CN111596978A - Web page display method, module and system for lithofacies classification by artificial intelligence - Google Patents

Web page display method, module and system for lithofacies classification by artificial intelligence Download PDF

Info

Publication number
CN111596978A
CN111596978A CN201811629031.8A CN201811629031A CN111596978A CN 111596978 A CN111596978 A CN 111596978A CN 201811629031 A CN201811629031 A CN 201811629031A CN 111596978 A CN111596978 A CN 111596978A
Authority
CN
China
Prior art keywords
data
seismic
lithofacies
module
artificial intelligence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201811629031.8A
Other languages
Chinese (zh)
Inventor
苗和平
高端民
赵红艳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Yingcai University
Original Assignee
Shandong Yingcai University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shandong Yingcai University filed Critical Shandong Yingcai University
Priority to CN201811629031.8A priority Critical patent/CN111596978A/en
Publication of CN111596978A publication Critical patent/CN111596978A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44Arrangements for executing specific programs
    • G06F9/451Execution arrangements for user interfaces
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/047Probabilistic or stochastic networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Human Computer Interaction (AREA)
  • Geophysics And Detection Of Objects (AREA)

Abstract

The invention discloses a method for explaining logging curve and seismic map data by artificial intelligence, wherein the input of the data and the output of the result are displayed in a network page form, so that remote deployment and data sharing are facilitated. The method comprises the steps of using part of collected sample data sets of known lithofacies classification as training data, using methods of machine learning and deep learning to automatically identify the lithofacies, and then dividing the lithofacies in the stratum of an unknown region. The invention comprises an artificial intelligence interpretation method, a module and a system. The server in which the present invention is deployed will exhibit the following functions: the mutual communication interface of the webpage end comprises a data entry part, a data verification part, a data preprocessing module, a model establishing module, a model training module, a model iteration module, a model using module, a result displaying module and the like. The invention covers the display of the artificial intelligence calculation result on the network page by using a Python frame.

