CN-116027399-B - Density prediction method, device, equipment and storage medium based on neural network

CN116027399BCN 116027399 BCN116027399 BCN 116027399BCN-116027399-B

Abstract

The application provides a density prediction method, device, equipment and storage medium based on a neural network, which comprises the steps of data preparation, namely acquiring depth domain actual depth domain longitudinal wave speed, depth domain transverse wave speed and density logging data of a research area; the method comprises the steps of establishing a sample set, constructing a density prediction model based on a depth feedforward neural network, training the model by utilizing the training sample set to obtain a nonlinear relation model between depth domain longitudinal wave speed and depth domain transverse wave speed and density, realizing a density prediction function, carrying out normalization pretreatment on predicted data of actual depth domain longitudinal wave speed and depth domain transverse wave speed, and inputting the predicted data after normalization pretreatment into the nonlinear relation model to predict the density.

Inventors

XIE WEI
BI CHENCHEN
HU HUAFENG
YAO MING
LEI CHAOYANG
ZHANG KEFEI

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石油物探技术研究院

Dates

Publication Date: 20260512
Application Date: 20211025

Claims (4)

1. A neural network-based density prediction method, comprising: s1, data preparation, namely acquiring depth domain actual measurement depth domain longitudinal wave speed, depth domain transverse wave speed and density logging data of a research area; S2, constructing a sample set, namely carrying out normalization pretreatment on the depth domain measured depth domain longitudinal wave speed, the depth domain transverse wave speed and the density logging data, so as to form a training sample set of the neural network; s3, model training, namely selecting a conjugate gradient method as an optimization algorithm of a neural network, constructing a density prediction model based on a depth feedforward neural network, training the density prediction model by utilizing the training sample set to obtain a nonlinear relation model between depth domain longitudinal wave speed and depth domain transverse wave speed and density, and realizing a density prediction function; s4, model application, namely carrying out normalization pretreatment on predicted data of the depth domain longitudinal wave speed and the depth domain transverse wave speed, and inputting the predicted data after normalization pretreatment into the nonlinear relation model to predict density; the neural network is a deep feedforward neural network, and the structure of the network of the deep feedforward neural network comprises an input layer Output layer And L-1 hidden layers, the number of neurons of an input layer of the network of the deep feed-forward neural network Input layer activation function Selecting a ReLU function, selecting a Sigmoid function by an activation function of an output layer and an implicit layer, and selecting the number of neurons of the output layer ; The deep feedforward neural network belongs to a fully connected neural network, neurons between adjacent layers are fully connected with each other, and neurons between the same hidden layer and the hidden layers which are separated from each other are not connected with each other And output data y (z) ∈r, the output of its hidden layer is: Parameters to be learned of the deep feedforward neural network are as follows: the depth domain actual measurement depth domain longitudinal wave speed, depth domain transverse wave speed and density logging data of the research area specifically comprise: the depth domain longitudinal wave velocity, depth domain transverse wave velocity and density data obtained from conventional well logging and full-wave train well logging; The depth domain actual measurement depth domain longitudinal wave speed and the depth domain transverse wave speed are used as input data of the neural network; the density is used as output data of the neural network; before the normalization preprocessing is performed on the depth domain actual measurement depth domain longitudinal wave speed, the depth domain transverse wave speed and the density logging data, the method further comprises the following steps: Removing outliers from the depth domain measured depth domain longitudinal wave speed, the depth domain transverse wave speed and the density logging data; the specific method for normalizing pretreatment comprises the following steps: wherein b (z) and a (z) are logging values before and after normalization respectively; And The maximum and minimum values of the log values before normalization, respectively.
2. A neural network-based density prediction apparatus for implementing the neural network-based density prediction method of claim 1, comprising: The system comprises a data preparation module, a sample set construction module, a model training module and a model application module; The data preparation module is used for acquiring depth domain measured depth domain longitudinal wave speed, depth domain transverse wave speed and density logging data of a research area; the sample set construction module is used for carrying out normalization pretreatment on the depth domain measured depth domain longitudinal wave speed, the depth domain transverse wave speed and the density logging data so as to form a training sample set of the neural network; The model training module is used for constructing a density prediction model based on the depth feedforward neural network, training the density prediction model by utilizing the training sample set to obtain a nonlinear relation model between the depth domain longitudinal wave speed and the depth domain transverse wave speed and the density, and realizing a density prediction function; and the model application module is used for carrying out normalization pretreatment on the predicted data of the depth domain longitudinal wave speed and the depth domain transverse wave speed, and inputting the predicted data after normalization pretreatment into the nonlinear relation model for density prediction.
3. A neural network based density prediction device comprising a memory and a processor, the memory having stored thereon a computer program which, when executed by the processor, performs the neural network based density prediction method of claim 1.
4. A storage medium storing a computer program executable by one or more processors for implementing the neural network-based density prediction method of claim 1.

