditional trial-and-error method is low in efficiency, optimal super-parameter configuration is difficult to obtain, and the classification performance of the model is limited. In order to achieve the aim of the invention, the invention adopts the following technical scheme: A lithology recognition method based on a neural network comprises the following steps: acquiring a logging curve, identifying missing values in the logging curve, inputting parameter data in the logging curve as sequence data into an established neural network model, predicting by the neural network model to obtain complement data corresponding to the missing values, and filling the missing values in the logging curve by using the complement data to obtain a complete logging data set, namely a first data set; performing equalization processing on lithology classification samples in the first data set to obtain an equalized logging data set, namely a second data set; optimizing the super parameters of the classification model by adopting a whale optimization algorithm to obtain target super parameter configuration, and configuring the classification model; Inputting the second data set into the classification model, and identifying and outputting lithology categories. As a preferred technical scheme, the neural network model comprises a time domain convolution network, a bidirectional gating circulation unit and an attention mechanism; The time domain convolution network is composed of a causal convolution layer, an expansion convolution layer and residual connection, wherein the causal convolution layer is used for extracting time sequence features in the logging curve, the expansion convolution layer is used for acquiring long sequence dependency relat