Description

Web page display method, module and system for lithofacies classification by artificial intelligence
The invention discloses a method for explaining logging curve and seismic map data by artificial intelligence, wherein the input of the data and the output of results are displayed in a network page form, so that cloud computing operation, cloud server deployment, cloud data clustering, security and confidentiality of computing results, storage, sharing and the like are facilitated. Logging data herein refers to physical properties of subsurface formation rock, formation fluids, and mixtures thereof; the seismic maps herein refer to two-dimensional, three-dimensional, four-dimensional, and multi-dimensional datasets obtained by artificially launching seismic waves. The method comprises the steps of using a part of collected sample data sets with known facies classification as training data, and using a deep learning method to identify the lithofacies. Facies in the formation are divided, the thickness of the identified layers being divided from 15 centimeters to hundreds of meters, the accuracy of which depends on the resolution of the input data and the processed data. The invention also comprises an artificial intelligence interpretation method, a module and a system. The server in which the present invention is deployed will perform the following steps: the mutual communication interface of the webpage end comprises a data entry part, a data verification part, a data preprocessing module, a model establishing module, a model training module, a model iteration module, a model using module, a result displaying module and the like. The invention also covers the display of the artificial intelligence calculation results on the network page by using Python frames, wherein the frames comprise Django, Pylons, Tornado, Bottle and flash. This lays the foundation for cloud deployment.
Description of the invention
A webpage display method, module and system for rock facies classification by artificial intelligence.
Technical Field
The invention relates to a framework of petroleum exploration, petroleum development, well logging data processing, seismic map processing, artificial intelligence, machine learning and operation modules, deep learning and operation modules and Python plus website webpages.
Background
In recent years, modern scientific technology is effectively utilized in the field of oil exploration and development, so that the rapid development of the oil industry is promoted, and meanwhile, great benefits are brought to national economy. However, as the level of oil exploration is continuously improved, it is increasingly difficult to find new oil and gas fields, which requires that the understanding level is continuously improved, unknown conditions of oil and gas are known and mastered by a scientific method, and more new information is developed from the existing geophysical, geological, reservoir development and other data to predict oil and gas reservoirs.
Particularly, in recent two years, petroleum industries at home and abroad are undergoing large adjustment, a plurality of new subjects are introduced into the field of petroleum exploration and development, and in the field of artificial intelligence, China is also laying out at present. Therefore, the application of the invention can be said to accelerate the informatization step of the petroleum industry in China. The exploration and development technology of petroleum is expected to have breakthrough development in the aspect of artificial intelligence in the near future, and the invention is just one of the applications. The invention comprises a software package which can be deployed on a cloud server for global customers to use and is a complete application system combining petroleum data and a computer module.
In the process of exploration and development of petroleum, identification and definition of lithofacies have extremely important significance, and the lithofacies can analyze microfacies and space-time evolution thereof in the lithofacies, establish a sedimentary model, and analyze reservoir formation factors and combinations such as biogenesis, storage and covering, and the like, establish the reservoir formation model, further discuss the relation between oil gas gathering and sedimentary microfacies, explain the spreading form of known sand bodies, guide sand body prediction of an uncontrolled area, and provide basis for determining geological reserves, predicting oil-gas-containing areas and well position deployment.
Lithofacies analysis is an important step in interpreting reservoir description seismic data. Lithofacies interpretation plays an important role in initial exploration prospect evaluation, reservoir description, and final field development. A facies is a stratigraphic unit or region having characteristic reflection patterns that are distinguishable from other regions. Regions of different lithofacies are often delineated using descriptive terms reflecting large-scale seismic patterns. Such as reflection amplitude, continuity, and internal configuration of reflectors defined by the formation field of view. The applications and scale of lithofacies analysis vary widely from basin-wide applications to detailed reservoir descriptions. Lithofacies analysis has been applied to hydrocarbon system research in the basin context for exploration of widely identified sources, reservoirs and sealed prone areas. These regions are typically identified based on their reflection geometry and amplitude intensity and continuity. The high amplitude, semi-continuous reflector of the zone is typically used to identify potential hydrocarbon-bearing reservoirs, such as deep water channels, while the low amplitude continuous to semi-continuous zone may be used to identify seal-prone cells.
Lithofacies analysis may also be applied in a single reservoir to help constrain detailed physical property descriptions. In these local scale applications, the definition of continuity and amplitude is typically not strictly defined, and the environment is calibrated or depositionally interpreted based on rock properties. Relationships between seismic features and physical properties can be demonstrated, and lithofacies volumes can then be used to predict rock property distributions and conditional geological models.
The standard technique for lithofacies analysis and mapping is a manual process in which a seismic interpreter makes visual decisions on the characteristics of seismic reflection data within an interval of interest and maps these data. The lithofacies are then used for various purposes, but primarily to explain the distribution of lithofacies and rock properties. Intuition and experience have made a significant contribution to the success of facies research, however, this approach can also cause facies analysis to be subjective, time consuming, and often a laborious, inefficient task. Several related techniques have been used in the oil industry to improve automation and enhance interpretation of facies from seismic data.
"can solve The seismic stratigraphic problem by using automatic pattern analysis and recognition" (The Leading Edge, Geophys explorer, vol. 5, No. 9, pp.51-55,1986), which lays a conceptual framework for training The algorithm of The seismic interpretation process, thereby realizing automation. However, these authors do not demonstrate any working prototype or any details describing possible attributes or classification algorithms. R, Vintner, k.mosegaard et al, "seismic texture classification: computer aided stratigraphic analysis methods "(SEG international association and 65 th annual meeting, paper SL14, 1995, 10 months 8 to 13 days) and r. Vintner, k. Mosegaard, Abatzis, c.anderson, VO Vebaek and PHNielson (3D seismic texture classification, society of petroleum engineers 35482,1996), discuss texture analysis of seismic data and a version of classification principal component analysis and probability distribution using texture attributes. These publications do not utilize probabilistic neural networks or dynamically use probability values to optimize classification when using texture analysis methods for seismic data. These methods also do not use interactive training programs and texture analysis is not guided. The process of stratigraphic layer guided computation defined by seismic reflector dip is called dip steering. "first and second order seismic texture" of Gao: quantitative seismic interpretation and significance of oil and gas exploration "(1999), the use of standard texture analysis to generate seismic texture attributes that quantify reflection intensity, continuity, and geometry is described. However, this abstract does not describe the classification method of the texture attributes. In particular, Gao (1999), does not use probabilistic neural networks, nor interactive interpreter training of neural networks.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an intelligent lithofacies identification method based on logging information and geological information, which can effectively solve the problems in the prior art. The invention is composed of three technical systems:
system 1, petroleum specialty system;
system 2, website web page creation system;
system 3. data exchange system between the two;
the invention aims to solve the problem of lithofacies identification in the petroleum industry, the shale gas industry and the geological mine industry. The invention uses artificial intelligence method to automatically identify lithofacies.
The invention relates to a method for identifying a logging dataset and a seismic dataset by using an artificial intelligence method. First, a plurality of initial texture attributes representing the seismic data volume are calculated. Next, a probabilistic neural network is constructed based on the computed initial texture attributes. The final texture attributes are then calculated throughout the seismic data volume. And finally, classifying the calculated texture attributes by using the constructed probabilistic neural network. The invention relates to a lithofacies identification and prediction method. The method is also applicable to other well log attributes and seismic attributes, particularly to identifying micro lithofacies in seismic amplitude data. Thus, the analysis of seismic texture mimics the vision-based analysis process of a seismic interpreter in a manner not available with traditional attribute analysis. The interpreter translates a set of traces into an image to present the classification. This different analysis method offers the potential to capture the reflection geometry over the entire investigation region. One such technique for quantizing image texture in texels employs an image transform that results in a gray level co-occurrence matrix. The gray level co-occurrence matrix describes the spatial relationship between pixels of a small area within a larger image, i.e. texels. In practice, the gray level co-occurrence matrix is computed with overlapping texels so that any transition between texture classes within the entire image can be fully observed. Overlapping texels scans the image until the whole image is processed. The gray level co-occurrence matrix is a matrix of size N × N, N being the number of gray levels used to quantize an image. Texture analysis by constructing a gray level co-occurrence matrix from image texels is in effect a two-dimensional (or three-dimensional) extension of one-dimensional markov chain analysis. The structure of the seismic derived gray level co-occurrence may heuristically understand the matrix. In the uniform region, defining uniformity or continuity in a given direction, the difference between pixel values will be low, so elements close to the diagonal of the gray level co-occurrence matrix will have higher values. Less uniform areas will produce higher differences between adjacent pixel values and therefore the resulting gray scale co-occurrence matrix will have values further away from the diagonal. The average pixel values are also represented in the gray level co-occurrence matrix. The low amplitude region has a gray level co-occurrence matrix whose values are concentrated near the center. On the other hand, regions with higher amplitudes have more distributed gray level co-occurrence matrix values, either along a diagonal continuous texture or in the entire gray level co-occurrence matrix in more non-continuous textures.
To approximately simulate the process followed by the seismic interpreter, the preferred 2D texture attributes are computed and then filtered in time slices to simulate full 3D operation. Alternatively, 3D texture properties may also be computed and used to characterize facies. The gray level co-occurrence matrix cannot be directly and efficiently interpreted and is more efficiently described by scalar statistical measures of texture attributes. Texture attributes can be divided into first and second order descriptors. The first order statistics quantify the global distribution of pixel values within the image and can be computed directly from the texels using standard statistical techniques, even without intermediate gray level co-occurrence matrix transformations. The mean absolute amplitude and standard deviation of amplitude values within texels are examples of first order texture properties and can be used to delineate amplitude anomalies and reflection intensities. Derived properties of instantaneous amplitude, phase and frequency can also be used to generate first order statistics. First order statistics are the starting method for detailed texture quantization, although some geophysical regions may be roughly defined from different pixel intervals-in general, the value of a single texel cannot be adequately described in terms of its first order statistic alone. For example, high amplitude chaotic regions of a seismic image are not necessarily separated from continuous regions of high or even medium amplitude using only average amplitude values. The second order statistics of the image quantizes the spatial relationship of pixels in the image and is calculated to a gray level co-occurrence matrix through intermediate transformation. Second-order statistics, statistics of gray level co-occurrence matrix, capture of track shape characteristics, reflection geometry and reflection continuity, and amplitude intensity. The second order statistics of texels are multi-track image properties that allow capturing the reflection geometry and continuity by analyzing the tilt-controlled gray level co-occurrence matrix.
These gray level co-occurrence matrices correspond to textures of organized and contrasted features, with only a few gray levels at the same distance and azimuth. A lower texture uniformity value will correspond to a larger value of the gray level co-occurrence matrix further away from the matrix diagonal, i.e. a number of different gray levels of the same distance and azimuth. These features make texture uniformity particularly useful for quantifying continuity. The first texture attribute, texture correlation, represents the degree of similarity of elements of the metric space gray level co-occurrence matrix in the row or column direction, and the magnitude of the correlation value reflects the correlation of local gray levels in the image. The correlation value is large when the matrix element values are uniform and small when the matrix element values are widely different. The second texture attribute, texture inertia, represents the contrast of the gray level co-occurrence matrix and is a measure opposite to texture homogeneity. For high contrast images, texture uniformity is low and texture inertia is high. The third texture attribute, texture entropy, measures the deficit of spatial organization within the computational window. When all elements of the gray level co-occurrence matrix are equal, the texture entropy is high, corresponding to a coarse texture, and the low texture is more uniform or smoother. The fourth texture attribute, texture energy, also represents organization within the spatial computation window. In this case, it is also possible to calculate all or most of the gray levels within the window, which is characteristic of the coarse texture. Conversely, the highest value of texture energy indicates that a high value of the gray level co-occurrence matrix exists. In this case only a few grey levels prevail. The area within the calculation window is more uniform or exhibits some regular characteristics.
Neural networks are interconnected components of simple processing elements. The processing power of the neural network is stored in the obtained connection strengths or weights by a process of adapting or learning a set of training patterns. One advantage of neural networks is the ability to train or modify the strength of connections within the network to produce desired results. In classification applications, neural networks can be considered a special case of supervised classification schemes, since the training of neural networks is a supervised exercise. Once the plurality of calibration images are sufficiently trained, the neural network may be applied to the remaining images in the data volume. In the calculation, the connectivity, the weight and the input vector of the modified attribute of the node in the general neural network are calculated and go to the next layer of the network through the modified value. The weights of the network are modified through training such that, in a particular set of training examples, modification of the input attribute vector produces the desired result. Training of the network and modification of the connection weights results in a decision surface being generated for the network.