Description

Density prediction method, device, equipment and storage medium based on neural network Technical Field The application relates to the field of geophysical prospecting for oil and gas, in particular to a density prediction method, device and equipment based on a neural network and a storage medium. Background In the field of oil and gas exploration, as a bridge connecting rock physical properties with seismic wave exploration, longitudinal wave velocity, transverse wave velocity and density have important application in the aspects of seismic data AVO analysis, pre-stack inversion, lithology, physical properties, fluid identification and the like of reservoirs. However, in actual production, density parameters are lost and incomplete due to various reasons, and the development of subsequent exploration work is influenced. In production applications, the density is typically fitted using acoustic (longitudinal) velocity, with the most common method being the Gardner empirical formula. The formula is a statistical fit to a large amount of rock density data, but if the data is refined to a specific region, obvious errors still occur, and the current requirement of high-precision seismic interpretation cannot be met. The students at home and abroad fit the coefficients of the Gardner empirical formula according to experimental data of different regions, but basically, the coefficients are only power functions of the longitudinal wave speed, and once the actual measurement data of the longitudinal wave speed have systematic errors or random errors, the obtained density can also have error accumulation and amplification. Therefore, a method is required to improve the prediction accuracy of the density. Aiming at the defects, the invention fully excavates the internal relation between rock physical parameters by using the depth feedforward neural network based on the longitudinal wave speed and the transverse wave speed, establishes a nonlinear relation model between the longitudinal wave speed and the transverse wave speed and the density, and obtains a density prediction model based on the depth feedforward neural network, thereby improving the density prediction precision. Disclosure of Invention In view of the above, the present application provides a method, apparatus, device and storage medium for predicting density based on neural network. The application provides a density prediction method based on a neural network, which comprises the following steps: s1, data preparation, namely acquiring depth domain actual measurement depth domain longitudinal wave speed, depth domain transverse wave speed and density logging data of a research area; S2, constructing a sample set, namely carrying out normalization pretreatment on the depth domain measured depth domain longitudinal wave speed, the depth domain transverse wave speed and the density logging data, so as to form a training sample set of the neural network; S3, model training, namely constructing a density prediction model based on a depth feedforward neural network, training the density prediction model by utilizing the training sample set to obtain a nonlinear relation model between depth domain longitudinal wave speed and depth domain transverse wave speed and density, and realizing a density prediction function; And S4, model application, namely carrying out normalization pretreatment on predicted data of the depth domain longitudinal wave speed and the depth domain transverse wave speed, and inputting the predicted data after normalization pretreatment into the nonlinear relation model to predict the density. In some embodiments, the depth domain measured depth domain longitudinal wave velocity, depth domain transverse wave velocity, and density log data of the investigation region specifically comprises: the depth domain longitudinal wave velocity, depth domain transverse wave velocity and density data obtained from conventional well logging and full-wave train well logging; The depth domain actual measurement depth domain longitudinal wave speed and the depth domain transverse wave speed are used as input data of the neural network; the density is used as output data of the neural network. In some embodiments, before the normalizing preprocessing of the depth domain measured depth domain longitudinal wave velocity, depth domain transverse wave velocity, and density log data, the method further comprises: and removing abnormal values of the depth domain actual measurement depth domain longitudinal wave speed, the depth domain transverse wave speed and the density logging data. In some embodiments, the specific method of normalizing pretreatment comprises: Wherein b (z) and a (z) are log values before and after normalization, respectively, and b (z) max and b (z) min are maximum and minimum values of the log values before normalization, respectively. In some embodiments, the neural network is a deep feed forward neural network. In some embodiments, the