One of the advantages of neural network algorithms over more standard classification schemes is the ability to generate non-linear boundaries. Typical classification or prediction problems are usually only three layers, the first being the input layer; a second "hidden" layer; the third layer is the output layer. Probabilistic neural networks are parallel implementations of standard bayesian classifiers. Probabilistic neural networks are three-layer networks that can efficiently perform pattern classification. Mathematically, these probabilistic neural networks are very similar to the kriging method, where the proximity to known points guides the classification and prediction of unknown points. In its standard form, probabilistic neural networks are not trained in the same manner as the more traditional neural networks described above. Instead, the training vector simply becomes a weight vector in the first layer of the network. This simpler approach allows a probabilistic neural network to have the advantage of not requiring extensive training. For example, in seismic texture analysis, the texture attributes of a training image provide a weight vector in the first layer of the network. This has a significant speed advantage during the training phase compared to traditional types of neural network architectures, such as fully connected back-propagation architectures. In addition, probabilistic neural networks tend to generalize well, whereas more traditional networks, even with large amounts of training data, do not guarantee convergence and generalisation to data not used in the training phase. When an input pattern is presented to a probabilistic neural network, the first layer (input layer) of the network is to compute the distance from the input vector to the training input vector and generate a vector whose elements indicate how close the input is to the training input. The second layer is to sum these contributions for each class of input to produce a probability vector as its net output, which is another advantage of using probabilistic neural networks. This is the ability to extract the classification probability directly from the second or hidden layer, in addition to the maximum probability of classification from the third or output layer.
This negative effect can be caused by using a single gray level co-occurrence matrix to calculate the window size for the entire volume. The results are improved by varying the window size of the entire volume. As the data frequency decreases with increasing depth, the window size becomes larger. This mode is used in conjunction with a dynamically adjusted window size based on user-defined confidence. In another alternative embodiment to deal with reduced seismic data quality, the data may first be filtered with a convolution or median filter to smooth the data prior to input.
If the confidence level is below the user-defined level, the calculation window size may be automatically adjusted until the confidence level rises above that in another alternative embodiment, the generation of the gray level co-occurrence matrix may be tilt-controlled, and the facies may be recalculated and reclassified accordingly. Despite the unconscious particular geological environment, stratigraphic frameworks are an important aspect that has been considered by seismic interpreters. Lithofacies interpreters not only consider the continuity of the time plane, but they also judge the continuity of stratigraphic layering defined by seismic reflector dip.
Texture analysis and construction of the gray level co-occurrence matrix of texels, where the pixels within a texel are correlated, depends on the viewing direction or azimuth. Texture analysis applied to seismic data is extremely sensitive to the stratigraphic framework of the texels and must also follow the stratigraphic dip of the reflectors to correctly mimic the process performed by human translation. After the dip in the gray level co-occurrence matrix calculation, the continuity of the image is maximized, as represented in the gray level co-occurrence matrix. The process of guiding the calculations by the formation dip is called dip steering. Texture analysis requires high resolution in the formation geometry to correctly handle the computation of the gray level co-occurrence matrix. To achieve the required resolution, the multi-trace, image, property of the texel is utilized and the dip within the image is estimated by a gradient-based technique.
The invention organically combines the processing module of petroleum data with the artificial intelligence technology and the website webpage display technology to form a unified system. The invention relates to lithofacies classification by artificial intelligence, which is a method for helping customers to process logging data and seismic data of the customers. The method and the device realize the remote quick input of the client data, the model operation of the shared server side and the independent display and distribution of the calculation result, improve the operation speed and improve the user experience.
The method can accurately identify underground oil and gas storage layers, particularly thin oil and gas storage layers with the thickness of less than 3 meters, and improve the geological reserves, which is one of the work and main challenges of researchers such as geology, exploration, development, well logging and the like. The currently used technique is to study petrophysical data of sharp changes along the wellbore, performing the study of lithofacies discontinuities. In the inventive software, an Artificial Neural Network (ANN) model is included, which can predict the characteristic vector of the oil reservoir by inputting 10 logging parameters as input, and can be regarded as the recognition standard for detecting the oil reservoir through learning training. The software of the invention uses a support vector machine of a machine learning algorithm to distribute lithofacies to logging data for training. And performing model training based on expert core description and logging data of 3-9 left and right wells. The data is used for training a support vector machine and intelligently identifying lithofacies based on logging data.
The artificial intelligence is adopted to identify the oil storage layer, so that the workload of geology, exploration and development researchers is greatly reduced, the accuracy and the efficiency of reservoir identification are greatly improved, and end-to-end butt joint is realized. The convolution layer and the sampling layer of the algorithm convolution neural network in the artificial intelligence are matched with each other, one is used for extracting features, the other is used for carrying out local non-overlapping sampling, and the reservoir identification result is more and more accurate after the above steps are repeated.
The identification essence of the oil and gas reservoir is the identification and evaluation of the rock pore fluid attribute and the saturation, the volume and the quality of the reservoir pore fluid only account for a very small part of the reservoir rock and are filled in the pores of the solid rock framework; the logging method can be distinguished, but the seismic response is very weak. Seismic records, if responsive to changes in rock pore fluid, may only be reflected in the fine structure of the seismic event. Wave equations describing the propagation of seismic waves are approximate equations obtained under certain assumptions (e.g., fully elastic media, etc.), and the dominant phase is well characterized by wave motion, but does not necessarily reflect the microphase of the pore fluid response. Seismic recordings are an objective reflection of the actual geological medium response, without any approximation. If the amplitude of the rock pore fluid seismic response is observable, it must be present in the seismic record. The key to the problem is how to identify the pore fluid response on the seismic record. The deep learning in the artificial intelligence can carry out automatic feature extraction, and the software of the invention applies the deep learning to the field of seismic exploration.
The software of the invention aims to replace manual judgment of the position of the hydrocarbon reservoir, because human eye recognition is tired, and phenomena of missed judgment, wrong judgment and inaccurate explanation often occur. Meanwhile, the position of an oil layer is judged by using a logging curve, more than ten curves need to be synthesized at one time, and the position cannot be judged at all sometimes by using manual identification.
Because the reservoir stratum under exploration is deeply buried underground, the characteristics of the reservoir stratum and the sedimentary facies marks can be analyzed and observed only through well logging and rock data in the analysis of the underground lithofacies. Well coring is generally discontinuous, and the heart rate of a whole well of an exploratory well is often only a few percent to a dozen percent, which causes great difficulty in the study of sedimentary facies. Although the logging phase analysis is performed by using the logging information, the continuous dephasing interpretation can be performed on the whole well, but the important information such as the stratum superposition mode, the shape of a sedimentary body and the like is not fully utilized, and the complete interpretation is very local even if the complete interpretation is correct. To further understand the planar distribution of dephasing, a large enough population of well data must be available, which is difficult to meet at the exploration stage. Therefore, a new means and method for better understanding the sedimentary facies plane change characteristics with only a small amount of drilling data is needed. The identification and prediction of lithofacies is being generated to meet the above desiderata.
Lithofacies refer to a range of three-dimensional seismic reflector units within a distribution in which the seismic characteristic parameters (e.g., reflector structure, geometry, amplitude, frequency, continuity, etc.) are different from those of adjacent units and represent the lithological composition, bedding and depositional characteristics of the sediment from which the reflections are generated. The lithofacies prediction is to identify and map lithofacies units according to certain procedures in sedimentary stratigraphic units according to seismic characteristic parameters and by combining with other underground and ground data, and lays a necessary foundation for comprehensively explaining the sedimentary environment and sedimentary system. Seismic data necessary for lithofacies prediction are essential basic data in oil exploration, can be obtained at the early stage of exploration and generally cover the whole work area, and contain extremely abundant stratum, structure and sedimentary facies information. The invention carries out identification and prediction of lithofacies, namely, for regional stratum interpretation, determining a sedimentary system, lithofacies characteristics and explaining a sedimentary development history, and finally, the lithofacies is converted into the sedimentary facies to be used as a basis for researching petroleum geology biogenesis, storage and cover combination and distribution rules thereof, thereby predicting favorable oil-producing areas and storage facies zones.
The key point of lithofacies prediction is to accurately divide lithofacies, and the traditional lithofacies division method is to observe images, namely, to observe and describe reflection characteristics on seismic sections through naked eyes, and to classify lithofacies with similar characteristics into one class. The pure manual mode is time-consuming, labor-consuming and highly subjective and is not beneficial to identifying unobtrusive abnormal reflection characteristics on the seismic section. With the continuous improvement of seismic data acquisition technology, the seismic information contained in the seismic profile is richer, and much information cannot be detected by naked eye observation and must be extracted and analyzed by means of seismic data processing technology and computer technology. At first, people used seismic structure attributes to divide lithofacies, but at that time, the method for extracting the seismic structure attributes is not mature, and the division result is limited by the signal-to-noise ratio of seismic data. Later, people extract seismic structure attributes by utilizing the gray level co-occurrence matrix, describe the amplitude distribution of seismic data in the statistical and mathematical sense, and improve the accuracy of dividing lithofacies by utilizing the structure attributes. However, the seismic structure attribute only represents a few physical parameters of the seismic signal, and the description of the general abnormality of the seismic signal is still lacked. In recent years, artificial intelligence neural network technology has been introduced into the division of lithofacies. The neural network has adaptive capacity, fault-tolerant capacity and large-scale parallel processing capacity, so that the lithofacies division precision is further improved. At present, methods for dividing lithofacies using neural networks can be classified into two main categories: one is unsupervised pattern recognition; another type is supervised pattern recognition. Unsupervised mode recognition divides the facies of the work area based on input data and classification numbers preset by interpreters; the supervised pattern recognition adds well drilling data as control information in the classification process, so that lithofacies division results have clear directionality, for example, a certain lithofacies represents a certain lithology or hydrocarbon-containing area.
Two common lithofacies prediction techniques are described below, which are based on unsupervised and supervised neural networks, respectively.
The first is a facies identification and prediction technique based on seismic waveform classification. Different sedimentary bodies are formed in different sedimentary environments, and the different sedimentary bodies are different in lithology, physical properties, oil-containing property and the like, and the change of the amplitude, the frequency and the phase of the seismic wave, namely the change of the seismic waveform, is reflected on seismic information. Therefore, the seismic channel waveform and the geological features reflected by the seismic channel waveform can be automatically identified and classified by utilizing the self-organizing feature mapping neural network, so that lithofacies identification and prediction are completed. The technology mainly comprises three steps: firstly, as for the whole work area, learning and training actual seismic channels in a target layer section by using a self-organizing neural network to obtain a series of model channels capable of reflecting the change of the seismic channels in the layer section, wherein the model channels are arranged in a shape gradual change mode, each model channel represents a type of lithofacies, and color and number are assigned in sequence; secondly, comparing each actual seismic channel in the target interval of the total work area with the model channel, classifying the actual seismic channel into the type of the model channel with the highest correlation degree and assigning corresponding color and number; and thirdly, drawing a lithofacies diagram of the target interval according to different colors and digital numbers. At this point, the lithofacies prediction of the target interval is complete.
The second is a lithofacies prediction technique based on seismic attributes and multi-layer perceptron neural networks. The seismic attribute contains abundant geological information such as stratum structure, lithology, physical property and the like, and the multi-layer sensor neural network is driven by the geological information, so that lithofacies of a research area can be accurately predicted under the condition of limited drilling. Early people successfully predicted lithofacies using this technique. The technology comprises the following steps: firstly, extracting seismic attributes which can clearly describe lithofacies characteristics of a target interval, such as seismic amplitude attributes, energy attributes, coherent body attributes and the like; secondly, taking seismic attributes as input of a neural network of the multilayer perceptron, taking well drilling data as control point information, and training the network by adopting an error back propagation algorithm; and thirdly, using the trained network for dividing the lithofacies of the target interval and drawing a lithofacies graph, thereby completing the prediction of the lithofacies of the interval.
Facies prediction techniques based on seismic waveform classification utilize self-organizing neural networks to partition facies. The disadvantages are as follows: 1. manually presetting the classification number of the lithofacies, which often causes inaccurate setting of the classification number and generally needs multiple calculations to estimate the parameter; 2. the classification result can be converged to an accurate result only by multiple iterative operations, and in practical application, 20-40 iterative operations are usually required to ensure the best network convergence; 3. the self-organizing neural network belongs to an unsupervised mode identification method, the classification result has ambiguous geological meaning, and a lithofacies diagram with unambiguous indication meaning can be obtained by further explanation by combining drilling data.
Lithofacies prediction techniques based on seismic attributes and multi-layered perceptron neural networks. The disadvantages are as follows: 1. the multi-layer perceptron neural network adopts an error back propagation algorithm to train the network, and the training method has low convergence rate and usually needs to consume a large amount of calculation time; 2. when the multilayer perceptron neural network is used for lithofacies division, the existing supervision information is divided into several types, lithofacies of reservoirs are predicted at one time, and the operation often has the condition of inaccurate classification. Because the supervised information is typically derived from geological, well log data, which are on a different scale than seismic data, the classification of the supervised information does not necessarily correspond exactly to the classification of the facies.
The invention aims to provide a reservoir identification system based on an artificial intelligence method. The method is essentially characterized in that an artificial intelligent deep learning method is applied to automatically extract the characteristics of a logging curve and a seismic map, and the expression form is hierarchical characteristic extraction. The low-level features belong to the locality features, the high-level features are nonlinear combinations of the low-level features and belong to abstract structural features, and the high-level features are more distinctive and category-indicative. The software innovatively introduces a machine learning and deep learning characteristic extraction method, extracts micro characteristics, weak seismic response characteristics and the like of the reservoir logging curve, can determine the reservoir characteristics more simply and efficiently, and improves the utilization rate of the logging curve and the identification precision of seismic exploration data.
The core function of the software of the present invention is reservoir image recognition, which refers to the use of computers to process, analyze and understand images to identify various patterns of targets and techniques. Image recognition is an important area of artificial intelligence. At present, the main image recognition methods include an image recognition method based on a neural network, an image recognition method based on a wavelet moment, and the like.
In order to develop a computer program for simulating human reservoir image recognition activities, different reservoir image recognition models are proposed here.
1. Template matching model
The model considers that a certain reservoir image is identified, and a memory mode of the reservoir image, called a template, must be found in past experience. If the current stimulus matches the template in the brain, the reservoir image is identified. For example, in a three-dimensional (3D) seismic reservoir image, a reservoir map has a feature of 'two curves crossing and bending down', a big data model has a template, and if the size, the orientation and the shape of the reservoir map are completely consistent with the template, the reservoir is identified. The model is simple and clear and can be easily applied to practical use. However, the model emphasizes that the reservoir images must be completely consistent with the templates in the big data to be identified, and the big data can not only identify the reservoir images completely consistent with the templates in the human brain, but also identify the reservoir images not completely consistent with the templates.
2. Prototype matching model
In order to solve the problems of the template matching model, a prototype matching model is provided. This model assumes that not the myriad of templates to be identified are stored in long-term memory, but rather some "similarity" of the reservoir images. The "similarity" abstracted from the reservoir images can be used as a prototype to examine the reservoir images to be identified. If a similar prototype can be found, the reservoir image is identified. This model is more appropriate than the template matching model both from a neural and memory search perspective, and also accounts for the identification of somewhat irregular, but in some ways similar reservoir images to prototypes. In general use in the petroleum industry, three-dimensional (3D) seismograms are adopted, and useful oil and gas layer information is identified after processing according to picture color differences and gray level differences by software.
The software of the invention is based on cloud storage and cloud computing of the server side, data input by the webpage side and result output by the webpage side. In order to achieve the above object, the present invention adopts the following scheme:
the artificial intelligence module comprises a machine learning module, a deep learning module and a Python Web framework. The software system of the invention comprises a client and a server, wherein the client is connected with the server through a network, and the method comprises the following steps:
1) opening a webpage and entering a data entry interface;
2) inputting a data set, and setting calculation parameters and module parameters;
3) entering a model training interface, operating a training module and debugging parameters to improve the model learning efficiency;
4) after the model training is finished, entering a data processing interface to identify and divide lithofacies;
5) and displaying and distributing the results, and adding the training model into the training set to be used next time.
The invention is realized by the following technical elements:
the invention includes using a portion of samples having known facies classifications as training data, dividing the training data into two subsets, a calibration set and a cross-validation set, using an automated supervised learning facies identification method to determine identification and prediction of facies in the subsurface formation of interest based on the calibration set, calculating a confusion matrix for the supervised learning facies identification method by comparing predicted and observed facies for the cross-validation set, calculating a facies transfer matrix characterizing changes between successive facies, and iteratively calculating using the preliminary identification, the facies transfer matrix, and the confusion matrix to perform facies identification.
The deep learning method based on the logging curve data and the seismic data carries out module operation by using a calculation method, and comprises the following steps of:
(1) calibrating a target zone by using logging, logging and synthetic seismic records;
(2) after the well log data is obtained, it is assigned and classified on a plurality of well log samples. For example by using a core description. It will be appreciated that the classification may have been previously specified, or may be specified by expert analysis as part of the implementation of the method. These designations are considered to be known phases. The portion of the log data having a known phase is selected and shifted out and set aside before further processing. That is, data with known facies is divided into a training subset and a testing subset, where the testing subset may be referred to as "ignored" or "cross-validation" data. The data to be ignored may be randomly selected and the percentage of the data to be ignored may be set as a parameter by the user or may be a fixed percentage. When there are many data samples, the percentage of data that is ignored may be close to 50%.
The method proceeds by implementing any conventional computer-implemented supervised pattern recognition or machine learning method for recognizing phases and training using a training set. It will be appreciated that there are a wide variety of such methods, including back propagation, neural networks, decision trees, and any number of additional supervised learning algorithms that may be applied to the well log data.
Once the machine learning method has been trained, it is used to predict the phases over all samples, including the ignored data, and generate a confusion matrix by comparing the output of the trained machine learning algorithm against the classifications previously specified for those portions of the data.
A phase transfer matrix is generated that characterizes changes between previously specified phases in the log data. A preliminary predicted phase transition matrix is generated that characterizes the changes between the phases in the preliminary prediction classification.
The transition matrix describes each pair of successive phases and the relationship of the pair of phases to each other. For example, when successive pairs show a change from shale to sandstone, the transfer matrix will capture the relationship and the change from sandstone back to shale, as shown in fig. 2.
Once the observed and preliminary predicted branch matrices are calculated, a target probability matrix may be formed. In this regard, the target probability may be calculated based on the prediction, or the transition probability matrix may be set based strictly on the observed transitions.
Extracting seismic data along the specified time window width of a target horizon to serve as training data of a deep learning pre-training model, wherein a single training data sample is formed by connecting specified time window data of each peripheral channel, and the moving distance of a time window is generally less than or equal to the length of the time window;
pre-training deep learning model parameters by using a limit Boltzmann machine (RBM) or a continuous limit Boltzmann machine (CRBM);
selecting the optimal model depth, the number of neuron nodes on each layer of the model, a neuron activation function and sparse limitation through experiments;
extracting well side channel seismic data along the specified time window width of the target horizon to serve as training data of a deep learning fine tuning model, wherein the category of the fine tuning model comprises oil, gas and water;
fine-tuning the parameters of the deep learning model by using a batch random gradient descent algorithm;
calculating each layer of base of the deep learning model, extracting a seismic response value of a target layer, and determining deep learning target characteristics by using the correlation of the sample and the base, wherein the characteristics can reflect the weak change of seismic signals, strengthen the oil and gas seismic response characteristics and strengthen the difference between a reservoir and a non-reservoir.
And extracting well-passing or well-connecting seismic data according to a training data extraction method, and inputting the well-passing or well-connecting seismic data into a trained deep learning model to obtain target characteristics.
And determining the difference of the seismic depth learning characteristics caused by different lithologies and fluids according to the well data, and extrapolating the different seismic depth learning characteristics caused by different lithologies and fluids to a well-free area so as to detect lithology and hydrocarbon.
The calculation of the deep learning high-level nonlinear features can be suitable for two-dimensional and three-dimensional data, the calculation modes are flexible and various, and time slices, along-layer slices and the like can be calculated according to actual requirements.
Theoretical basis of software: deep learning in artificial intelligence forms the theoretical basis of the software. The method is characterized in that a convolution neural network is combined with a support vector machine method, abnormal values possibly existing in the generated multi-wave seismic attributes are eliminated by utilizing the Lett criterion, and the number of network variants is reduced. And then, constructing a hidden layer through a clustering algorithm and an unsupervised learning algorithm which can highlight the characteristics of the oil and gas reservoir, and using the hidden layer to increase network sharing and extract the oil and gas characteristics. Finally, the known well point sample after the network penalty value is added is used as an input sample for the support vector machine prediction, the sampled convolutional layer attribute is used as a learning set, the prediction from the known oil and gas reservoir to the unknown oil and gas reservoir is further carried out, and the intelligent prompting method for searching is characterized in that the step of calculating the statistical frequency of the keywords in the step S511 comprises the step of weighting the calculated statistical frequency according to time.
The software realizes the functions:
1) extracting hydrocarbon reservoir characteristics of a three-dimensional (3D) seismic reservoir image;
2) modeling of hydrocarbon reservoir characteristics;
3) triggering, realizing and debugging model parameters;
4) training of reservoir image recognition: fast learning and iteration;
5) boundary processing of a reservoir image identification model;
6) experimental application of machine-automated interpretation;
7) commercial application of reservoir interpretation for large data platforms;
8) the working scheme includes finding 2-3 petroleum companies for cooperation, so that the results are quickly converted into productivity
Application scenarios
With the rapid development of petroleum exploration and development and three-dimensional seismic technology, the oil and gas reserves of the domestic ascertained concealed oil and gas reservoirs exceed half of the oil and gas reserves of the domestic ascertained oil and gas reservoirs, and the petroleum exploration and development are shifted from the early constructed oil and gas reservoirs to the concealed oil and gas reservoir exploration and development stage which mainly comprises lithologic or stratum oil and gas reservoirs. The hidden oil and gas reservoir identification difficulty is high, the control factors are many, the reservoir formation rule is complex, and the requirement on the geophysical prospecting technology and the recognition degree of the oil reservoir is high. The oil reservoir geophysical technology organically combines the seismic exploration technology and the oil reservoir research technology together, meets the requirements of longitudinal resolution and plane resolution for oil reservoir recognition, and has strong pertinence on the research of the middle and later stages of hidden oil and gas reservoir development and the production of increasing storage. Most of the hidden oil and gas reservoirs discovered in the early exploration become old oil fields in the middle and later development stages, the production data is rich, the extraction degree is high, the water content rises quickly, and the yield decrease is urgently needed to be compensated through encryption adjustment and edge expansion excavation. Therefore, the recognization of the concealed oil and gas reservoir and the research of movable residual reserve distribution for many years of exploitation are a brand-new field of the seismic identification technology, and the oil reservoir geophysical technology organically combines seismic data with static and dynamic data of the oil reservoir, so that the residual reserve distribution characteristics of the old oil field can be effectively realized, the production well position deployment is guided, and the yield is further increased and the potential is excavated.
In oil exploitation, reservoir prediction is one of the tasks of finding oil and gas resources and evaluating the key points of oil and gas reserves. Due to the complexity of underground geological structures and the fuzziness of the distribution of logging parameters, the traditional lithology identification method has limited identification precision, and the interpretation result is not satisfactory in many times. The main reason is that during logging, a wellbore curve is drawn by using geophysical methods such as resistivity, ultrasonic waves, radioactivity, nuclear magnetic resonance and the like, and whether oil gas exists at a certain position in the wellbore or not is judged according to a comparison curve, so that judgment deviation, omission, errors and the like are inevitably caused. Later, people evaluate formation parameters such as formation lithology, electrical property, porosity, saturation, permeability and the like by means of computer-aided calculation and logging information, the precision is improved, and lithology identification results play an important role in searching hydrocarbon reservoir resources. Because the influence factors of the measured values of the logging curves are numerous, the change of the lithology of strata in different areas is large, the lithology types are various, the geological structure is complex, and the like, the existing identification method has the problems of low identification accuracy, time and labor waste due to the fact that different models need to be established in different areas, and the like, and therefore, how to accurately identify the lithology becomes a key problem in the logging process.
On the other hand, a problem that plagues the field of oil exploration and development is the accurate interpretation of 3D seismic images. In the first aspect, the interpretation of the reservoir image is ultimately determined by human, and the human judgment is determined by the expert experience of the reservoir image, the expert experience is the reflection of the human brain, and how to map the experience of identifying the oil layer judgment in the human brain by using a computer model is one of the targets of the subject. In the second aspect, the position of the hydrocarbon reservoir is judged manually, and phenomena of missing judgment, wrong judgment and inaccurate explanation can often occur due to fatigue of human eye recognition. In the third aspect, logging curves are used for judging the position of a hydrocarbon reservoir, more than ten curves need to be synthesized at one time, and the problem also exists in manual identification.
With the emergence of advanced technologies such as big data, cloud computing, artificial intelligence and the like, especially the successful application of face recognition, a new way is opened up for machine recognition of oil and gas reservoirs of seismic reservoir images. The basic principle is as follows: the existing reservoir images of the hydrocarbon reservoir are used, the reservoir image characteristics of the reservoir images are stored into a data model as a single face, and then a machine is enabled to automatically search for similar hydrocarbon reservoirs in the 3D seismic reservoir image of a client, so that the processing and explaining process is not only quick but also accurate; and manual inspection can be added at the beginning, so that the machine can continuously learn and iterate repeatedly to reach an ideal state. The processing method can overcome the defects of unstable and inaccurate manual interpretation, realize the application of the artificial intelligence technology in the petroleum field, and the petroleum industry urgently needs the software to be capable of landing practically at present.
The specific main application scenarios are as follows:
firstly, geological conditions are complex, the difficulty in structure interpretation is high, most of exploration areas have complex underground structures, fracture zones and small fault blocks develop, and the underground speed is changed rapidly in the transverse direction; reservoir lithology is complicated, potential reservoir types are various, gas reservoir types are complicated and various, and the fine description difficulty is large.
Secondly, the utilization rate of the existing seismic data is low, the existing seismic data is limited by software and hardware conditions, high-precision seismic data acquired by customers at a high cost can only be used for developing structural interpretation and simple prediction of a post-stack reservoir, particularly exploration work of a low-permeability oil-gas reservoir, the acquired high-precision omnibearing three-dimensional seismic data cannot independently and effectively develop prediction of a pre-stack reservoir and a crack, the exploration process of a low-permeability block is severely restricted, the exploration risk cannot be reduced by deeply excavating latent data information, and data waste to a certain extent is caused.
Thirdly, the seismic exploration project is limited by conditions such as professional seismic software, professional technologies and the like, exploration deployment personnel and decision makers are difficult to master real, reliable and scientific evaluation data at the first time, exploration risks are increased to a certain extent, meanwhile, scientific research personnel of customers also lose the opportunity of further researching and mastering exploration core technologies, and the establishment of a core technology system of the oil-gas exploration assets of the customers is restricted.
The invention can construct an exploration and development geophysical integrated platform, has great significance for the construction of a customer promotion technical system, needs the oil and gas exploration and development integrated platform to carry out fine seismic data interpretation (stratum contrast, structure interpretation) and sedimentary facies research, carries out pre-stack and post-stack seismic inversion and reservoir prediction, determines the distribution range of favorable reservoirs and provides an exploration and development target. The geological information rich in the existing massive seismic data can be mined to the greatest extent, the risk of oil-gas exploration and development is effectively reduced, and the method contributes to the mining of geological reserves by customers.
Necessity and demand analysis
The invention can be used for carrying out construction explanation and comprehensive evaluation on oil and gas exploration and development projects. The comprehensive integrated platform can be constructed, and has the deep domain interpretation capability.
The invention has the advantages that:
(1) an integrated data management platform can be constructed to provide the functions of establishing a work area and integrated data management and analysis. The resource tree mode is adopted to manage various data, so that the data can be conveniently and quickly inquired, and Chinese and English are supported. The data loading has wide adaptability and high intelligence, and supports well data loading of an Excel table.
(2) A seismic data optimization processing toolkit. The platform provides a variety of seismic data optimization processing tools. Provide a high signal-to-noise ratio or high resolution seismic profile for seismic interpretation.
(3) Wavelet estimation and well seismic calibration. The method has various advanced wavelet estimation methods, and helps a user to extract the optimal wavelet to make a synthetic record (vertical well and inclined well).
(4) And (4) multi-well comprehensive geological interpretation. The geological and logging data are utilized to help geological engineers to carry out single-well geological comprehensive interpretation, namely well-connecting analysis, and carry out comprehensive interpretation and mapping of well-connecting stratums, oil reservoirs and sedimentary facies profiles; and meanwhile, a dominant phase automatic lifting and sedimentation cycle division oil test data display analysis tool and a stratum leveling tool are provided.
(5) Rock physics analysis and prestack earthquake forward modeling. Providing a single well attenuation coefficient curve, a prestack elasticity and viscoelasticity seismic forward modeling result and a seismic forward modeling gather attribute analysis result; meanwhile, 2D rock physical models and fluid replacement rock physical simulation can be performed, and the 2D prestack incident angle gather forward modeling and attribute analysis can be performed.
(6) And constructing virtual logging data. Based on the prestack angle gathers and the raw velocity field data, a genetic algorithm is used to create a virtual well curve, simulating compressional velocity, shear velocity, and density. The well constraint condition is provided for seismic inversion under the condition of no well or few wells, and the purpose of reservoir prediction is achieved.
(7) And (5) construction explanation. The method can directly perform 2D and 3D forward and reverse fault structure interpretation and forward and reverse fault direct plane mapping, support a single-point multi-time-value structure, can take multiple points on cdp, namely one layer name can contain two or more time values, and simultaneously map the upper and lower plate structure plane maps of the reverse fault at one time. And the multi-user collaborative interpretation and deep domain interpretation capabilities are simultaneously realized.
(8) Speed modeling and time-depth conversion. Speed modeling under the constraint of construction information; the generated velocity field not only keeps vertical velocity change on the well, but also has the transverse velocity change characteristic of a velocity spectrum; has the functions of correction and multi-azimuth quality control analysis. The method can have speed modeling modes of block filling and block interpolation, and comprises three interpolation modeling modes of reverse distance weighting, kriging interpolation and collaborative kriging. By the fine speed modeling mode, the purpose of fine time-depth conversion of horizon, fault, seismic data and grids and scattered points is achieved.
(9) And (5) seismic inversion. And establishing a seismic inversion initial model by using the well data, the seismic data and the construction interpretation data. The phase control-volume control modeling algorithms such as linear interpolation, fractal interpolation and cooperative kriging are provided, the complex construction system of the normal fault and the reverse fault can be processed, and the constraints of various deposition modes are considered. And (3) obtaining high-resolution acoustic impedance/speed data by adopting a global optimization fast inversion algorithm and iterative optimization, and performing reservoir interpretation and reservoir parameter inversion.
(10) Lithologic body interpretation. The module provides a complete set of tools that enable lithologic fine interpretation and mapping of the target geologic body.
(11) And predicting reservoir parameters. And multiple mathematical methods are adopted to comprehensively predict the multi-parameter reservoir, so that the geological target evaluation means is enriched.
(12) And calculating and analyzing seismic attributes. The method can provide a high-resolution time-frequency spectrum and transient spectrum analysis algorithm based on wavelet transformation and three-parameter wavelet transformation, calculates multiple seismic attributes such as energy attribute, attenuation attribute, frequency attribute and transient attribute, and can be applied to reservoir prediction and fluid detection.
(13) Pre-stack seismic inversion. Two pre-stack inversion modes are provided, namely pre-stack elastic wave impedance inversion and pre-stack expansion elastic wave impedance inversion.
Elastic parameters such as longitudinal wave impedance, transverse wave impedance, Lame coefficient, shear modulus, Poisson's ratio, elastic gradient, Young modulus, brittleness index and the like can be simultaneously obtained on the basis of prestack wave impedance inversion. The prestack expanded elastic wave impedance inversion part comprises correlation analysis of an in-well expanded elastic impedance curve and a target curve and determination of an optimal rotation angle of the in-well expanded elastic impedance curve and the target curve, and comprises calculation of optimal rotation angle projection seismic data, including prestack expanded elastic wave impedance modeling and inversion results.
(14) Mapping and data analysis. Can provide abundant seismic interpretation, oil reservoir data plane analysis and powerful comprehensive drawing compiling function. By this function, data can be filtered, gridded, smoothed, etc.; meanwhile, drawing work of various construction diagrams, thickness diagrams, attribute distribution diagrams, sedimentary facies plane diagrams and the like which accord with the industry standard can be completed; various scattered point data, grid data, contour line data, survey area data, boundary data and fault polygon data are managed in a layer mode; the distance and the area of the trap can be calculated, and a built-in grapple tool should be provided.
The specific implementation mode is as follows:
this example demonstrates the prediction of lithofacies from well log data. The data set used here is log data from 9 wells that have been tagged with lithofacies types based on observations of the core. We will use this test data to train a support vector machine to classify lithofacies types. A support vector machine (or SVM) is a supervised learning model that can train data to perform classification and regression tasks. The SVM algorithm uses the training data to best fit between the different classes in a hyperplane. We will use SVM in scimit-spare for implementation.
Firstly, sorting a data set: the 9 well training data was loaded, cross-plots were created to see the changes in the data, and model parameter selection was performed using the cross-validation set.
Step two, establishing and adjusting a classifier: trained models are applied to classify facies in wells without labels, where classifiers are applied to two wells, and in the future classifiers may be applied to any number of wells in the area.
And step three, implementing: the data set was from 9 wells (with 8722 instances) consisting of a set of seven predictors and a facies (class), each instance vector and validation (test) data (830 instances from two wells) having the same seven predictors in the feature vector. The phase is based on the detection of cores from 9 wells tested vertically at half-foot intervals. The predicted variables include five geological constraint variables from wireline log measurements and two from geological knowledge. The seven predictor variables are: the five log plot includes 1, Gamma Ray (GR) 2, resistivity log (ILD _ log 10) 3, Photoelectric Effect (PE) 4, neutron density porosity difference 5, mean neutron density porosity (DeltaPHI and PHIND) note that some wells do not have PE these lithofacies are not discrete, but gradually merge with each other. Some adjacent phases. False marks can be expected to occur within these adjacent neighbors. The following table lists facies, their abbreviated labels and their approximate neighborhoods. Facies labels are adjacent.
Drawings
For a more clear description of the embodiments of the present invention or of the technical solutions in the prior art, it will become apparent to those skilled in the art. The drawings that accompany the detailed description can be briefly described as follows, wherein the drawings are only examples of the invention.
Description of the drawings fig. 1: the system constructs a framework.
Description of the drawings fig. 2: machine learning partial module relationships.
Description of the drawings figure 3: lithofacies classification map: the data set in this example is from 9 wells (with 4149 samples), consisting of a set of seven predictors and a facies (class), with each instance vector and validation (test) data (830 from two wells) having the same seven predictors in the feature vector. The phase was based on the detection of cores from 9 wells tested vertically at 15.24 cm intervals. The predicted variables include five geological constraint variables from wireline log measurements and two from geological knowledge. These are essentially continuous variables sampled at a 15.24 cm sampling rate:
the seven predictor variables are:
the five log plot includes 1, Gamma Ray (GR) 2, resistivity log (ILD _ log 10) 3, Photoelectric Effect (PE) 4, neutron density porosity difference 5, mean neutron density porosity (DeltaPHI and PHIND) note that some wells do not have PE:
two geologically constrained variables: non-oceanic indices (NM _ M) and relative position (RELPOS)
Nine discrete phases (rock classes) are: 1. Non-marine sandstone (FS) 2, non-marine coarse siltstone (XFS) 3, non-marine fine siltstone (XFS) 4, marine siltstone and shale (CJ) 5, mudstone (limestone MS) 6, basalt sandstone (limestone WS) 7, dolomite (NY) 8, mudstone and granitic limestone (limestone PS) 9, foliate algal reef (limestone BS).
Description of the drawings fig. 4: log sample curve 1.
Description of the drawings fig. 5: log sample curve 2.
Description of the drawings fig. 6: seismic samples figure 1.
Description of the drawings fig. 7: seismic samples figure 2.
Description of the drawings figure 8: lithofacies classification maps.
Although the present invention has been described in detail with reference to the embodiments, the present invention is not limited to the embodiments described in the following description. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (6)

1. The invention discloses a web page display method for lithofacies classification by using artificial intelligence, which is a method for explaining well logging curve and seismic map data by using an artificial intelligence method, and the input of the data and the output of results can both adopt a remote login network page form, thereby realizing the cloud computing operation, cloud server deployment, cloud data clustering and cloud storage of computing results.
2. The invention relates to an integration method for integrating the input and output of petroleum data and a website webpage, a composition and calculation method of each module of artificial intelligence, a display, storage and integration method of data results and the like.
3. The method as claimed in claim 2, wherein the results of the module calculation of artificial intelligence machine learning and deep learning are displayed to the web (web page, website) end through various frames of python, including constructing ideas, skeleton diagrams, modules and methods.
4. The method of claim 2, wherein the Django framework and module combination technology, method and output form and use skill of the website, webpage display mode, color, layout, etc.
5. The method of claim 2, wherein the data set may consist of seven signatures (five wireline measurements and two indicator variables) and phase labels at intervals of depth, seven predictor variables (or input variables) being:
the five logs included:
gamma Ray (GR)
Resistivity log (ILD _ log 10)
Photoelectric Effect (PE)
Neutron density porosity differential (DeltaPHI)
Average neutron density Porosity (PHIND)
Two geologically constrained variables:
non-marine indicator (NM _ M);
relative position (RELPOS);
in practice, some wells may not have PE.
6. The method of claim 2, wherein the design flow of the computational model is: the method comprises the steps of data inputting- > splitting the data into a training and testing set- > preprocessing the data- > constructing an activity code- > constructing a vector flow (Tensorflow) - > constructing an execution computation graph- > constructing a visual computation graph- > conducting testing data prediction- > evaluating computation.
CN201811629031.8A 2019-03-03 2019-03-03 Web page display method, module and system for lithofacies classification by artificial intelligence Pending CN111596978A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811629031.8A CN111596978A (en) 2019-03-03 2019-03-03 Web page display method, module and system for lithofacies classification by artificial intelligence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811629031.8A CN111596978A (en) 2019-03-03 2019-03-03 Web page display method, module and system for lithofacies classification by artificial intelligence

Publications (1)

Publication Number Publication Date
CN111596978A true CN111596978A (en) 2020-08-28

Family

ID=72181289

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811629031.8A Pending CN111596978A (en) 2019-03-03 2019-03-03 Web page display method, module and system for lithofacies classification by artificial intelligence

Country Status (1)

Country Link
CN (1) CN111596978A (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112034512A (en) * 2020-09-02 2020-12-04 中海石油(中国)有限公司 Seismic data discontinuity detection method and system based on deep learning model
CN112083144A (en) * 2020-09-01 2020-12-15 中国科学院地质与地球物理研究所 Fault on-off prediction method and device, computer equipment and storage medium
CN112507615A (en) * 2020-12-01 2021-03-16 西南石油大学 Intelligent identification and visualization method for lithofacies of continental tight reservoir
CN112578475A (en) * 2020-11-23 2021-03-30 中海石油(中国)有限公司 Compact reservoir dual-dessert identification method based on data mining
CN112766321A (en) * 2020-12-31 2021-05-07 中国地质调查局成都地质调查中心 Geological feature detection and identification method and system based on deep learning
CN112784980A (en) * 2021-01-05 2021-05-11 中国石油天然气集团有限公司 Intelligent logging horizon division method
CN112882095A (en) * 2021-01-15 2021-06-01 中国海洋石油集团有限公司 Lithology identification method and system for lake-facies carbonate rock under salt
CN112949682A (en) * 2021-01-27 2021-06-11 重庆交通大学 SAR image classification method for feature level statistical description learning
CN113722412A (en) * 2021-09-01 2021-11-30 天津大学 Method for inquiring and predicting rock parameters in spatial dimension
CN115012903A (en) * 2022-05-31 2022-09-06 中国石油大学(华东) Logging evaluation method for judging shale bedding structure development
CN115099682A (en) * 2022-07-18 2022-09-23 同济大学 Shield tunnel face soft and hard classification and excavation risk classification method
CN115421181A (en) * 2022-07-27 2022-12-02 北京超维创想信息技术有限公司 Three-dimensional geological model phase control attribute modeling method based on deep learning
CN116522688A (en) * 2023-06-29 2023-08-01 北京城建勘测设计研究院有限责任公司 Well control multi-information fusion engineering geological modeling method and device

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102037380A (en) * 2008-04-07 2011-04-27 雪佛龙美国公司 Lithofacies classification system and method
CN104297785A (en) * 2014-09-29 2015-01-21 中国石油天然气股份有限公司 Lithofacies constrained reservoir physical property parameter inversion method and device
CN106062310A (en) * 2014-02-28 2016-10-26 界标制图有限公司 Facies definition using unsupervised classification procedures
CN106066493A (en) * 2016-05-24 2016-11-02 中国石油大学(北京) Bayes's petrofacies method of discrimination and device
CN106547538A (en) * 2016-10-09 2017-03-29 广州市佳众联科技有限公司 A kind of intelligent operation management system based on Django
CN107733704A (en) * 2017-09-29 2018-02-23 中国石油化工股份有限公司 A kind of system and method for the exploration and development cloud based on virtualization and container technique
CN108961246A (en) * 2018-07-10 2018-12-07 吉林大学 A kind of scanning electron microscope image hole recognition methods based on artificial intelligence
CN111638926A (en) * 2019-05-29 2020-09-08 山东英才学院 Method for realizing artificial intelligence in Django framework
CN114707597A (en) * 2022-03-31 2022-07-05 中国石油大学(北京) River facies tight sandstone reservoir complex lithofacies intelligent identification method and system

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102037380A (en) * 2008-04-07 2011-04-27 雪佛龙美国公司 Lithofacies classification system and method
CN106062310A (en) * 2014-02-28 2016-10-26 界标制图有限公司 Facies definition using unsupervised classification procedures
CN104297785A (en) * 2014-09-29 2015-01-21 中国石油天然气股份有限公司 Lithofacies constrained reservoir physical property parameter inversion method and device
CN106066493A (en) * 2016-05-24 2016-11-02 中国石油大学(北京) Bayes's petrofacies method of discrimination and device
CN106547538A (en) * 2016-10-09 2017-03-29 广州市佳众联科技有限公司 A kind of intelligent operation management system based on Django
CN107733704A (en) * 2017-09-29 2018-02-23 中国石油化工股份有限公司 A kind of system and method for the exploration and development cloud based on virtualization and container technique
CN108961246A (en) * 2018-07-10 2018-12-07 吉林大学 A kind of scanning electron microscope image hole recognition methods based on artificial intelligence
CN111638926A (en) * 2019-05-29 2020-09-08 山东英才学院 Method for realizing artificial intelligence in Django framework
CN114707597A (en) * 2022-03-31 2022-07-05 中国石油大学(北京) River facies tight sandstone reservoir complex lithofacies intelligent identification method and system

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
JUNLAN ZHAO等: ""Research on artificial slope deformation monitoring technologies based on GIS"", 《2011 INTERNATIONAL CONFERENCE ON ELECTRIC TECHNOLOGY AND CIVIL ENGINEERING (ICETCE)》 *
李文昊: ""基于统计学习方法的储层岩相分类研究"", 《中国优秀硕士学位论文全文数据库基础科学辑》, pages 011 - 153 *
苗和平等: ""深度学习在油井管理上的应用"", pages 259 - 260 *
赵红艳: ""基于WEB的专业评估数据采集系统的设计与实现"", 《赤子(上中旬)》, pages 160 - 162 *
郑阳: ""基于深度学习的岩性识别研究"", pages 019 - 405 *

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112083144A (en) * 2020-09-01 2020-12-15 中国科学院地质与地球物理研究所 Fault on-off prediction method and device, computer equipment and storage medium
CN112034512A (en) * 2020-09-02 2020-12-04 中海石油(中国)有限公司 Seismic data discontinuity detection method and system based on deep learning model
CN112034512B (en) * 2020-09-02 2023-03-14 中海石油(中国)有限公司 Seismic data discontinuity detection method and system based on deep learning model
CN112578475A (en) * 2020-11-23 2021-03-30 中海石油(中国)有限公司 Compact reservoir dual-dessert identification method based on data mining
CN112507615A (en) * 2020-12-01 2021-03-16 西南石油大学 Intelligent identification and visualization method for lithofacies of continental tight reservoir
CN112766321A (en) * 2020-12-31 2021-05-07 中国地质调查局成都地质调查中心 Geological feature detection and identification method and system based on deep learning
CN112766321B (en) * 2020-12-31 2024-05-17 中国地质调查局成都地质调查中心 Geological feature detection and recognition method and system based on deep learning
CN112784980A (en) * 2021-01-05 2021-05-11 中国石油天然气集团有限公司 Intelligent logging horizon division method
CN112784980B (en) * 2021-01-05 2024-05-28 中国石油天然气集团有限公司 Intelligent logging horizon dividing method
CN112882095B (en) * 2021-01-15 2022-08-02 中国海洋石油集团有限公司 Lithology identification method and system for lake-facies carbonate rock under salt
CN112882095A (en) * 2021-01-15 2021-06-01 中国海洋石油集团有限公司 Lithology identification method and system for lake-facies carbonate rock under salt
CN112949682B (en) * 2021-01-27 2022-05-20 重庆交通大学 SAR image classification method for feature level statistical description learning
CN112949682A (en) * 2021-01-27 2021-06-11 重庆交通大学 SAR image classification method for feature level statistical description learning
CN113722412A (en) * 2021-09-01 2021-11-30 天津大学 Method for inquiring and predicting rock parameters in spatial dimension
CN113722412B (en) * 2021-09-01 2023-09-08 天津大学 Method for inquiring and predicting rock parameters in space dimension
CN115012903A (en) * 2022-05-31 2022-09-06 中国石油大学(华东) Logging evaluation method for judging shale bedding structure development
CN115012903B (en) * 2022-05-31 2023-06-27 中国石油大学(华东) Logging evaluation method for judging shale bedding structure development
CN115099682A (en) * 2022-07-18 2022-09-23 同济大学 Shield tunnel face soft and hard classification and excavation risk classification method
CN115421181A (en) * 2022-07-27 2022-12-02 北京超维创想信息技术有限公司 Three-dimensional geological model phase control attribute modeling method based on deep learning
CN115421181B (en) * 2022-07-27 2023-10-20 北京超维创想信息技术有限公司 Three-dimensional geological model phase control attribute modeling method based on deep learning
CN116522688A (en) * 2023-06-29 2023-08-01 北京城建勘测设计研究院有限责任公司 Well control multi-information fusion engineering geological modeling method and device
CN116522688B (en) * 2023-06-29 2023-09-15 北京城建勘测设计研究院有限责任公司 Well control multi-information fusion engineering geological modeling method and device

Similar Documents

Publication Publication Date Title
US11520077B2 (en) Automated reservoir modeling using deep generative networks
US11668853B2 (en) Petrophysical inversion with machine learning-based geologic priors
CN111596978A (en) Web page display method, module and system for lithofacies classification by artificial intelligence
US11048000B2 (en) Seismic image data interpretation system
AU2023200300B2 (en) Detecting fluid types using petrophysical inversion
US20150066460A1 (en) Stratigraphic function
US20230161061A1 (en) Structured representations of subsurface features for hydrocarbon system and geological reasoning
US20230125277A1 (en) Integration of upholes with inversion-based velocity modeling
Bhattacharya Unsupervised time series clustering, class-based ensemble machine learning, and petrophysical modeling for predicting shear sonic wave slowness in heterogeneous rocks
US11561315B2 (en) Systems and methods for identifying geostructural properties as a function of position in a subsurface region of interest
Liu et al. Machine learning assisted recovery of subsurface energy: a review
CN115880455A (en) Three-dimensional intelligent interpolation method based on deep learning
US20240052734A1 (en) Machine learning framework for sweep efficiency quantification
Mansoor Multi-Attribute Seismic Analysis Using Unsupervised Machine Learning Method: Self-Organizing Maps
Hidayat et al. The Pematang Group Sand Analysis Using Growing Neural Network Machine Learning
Heydarpour et al. Applying deep learning method to develop a fracture modeling for a fractured carbonate reservoir using geologic, seismic and petrophysical data
WO2024015895A2 (en) Seismic horizon tracking framework
WO2024206578A1 (en) Subsurface discontinuity modeling framework

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20200828

WD01 Invention patent application deemed withdrawn after publication